GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer > Struct Template Reference

GridwiseGemm_wmma_cshuffle_v3_base&lt; ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer &gt; Struct Template Reference#

Composable Kernel: ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer > Struct Template Reference
Public Types | Static Public Member Functions | Static Public Attributes | List of all members
ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer > Struct Template Reference

#include <gridwise_gemm_wmma_cshuffle_v3_common.hpp>

Inheritance diagram for ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >:
ck::GridwiseGemm_wmma_cshuffle_v3< ALayout, BLayout, Tuple<>, CLayout, Tuple< ADataType >, Tuple< BDataType >, AccDataType, CShuffleDataType, Tuple<>, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, false, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, false, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, Sequence< CShuffleBlockTransferScalarPerVector_NPerBlock >, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, false, false > ck::GridwiseGemm_wmma_cshuffle_v3< ALayout, BLayout, DsLayout, HLayout, Tuple< ADataType >, Tuple< BDataType >, AccDataType, CShuffleDataType, DsDataType, EMeanVarDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, false, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, false, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB > ck::GridwiseGemm_wmma_cshuffle_v3< ALayout, BLayout, DsLayout, ELayout, Tuple< ADataType >, Tuple< BDataType >, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, false, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, false, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB > ck::GridwiseGemm_wmma_cshuffle_v3< ALayout, BLayout, Tuple<>, CLayout, Tuple< ADataType >, Tuple< BDataType >, AccDataType, CShuffleDataType, Tuple<>, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, false, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, false, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, Sequence< CShuffleBlockTransferScalarPerVector_NPerBlock >, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB > ck::GridwiseGemm_wmma_cshuffle_v3< ALayout, BLayout, Tuple<>, CLayout, Tuple< ADataType >, Tuple< BDataType >, GemmAccDataType, ReduceDataType, Tuple<>, ReduceDataType, AElementwiseOperation, BElementwiseOperation, PassThrough, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, false, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, false, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, Sequence< CShuffleBlockTransferScalarPerVector_NPerBlock >, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, false, false > ck::GridwiseGemm_wmma_cshuffle_v3_b_scale< ALayout, BLayout, Tuple<>, CLayout, Tuple< ADataType >, Tuple< BDataType >, BScaleDataType, AccDataType, CShuffleDataType, Tuple<>, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, BlockSize, ScaleBlockN, ScaleBlockK, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, false, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, false, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, Sequence< CShuffleBlockTransferScalarPerVector_NPerBlock >, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >

Public Types

using LDSTypeA
using LDSTypeB
using ThisThreadBlock = ThisThreadBlock<BlockSize>
using ATransfer
using BTransfer
using Block2CTileMap = BlockToCTileMap_Grouped_M00_N0_M01Adapt<8, MPerBlock, NPerBlock>
using AsGridPointer = decltype(MakeAsGridPointer())
using BsGridPointer = decltype(MakeBsGridPointer())
using DsGridPointer = decltype(MakeDsGridPointer())
using BlockwiseGemmPipe
using EpilogueCShuffle
using EpilogueWelfordCShuffle

Static Public Member Functions

static __host__ auto CalculateGridSize (index_t M, index_t N, index_t KBatch)
static __host__ auto CalculateMPadded (index_t M)
static __host__ auto CalculateNPadded (index_t N)
static __host__ auto CalculateKPadded (index_t K)
static __host__ auto CalculateAK0Padded (index_t K, index_t K_Batch=1)
static __host__ auto CalculateBK0Padded (index_t K, index_t K_Batch=1)
static __host__ auto CalculateKPadded (index_t K, index_t K_Batch=1)
static __host__ auto CalculateKRead (index_t K, index_t K_Batch=1)
static __host__ auto CalculateMBlock (index_t M)
static __host__ auto CalculateNBlock (index_t N)
static constexpr auto MakeAsGridPointer ()
static constexpr auto MakeBsGridPointer ()
__host__ static __device__ auto MakeAGridDescriptor_M_K (index_t M, index_t K, index_t StrideA)
__host__ static __device__ auto MakeBGridDescriptor_N_K (index_t N, index_t K, index_t StrideB)
__host__ static __device__ auto MakeAsGridDescriptor_AK0_M_AK1 (const index_t M, const index_t MPad, const index_t K, const index_t KPad, const std::array< index_t, NumATensor > &StrideAs, const index_t AK0)
__host__ static __device__ auto MakeBsGridDescriptor_BK0_N_BK1 (const index_t K, const index_t KPad, const index_t N, const index_t NPad, const std::array< index_t, NumBTensor > &StrideBs, const index_t BK0)
__host__ static __device__ constexpr auto MakeAWmmaTileDescriptor ()
__host__ static __device__ constexpr auto MakeBWmmaTileDescriptor ()
template<typename DELayout>
__host__ static __device__ auto MakeDEGridDescriptor_M_N (index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideDE)
static constexpr auto MakeDsGridPointer ()
__host__ static __device__ auto MakeDsGridDescriptor_M_N (index_t M, index_t MPad, index_t N, index_t NPad, std::array< index_t, NumDTensor > StrideDs)
template<typename DsGridDesc>
static __device__ constexpr auto MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock (const DsGridDesc &ds_grid_desc_m_n, index_t MBlock, index_t NBlock)
template<typename DEGridDesc>
static __device__ constexpr auto MakeDEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock (const DEGridDesc &de_grid_desc_m_n, index_t MBlock, index_t NBlock)
template<typename Argument>
static __host__ constexpr bool CheckValidity (const Argument &karg)
static __host__ constexpr bool CalculateHasMainKBlockLoop (index_t K)
static __host__ constexpr TailNumber CalculateKBlockLoopTailNum (index_t K)
template<typename EpilogueType>
static __device__ constexpr index_t GetSharedMemoryNumberOfByte ()
template<index_t numElements, typename Type>
__device__ static __forceinline__ auto get_first_element_workaround (Type &array)
template<typename AGridDesc_AK0_M_K1, typename BGridDesc_BK0_N_K1, typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, typename BScaleStruct, typename EpilogueArgument, bool HasMainKBlockLoop, InMemoryDataOperationEnum EGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>
static __device__ void Run (AsGridPointer p_as_grid, BsGridPointer p_bs_grid, DsGridPointer p_ds_grid, EDataType *p_e_grid, void *p_shared, const AGridDesc_AK0_M_K1 &as_grid_desc_ak0_m_ak1, const BGridDesc_BK0_N_K1 &bs_grid_desc_bk0_n_bk1, const DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock &ds_grid_desc_mblock_mperblock_nblock_nperblock, const EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock &e_grid_desc_mblock_mperblock_nblock_nperblock, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, CDEElementwiseOperation cde_element_op, const index_t &block_m_id, const index_t &block_n_id, const index_t &num_k_block_per_scale, BScaleStruct &b_scale_struct, EpilogueArgument &epilogue_args)

Static Public Attributes

static constexpr auto I0 = Number<0>{}
static constexpr auto I1 = Number<1>{}
static constexpr auto I2 = Number<2>{}
static constexpr auto I3 = Number<3>{}
static constexpr auto I4 = Number<4>{}
static constexpr auto I5 = Number<5>{}
static constexpr auto I6 = Number<6>{}
static constexpr auto I7 = Number<7>{}
static constexpr index_t NumATensor = AsDataType::Size()
static constexpr index_t NumBTensor = BsDataType::Size()
static constexpr auto EShuffleBlockTransferScalarPerVector
static constexpr auto AK0Number = Number<KPerBlock / AK1Value>{}
static constexpr auto BK0Number = Number<KPerBlock / BK1Value>{}
static constexpr auto AK1Number = Number<AK1Value>{}
static constexpr auto BK1Number = Number<BK1Value>{}
static constexpr index_t KPack
static constexpr index_t APackedSize
static constexpr index_t BPackedSize
static constexpr bool IsAWaveTransferApplicable = false
static constexpr bool IsBWaveTransferApplicable = false
static constexpr index_t WaveSize
static constexpr bool UseBlockPaddingA
static constexpr bool UseBlockPaddingB
static constexpr index_t NumDTensor = DsDataType::Size()

Member Typedef Documentation

◆ AsGridPointer

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::AsGridPointer = decltype(MakeAsGridPointer())

◆ ATransfer

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::ATransfer
Initial value:
typename std::conditional<
IsAWaveTransferApplicable,
ABTransferWaveTiles<ALayout,
tensor_layout::gemm::RowMajor,
LDSTypeA,
BlockSize,
MPerBlock,
KPerBlock,
MPerWmma,
KPack,
AK1Value,
WaveSize>,
ABTransferThreadTiles<ALayout,
tensor_layout::gemm::RowMajor,
LDSTypeA,
BlockSize,
MPerBlock,
KPerBlock,
MPerWmma,
AK1Value,
UseBlockPaddingA,
PermuteA,
ABlockTransferThreadClusterLengths_AK0_M_AK1,
ABlockTransferThreadClusterArrangeOrder,
ABlockTransferSrcAccessOrder,
ABlockTransferSrcVectorDim,
ABlockTransferSrcScalarPerVector,
ABlockTransferDstScalarPerVector_AK1,
AThreadTransferSrcResetCoordinateAfterRun>>::type
ck::ABTransferThreadTiles
Definition gridwise_ab_transfer_thread_tiles.hpp:30
ck::ABTransferWaveTiles
Definition gridwise_ab_transfer_wave_tiles.hpp:23
ck::GridwiseGemm_wmma_cshuffle_v3_base::IsAWaveTransferApplicable
static constexpr bool IsAWaveTransferApplicable
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:192
ck::GridwiseGemm_wmma_cshuffle_v3_base::WaveSize
static constexpr index_t WaveSize
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:196
ck::GridwiseGemm_wmma_cshuffle_v3_base::LDSTypeA
typename std::conditional<(NumATensor > 1), ComputeTypeA, remove_cvref_t< tuple_element_t< 0, AsDataType > > >::type LDSTypeA
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:136
ck::GridwiseGemm_wmma_cshuffle_v3_base::KPack
static constexpr index_t KPack
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:154
ck::GridwiseGemm_wmma_cshuffle_v3_base::UseBlockPaddingA
static constexpr bool UseBlockPaddingA
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:199
ck::tensor_layout::gemm::RowMajor
Definition tensor_operation/gpu/device/tensor_layout.hpp:26

◆ Block2CTileMap

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::Block2CTileMap = BlockToCTileMap_Grouped_M00_N0_M01Adapt<8, MPerBlock, NPerBlock>

◆ BlockwiseGemmPipe

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::BlockwiseGemmPipe
Initial value:
remove_cvref_t<decltype(BlockGemmPipeline_Selector<BlkGemmPipelineVer,
BlkGemmPipeSched,
BlockSize,
LDSTypeA,
LDSTypeB,
ComputeTypeA,
ComputeTypeB,
AccDataType,
decltype(MakeAWmmaTileDescriptor()),
decltype(MakeBWmmaTileDescriptor()),
ABlockTransferSrcScalarPerVector,
BBlockTransferSrcScalarPerVector,
MPerBlock,
NPerBlock,
KPerBlock,
MPerWmma,
NPerWmma,
MRepeat,
NRepeat,
KPack>())>
ck::remove_cvref_t
remove_cv_t< remove_reference_t< T > > remove_cvref_t
Definition type.hpp:297
ck::BlockGemmPipeline_Selector
constexpr auto BlockGemmPipeline_Selector()
Definition blockwise_gemm_pipeline_wmma_selector.hpp:32
ck::GridwiseBatchedGemmGemm_wmma_cshuffle_v3::KPack
static constexpr auto KPack
Definition gridwise_batched_gemm_gemm_wmma_cshuffle_v3.hpp:113
ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeBWmmaTileDescriptor
__host__ static __device__ constexpr auto MakeBWmmaTileDescriptor()
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:437
ck::GridwiseGemm_wmma_cshuffle_v3_base::LDSTypeB
typename std::conditional<(NumBTensor > 1), ComputeTypeB, remove_cvref_t< tuple_element_t< 0, BsDataType > > >::type LDSTypeB
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:140
ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeAWmmaTileDescriptor
__host__ static __device__ constexpr auto MakeAWmmaTileDescriptor()
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:430

◆ BsGridPointer

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::BsGridPointer = decltype(MakeBsGridPointer())

◆ BTransfer

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::BTransfer
Initial value:
typename std::conditional<
IsBWaveTransferApplicable,
ABTransferWaveTiles<BLayout,
tensor_layout::gemm::ColumnMajor,
LDSTypeB,
BlockSize,
NPerBlock,
KPerBlock,
NPerWmma,
KPack,
BK1Value,
WaveSize>,
ABTransferThreadTiles<BLayout,
tensor_layout::gemm::ColumnMajor,
LDSTypeB,
BlockSize,
NPerBlock,
KPerBlock,
NPerWmma,
BK1Value,
UseBlockPaddingB,
PermuteB,
BBlockTransferThreadClusterLengths_BK0_N_BK1,
BBlockTransferThreadClusterArrangeOrder,
BBlockTransferSrcAccessOrder,
BBlockTransferSrcVectorDim,
BBlockTransferSrcScalarPerVector,
BBlockTransferDstScalarPerVector_BK1,
BThreadTransferSrcResetCoordinateAfterRun>>::type
ck::GridwiseGemm_wmma_cshuffle_v3_base::IsBWaveTransferApplicable
static constexpr bool IsBWaveTransferApplicable
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:193
ck::GridwiseGemm_wmma_cshuffle_v3_base::UseBlockPaddingB
static constexpr bool UseBlockPaddingB
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:231
ck::tensor_layout::gemm::ColumnMajor
Definition tensor_operation/gpu/device/tensor_layout.hpp:31

◆ DsGridPointer

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::DsGridPointer = decltype(MakeDsGridPointer())

◆ EpilogueCShuffle

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::EpilogueCShuffle
Initial value:
EpilogueCShuffle<DsDataType,
EDataType,
AccDataType,
CShuffleDataType,
MPerBlock,
NPerBlock,
MPerWmma,
NPerWmma,
MRepeat,
NRepeat,
CShuffleMRepeatPerShuffle,
CShuffleNRepeatPerShuffle,
CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
CDEShuffleBlockTransferScalarPerVectors,
CDEElementwiseOperation,
ThisThreadBlock,
BlockwiseGemmPipe>
ck::GridwiseGemm_wmma_cshuffle_v3_base::ThisThreadBlock
ThisThreadBlock< BlockSize > ThisThreadBlock
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:159
ck::GridwiseGemm_wmma_cshuffle_v3_base::BlockwiseGemmPipe
remove_cvref_t< decltype(BlockGemmPipeline_Selector< BlkGemmPipelineVer, BlkGemmPipeSched, BlockSize, LDSTypeA, LDSTypeB, ComputeTypeA, ComputeTypeB, AccDataType, decltype(MakeAWmmaTileDescriptor()), decltype(MakeBWmmaTileDescriptor()), ABlockTransferSrcScalarPerVector, BBlockTransferSrcScalarPerVector, MPerBlock, NPerBlock, KPerBlock, MPerWmma, NPerWmma, MRepeat, NRepeat, KPack >())> BlockwiseGemmPipe
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:546
ck::GridwiseGemm_wmma_cshuffle_v3_base::EpilogueCShuffle
EpilogueCShuffle< DsDataType, EDataType, AccDataType, CShuffleDataType, MPerBlock, NPerBlock, MPerWmma, NPerWmma, MRepeat, NRepeat, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, CDEElementwiseOperation, ThisThreadBlock, BlockwiseGemmPipe > EpilogueCShuffle
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:569

◆ EpilogueWelfordCShuffle

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::EpilogueWelfordCShuffle
Initial value:
EpilogueWelfordCShuffle<
DsDataType,
EDataType,
AccDataType,
CShuffleDataType,
MPerBlock,
NPerBlock,
MPerWmma,
NPerWmma,
MRepeat,
NRepeat,
CShuffleMRepeatPerShuffle,
CShuffleNRepeatPerShuffle,
CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
CDEShuffleBlockTransferScalarPerVectors,
CDEElementwiseOperation,
ThisThreadBlock,
BlockwiseGemmPipe,
BlockSize>
ck::GridwiseGemm_wmma_cshuffle_v3_base::EpilogueWelfordCShuffle
EpilogueWelfordCShuffle< DsDataType, EDataType, AccDataType, CShuffleDataType, MPerBlock, NPerBlock, MPerWmma, NPerWmma, MRepeat, NRepeat, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, CDEElementwiseOperation, ThisThreadBlock, BlockwiseGemmPipe, BlockSize > EpilogueWelfordCShuffle
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:588

◆ LDSTypeA

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::LDSTypeA
Initial value:
typename std::conditional<(NumATensor > 1),
ComputeTypeA,
remove_cvref_t<tuple_element_t<0, AsDataType>>>::type
ck::tuple_element_t
typename tuple_element< I, TTuple >::type tuple_element_t
Definition utility/tuple.hpp:208
ck::GridwiseGemm_wmma_cshuffle_v3_base::NumATensor
static constexpr index_t NumATensor
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:133

◆ LDSTypeB

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::LDSTypeB
Initial value:
typename std::conditional<(NumBTensor > 1),
ComputeTypeB,
remove_cvref_t<tuple_element_t<0, BsDataType>>>::type
ck::GridwiseGemm_wmma_cshuffle_v3_base::NumBTensor
static constexpr index_t NumBTensor
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:134

◆ ThisThreadBlock

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
using ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::ThisThreadBlock = ThisThreadBlock<BlockSize>

Member Function Documentation

◆ CalculateAK0Padded()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateAK0Padded ( index_t K,
index_t K_Batch = 1 )
inlinestatic

◆ CalculateBK0Padded()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateBK0Padded ( index_t K,
index_t K_Batch = 1 )
inlinestatic

◆ CalculateGridSize()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateGridSize ( index_t M,
index_t N,
index_t KBatch )
inlinestatic

◆ CalculateHasMainKBlockLoop()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ constexpr bool ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateHasMainKBlockLoop ( index_t K)
inlinestaticconstexpr

◆ CalculateKBlockLoopTailNum()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ constexpr TailNumber ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateKBlockLoopTailNum ( index_t K)
inlinestaticconstexpr

◆ CalculateKPadded() [1/2]

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateKPadded ( index_t K)
inlinestatic

◆ CalculateKPadded() [2/2]

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateKPadded ( index_t K,
index_t K_Batch = 1 )
inlinestatic

◆ CalculateKRead()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateKRead ( index_t K,
index_t K_Batch = 1 )
inlinestatic

◆ CalculateMBlock()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateMBlock ( index_t M)
inlinestatic

◆ CalculateMPadded()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateMPadded ( index_t M)
inlinestatic

◆ CalculateNBlock()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateNBlock ( index_t N)
inlinestatic

◆ CalculateNPadded()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CalculateNPadded ( index_t N)
inlinestatic

◆ CheckValidity()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<typename Argument>
__host__ constexpr bool ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::CheckValidity ( const Argument & karg)
inlinestaticconstexpr

◆ get_first_element_workaround()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<index_t numElements, typename Type>
__device__ static __forceinline__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::get_first_element_workaround ( Type & array)
inlinestatic

◆ GetSharedMemoryNumberOfByte()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<typename EpilogueType>
__device__ constexpr index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::GetSharedMemoryNumberOfByte ( )
inlinestaticconstexpr

◆ MakeAGridDescriptor_M_K()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeAGridDescriptor_M_K ( index_t M,
index_t K,
index_t StrideA )
inlinestatic

◆ MakeAsGridDescriptor_AK0_M_AK1()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeAsGridDescriptor_AK0_M_AK1 ( const index_t M,
const index_t MPad,
const index_t K,
const index_t KPad,
const std::array< index_t, NumATensor > & StrideAs,
const index_t AK0 )
inlinestatic

◆ MakeAsGridPointer()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeAsGridPointer ( )
inlinestaticconstexpr

◆ MakeAWmmaTileDescriptor()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeAWmmaTileDescriptor ( )
inlinestaticconstexpr

◆ MakeBGridDescriptor_N_K()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeBGridDescriptor_N_K ( index_t N,
index_t K,
index_t StrideB )
inlinestatic

◆ MakeBsGridDescriptor_BK0_N_BK1()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeBsGridDescriptor_BK0_N_BK1 ( const index_t K,
const index_t KPad,
const index_t N,
const index_t NPad,
const std::array< index_t, NumBTensor > & StrideBs,
const index_t BK0 )
inlinestatic

◆ MakeBsGridPointer()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeBsGridPointer ( )
inlinestaticconstexpr

◆ MakeBWmmaTileDescriptor()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeBWmmaTileDescriptor ( )
inlinestaticconstexpr

◆ MakeDEGridDescriptor_M_N()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<typename DELayout>
__host__ static __device__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeDEGridDescriptor_M_N ( index_t M,
index_t MPad,
index_t N,
index_t NPad,
index_t StrideDE )
inlinestatic

◆ MakeDEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<typename DEGridDesc>
__device__ constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeDEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock ( const DEGridDesc & de_grid_desc_m_n,
index_t MBlock,
index_t NBlock )
inlinestaticconstexpr

◆ MakeDsGridDescriptor_M_N()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
__host__ static __device__ auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeDsGridDescriptor_M_N ( index_t M,
index_t MPad,
index_t N,
index_t NPad,
std::array< index_t, NumDTensor > StrideDs )
inlinestatic

◆ MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<typename DsGridDesc>
__device__ constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock ( const DsGridDesc & ds_grid_desc_m_n,
index_t MBlock,
index_t NBlock )
inlinestaticconstexpr

◆ MakeDsGridPointer()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
constexpr auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::MakeDsGridPointer ( )
inlinestaticconstexpr

◆ Run()

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
template<typename AGridDesc_AK0_M_K1, typename BGridDesc_BK0_N_K1, typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, typename BScaleStruct, typename EpilogueArgument, bool HasMainKBlockLoop, InMemoryDataOperationEnum EGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>
__device__ void ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::Run ( AsGridPointer p_as_grid,
BsGridPointer p_bs_grid,
DsGridPointer p_ds_grid,
EDataType * p_e_grid,
void * p_shared,
const AGridDesc_AK0_M_K1 & as_grid_desc_ak0_m_ak1,
const BGridDesc_BK0_N_K1 & bs_grid_desc_bk0_n_bk1,
const DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock & ds_grid_desc_mblock_mperblock_nblock_nperblock,
const EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock & e_grid_desc_mblock_mperblock_nblock_nperblock,
AElementwiseOperation a_element_op,
BElementwiseOperation b_element_op,
CDEElementwiseOperation cde_element_op,
const index_t & block_m_id,
const index_t & block_n_id,
const index_t & num_k_block_per_scale,
BScaleStruct & b_scale_struct,
EpilogueArgument & epilogue_args )
inlinestatic

Member Data Documentation

◆ AK0Number

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::AK0Number = Number<KPerBlock / AK1Value>{}
staticconstexpr

◆ AK1Number

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::AK1Number = Number<AK1Value>{}
staticconstexpr

◆ APackedSize

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::APackedSize
staticconstexpr
Initial value:
= []() {
if constexpr(is_same_v<remove_cvref_t<LDSTypeA>, pk_i4_t>)
return 2;
else
return 1;
}()
ck::is_same_v
constexpr bool is_same_v
Definition type.hpp:283
ck::pk_i4_t
Definition data_type.hpp:187

◆ BK0Number

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::BK0Number = Number<KPerBlock / BK1Value>{}
staticconstexpr

◆ BK1Number

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::BK1Number = Number<BK1Value>{}
staticconstexpr

◆ BPackedSize

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::BPackedSize
staticconstexpr
Initial value:
= []() {
if constexpr(is_same_v<remove_cvref_t<LDSTypeB>, pk_i4_t>)
return 2;
else
return 1;
}()

◆ EShuffleBlockTransferScalarPerVector

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::EShuffleBlockTransferScalarPerVector
staticconstexpr
Initial value:
=
CDEShuffleBlockTransferScalarPerVectors{}[I0]
ck::GridwiseGemm_wmma_cshuffle_v3_base::I0
static constexpr auto I0
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:124

◆ I0

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I0 = Number<0>{}
staticconstexpr

◆ I1

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I1 = Number<1>{}
staticconstexpr

◆ I2

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I2 = Number<2>{}
staticconstexpr

◆ I3

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I3 = Number<3>{}
staticconstexpr

◆ I4

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I4 = Number<4>{}
staticconstexpr

◆ I5

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I5 = Number<5>{}
staticconstexpr

◆ I6

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I6 = Number<6>{}
staticconstexpr

◆ I7

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
auto ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::I7 = Number<7>{}
staticconstexpr

◆ IsAWaveTransferApplicable

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
bool ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::IsAWaveTransferApplicable = false
staticconstexpr

◆ IsBWaveTransferApplicable

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
bool ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::IsBWaveTransferApplicable = false
staticconstexpr

◆ KPack

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::KPack
staticconstexpr
Initial value:
= math::max(
math::lcm(AK1Number, BK1Number),
WmmaSelector<ComputeTypeA, ComputeTypeB, AccDataType, MPerWmma, NPerWmma>::selected_wmma
.k_per_wmma)
ck::math::max
__host__ __device__ constexpr T max(T x)
Definition utility/math.hpp:84
ck::math::lcm
__host__ __device__ constexpr auto lcm(X x, Y y)
Definition utility/math.hpp:198
ck::GridwiseGemm_wmma_cshuffle_v3_base::AK1Number
static constexpr auto AK1Number
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:151
ck::GridwiseGemm_wmma_cshuffle_v3_base::BK1Number
static constexpr auto BK1Number
Definition gridwise_gemm_wmma_cshuffle_v3_common.hpp:152
ck::WmmaSelector::selected_wmma
static constexpr auto selected_wmma
Definition wmma_gemm.hpp:636

◆ NumATensor

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::NumATensor = AsDataType::Size()
staticconstexpr

◆ NumBTensor

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::NumBTensor = BsDataType::Size()
staticconstexpr

◆ NumDTensor

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::NumDTensor = DsDataType::Size()
staticconstexpr

◆ UseBlockPaddingA

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
bool ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::UseBlockPaddingA
staticconstexpr
Initial value:
=
ABlockLdsExtraM || BlkGemmPipelineVer == BlockGemmPipelineVersion::v4
ck::BlockGemmPipelineVersion::v4
@ v4
Definition blkgemmpipe_scheduler.hpp:17

◆ UseBlockPaddingB

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
bool ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::UseBlockPaddingB
staticconstexpr
Initial value:
=
BBlockLdsExtraN || BlkGemmPipelineVer == BlockGemmPipelineVersion::v4

◆ WaveSize

template<typename ALayout, typename BLayout, typename DsLayout, typename ELayout, typename AsDataType, typename BsDataType, typename AccDataType, typename CShuffleDataType, typename DsDataType, typename EDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CDEElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerWmma, index_t NPerWmma, index_t MRepeat, index_t NRepeat, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool AThreadTransferSrcResetCoordinateAfterRun, index_t ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BThreadTransferSrcResetCoordinateAfterRun, index_t BBlockLdsExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, typename CDEShuffleBlockTransferScalarPerVectors, BlockGemmPipelineScheduler BlkGemmPipeSched, BlockGemmPipelineVersion BlkGemmPipelineVer, typename ComputeTypeA, typename ComputeTypeB, bool PermuteA, bool PermuteB, bool ForceThreadTileTransfer = false>
index_t ck::GridwiseGemm_wmma_cshuffle_v3_base< ALayout, BLayout, DsLayout, ELayout, AsDataType, BsDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CDEShuffleBlockTransferScalarPerVectors, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB, ForceThreadTileTransfer >::WaveSize
staticconstexpr
Initial value:
=
WmmaSelector<ComputeTypeA, ComputeTypeB, AccDataType, MPerWmma, NPerWmma>::selected_wmma
.wave_size
The documentation for this struct was generated from the following file: