DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer > Struct Template Reference

DeviceGemmWmma_CShuffle&lt; ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer &gt; Struct Template Reference#

Composable Kernel: ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer > Struct Template Reference
Classes | Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members
ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer > Struct Template Reference

#include <device_gemm_wmma.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >:
ck::tensor_operation::device::DeviceGemm< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation > ck::tensor_operation::device::BaseOperator

Classes

struct  Argument
 
struct  Invoker
 

Public Types

using AGridDesc = decltype(MakeAGridDescriptor(1, 1, 1))
 
using BGridDesc = decltype(MakeBGridDescriptor(1, 1, 1))
 
using CGridDesc_M_N = decltype(MakeCGridDescriptor_M_N(1, 1, 1))
 
using GridwiseGemm = GridwiseGemm_Wmma< BlockSize, ADataType, BDataType, AccDataType, CShuffleDataType, CDataType, InMemoryDataOperationEnum::Set, AGridDesc, BGridDesc, CGridDesc_M_N, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, MPerBlock, NPerBlock, KPerBlock, MPerWmma, NPerWmma, K1, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, AEnableLds, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BEnableLds, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, NumPrefetch, LoopSched, PipelineVer >
 

Public Member Functions

bool IsSupportedArgument (const BaseArgument *p_arg) override
 
std::unique_ptr< BaseArgument > MakeArgumentPointer (const void *p_a, const void *p_b, void *p_c, index_t M, index_t N, index_t K, index_t StrideA, index_t StrideB, index_t StrideC, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, CElementwiseOperation c_element_op) override
 
std::unique_ptr< BaseInvoker > MakeInvokerPointer () override
 
std::string GetTypeString () const override
 
- Public Member Functions inherited from ck::tensor_operation::device::BaseOperator
 BaseOperator ()=default
 
 BaseOperator (const BaseOperator &)=default
 
BaseOperator & operator= (const BaseOperator &)=default
 
virtual std::string GetTypeIdName () const
 
virtual std::optional< std::string > GetObjectName () const
 
virtual std::optional< std::string > GetTemplateInfo () const
 
virtual std::string GetTypeIdHashCode () const
 
virtual size_t GetWorkSpaceSize (const BaseArgument *) const
 
virtual void SetWorkSpacePointer (BaseArgument *p_arg, void *p_workspace, const StreamConfig &=StreamConfig{}) const
 
virtual ~BaseOperator ()
 

Static Public Member Functions

static auto MakeAGridDescriptor (index_t MRaw, index_t KRaw, index_t StrideA)
 
static auto MakeBGridDescriptor (index_t KRaw, index_t NRaw, index_t StrideB)
 
static auto MakeCGridDescriptor_M_N (index_t MRaw, index_t NRaw, index_t StrideC)
 
static constexpr bool IsValidCompilationParameter ()
 
static bool IsSupportedArgument (const Argument &arg)
 
static auto MakeArgument (const ADataType *p_a, const BDataType *p_b, CDataType *p_c, index_t M, index_t N, index_t K, index_t StrideA, index_t StrideB, index_t StrideC, AElementwiseOperation a_element_op, BElementwiseOperation b_element_op, CElementwiseOperation c_element_op)
 
static auto MakeInvoker ()
 

Static Public Attributes

static constexpr auto I0 = Number<0>{}
 
static constexpr auto I1 = Number<1>{}
 
static constexpr auto I2 = Number<2>{}
 
static constexpr auto I3 = Number<3>{}
 
static constexpr auto I4 = Number<4>{}
 
static constexpr auto I5 = Number<5>{}
 
static constexpr auto I6 = Number<6>{}
 
static constexpr auto K1Number = Number<K1>{}
 
static constexpr auto MWaves = MPerBlock / (MRepeat * MPerWmma)
 
static constexpr auto NWaves = NPerBlock / (NRepeat * NPerWmma)
 
static constexpr auto WmmaK = K1 == 16 ? 32 : 16
 
static constexpr auto MaxVectorLoadA = K1 * sizeof(ADataType) == 16 ? true : false
 
static constexpr auto MaxVectorLoadB = K1 * sizeof(BDataType) == 16 ? true : false
 
static constexpr auto AEnableLds_auto
 
static constexpr auto BEnableLds_auto
 
static constexpr auto AEnableLds_manu = false
 
static constexpr auto BEnableLds_manu = false
 
static constexpr auto AEnableLds = AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1)
 
static constexpr auto BEnableLds = BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1)
 
static constexpr auto matrix_padder
 

Member Typedef Documentation

◆ AGridDesc

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
using ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::AGridDesc = decltype(MakeAGridDescriptor(1, 1, 1))

◆ BGridDesc

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
using ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::BGridDesc = decltype(MakeBGridDescriptor(1, 1, 1))

◆ CGridDesc_M_N

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
using ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::CGridDesc_M_N = decltype(MakeCGridDescriptor_M_N(1, 1, 1))

◆ GridwiseGemm

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
using ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::GridwiseGemm = GridwiseGemm_Wmma<BlockSize, ADataType, BDataType, AccDataType, CShuffleDataType, CDataType, InMemoryDataOperationEnum::Set, AGridDesc, BGridDesc, CGridDesc_M_N, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, MPerBlock, NPerBlock, KPerBlock, MPerWmma, NPerWmma, K1, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, AEnableLds, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BEnableLds, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, NumPrefetch, LoopSched, PipelineVer>

Member Function Documentation

◆ GetTypeString()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
std::string ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::GetTypeString ( ) const
inlineoverridevirtual

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

◆ IsSupportedArgument() [1/2]

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static bool ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::IsSupportedArgument ( const Argument &  arg)
inlinestatic

◆ IsSupportedArgument() [2/2]

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
bool ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::IsSupportedArgument ( const BaseArgument *  p_arg)
inlineoverridevirtual

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

◆ IsValidCompilationParameter()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static constexpr bool ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::IsValidCompilationParameter ( )
inlinestaticconstexpr

◆ MakeAGridDescriptor()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeAGridDescriptor ( index_t  MRaw,
index_t  KRaw,
index_t  StrideA 
)
inlinestatic

◆ MakeArgument()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeArgument ( const ADataType *  p_a,
const BDataType *  p_b,
CDataType *  p_c,
index_t  M,
index_t  N,
index_t  K,
index_t  StrideA,
index_t  StrideB,
index_t  StrideC,
AElementwiseOperation  a_element_op,
BElementwiseOperation  b_element_op,
CElementwiseOperation  c_element_op 
)
inlinestatic

◆ MakeArgumentPointer()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
std::unique_ptr<BaseArgument> ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeArgumentPointer ( const void *  p_a,
const void *  p_b,
void *  p_c,
index_t  M,
index_t  N,
index_t  K,
index_t  StrideA,
index_t  StrideB,
index_t  StrideC,
AElementwiseOperation  a_element_op,
BElementwiseOperation  b_element_op,
CElementwiseOperation  c_element_op 
)
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceGemm< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation >.

◆ MakeBGridDescriptor()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeBGridDescriptor ( index_t  KRaw,
index_t  NRaw,
index_t  StrideB 
)
inlinestatic

◆ MakeCGridDescriptor_M_N()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeCGridDescriptor_M_N ( index_t  MRaw,
index_t  NRaw,
index_t  StrideC 
)
inlinestatic

◆ MakeInvoker()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
static auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeInvoker ( )
inlinestatic

◆ MakeInvokerPointer()

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
std::unique_ptr<BaseInvoker> ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MakeInvokerPointer ( )
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceGemm< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation >.

Member Data Documentation

◆ AEnableLds

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::AEnableLds = AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1)
staticconstexpr

◆ AEnableLds_auto

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::AEnableLds_auto
staticconstexpr
Initial value:
= (NWaves == 1 && (MaxVectorLoadA || MRepeat == 1) &&
is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
? false
: true
ck::integral_constant< bool, false >::value
static constexpr bool value
Definition: integral_constant.hpp:21
ck::tensor_operation::device::DeviceGemmWmma_CShuffle::MaxVectorLoadA
static constexpr auto MaxVectorLoadA
Definition: device_gemm_wmma.hpp:90
ck::tensor_operation::device::DeviceGemmWmma_CShuffle::NWaves
static constexpr auto NWaves
Definition: device_gemm_wmma.hpp:88

◆ AEnableLds_manu

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::AEnableLds_manu = false
staticconstexpr

◆ BEnableLds

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::BEnableLds = BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1)
staticconstexpr

◆ BEnableLds_auto

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::BEnableLds_auto
staticconstexpr
Initial value:
=
(MWaves == 1 && (MaxVectorLoadB || NRepeat == 1) &&
is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
? false
: true
ck::tensor_operation::device::DeviceGemmWmma_CShuffle::MWaves
static constexpr auto MWaves
Definition: device_gemm_wmma.hpp:87
ck::tensor_operation::device::DeviceGemmWmma_CShuffle::MaxVectorLoadB
static constexpr auto MaxVectorLoadB
Definition: device_gemm_wmma.hpp:91

◆ BEnableLds_manu

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::BEnableLds_manu = false
staticconstexpr

◆ I0

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I0 = Number<0>{}
staticconstexpr

◆ I1

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I1 = Number<1>{}
staticconstexpr

◆ I2

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I2 = Number<2>{}
staticconstexpr

◆ I3

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I3 = Number<3>{}
staticconstexpr

◆ I4

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I4 = Number<4>{}
staticconstexpr

◆ I5

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I5 = Number<5>{}
staticconstexpr

◆ I6

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::I6 = Number<6>{}
staticconstexpr

◆ K1Number

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::K1Number = Number<K1>{}
staticconstexpr

◆ matrix_padder

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::matrix_padder
staticconstexpr
Initial value:
=
MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock}

◆ MaxVectorLoadA

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MaxVectorLoadA = K1 * sizeof(ADataType) == 16 ? true : false
staticconstexpr

◆ MaxVectorLoadB

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MaxVectorLoadB = K1 * sizeof(BDataType) == 16 ? true : false
staticconstexpr

◆ MWaves

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::MWaves = MPerBlock / (MRepeat * MPerWmma)
staticconstexpr

◆ NWaves

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::NWaves = NPerBlock / (NRepeat * NPerWmma)
staticconstexpr

◆ WmmaK

template<typename ALayout , typename BLayout , typename CLayout , typename ADataType , typename BDataType , typename CDataType , typename AccDataType , typename CShuffleDataType , typename AElementwiseOperation , typename BElementwiseOperation , typename CElementwiseOperation , GemmSpecialization GemmSpec, ck::index_t NumPrefetch, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, ck::index_t K1, ck::index_t MPerWmma, ck::index_t NPerWmma, ck::index_t MRepeat, ck::index_t NRepeat, typename ABlockTransferThreadClusterLengths_K0_M_K1 , typename ABlockTransferThreadClusterArrangeOrder , typename ABlockTransferSrcAccessOrder , ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1 , typename BBlockTransferThreadClusterArrangeOrder , typename BBlockTransferSrcAccessOrder , ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, index_t CShuffleMRepeatPerShuffle, index_t CShuffleNRepeatPerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock , index_t CShuffleBlockTransferScalarPerVector_NPerBlock, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
constexpr auto ck::tensor_operation::device::DeviceGemmWmma_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, GemmSpec, NumPrefetch, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, MPerWmma, NPerWmma, MRepeat, NRepeat, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CShuffleMRepeatPerShuffle, CShuffleNRepeatPerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, LoopSched, PipelineVer >::WmmaK = K1 == 16 ? 32 : 16
staticconstexpr
The documentation for this struct was generated from the following file:
  • /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/docs-7.0.1/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp