Composable Kernel: ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched > Struct Template Reference

DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched > Struct Template Reference

DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched > Struct Template Reference#

Composable Kernel: ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched > Struct Template Reference

ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched > Struct Template Reference

#include <device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >:

Classes
struct	ComputeBasePtrOfStridedBatch
struct	Argument
struct	Invoker

Public Types
using	DeviceOp = DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
using	Transform
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	C1GridDesc_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	C1GridDesc_G_M_N = decltype(Transform::MakeCGridDescriptor_G_M_N({}, {}))
using	D0sGridDesc_M_N = decltype(MakeD0sGridDescriptor_M_N({}, {}))
using	D0sGridDesc_G_M_N = decltype(MakeD0sGridDescriptor_G_M_N({}, {}))
using	C0MatrixMask = C0MatrixMask_impl<decltype(make_MaskOutPredicate())>
template<index_t MXdlPerWave_>
using	GridwiseGemmBase
using	GridwiseGemm64 = GridwiseGemmBase<math::max(MXdlPerWave64, 1)>
using	GridwiseGemm32 = GridwiseGemmBase<MXdlPerWave32>

Public Member Functions
bool	IsSupportedArgument (const BaseArgument *p_arg) override
std::unique_ptr< BaseArgument >	MakeArgumentPointer (const void p_a, const void p_b, const void p_b1, void p_c, const std::array< void , NumD0Tensor > p_acc0_biases, const std::array< void , NumD1Tensor > p_acc1_biases, const std::vector< index_t > &a_gs_ms_ks_lengths, const std::vector< index_t > &a_gs_ms_ks_strides, const std::vector< index_t > &b_gs_ns_ks_lengths, const std::vector< index_t > &b_gs_ns_ks_strides, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_lengths, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_strides, const std::vector< index_t > &c_gs_ms_gemm1ns_lengths, const std::vector< index_t > &c_gs_ms_gemm1ns_strides, const std::array< std::vector< ck::index_t >, NumD0Tensor > acc0_biases_gs_ms_ns_lengths, const std::array< std::vector< ck::index_t >, NumD0Tensor > acc0_biases_gs_ms_ns_strides, const std::array< std::vector< ck::index_t >, NumD1Tensor > acc1_biases_gs_ms_gemm1ns_lengths, const std::array< std::vector< ck::index_t >, NumD1Tensor > acc1_biases_gs_ms_gemm1ns_strides, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, C0DEElementwiseOperation c0de_element_op, B1ElementwiseOperation b1_element_op, C1DEElementwiseOperation c1de_element_op) override
std::unique_ptr< BaseInvoker >	MakeInvokerPointer () override
std::string	GetTypeString () const override
Public Member Functions inherited from ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, MaskingSpec >
virtual std::unique_ptr< BaseArgument >	MakeArgumentPointer (const void p_a, const void p_b0, const void p_b1, void p_c, const std::array< void , NumAcc0Bias > p_acc0_biases, const std::array< void , NumAcc1Bias > p_acc1_biases, const std::vector< index_t > &a_gs_ms_ks_lengths, const std::vector< index_t > &a_gs_ms_ks_strides, const std::vector< index_t > &b_gs_ns_ks_lengths, const std::vector< index_t > &b_gs_ns_ks_strides, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_lengths, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_strides, const std::vector< index_t > &c_gs_ms_gemm1ns_lengths, const std::vector< index_t > &c_gs_ms_gemm1ns_strides, const std::array< std::vector< index_t >, NumAcc0Bias > acc0_biases_gs_ms_ns_lengths, const std::array< std::vector< index_t >, NumAcc0Bias > acc0_biases_gs_ms_ns_strides, const std::array< std::vector< index_t >, NumAcc1Bias > acc1_biases_gs_ms_gemm1ns_lengths, const std::array< std::vector< index_t >, NumAcc1Bias > acc1_biases_gs_ms_gemm1ns_strides, AElementwiseOperation a_element_op, BElementwiseOperation b0_element_op, C0DEElementwiseOperation c0de_element_op, B1ElementwiseOperation b1_element_op, C1DEElementwiseOperation c1de_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	GetInstanceString () const
virtual std::string	GetTypeIdName () const
virtual std::optional< std::string >	GetObjectName () const
virtual std::optional< std::string >	GetTemplateInfo () const
virtual std::string	GetTypeIdHashCode () const
virtual size_t	GetWorkSpaceSize (const BaseArgument *) const
virtual void	SetWorkSpacePointer (BaseArgument p_arg, void p_workspace, const StreamConfig &=StreamConfig{}) const
virtual	~BaseOperator ()

Static Public Member Functions
static 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)
static auto	MakeD0sGridDescriptor_M_N (const std::array< std::vector< ck::index_t >, NumD0Tensor > &acc0_biases_gs_ms_ns_lengths, const std::array< std::vector< ck::index_t >, NumD0Tensor > &acc0_biases_gs_ms_ns_strides)
static auto	MakeD0sGridDescriptor_G_M_N (const std::array< std::vector< ck::index_t >, NumD0Tensor > &acc0_biases_gs_ms_ns_lengths, const std::array< std::vector< ck::index_t >, NumD0Tensor > &acc0_biases_gs_ms_ns_strides)
static constexpr auto	make_MaskOutPredicate ()
static constexpr bool	IsValidCompilationParameter ()
static bool	IsSupportedArgument (const Argument &arg)
static auto	MakeArgument (const ADataType p_a, const BDataType p_b, const B1DataType p_b1, CDataType p_c, const std::array< void , NumD0Tensor > p_acc0_biases, const std::array< void , NumD1Tensor > p_acc1_biases, const std::vector< index_t > &a_gs_ms_ks_lengths, const std::vector< index_t > &a_gs_ms_ks_strides, const std::vector< index_t > &b_gs_ns_ks_lengths, const std::vector< index_t > &b_gs_ns_ks_strides, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_lengths, const std::vector< index_t > &b1_gs_gemm1ns_gemm1ks_strides, const std::vector< index_t > &c_gs_ms_gemm1ns_lengths, const std::vector< index_t > &c_gs_ms_gemm1ns_strides, const std::array< std::vector< ck::index_t >, NumD0Tensor > acc0_biases_gs_ms_ns_lengths, const std::array< std::vector< ck::index_t >, NumD0Tensor > acc0_biases_gs_ms_ns_strides, const std::array< std::vector< ck::index_t >, NumD1Tensor > acc1_biases_gs_ms_gemm1ns_lengths, const std::array< std::vector< ck::index_t >, NumD1Tensor > acc1_biases_gs_ms_gemm1ns_strides, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, C0DEElementwiseOperation c0de_element_op, B1ElementwiseOperation b1_element_op, C1DEElementwiseOperation c1de_element_op)
static auto	MakeInvoker ()

Static Public Attributes
static constexpr auto	MXdlPerWave64
static constexpr auto	MXdlPerWave32
static constexpr index_t	NumD0Tensor = D0sDataType::Size()
static constexpr index_t	NumD1Tensor = D1sDataType::Size()
static constexpr auto	I0 = Number<0>{}
static constexpr auto	I1 = Number<1>{}
static constexpr auto	I2 = Number<2>{}
Static Public Attributes inherited from ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, MaskingSpec >
static constexpr index_t	NumAcc0Bias
static constexpr index_t	NumAcc1Bias

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::BGridDesc_G_N_K = decltype(Transform::MakeB0GridDescriptor_G_N_K({}, {}))

◆ 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::C0MatrixMask = C0MatrixMask_impl<decltype(make_MaskOutPredicate())>

◆ C1GridDesc_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::C1GridDesc_G_M_N = decltype(Transform::MakeCGridDescriptor_G_M_N({}, {}))

◆ C1GridDesc_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::C1GridDesc_M_N = decltype(Transform::MakeCGridDescriptor_M_N({}, {}))

◆ D0sGridDesc_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::D0sGridDesc_G_M_N = decltype(MakeD0sGridDescriptor_G_M_N({}, {}))

◆ D0sGridDesc_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::D0sGridDesc_M_N = decltype(MakeD0sGridDescriptor_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::DeviceOp = DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle

◆ GridwiseGemm32

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::GridwiseGemm32 = GridwiseGemmBase<MXdlPerWave32>

◆ GridwiseGemm64

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::GridwiseGemm64 = GridwiseGemmBase<math::max(MXdlPerWave64, 1)>

◆ GridwiseGemmBase

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

template<index_t MXdlPerWave_>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::GridwiseGemmBase

◆ 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

using ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::Transform

Initial value:

 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

std::string ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::GetTypeString ( ) 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

bool ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

bool ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

constexpr bool ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

constexpr auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MakeArgument	(	const ADataType *	p_a,
		const BDataType *	p_b,
		const B1DataType *	p_b1,
		CDataType *	p_c,
		const std::array< void *, NumD0Tensor >	p_acc0_biases,
		const std::array< void *, NumD1Tensor >	p_acc1_biases,
		const std::vector< index_t > &	a_gs_ms_ks_lengths,
		const std::vector< index_t > &	a_gs_ms_ks_strides,
		const std::vector< index_t > &	b_gs_ns_ks_lengths,
		const std::vector< index_t > &	b_gs_ns_ks_strides,
		const std::vector< index_t > &	b1_gs_gemm1ns_gemm1ks_lengths,
		const std::vector< index_t > &	b1_gs_gemm1ns_gemm1ks_strides,
		const std::vector< index_t > &	c_gs_ms_gemm1ns_lengths,
		const std::vector< index_t > &	c_gs_ms_gemm1ns_strides,
		const std::array< std::vector< ck::index_t >, NumD0Tensor >	acc0_biases_gs_ms_ns_lengths,
		const std::array< std::vector< ck::index_t >, NumD0Tensor >	acc0_biases_gs_ms_ns_strides,
		const std::array< std::vector< ck::index_t >, NumD1Tensor >	acc1_biases_gs_ms_gemm1ns_lengths,
		const std::array< std::vector< ck::index_t >, NumD1Tensor >	acc1_biases_gs_ms_gemm1ns_strides,
		AElementwiseOperation	a_element_op,
		BElementwiseOperation	b_element_op,
		C0DEElementwiseOperation	c0de_element_op,
		B1ElementwiseOperation	b1_element_op,
		C1DEElementwiseOperation	c1de_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

std::unique_ptr< BaseArgument > ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MakeArgumentPointer	(	const void *	p_a,
		const void *	p_b,
		const void *	p_b1,
		void *	p_c,
		const std::array< void *, NumD0Tensor >	p_acc0_biases,
		const std::array< void *, NumD1Tensor >	p_acc1_biases,
		const std::vector< index_t > &	a_gs_ms_ks_lengths,
		const std::vector< index_t > &	a_gs_ms_ks_strides,
		const std::vector< index_t > &	b_gs_ns_ks_lengths,
		const std::vector< index_t > &	b_gs_ns_ks_strides,
		const std::vector< index_t > &	b1_gs_gemm1ns_gemm1ks_lengths,
		const std::vector< index_t > &	b1_gs_gemm1ns_gemm1ks_strides,
		const std::vector< index_t > &	c_gs_ms_gemm1ns_lengths,
		const std::vector< index_t > &	c_gs_ms_gemm1ns_strides,
		const std::array< std::vector< ck::index_t >, NumD0Tensor >	acc0_biases_gs_ms_ns_lengths,
		const std::array< std::vector< ck::index_t >, NumD0Tensor >	acc0_biases_gs_ms_ns_strides,
		const std::array< std::vector< ck::index_t >, NumD1Tensor >	acc1_biases_gs_ms_gemm1ns_lengths,
		const std::array< std::vector< ck::index_t >, NumD1Tensor >	acc1_biases_gs_ms_gemm1ns_strides,
		AElementwiseOperation	a_element_op,
		BElementwiseOperation	b_element_op,
		C0DEElementwiseOperation	c0de_element_op,
		B1ElementwiseOperation	b1_element_op,
		C1DEElementwiseOperation	c1de_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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

◆ MakeD0sGridDescriptor_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MakeD0sGridDescriptor_G_M_N	(	const std::array< std::vector< ck::index_t >, NumD0Tensor > &	acc0_biases_gs_ms_ns_lengths,
		const std::array< std::vector< ck::index_t >, NumD0Tensor > &	acc0_biases_gs_ms_ns_strides )

inlinestatic

◆ MakeD0sGridDescriptor_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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MakeD0sGridDescriptor_M_N	(	const std::array< std::vector< ck::index_t >, NumD0Tensor > &	acc0_biases_gs_ms_ns_lengths,
		const std::array< std::vector< ck::index_t >, NumD0Tensor > &	acc0_biases_gs_ms_ns_strides )

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

std::unique_ptr< BaseInvoker > ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MakeInvokerPointer ( )

inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, 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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::I2 = Number<2>{}

staticconstexpr

◆ MXdlPerWave32

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MXdlPerWave32

staticconstexpr

Initial value:

=

GetNXdlPerWave2<BlockSize, NPerBlock, MPerBlock, NPerXDL, MPerXDL, NXdlPerWave, false>()

◆ MXdlPerWave64

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

auto ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::MXdlPerWave64

staticconstexpr

Initial value:

=

GetNXdlPerWave2<BlockSize, NPerBlock, MPerBlock, NPerXDL, MPerXDL, NXdlPerWave, true>()

◆ NumD0Tensor

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

index_t ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::NumD0Tensor = D0sDataType::Size()

staticconstexpr

◆ NumD1Tensor

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 D0sDataType, typename D1sDataType, typename GemmAccDataType, typename CShuffleDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename C0DEElementwiseOperation, typename B1ElementwiseOperation, typename C1DEElementwiseOperation, 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, int D0sTransferSrcScalarPerVector = 4, LoopScheduler LoopSched = LoopScheduler::Default>

index_t ck::tensor_operation::device::DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle< NumDimG, NumDimM, NumDimN, NumDimK, NumDimO, ADataType, BDataType, B1DataType, CDataType, D0sDataType, D1sDataType, GemmAccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, C0DEElementwiseOperation, B1ElementwiseOperation, C1DEElementwiseOperation, 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, D0sTransferSrcScalarPerVector, LoopSched >::NumD1Tensor = D1sDataType::Size()

staticconstexpr

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

device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp

Classes

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Member Typedef Documentation

◆ AGridDesc_AK0_M_AK1

◆ AGridDesc_G_M_K

◆ B1GridDesc_BK0_N_BK1

◆ B1GridDesc_G_N_K

◆ BGridDesc_BK0_N_BK1

◆ BGridDesc_G_N_K

◆ C0MatrixMask

◆ C1GridDesc_G_M_N

◆ C1GridDesc_M_N

◆ D0sGridDesc_G_M_N

◆ D0sGridDesc_M_N

◆ DeviceOp

◆ GridwiseGemm32

◆ GridwiseGemm64

◆ GridwiseGemmBase

◆ Transform

Member Function Documentation

◆ GetTypeString()

◆ IsSupportedArgument() [1/2]

◆ IsSupportedArgument() [2/2]

◆ IsValidCompilationParameter()

◆ make_MaskOutPredicate()

◆ MakeAGridDescriptor_AK0_M_AK1()

◆ MakeArgument()

◆ MakeArgumentPointer()

◆ MakeB1GridDescriptor_BK0_N_BK1()

◆ MakeBGridDescriptor_BK0_N_BK1()

◆ MakeD0sGridDescriptor_G_M_N()

◆ MakeD0sGridDescriptor_M_N()

◆ MakeInvoker()

◆ MakeInvokerPointer()

Member Data Documentation

◆ I0

◆ I1

◆ I2

◆ MXdlPerWave32

◆ MXdlPerWave64

◆ NumD0Tensor

◆ NumD1Tensor