DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched > Struct Template Reference

DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle&lt; NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched &gt; Struct Template Reference#

Composable Kernel: ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched > Struct Template Reference
Classes | Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members
ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched > Struct Template Reference

#include <device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >:
ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, MaskingSpec > ck::tensor_operation::device::BaseOperator

Classes

struct  Argument
 
struct  ComputeBasePtrOfStridedBatch
 
struct  GroupDeviceArg
 
struct  GroupKernelArg
 
struct  Invoker
 

Public Types

using DeviceOp = DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
 
using ProblemDesc = typename DeviceGroupedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, MaskingSpec >::ProblemDesc
 
using Transform = TransformBatchedContractionContractionToBatchedGemmGemm< Sequence< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO >, Sequence< MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock >, GemmSpec, ASpec, BSpec, B1Spec, CSpec >
 
using AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1({}, {}))
 
using BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1({}, {}))
 
using B1GridDesc_BK0_N_BK1 = decltype(MakeB1GridDescriptor_BK0_N_BK1({}, {}))
 
using CGridDesc_M_N = decltype(Transform::MakeCGridDescriptor_M_N({}, {}))
 
using AGridDesc_G_M_K = decltype(Transform::MakeAGridDescriptor_G_M_K({}, {}))
 
using BGridDesc_G_N_K = decltype(Transform::MakeB0GridDescriptor_G_N_K({}, {}))
 
using B1GridDesc_G_N_K = decltype(Transform::MakeB1GridDescriptor_G_N_K({}, {}))
 
using CGridDesc_G_M_N = decltype(Transform::MakeCGridDescriptor_G_M_N({}, {}))
 
using C0MatrixMask = C0MatrixMask_impl< decltype(make_MaskOutPredicate())>
 
using GridwiseGemm = GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle< ADataType, GemmAccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, InMemoryDataOperationEnum::Set, AGridDesc_AK0_M_AK1, BGridDesc_BK0_N_BK1, B1GridDesc_BK0_N_BK1, CGridDesc_M_N, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, true, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, true, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, false, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, Transform::matrix_padder.PadN, MaskingSpec==MaskingSpecialization::MaskOutUpperTriangle >
 
using Block2CTileMap = OffsettedBlockToCTileMap< typename GridwiseGemm::DefaultBlock2CTileMap >
 

Public Member Functions

bool IsSupportedArgument (const BaseArgument *p_arg) override
 
std::unique_ptr< BaseArgument > MakeArgumentPointer (std::vector< const void * > p_a_vec, std::vector< const void * > p_b_vec, std::vector< const void * > p_b1_vec, std::vector< void * > p_c_vec, std::vector< std::vector< const void * >> p_acc0_biases_vec, std::vector< std::vector< const void * >> p_acc1_biases_vec, std::vector< ProblemDesc > problem_desc_vec, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, AccElementwiseOperation acc_element_op, B1ElementwiseOperation b1_element_op, CElementwiseOperation c_element_op) override
 
std::unique_ptr< BaseInvoker > MakeInvokerPointer () override
 
std::string GetTypeString () const override
 
size_t GetWorkSpaceSize (const BaseArgument *p_arg) const override
 
- Public Member Functions inherited from ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, MaskingSpec >
virtual std::unique_ptr< BaseArgument > MakeArgumentPointer (std::vector< const void * > p_a_vec, std::vector< const void * > p_b0_vec, std::vector< const void * > p_b1_vec, std::vector< void * > p_c_vec, std::vector< std::vector< const void * >> p_acc0_biases_vec, std::vector< std::vector< const void * >> p_acc1_biases_vec, std::vector< ProblemDesc > problem_desc_vec, AElementwiseOperation a_element_op, BElementwiseOperation b0_element_op, AccElementwiseOperation acc0_element_op, B1ElementwiseOperation b1_element_op, CElementwiseOperation c_element_op)=0
 
- 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 void SetWorkSpacePointer (BaseArgument *p_arg, void *p_workspace, const StreamConfig &=StreamConfig{}) const
 
virtual ~BaseOperator ()
 

Static Public Member Functions

static auto MakeAGridDescriptor_AK0_M_AK1 (const std::vector< index_t > &a_gs_ms_ks_lengths_vec, const std::vector< index_t > &a_gs_ms_ks_strides_vec)
 
static auto MakeBGridDescriptor_BK0_N_BK1 (const std::vector< index_t > &b_gs_ns_ks_lengths_vec, const std::vector< index_t > &b_gs_ns_ks_strides_vec)
 
static auto MakeB1GridDescriptor_BK0_N_BK1 (const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_lengths_vec, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_strides_vec)
 
constexpr static auto make_MaskOutPredicate ()
 
static constexpr bool IsValidCompilationParameter ()
 
static bool IsSupportedArgument (const Argument &arg)
 
static auto MakeArgument (std::vector< const void * > p_a_vec, std::vector< const void * > p_b_vec, std::vector< const void * > p_b1_vec, std::vector< void * > p_c_vec, std::vector< std::vector< const void * >> p_acc0_biases_vec, std::vector< std::vector< const void * >> p_acc1_biases_vec, std::vector< ProblemDesc > problem_desc_vec, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, AccElementwiseOperation acc_element_op, B1ElementwiseOperation b1_element_op, CElementwiseOperation c_element_op)
 
static auto MakeInvoker ()
 

Static Public Attributes

static constexpr index_t NumAcc0Bias = Acc0BiasDataType::Size()
 
static constexpr index_t NumAcc1Bias = Acc1BiasDataType::Size()
 
static constexpr auto I0 = Number<0>{}
 
static constexpr auto I1 = Number<1>{}
 
static constexpr auto I2 = Number<2>{}
 

Member Typedef Documentation

◆ AGridDesc_AK0_M_AK1

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1({}, {}))

◆ AGridDesc_G_M_K

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::AGridDesc_G_M_K = decltype(Transform::MakeAGridDescriptor_G_M_K({}, {}))

◆ B1GridDesc_BK0_N_BK1

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::B1GridDesc_BK0_N_BK1 = decltype(MakeB1GridDescriptor_BK0_N_BK1({}, {}))

◆ B1GridDesc_G_N_K

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::B1GridDesc_G_N_K = decltype(Transform::MakeB1GridDescriptor_G_N_K({}, {}))

◆ BGridDesc_BK0_N_BK1

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1({}, {}))

◆ BGridDesc_G_N_K

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::BGridDesc_G_N_K = decltype(Transform::MakeB0GridDescriptor_G_N_K({}, {}))

◆ Block2CTileMap

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::Block2CTileMap = OffsettedBlockToCTileMap<typename GridwiseGemm::DefaultBlock2CTileMap>

◆ C0MatrixMask

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::C0MatrixMask = C0MatrixMask_impl<decltype(make_MaskOutPredicate())>

◆ CGridDesc_G_M_N

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::CGridDesc_G_M_N = decltype(Transform::MakeCGridDescriptor_G_M_N({}, {}))

◆ CGridDesc_M_N

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::CGridDesc_M_N = decltype(Transform::MakeCGridDescriptor_M_N({}, {}))

◆ DeviceOp

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::DeviceOp = DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle

◆ GridwiseGemm

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::GridwiseGemm = GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle< ADataType, GemmAccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, InMemoryDataOperationEnum::Set, AGridDesc_AK0_M_AK1, BGridDesc_BK0_N_BK1, B1GridDesc_BK0_N_BK1, CGridDesc_M_N, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, true, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, true, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, false, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, Transform::matrix_padder.PadN, MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle>

◆ ProblemDesc

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::ProblemDesc = typename DeviceGroupedGemmSoftmaxGemmPermute<NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, MaskingSpec>::ProblemDesc

◆ Transform

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
using ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::Transform = TransformBatchedContractionContractionToBatchedGemmGemm< Sequence<NumDimG, NumDimM, NumDimN, NumDimK, NumDimO>, Sequence<MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock>, GemmSpec, ASpec, BSpec, B1Spec, CSpec>

Member Function Documentation

◆ GetTypeString()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
std::string ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::GetTypeString ( ) const
inlineoverridevirtual

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

◆ GetWorkSpaceSize()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
size_t ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::GetWorkSpaceSize ( const BaseArgument *  p_arg) const
inlineoverridevirtual

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

◆ IsSupportedArgument() [1/2]

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static bool ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::IsSupportedArgument ( const Argument &  arg)
inlinestatic

◆ IsSupportedArgument() [2/2]

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
bool ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::IsSupportedArgument ( const BaseArgument *  p_arg)
inlineoverridevirtual

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

◆ IsValidCompilationParameter()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static constexpr bool ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::IsValidCompilationParameter ( )
inlinestaticconstexpr

◆ make_MaskOutPredicate()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
constexpr static auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::make_MaskOutPredicate ( )
inlinestaticconstexpr

◆ MakeAGridDescriptor_AK0_M_AK1()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeAGridDescriptor_AK0_M_AK1 ( const std::vector< index_t > &  a_gs_ms_ks_lengths_vec,
const std::vector< index_t > &  a_gs_ms_ks_strides_vec 
)
inlinestatic

◆ MakeArgument()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeArgument ( std::vector< const void * >  p_a_vec,
std::vector< const void * >  p_b_vec,
std::vector< const void * >  p_b1_vec,
std::vector< void * >  p_c_vec,
std::vector< std::vector< const void * >>  p_acc0_biases_vec,
std::vector< std::vector< const void * >>  p_acc1_biases_vec,
std::vector< ProblemDesc >  problem_desc_vec,
AElementwiseOperation  a_element_op,
BElementwiseOperation  b_element_op,
AccElementwiseOperation  acc_element_op,
B1ElementwiseOperation  b1_element_op,
CElementwiseOperation  c_element_op 
)
inlinestatic

◆ MakeArgumentPointer()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
std::unique_ptr<BaseArgument> ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeArgumentPointer ( std::vector< const void * >  p_a_vec,
std::vector< const void * >  p_b_vec,
std::vector< const void * >  p_b1_vec,
std::vector< void * >  p_c_vec,
std::vector< std::vector< const void * >>  p_acc0_biases_vec,
std::vector< std::vector< const void * >>  p_acc1_biases_vec,
std::vector< ProblemDesc >  problem_desc_vec,
AElementwiseOperation  a_element_op,
BElementwiseOperation  b_element_op,
AccElementwiseOperation  acc_element_op,
B1ElementwiseOperation  b1_element_op,
CElementwiseOperation  c_element_op 
)
inlineoverride

◆ MakeB1GridDescriptor_BK0_N_BK1()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeB1GridDescriptor_BK0_N_BK1 ( const std::vector< index_t > &  b1_gs_gemm1ns_gemm1ks_lengths_vec,
const std::vector< index_t > &  b1_gs_gemm1ns_gemm1ks_strides_vec 
)
inlinestatic

◆ MakeBGridDescriptor_BK0_N_BK1()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeBGridDescriptor_BK0_N_BK1 ( const std::vector< index_t > &  b_gs_ns_ks_lengths_vec,
const std::vector< index_t > &  b_gs_ns_ks_strides_vec 
)
inlinestatic

◆ MakeInvoker()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
static auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeInvoker ( )
inlinestatic

◆ MakeInvokerPointer()

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
std::unique_ptr<BaseInvoker> ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::MakeInvokerPointer ( )
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, MaskingSpec >.

Member Data Documentation

◆ I0

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
constexpr auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::I0 = Number<0>{}
staticconstexpr

◆ I1

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
constexpr auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::I1 = Number<1>{}
staticconstexpr

◆ I2

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
constexpr auto ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::I2 = Number<2>{}
staticconstexpr

◆ NumAcc0Bias

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
constexpr index_t ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::NumAcc0Bias = Acc0BiasDataType::Size()
staticconstexpr

◆ NumAcc1Bias

template<index_t NumDimG, index_t NumDimM, index_t NumDimN, index_t NumDimK, index_t NumDimO, typename ADataType , typename BDataType , typename B1DataType , typename CDataType , typename Acc0BiasDataType , typename Acc1BiasDataType , typename GemmAccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename AccElementwiseOperation , typename B1ElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, TensorSpecialization ASpec, TensorSpecialization BSpec, TensorSpecialization B1Spec, TensorSpecialization CSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t Gemm1NPerBlock, index_t Gemm1KPerBlock, index_t AK1, index_t BK1, index_t B1K1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, index_t Gemm1NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, typename B1BlockTransferThreadClusterLengths_BK0_N_BK1 , typename B1BlockTransferThreadClusterArrangeOrder , typename B1BlockTransferSrcAccessOrder , index_t B1BlockTransferSrcVectorDim, index_t B1BlockTransferSrcScalarPerVector, index_t B1BlockTransferDstScalarPerVector_BK1, bool B1BlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpecialization MaskingSpec, LoopScheduler LoopSched = LoopScheduler::Default>
constexpr index_t ck::tensor_operation::device::DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, Acc0BiasDataType, Acc1BiasDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, AccElementwiseOperation, B1ElementwiseOperation, CElementwiseOperation, GemmSpec, ASpec, BSpec, B1Spec, CSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, Gemm1NPerBlock, Gemm1KPerBlock, AK1, BK1, B1K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, Gemm1NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, B1BlockTransferThreadClusterLengths_BK0_N_BK1, B1BlockTransferThreadClusterArrangeOrder, B1BlockTransferSrcAccessOrder, B1BlockTransferSrcVectorDim, B1BlockTransferSrcScalarPerVector, B1BlockTransferDstScalarPerVector_BK1, B1BlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, MaskingSpec, LoopSched >::NumAcc1Bias = Acc1BiasDataType::Size()
staticconstexpr
The documentation for this struct was generated from the following file: