Composable Kernel: ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector > Struct Template Reference

DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector > Struct Template Reference

DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector > Struct Template Reference#

Composable Kernel: ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector > Struct Template Reference

ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector > Struct Template Reference

#include <device_grouped_conv_bwd_weight_xdl_cshuffle.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >:

Classes
struct	ActiveWorkgroupsPerCU
struct	Argument
struct	Invoker

Public Types
using	DeviceOp = DeviceGroupedConvBwdWeight_Xdl_CShuffle
using	ADataType = OutDataType
using	BDataType = InDataType
using	CDataType = WeiDataType
using	AElementwiseOperation = OutElementwiseOperation
using	BElementwiseOperation = InElementwiseOperation
using	CElementwiseOperation = WeiElementwiseOperation
using	ABDataType = InDataType
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	Block2TileMapTranspose = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>
using	NGCHWTransposeDescType
using	NHWGCTransposeDescType
using	GKCYXTransposeDescType
using	GKYXCTransposeDescType
using	GridwiseInOutTranspose
using	GridwiseElementwiseWeightTranspose
template<index_t NXdlPerWave_>
using	GridwiseGemmBase
using	GridwiseGemm64 = GridwiseGemmBase<math::max(NXdlPerWave64, 1)>
using	GridwiseGemm32 = GridwiseGemmBase<NXdlPerWave32>
using	CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
using	Block2CTileMap

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==1, bool >::type = false>
static auto	GetABCGridDesc ()
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 ()
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
static constexpr auto	BankLength = 128
static constexpr auto	ElePerBank = BankLength / sizeof(ADataType)
static constexpr auto	ABlockLdsM1PerBlock = ElePerBank / K1
static constexpr auto	ABlockLdsM0PerBlock = MPerBlock / ABlockLdsM1PerBlock
static constexpr auto	ABlockLdsM1Padding = 4
static constexpr auto	BBlockLdsN1PerBlock = ElePerBank / K1
static constexpr auto	BBlockLdsN0PerBlock = NPerBlock / BBlockLdsN1PerBlock
static constexpr auto	BBlockLdsN1Padding = 4
static constexpr index_t	ClusterLengthMPerBlock
static constexpr index_t	ClusterLengthNPerBlock
static constexpr auto	conv_ngchw_to_nhwgc_transformer
static constexpr index_t	TransposeTransferSrcScalarPerVectorAligned
static constexpr index_t	TransposeTransferDstScalarPerVectorAligned

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>

◆ Block2CTileMap

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::Block2CTileMap

Initial value:

decltype(GridwiseGemm64::MakeCBlockClusterAdaptor(CGridDesc_M_N{}, 1, 1, 1))

ck::GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight< BlockSize, ADataType, BDataType, AccDataType, CDataType, InMemoryDataOperationEnum::AtomicAdd, AGridDesc_K0_M_K1, BGridDesc_K0_N_K1, CGridDesc_M_N, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, MPerBlock, NPerBlock, K0PerBlock, MPerXDL, NPerXDL, K1, MXdlPerWave, NXdlPerWave_, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, ABlockLdsAddExtraM, ABlockLdsM1PerBlock, ABlockLdsM0PerBlock, ABlockLdsM1Padding, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BBlockLdsAddExtraN, BBlockLdsN1PerBlock, BBlockLdsN0PerBlock, BBlockLdsN1Padding, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferScalarPerVector_NWaveNPerXdl, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, true, true, 1, PipelineVersion::v1, ComputeTypeA, ComputeTypeB >::MakeCBlockClusterAdaptor

__host__ static __device__ constexpr auto MakeCBlockClusterAdaptor(const CGridDesc_M_N &c_m_n_grid_desc, index_t M01, index_t N01, index_t KBatch)

Definition gridwise_gemm_xdlops_bwd_weight.hpp:625

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

◆ Block2TileMapTranspose

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::Block2TileMapTranspose = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>

◆ CDataType

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::CDataType = WeiDataType

◆ CElementwiseOperation

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::CElementwiseOperation = WeiElementwiseOperation

◆ 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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock

Initial value:

decltype(GridwiseGemm64::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))

ck::GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight< BlockSize, ADataType, BDataType, AccDataType, CDataType, InMemoryDataOperationEnum::AtomicAdd, AGridDesc_K0_M_K1, BGridDesc_K0_N_K1, CGridDesc_M_N, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, MPerBlock, NPerBlock, K0PerBlock, MPerXDL, NPerXDL, K1, MXdlPerWave, NXdlPerWave_, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, ABlockLdsAddExtraM, ABlockLdsM1PerBlock, ABlockLdsM0PerBlock, ABlockLdsM1Padding, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BBlockLdsAddExtraN, BBlockLdsN1PerBlock, BBlockLdsN0PerBlock, BBlockLdsN1Padding, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferScalarPerVector_NWaveNPerXdl, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, true, true, 1, PipelineVersion::v1, ComputeTypeA, ComputeTypeB >::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock

__host__ static __device__ constexpr auto MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc_M_N &c_m_n_grid_desc)

Definition gridwise_gemm_xdlops_bwd_weight.hpp:608

◆ 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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::DeviceOp = DeviceGroupedConvBwdWeight_Xdl_CShuffle

◆ 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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GKCYXTransposeDescType

Initial value:

 
        remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer
                                    .template MakeGKCYXTransposeDesc<NDimSpatial>({}, {}))>

◆ 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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GKYXCTransposeDescType

Initial value:

 
        remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer
                                    .template MakeGKYXCTransposeDesc<NDimSpatial>({}, {}))>

◆ GridwiseElementwiseWeightTranspose

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GridwiseElementwiseWeightTranspose

Initial value:

 
        GridwiseElementwise<Tuple<GKYXCTransposeDescType>,
                            Tuple<GKCYXTransposeDescType>,
                            Tuple<const CDataType*>,
                            Tuple<CDataType*>,
                            Block2TileMapTranspose,
                            element_wise::PassThrough,
                            BlockSize,
                            MPerBlock,
                            NPerBlock,
                            MPerBlock / ClusterLengthMPerBlock,
                            NPerBlock / ClusterLengthNPerBlock,
                            Sequence<1, 0>,
                            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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

template<index_t NXdlPerWave_>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GridwiseGemmBase

◆ GridwiseInOutTranspose

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GridwiseInOutTranspose

Initial value:

 
        GridwiseElementwise<Tuple<NGCHWTransposeDescType>,
                            Tuple<NHWGCTransposeDescType>,
                            Tuple<const ADataType*>,
                            Tuple<ADataType*>,
                            Block2TileMapTranspose,
                            element_wise::PassThrough,
                            BlockSize,
                            MPerBlock,
                            NPerBlock,
                            MPerBlock / ClusterLengthMPerBlock,
                            NPerBlock / ClusterLengthNPerBlock,
                            Sequence<1, 0>,
                            Sequence<TransposeTransferSrcScalarPerVectorAligned>,
                            Sequence<TransposeTransferDstScalarPerVectorAligned>,
                            I1,
                            I0>

◆ 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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::NHWGCTransposeDescType

Initial value:

 
        remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer
                                    .template MakeNHWGCTransposeDesc<NDimSpatial>({}, {}))>

Member Function Documentation

◆ GetABCGridDesc() [1/3]

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

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

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GetABCGridDesc ( )

inlinestatic

◆ GetABCGridDesc() [2/3]

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

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

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GetABCGridDesc ( )

inlinestatic

◆ GetABCGridDesc() [3/3]

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

template<ck::index_t NDim, typename ck::enable_if< NDim==1, bool >::type = false>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::GetABCGridDesc ( )

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

std::string ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

size_t ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

bool ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

bool ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

constexpr bool ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

std::unique_ptr< BaseArgument > ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

std::unique_ptr< BaseInvoker > ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

void ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::SetWorkSpacePointer	(	BaseArgument *	p_arg,
		void *	p_workspace,
		const StreamConfig &	= StreamConfig{} ) const

inlineoverridevirtual

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

Member Data Documentation

◆ ABlockLdsM0PerBlock

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::ABlockLdsM0PerBlock = MPerBlock / ABlockLdsM1PerBlock

staticconstexpr

◆ ABlockLdsM1Padding

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::ABlockLdsM1Padding = 4

staticconstexpr

◆ ABlockLdsM1PerBlock

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::ABlockLdsM1PerBlock = ElePerBank / K1

staticconstexpr

◆ BankLength

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::BankLength = 128

staticconstexpr

◆ BBlockLdsN0PerBlock

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::BBlockLdsN0PerBlock = NPerBlock / BBlockLdsN1PerBlock

staticconstexpr

◆ BBlockLdsN1Padding

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::BBlockLdsN1Padding = 4

staticconstexpr

◆ BBlockLdsN1PerBlock

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::BBlockLdsN1PerBlock = ElePerBank / K1

staticconstexpr

◆ 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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

index_t ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

index_t ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::conv_ngchw_to_nhwgc_transformer

staticconstexpr

Initial value:

=
        TransformConvNGCHWToNHWGC<InLayout,
                                  WeiLayout,
                                  OutLayout,
                                  NDimSpatial,
                                  MPerBlock / ClusterLengthMPerBlock,
                                  NPerBlock / ClusterLengthNPerBlock>{}

◆ conv_to_gemm_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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::conv_to_gemm_transformer

staticconstexpr

Initial value:

=
        TransformConvBwdWeightToGemm<NDimSpatial,
                                     MPerBlock,
                                     NPerBlock,
                                     K1Number,
                                     K0PerBlock,
                                     ConvBackwardWeightSpecialization>{}

◆ ElePerBank

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::ElePerBank = BankLength / sizeof(ADataType)

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

GET_NXDL_PER_WAVE_IMPL constexpr auto ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::NXdlPerWave64 = GetNXdlPerWave<true>()

staticconstexpr

◆ TransposeTransferDstScalarPerVectorAligned

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

index_t ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::TransposeTransferDstScalarPerVectorAligned

staticconstexpr

Initial value:

=

std::min(MPerBlock / ClusterLengthMPerBlock, MaxTransposeTransferDstScalarPerVector)

◆ TransposeTransferSrcScalarPerVectorAligned

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 K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t MaxTransposeTransferSrcScalarPerVector = 1, index_t MaxTransposeTransferDstScalarPerVector = 1>

index_t ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, ComputeTypeA, ComputeTypeB, MaxTransposeTransferSrcScalarPerVector, MaxTransposeTransferDstScalarPerVector >::TransposeTransferSrcScalarPerVectorAligned

staticconstexpr

Initial value:

=

std::min(NPerBlock / ClusterLengthNPerBlock, MaxTransposeTransferSrcScalarPerVector)

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

device_grouped_conv_bwd_weight_xdl_cshuffle.hpp

Classes

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Member Typedef Documentation

◆ ABCGridDescs

◆ ABDataType

◆ ADataType

◆ AElementwiseOperation

◆ AGridDesc_K0_M_K1

◆ BDataType

◆ BElementwiseOperation

◆ BGridDesc_K0_N_K1

◆ Block2CTileMap

◆ Block2TileMapTranspose

◆ CDataType

◆ CElementwiseOperation

◆ CGridDesc_M_N

◆ CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock

◆ DeviceOp

◆ GKCYXTransposeDescType

◆ GKYXCTransposeDescType

◆ GridwiseElementwiseWeightTranspose

◆ GridwiseGemm32

◆ GridwiseGemm64

◆ GridwiseGemmBase

◆ GridwiseInOutTranspose

◆ NGCHWTransposeDescType

◆ NHWGCTransposeDescType

Member Function Documentation

◆ GetABCGridDesc() [1/3]

◆ GetABCGridDesc() [2/3]

◆ GetABCGridDesc() [3/3]

◆ GetTypeString()

◆ GetWorkSpaceSize()

◆ IsSupportedArgument() [1/2]

◆ IsSupportedArgument() [2/2]

◆ IsValidCompilationParameter()

◆ MakeArgument()

◆ MakeArgumentPointer()

◆ MakeInvoker()

◆ MakeInvokerPointer()

◆ SetWorkSpacePointer()

Member Data Documentation

◆ ABlockLdsM0PerBlock

◆ ABlockLdsM1Padding

◆ ABlockLdsM1PerBlock

◆ BankLength

◆ BBlockLdsN0PerBlock

◆ BBlockLdsN1Padding

◆ BBlockLdsN1PerBlock

◆ ClusterLengthMPerBlock

◆ ClusterLengthNPerBlock

◆ conv_ngchw_to_nhwgc_transformer

◆ conv_to_gemm_transformer

◆ ElePerBank

◆ I0

◆ I1

◆ I2

◆ I3

◆ I4

◆ I5

◆ K1Number

◆ NXdlPerWave32

◆ NXdlPerWave64

◆ TransposeTransferDstScalarPerVectorAligned

◆ TransposeTransferSrcScalarPerVectorAligned