DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector > Struct Template Reference

DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K&lt; InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector &gt; Struct Template Reference#

Composable Kernel: ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector > Struct Template Reference
Classes | Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members
ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector > Struct Template Reference

#include <device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >:
ck::tensor_operation::device::DeviceConvFwd< InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation > ck::tensor_operation::device::BaseOperator

Classes

struct  Argument
 
struct  Invoker
 

Public Types

using DeviceOp = DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K
 
using ADataType = InDataType
 
using BDataType = WeiDataType
 
using CDataType = OutDataType
 
using ABDataType = InDataType
 
using ABCGridDescs = remove_cvref_t< decltype(MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(1, 1, 1, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}))>
 
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 GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3< BlockSize, InDataType, AccDataType, OutDataType, InMemoryDataOperationEnum::Set, AGridDesc_K0_M_K1, BGridDesc_K0_N_K1, CGridDesc_M_N, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, MPerBlock, NPerBlock, K0PerBlock, MPerXDL, NPerXDL, K1, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, Sequence< 1, 0, 2 >, Sequence< 1, 0, 2 >, 2, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, Sequence< 1, 0, 2 >, Sequence< 1, 0, 2 >, 2, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BBlockLdsAddExtraN, Sequence< 2, 3, 0, 1, 7, 5, 4, 6 >, 7, CThreadTransferDstScalarPerVector >
 
using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 = decltype(GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}))
 
using Block2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap
 

Public Member Functions

bool IsSupportedArgument (const BaseArgument *p_arg) override
 
std::unique_ptr< BaseArgument > MakeArgumentPointer (const void *p_in, const void *p_wei, void *p_out, const index_t N, const index_t K, const index_t C, std::vector< ck::index_t > input_spatial_lengths, std::vector< ck::index_t > filter_spatial_lengths, std::vector< ck::index_t > output_spatial_lengths, std::vector< ck::index_t > conv_filter_strides, std::vector< ck::index_t > conv_filter_dilations, std::vector< ck::index_t > input_left_pads, std::vector< ck::index_t > input_right_pads, InElementwiseOperation in_element_op, WeiElementwiseOperation wei_element_op, OutElementwiseOperation out_element_op) override
 
std::unique_ptr< BaseInvoker > MakeInvokerPointer () override
 
std::string GetTypeString () 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 GetTypeIdName () const
 
virtual std::optional< std::string > GetObjectName () const
 
virtual std::optional< std::string > GetTemplateInfo () const
 
virtual std::string GetTypeIdHashCode () const
 
virtual size_t GetWorkSpaceSize (const BaseArgument *) const
 
virtual void SetWorkSpacePointer (BaseArgument *p_arg, void *p_workspace, const StreamConfig &=StreamConfig{}) const
 
virtual ~BaseOperator ()
 

Static Public Member Functions

static index_t GetMaxAllowableSubBatchSize (const index_t N, const index_t K, const index_t C, std::vector< ck::index_t > input_spatial_lengths, std::vector< ck::index_t > output_spatial_lengths)
 
static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N (const index_t N, const index_t K, const index_t C, std::vector< ck::index_t > input_spatial_lengths, std::vector< ck::index_t > filter_spatial_lengths, std::vector< ck::index_t > output_spatial_lengths, std::vector< ck::index_t > conv_filter_strides, std::vector< ck::index_t > conv_filter_dilations, std::vector< ck::index_t > input_left_pads, std::vector< ck::index_t > input_right_pads)
 
static constexpr bool IsValidCompilationParameter ()
 
static bool IsSupportedArgument (const Argument &arg)
 
static auto MakeArgument (const InDataType *p_in, const WeiDataType *p_wei, OutDataType *p_out, const index_t N, const index_t K, const index_t C, std::vector< ck::index_t > input_spatial_lengths, std::vector< ck::index_t > filter_spatial_lengths, std::vector< ck::index_t > output_spatial_lengths, std::vector< ck::index_t > conv_filter_strides, std::vector< ck::index_t > conv_filter_dilations, std::vector< ck::index_t > input_left_pads, std::vector< ck::index_t > input_right_pads, InElementwiseOperation in_element_op, WeiElementwiseOperation wei_element_op, OutElementwiseOperation out_element_op)
 
static auto MakeInvoker ()
 

Static Public Attributes

static constexpr auto I0 = Number<0>{}
 
static constexpr auto I1 = Number<1>{}
 
static constexpr auto I2 = Number<2>{}
 
static constexpr auto I3 = Number<3>{}
 

Member Typedef Documentation

◆ ABCGridDescs

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::ABCGridDescs = remove_cvref_t<decltype(MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N( 1, 1, 1, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}))>

◆ ABDataType

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::ABDataType = InDataType

◆ ADataType

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::ADataType = InDataType

◆ AGridDesc_K0_M_K1

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>

◆ BDataType

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::BDataType = WeiDataType

◆ BGridDesc_K0_N_K1

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>

◆ Block2CTileMap

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::Block2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap

◆ CDataType

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::CDataType = OutDataType

◆ CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 = decltype(GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}))

◆ CGridDesc_M_N

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::CGridDesc_M_N = remove_cvref_t<decltype(ABCGridDescs{}[I2])>

◆ DeviceOp

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::DeviceOp = DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K

◆ GridwiseGemm

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
using ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3< BlockSize, InDataType, AccDataType, OutDataType, InMemoryDataOperationEnum::Set, AGridDesc_K0_M_K1, BGridDesc_K0_N_K1, CGridDesc_M_N, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, MPerBlock, NPerBlock, K0PerBlock, MPerXDL, NPerXDL, K1, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, Sequence<1, 0, 2>, Sequence<1, 0, 2>, 2, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, Sequence<1, 0, 2>, Sequence<1, 0, 2>, 2, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BBlockLdsAddExtraN, Sequence<2, 3, 0, 1, 7, 5, 4, 6>, 7, CThreadTransferDstScalarPerVector>

Member Function Documentation

◆ GetMaxAllowableSubBatchSize()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
static index_t ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::GetMaxAllowableSubBatchSize ( const index_t  N,
const index_t  K,
const index_t  C,
std::vector< ck::index_t >  input_spatial_lengths,
std::vector< ck::index_t >  output_spatial_lengths 
)
inlinestatic

◆ GetTypeString()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
std::string ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::GetTypeString ( ) const
inlineoverridevirtual

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

◆ IsSupportedArgument() [1/2]

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
static bool ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::IsSupportedArgument ( const Argument &  arg)
inlinestatic

◆ IsSupportedArgument() [2/2]

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
bool ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::IsSupportedArgument ( const BaseArgument *  p_arg)
inlineoverridevirtual

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

◆ IsValidCompilationParameter()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
static constexpr bool ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::IsValidCompilationParameter ( )
inlinestaticconstexpr

◆ MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
static auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N ( const index_t  N,
const index_t  K,
const index_t  C,
std::vector< ck::index_t >  input_spatial_lengths,
std::vector< ck::index_t >  filter_spatial_lengths,
std::vector< ck::index_t >  output_spatial_lengths,
std::vector< ck::index_t >  conv_filter_strides,
std::vector< ck::index_t >  conv_filter_dilations,
std::vector< ck::index_t >  input_left_pads,
std::vector< ck::index_t >  input_right_pads 
)
inlinestatic

◆ MakeArgument()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
static auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::MakeArgument ( const InDataType *  p_in,
const WeiDataType *  p_wei,
OutDataType *  p_out,
const index_t  N,
const index_t  K,
const index_t  C,
std::vector< ck::index_t >  input_spatial_lengths,
std::vector< ck::index_t >  filter_spatial_lengths,
std::vector< ck::index_t >  output_spatial_lengths,
std::vector< ck::index_t >  conv_filter_strides,
std::vector< ck::index_t >  conv_filter_dilations,
std::vector< ck::index_t >  input_left_pads,
std::vector< ck::index_t >  input_right_pads,
InElementwiseOperation  in_element_op,
WeiElementwiseOperation  wei_element_op,
OutElementwiseOperation  out_element_op 
)
inlinestatic

◆ MakeArgumentPointer()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
std::unique_ptr<BaseArgument> ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::MakeArgumentPointer ( const void *  p_in,
const void *  p_wei,
void *  p_out,
const index_t  N,
const index_t  K,
const index_t  C,
std::vector< ck::index_t >  input_spatial_lengths,
std::vector< ck::index_t >  filter_spatial_lengths,
std::vector< ck::index_t >  output_spatial_lengths,
std::vector< ck::index_t >  conv_filter_strides,
std::vector< ck::index_t >  conv_filter_dilations,
std::vector< ck::index_t >  input_left_pads,
std::vector< ck::index_t >  input_right_pads,
InElementwiseOperation  in_element_op,
WeiElementwiseOperation  wei_element_op,
OutElementwiseOperation  out_element_op 
)
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceConvFwd< InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation >.

◆ MakeInvoker()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
static auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::MakeInvoker ( )
inlinestatic

◆ MakeInvokerPointer()

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
std::unique_ptr<BaseInvoker> ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::MakeInvokerPointer ( )
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceConvFwd< InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation >.

Member Data Documentation

◆ I0

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
constexpr auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::I0 = Number<0>{}
staticconstexpr

◆ I1

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
constexpr auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::I1 = Number<1>{}
staticconstexpr

◆ I2

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
constexpr auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::I2 = Number<2>{}
staticconstexpr

◆ I3

template<typename InDataType , typename WeiDataType , typename OutDataType , typename AccDataType , typename InElementwiseOperation , typename WeiElementwiseOperation , typename OutElementwiseOperation , ConvolutionForwardSpecialization ConvForwardSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector>
constexpr auto ck::tensor_operation::device::DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvForwardSpecialization, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector >::I3 = Number<3>{}
staticconstexpr
The documentation for this struct was generated from the following file: