GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB > Struct Template Reference

• ck
• GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle

Classes | Public Types | Static Public Member Functions | Static Public Attributes | List of all members

ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB > Struct Template Reference

#include <gridwise_gemm_xdl_cshuffle_v3_mx_bpreshuffle.hpp>

Classes
struct	Problem
struct	Argument
struct	SplitKBatchOffset

Public Types
using	ThisThreadBlock = ThisThreadBlock<BlockSize>
using	mx_scale_t = e8m0_bexp_t
using	BlockwiseGemmPipe
using	Block2CTileMap = BlockToCTileMap_Grouped_M00_N0_M01Adapt<8, MPerBlock, NPerBlock>

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)
__host__ static __device__ auto	CalculateBN0Shuffled (index_t N)
__host__ static __device__ auto	CalculateBK0Shuffled (index_t K)
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)
template<index_t MNXdlPerWave, index_t MNWaves, index_t MNXdlPack, index_t MNPerXdl, bool IsXor, typename TileDesc_K0_MN_K1>
__host__ static __device__ constexpr auto	MakeGemmMmaTileDescriptor (const TileDesc_K0_MN_K1 &)
__host__ static __device__ auto	MakeAGridDescriptor_AK0_M_AK1 (index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA, index_t AK0)
__host__ static __device__ auto	MakeBGridDescriptor_Preshuffled (index_t N0, index_t K0)
__host__ static __device__ auto	MakeBGridDescriptor_BK0_N_BK1 (index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB, index_t BK0)
template<typename ABlockDesc_AK0_M_AK1>
__host__ static __device__ constexpr auto	MakeAMmaTileDescriptor_M0_M1_M2_M3_K (const ABlockDesc_AK0_M_AK1 &)
template<typename BBlockDesc_BK0_N_BK1>
__host__ static __device__ constexpr auto	MakeBMmaTileDescriptor_N0_N1_N2_N3_K (const BBlockDesc_BK0_N_BK1 &)
__host__ static __device__ auto	MakeCGridDescriptor_M_N (index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC)
static __device__ constexpr auto	GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1 ()
static __device__ constexpr auto	GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1 ()
static __device__ constexpr auto	GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock ()
static __device__ constexpr index_t	GetSharedMemoryNumberOfByte ()
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 CGridDesc>
__host__ static __device__ constexpr auto	MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock (const CGridDesc &c_grid_desc_m_n, index_t MBlock, index_t NBlock)
template<typename AGridDesc_AK0_M_K1, typename AScaleGridDesc_AM_AK, typename BGridDesc_BK0_N_K1, typename BScaleGridDesc_BN_AK, typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>
static __device__ void	Run (const ADataType *p_a_grid, const AScaleDataType *p_a_scale_grid, const BDataType *p_b_grid, const BScaleDataType *p_b_scale_grid, CDataType *p_c_grid, void *p_shared, const Problem &problem, const AGridDesc_AK0_M_K1 &a_grid_desc_ak0_m_ak1, const AScaleGridDesc_AM_AK &a_scale_grid_desc_am_ak, const BGridDesc_BK0_N_K1 &b_grid_desc_bk0_n_bk1, const BScaleGridDesc_BN_AK &b_scale_grid_desc_bn_ak, const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock &c_grid_desc_mblock_mperblock_nblock_nperblock)
template<bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>
static __device__ void	Run (const ADataType *p_a_grid, const AScaleDataType *p_a_scale_grid, const BDataType *p_b_grid, const BScaleDataType *p_b_scale_grid, CDataType *p_c_grid, void *p_shared, const Problem &problem)
template<typename AGridDesc_AK0_M_K1, typename AScaleGridDesc_AM_AK, typename BGridDesc_BK0_N_K1, typename BScaleGridDesc_BN_AK, typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>
static __device__ void	Run_2Lds (const ADataType *p_a_grid, const AScaleDataType *p_a_scale_grid, const BDataType *p_b_grid, const BScaleDataType *p_b_scale_grid, CDataType *p_c_grid, void *p_shared_0, void *p_shared_1, const Problem &problem, const AGridDesc_AK0_M_K1 &a_grid_desc_ak0_m_ak1, const AScaleGridDesc_AM_AK &a_scale_grid_desc_am_ak, const BGridDesc_BK0_N_K1 &b_grid_desc_bk0_n_bk1, const BScaleGridDesc_BN_AK &b_scale_grid_desc_bn_ak, const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock &c_grid_desc_mblock_mperblock_nblock_nperblock)
template<bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>
static __device__ void	Run_2Lds (const ADataType *p_a_grid, const AScaleDataType *p_a_scale_grid, const BDataType *p_b_grid, const BScaleDataType *p_b_scale_grid, CDataType *p_c_grid, void *p_shared_0, void *p_shared_1, const Problem &problem)

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 auto	I8 = Number<8>{}
static constexpr auto	I9 = Number<9>{}
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 auto	lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number)
static constexpr bool	is_single_rate_mfma = false
static constexpr auto	is_scale_mfma = true
static constexpr auto	MXdlPack = 2
static constexpr auto	NXdlPack = 2
static constexpr auto	KXdlPack = 2
static constexpr index_t	APackedSize = packed_size_v<ADataType>
static constexpr index_t	BPackedSize = packed_size_v<BDataType>
static constexpr index_t	KPack
static constexpr index_t	NLane = NPerXdl
static constexpr index_t	KLane = 64 / NLane
static constexpr index_t	NWave = NPerBlock / NPerXdl / NXdlPerWave
static constexpr index_t	KRepeat = KPerBlock / KLane / KPack
static constexpr index_t	scale_pack_size_a = sizeof(AScaleDataType) / sizeof(mx_scale_t)
static constexpr index_t	scale_pack_size_b = sizeof(BScaleDataType) / sizeof(mx_scale_t)

Member Typedef Documentation

◆ Block2CTileMap

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

using ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::Block2CTileMap = BlockToCTileMap_Grouped_M00_N0_M01Adapt<8, MPerBlock, NPerBlock>

◆ BlockwiseGemmPipe

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

using ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::BlockwiseGemmPipe

Initial value:

remove_cvref_t<decltype(BlockGemmMXBPreshufflePipeline_Selector<

BlkGemmPipelineVer,

BlkGemmPipeSched,

BlockSize,

ScaleBlockSize,

ADataType,

AScaleDataType,

BDataType,

BScaleDataType,

ComputeTypeA,

AccDataType,

decltype(GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()),

decltype(GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()),

decltype(MakeAMmaTileDescriptor_M0_M1_M2_M3_K(

GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1())),

decltype(MakeBMmaTileDescriptor_N0_N1_N2_N3_K(

GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1())),

ABlockTransferSrcScalarPerVector,

BBlockTransferSrcScalarPerVector,

MPerBlock,

NPerBlock,

KPerBlock,

MPerXdl,

NPerXdl,

MXdlPerWave,

NXdlPerWave,

KPack>())>

ck::remove_cvref_t

remove_cv_t< remove_reference_t< T > > remove_cvref_t

Definition type.hpp:297

ck::BlockGemmMXBPreshufflePipeline_Selector

constexpr auto BlockGemmMXBPreshufflePipeline_Selector()

Definition blockwise_gemm_pipeline_xdlops_b_preshuffle_mx_moe_selector.hpp:37

ck::GridwiseBatchedGemmGemm_wmma_cshuffle_v3::KPack

static constexpr auto KPack

Definition gridwise_batched_gemm_gemm_wmma_cshuffle_v3.hpp:113

ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle::MakeAMmaTileDescriptor_M0_M1_M2_M3_K

__host__ static __device__ constexpr auto MakeAMmaTileDescriptor_M0_M1_M2_M3_K(const ABlockDesc_AK0_M_AK1 &)

Definition gridwise_gemm_xdl_cshuffle_v3_mx_bpreshuffle.hpp:569

ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle::GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1

static __device__ constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()

Definition gridwise_gemm_xdl_cshuffle_v3_mx_bpreshuffle.hpp:958

ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle::GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1

static __device__ constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()

Definition gridwise_gemm_xdl_cshuffle_v3_mx_bpreshuffle.hpp:838

ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle::MakeBMmaTileDescriptor_N0_N1_N2_N3_K

__host__ static __device__ constexpr auto MakeBMmaTileDescriptor_N0_N1_N2_N3_K(const BBlockDesc_BK0_N_BK1 &)

Definition gridwise_gemm_xdl_cshuffle_v3_mx_bpreshuffle.hpp:579

◆ mx_scale_t

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

using ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::mx_scale_t = e8m0_bexp_t

◆ ThisThreadBlock

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

using ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::ThisThreadBlock = ThisThreadBlock<BlockSize>

Member Function Documentation

◆ CalculateAK0Padded()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateAK0Padded ( index_t K, index_t K_Batch = 1 )

inlinestatic

◆ CalculateBK0Padded()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateBK0Padded ( index_t K, index_t K_Batch = 1 )

inlinestatic

◆ CalculateBK0Shuffled()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ static __device__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateBK0Shuffled ( index_t K )

inlinestatic

◆ CalculateBN0Shuffled()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ static __device__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateBN0Shuffled ( index_t N )

inlinestatic

◆ CalculateGridSize()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateGridSize ( index_t M, index_t N, index_t KBatch )

inlinestatic

◆ CalculateHasMainKBlockLoop()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ constexpr bool ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateHasMainKBlockLoop ( index_t K )

inlinestaticconstexpr

◆ CalculateKBlockLoopTailNum()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ constexpr TailNumber ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateKBlockLoopTailNum ( index_t K )

inlinestaticconstexpr

◆ CalculateKPadded() [1/2]

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateKPadded ( index_t K )

inlinestatic

◆ CalculateKPadded() [2/2]

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateKPadded ( index_t K, index_t K_Batch = 1 )

inlinestatic

◆ CalculateKRead()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateKRead ( index_t K, index_t K_Batch = 1 )

inlinestatic

◆ CalculateMBlock()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateMBlock ( index_t M )

inlinestatic

◆ CalculateMPadded()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateMPadded ( index_t M )

inlinestatic

◆ CalculateNBlock()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateNBlock ( index_t N )

inlinestatic

◆ CalculateNPadded()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CalculateNPadded ( index_t N )

inlinestatic

◆ CheckValidity()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ constexpr bool ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::CheckValidity ( const Argument & karg )

inlinestaticconstexpr

◆ GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1 ( )

inlinestaticconstexpr

◆ GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1 ( )

inlinestaticconstexpr

◆ GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock ( )

inlinestaticconstexpr

◆ GetSharedMemoryNumberOfByte()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__device__ constexpr index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::GetSharedMemoryNumberOfByte ( )

inlinestaticconstexpr

◆ MakeAGridDescriptor_AK0_M_AK1()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ static __device__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeAGridDescriptor_AK0_M_AK1 ( index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA, index_t AK0 )

inlinestatic

◆ MakeAMmaTileDescriptor_M0_M1_M2_M3_K()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<typename ABlockDesc_AK0_M_AK1>

__host__ static __device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeAMmaTileDescriptor_M0_M1_M2_M3_K ( const ABlockDesc_AK0_M_AK1 & )

inlinestaticconstexpr

◆ MakeBGridDescriptor_BK0_N_BK1()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ static __device__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeBGridDescriptor_BK0_N_BK1 ( index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB, index_t BK0 )

inlinestatic

◆ MakeBGridDescriptor_Preshuffled()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ static __device__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeBGridDescriptor_Preshuffled ( index_t N0, index_t K0 )

inlinestatic

◆ MakeBMmaTileDescriptor_N0_N1_N2_N3_K()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<typename BBlockDesc_BK0_N_BK1>

__host__ static __device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeBMmaTileDescriptor_N0_N1_N2_N3_K ( const BBlockDesc_BK0_N_BK1 & )

inlinestaticconstexpr

◆ MakeCGridDescriptor_M_N()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

__host__ static __device__ auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeCGridDescriptor_M_N ( index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC )

inlinestatic

◆ MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<typename CGridDesc>

__host__ static __device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock ( const CGridDesc & c_grid_desc_m_n, index_t MBlock, index_t NBlock )

inlinestaticconstexpr

◆ MakeGemmMmaTileDescriptor()

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<index_t MNXdlPerWave, index_t MNWaves, index_t MNXdlPack, index_t MNPerXdl, bool IsXor, typename TileDesc_K0_MN_K1>

__host__ static __device__ constexpr auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MakeGemmMmaTileDescriptor ( const TileDesc_K0_MN_K1 & )

inlinestaticconstexpr

◆ Run() [1/2]

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>

__device__ void ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::Run ( const ADataType * p_a_grid, const AScaleDataType * p_a_scale_grid, const BDataType * p_b_grid, const BScaleDataType * p_b_scale_grid, CDataType * p_c_grid, void * p_shared, const Problem & problem )

inlinestatic

◆ Run() [2/2]

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<typename AGridDesc_AK0_M_K1, typename AScaleGridDesc_AM_AK, typename BGridDesc_BK0_N_K1, typename BScaleGridDesc_BN_AK, typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>

__device__ void ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::Run ( const ADataType * p_a_grid, const AScaleDataType * p_a_scale_grid, const BDataType * p_b_grid, const BScaleDataType * p_b_scale_grid, CDataType * p_c_grid, void * p_shared, const Problem & problem, const AGridDesc_AK0_M_K1 & a_grid_desc_ak0_m_ak1, const AScaleGridDesc_AM_AK & a_scale_grid_desc_am_ak, const BGridDesc_BK0_N_K1 & b_grid_desc_bk0_n_bk1, const BScaleGridDesc_BN_AK & b_scale_grid_desc_bn_ak, const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock & c_grid_desc_mblock_mperblock_nblock_nperblock )

inlinestatic

◆ Run_2Lds() [1/2]

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>

__device__ void ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::Run_2Lds ( const ADataType * p_a_grid, const AScaleDataType * p_a_scale_grid, const BDataType * p_b_grid, const BScaleDataType * p_b_scale_grid, CDataType * p_c_grid, void * p_shared_0, void * p_shared_1, const Problem & problem )

inlinestatic

◆ Run_2Lds() [2/2]

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

template<typename AGridDesc_AK0_M_K1, typename AScaleGridDesc_AM_AK, typename BGridDesc_BK0_N_K1, typename BScaleGridDesc_BN_AK, typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock, bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation, TailNumber TailNum = TailNumber::Odd>

__device__ void ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::Run_2Lds ( const ADataType * p_a_grid, const AScaleDataType * p_a_scale_grid, const BDataType * p_b_grid, const BScaleDataType * p_b_scale_grid, CDataType * p_c_grid, void * p_shared_0, void * p_shared_1, const Problem & problem, const AGridDesc_AK0_M_K1 & a_grid_desc_ak0_m_ak1, const AScaleGridDesc_AM_AK & a_scale_grid_desc_am_ak, const BGridDesc_BK0_N_K1 & b_grid_desc_bk0_n_bk1, const BScaleGridDesc_BN_AK & b_scale_grid_desc_bn_ak, const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock & c_grid_desc_mblock_mperblock_nblock_nperblock )

inlinestatic

Member Data Documentation

◆ AK0Number

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::AK0Number = Number<KPerBlock / AK1Value>{}

staticconstexpr

◆ AK1Number

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::AK1Number = Number<AK1Value>{}

staticconstexpr

◆ APackedSize

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::APackedSize = packed_size_v<ADataType>

staticconstexpr

◆ BK0Number

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::BK0Number = Number<KPerBlock / BK1Value>{}

staticconstexpr

◆ BK1Number

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::BK1Number = Number<BK1Value>{}

staticconstexpr

◆ BPackedSize

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::BPackedSize = packed_size_v<BDataType>

staticconstexpr

◆ I0

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I0 = Number<0>{}

staticconstexpr

◆ I1

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I1 = Number<1>{}

staticconstexpr

◆ I2

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I2 = Number<2>{}

staticconstexpr

◆ I3

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I3 = Number<3>{}

staticconstexpr

◆ I4

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I4 = Number<4>{}

staticconstexpr

◆ I5

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I5 = Number<5>{}

staticconstexpr

◆ I6

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I6 = Number<6>{}

staticconstexpr

◆ I7

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I7 = Number<7>{}

staticconstexpr

◆ I8

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I8 = Number<8>{}

staticconstexpr

◆ I9

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::I9 = Number<9>{}

staticconstexpr

◆ is_scale_mfma

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::is_scale_mfma = true

staticconstexpr

◆ is_single_rate_mfma

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

bool ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::is_single_rate_mfma = false

staticconstexpr

◆ KLane

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::KLane = 64 / NLane

staticconstexpr

◆ KPack

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::KPack

staticconstexpr

Initial value:

=

math::max(lcm_AK1_BK1,

MfmaSelector<ComputeTypeA,

MPerXdl,

NPerXdl,

ComputeTypeB,

is_single_rate_mfma,

is_scale_mfma>::selected_mfma.k_per_blk /

APackedSize)

ck::math::max

__host__ __device__ constexpr T max(T x)

Definition utility/math.hpp:84

ck::GridwiseGemm_xdl_cshuffle_conv_v3::is_single_rate_mfma

static constexpr bool is_single_rate_mfma

Definition gridwise_gemm_xdl_cshuffle_conv_v3.hpp:83

ck::GridwiseGemm_xdl_cshuffle_conv_v3::is_scale_mfma

static constexpr auto is_scale_mfma

Definition gridwise_gemm_xdl_cshuffle_conv_v3.hpp:91

ck::GridwiseGemm_xdl_cshuffle_conv_v3::lcm_AK1_BK1

static constexpr auto lcm_AK1_BK1

Definition gridwise_gemm_xdl_cshuffle_conv_v3.hpp:82

ck::GridwiseGemmMX_xdl_cshuffle_v3::APackedSize

static constexpr index_t APackedSize

Definition gridwise_gemm_xdl_cshuffle_v3_mx.hpp:189

ck::MfmaSelector

Selects the appropriate MFMA instruction type and configuration for given data types and tile sizes o...

Definition xdlops_gemm.hpp:1208

◆ KRepeat

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::KRepeat = KPerBlock / KLane / KPack

staticconstexpr

◆ KXdlPack

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::KXdlPack = 2

staticconstexpr

◆ lcm_AK1_BK1

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::lcm_AK1_BK1 = math::lcm(AK1Number, BK1Number)

staticconstexpr

◆ MXdlPack

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::MXdlPack = 2

staticconstexpr

◆ NLane

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::NLane = NPerXdl

staticconstexpr

◆ NWave

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::NWave = NPerBlock / NPerXdl / NXdlPerWave

staticconstexpr

◆ NXdlPack

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

auto ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::NXdlPack = 2

staticconstexpr

◆ scale_pack_size_a

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::scale_pack_size_a = sizeof(AScaleDataType) / sizeof(mx_scale_t)

staticconstexpr

◆ scale_pack_size_b

template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename AScaleDataType, typename BDataType, typename BScaleDataType, typename AccDataType, typename CShuffleDataType, typename CDataType, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, tensor_operation::device::GemmSpecialization GemmSpec, index_t ScaleBlockSize, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1Value, index_t BK1Value, index_t MPerXdl, index_t NPerXdl, index_t MXdlPerWave, index_t NXdlPerWave, 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 CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v3, typename ComputeTypeA = ADataType, typename ComputeTypeB = BDataType, bool PermuteA = false, bool PermuteB = false>

index_t ck::GridwiseGemmMX_xdl_cshuffle_v3_bpreshuffle< ALayout, BLayout, CLayout, ADataType, AScaleDataType, BDataType, BScaleDataType, AccDataType, CShuffleDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, ScaleBlockSize, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1Value, BK1Value, MPerXdl, NPerXdl, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, AThreadTransferSrcResetCoordinateAfterRun, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BThreadTransferSrcResetCoordinateAfterRun, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB, PermuteA, PermuteB >::scale_pack_size_b = sizeof(BScaleDataType) / sizeof(mx_scale_t)

staticconstexpr

---

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

• gridwise_gemm_xdl_cshuffle_v3_mx_bpreshuffle.hpp