DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched > Struct Template Reference#
Classes |
Public Types |
Public Member Functions |
Static Public Member Functions |
Static Public Attributes |
List of all members
ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched > Struct Template Reference
#include <device_batched_gemm_reduce_xdl_cshuffle.hpp>
Inheritance diagram for ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >:
Classes | |
| struct | ComputeBasePtrOfStridedBatch |
| struct | Argument |
| struct | Invoker |
Public Types | |
| using | DeviceOp = DeviceBatchedGemmReduce_Xdl_CShuffle |
| using | AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1(1, 1, 1)) |
| using | BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1(1, 1, 1)) |
| using | CGridDesc_M_N = decltype(MakeCGridDescriptor_M_N(1, 1, 1)) |
| using | ReduceGridDesc_M = decltype(MakeReduceGridDescriptor_M(1)) |
| template<index_t NXdlPerWave_> | |
| using | GridwiseGemmBase |
| using | GridwiseGemm64 = GridwiseGemmBase<math::max(NXdlPerWave64, 1)> |
| using | GridwiseGemm32 = GridwiseGemmBase<NXdlPerWave32> |
Public Member Functions | |
| bool | IsSupportedArgument (const BaseArgument *p_arg) override |
| std::unique_ptr< BaseArgument > | MakeArgumentPointer (const void *p_a, const void *p_b, const void *p_bias, std::array< const void *, 0 > p_ds, void *p_c, std::array< void *, NumReduce > p_reduces, ck::index_t M, ck::index_t N, ck::index_t K, ck::index_t StrideA, ck::index_t StrideB, ck::index_t StrideC, std::array< ck::index_t, 0 > StrideDs, std::array< void *, 3 > gemm_element_ops, std::array< void *, 0 > d_element_ops, std::array< void *, NumReduce > reduce_in_element_op, std::array< void *, NumReduce > reduce_out_element_op, index_t Batch=1) override |
| std::unique_ptr< BaseInvoker > | MakeInvokerPointer () override |
| std::string | GetTypeString () const override |
| Public Member Functions inherited from ck::tensor_operation::device::DeviceGemmReduce< 0, ReduceOperations::Size()> | |
| virtual std::unique_ptr< BaseArgument > | MakeArgumentPointer (const void *p_a, const void *p_b, const void *p_bias, std::array< const void *, NumDTensor > p_ds, void *p_c, std::array< void *, NumReduce > p_reduces, ck::index_t M, ck::index_t N, ck::index_t K, ck::index_t StrideA, ck::index_t StrideB, ck::index_t StrideC, std::array< ck::index_t, NumDTensor > StrideDs, std::array< void *, 3 > gemm_element_ops, std::array< void *, NumDTensor > d_element_ops, std::array< void *, NumReduce > reduce_in_element_ops, std::array< void *, NumReduce > reduce_out_element_ops, ck::index_t BatchCount=1)=0 |
| Public Member Functions inherited from ck::tensor_operation::device::BaseOperator | |
| BaseOperator ()=default | |
| BaseOperator (const BaseOperator &)=default | |
| BaseOperator & | operator= (const BaseOperator &)=default |
| virtual std::string | GetInstanceString () const |
| virtual std::string | GetTypeIdName () const |
| virtual std::optional< std::string > | GetObjectName () const |
| virtual std::optional< std::string > | GetTemplateInfo () const |
| virtual std::string | GetTypeIdHashCode () const |
| virtual size_t | GetWorkSpaceSize (const BaseArgument *) const |
| virtual void | SetWorkSpacePointer (BaseArgument *p_arg, void *p_workspace, const StreamConfig &=StreamConfig{}) const |
| virtual | ~BaseOperator () |
Static Public Member Functions | |
| static auto | MakeAGridDescriptor_AK0_M_AK1 (index_t MRaw, index_t KRaw, index_t StrideA) |
| static auto | MakeBGridDescriptor_BK0_N_BK1 (index_t KRaw, index_t NRaw, index_t StrideB) |
| static auto | MakeCGridDescriptor_M_N (index_t MRaw, index_t NRaw, index_t StrideC) |
| static auto | MakeReduceGridDescriptor_M (index_t MRaw) |
| static constexpr bool | IsValidCompilationParameter () |
| static bool | IsSupportedArgument (const Argument &arg) |
| static auto | MakeArgument (const void *p_a, const void *p_b, const void *p_bias, std::array< const void *, 0 > p_ds, void *p_c, std::array< void *, NumReduce > p_reduces, ck::index_t M, ck::index_t N, ck::index_t K, ck::index_t StrideA, ck::index_t StrideB, ck::index_t StrideC, std::array< ck::index_t, 0 > StrideDs, std::array< void *, 3 > gemm_element_ops, std::array< void *, 0 > d_element_ops, std::array< void *, NumReduce > reduce_in_element_op, std::array< void *, NumReduce > reduce_out_element_op, index_t Batch) |
| static auto | MakeInvoker () |
Static Public Attributes | |
| static GET_NXDL_PER_WAVE_IMPL constexpr auto | NXdlPerWave64 = GetNXdlPerWave<true>() |
| static constexpr auto | NXdlPerWave32 = GetNXdlPerWave<false>() |
| static constexpr auto | I0 = Number<0>{} |
| static constexpr auto | I1 = Number<1>{} |
| static constexpr auto | I2 = Number<2>{} |
| static constexpr int | NumReduce = ReduceOperations::Size() |
Member Typedef Documentation
◆ AGridDesc_AK0_M_AK1
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1(1, 1, 1)) |
◆ BGridDesc_BK0_N_BK1
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1(1, 1, 1)) |
◆ CGridDesc_M_N
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::CGridDesc_M_N = decltype(MakeCGridDescriptor_M_N(1, 1, 1)) |
◆ DeviceOp
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::DeviceOp = DeviceBatchedGemmReduce_Xdl_CShuffle |
◆ GridwiseGemm32
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::GridwiseGemm32 = GridwiseGemmBase<NXdlPerWave32> |
◆ GridwiseGemm64
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::GridwiseGemm64 = GridwiseGemmBase<math::max(NXdlPerWave64, 1)> |
◆ GridwiseGemmBase
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
template<index_t NXdlPerWave_>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::GridwiseGemmBase |
◆ ReduceGridDesc_M
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
| using ck::tensor_operation::device::DeviceBatchedGemmReduce_Xdl_CShuffle< ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, GemmAccDataType, CShuffleDataType, ReduceAccDataType, ReducePtrsGlobal, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, ReduceOperations, ReduceInElementwiseOperations, ReduceAccElementwiseOperations, ReduceGlobalMemoryDataOperation, GemmSpec, NumGemmKPrefetchStage, BlockSize, MPerBlock, NPerBlock, KPerBlock, AK1, BK1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_AK0_M_AK1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_AK1, ABlockLdsExtraM, BBlockTransferThreadClusterLengths_BK0_N_BK1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_BK1, BBlockLdsExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CShuffleBlockTransferScalarPerVector_NPerBlock, CReduceThreadClusterLengths_MPerBlock_NPerBlock, CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopSched >::ReduceGridDesc_M = decltype(MakeReduceGridDescriptor_M(1)) |
Member Function Documentation
◆ GetTypeString()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
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 GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
◆ IsSupportedArgument() [2/2]
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlineoverridevirtual |
Reimplemented from ck::tensor_operation::device::BaseOperator.
◆ IsValidCompilationParameter()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestaticconstexpr |
◆ MakeAGridDescriptor_AK0_M_AK1()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
◆ MakeArgument()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
◆ MakeArgumentPointer()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlineoverride |
◆ MakeBGridDescriptor_BK0_N_BK1()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
◆ MakeCGridDescriptor_M_N()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
◆ MakeInvoker()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
◆ MakeInvokerPointer()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlineoverridevirtual |
◆ MakeReduceGridDescriptor_M()
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
inlinestatic |
Member Data Documentation
◆ I0
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
staticconstexpr |
◆ I1
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
staticconstexpr |
◆ I2
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
staticconstexpr |
◆ NumReduce
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
staticconstexpr |
◆ NXdlPerWave32
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
staticconstexpr |
◆ NXdlPerWave64
template<typename ALayout, typename BLayout, typename CLayout, typename ADataType, typename BDataType, typename CDataType, typename GemmAccDataType, typename CShuffleDataType, typename ReduceAccDataType, typename ReducePtrsGlobal, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, typename ReduceOperations, typename ReduceInElementwiseOperations, typename ReduceAccElementwiseOperations, typename ReduceGlobalMemoryDataOperation, GemmSpecialization GemmSpec, index_t NumGemmKPrefetchStage, index_t BlockSize, index_t MPerBlock, index_t NPerBlock, index_t KPerBlock, index_t AK1, index_t BK1, index_t MPerXDL, index_t NPerXDL, index_t MXdlPerWave, index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_AK0_M_AK1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, index_t ABlockTransferSrcVectorDim, index_t ABlockTransferSrcScalarPerVector, index_t ABlockTransferDstScalarPerVector_AK1, bool ABlockLdsExtraM, typename BBlockTransferThreadClusterLengths_BK0_N_BK1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, index_t BBlockTransferSrcVectorDim, index_t BBlockTransferSrcScalarPerVector, index_t BBlockTransferDstScalarPerVector_BK1, bool BBlockLdsExtraN, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CShuffleBlockTransferScalarPerVector_NPerBlock, typename CReduceThreadClusterLengths_MPerBlock_NPerBlock, index_t CReduceThreadLds2VGprCopySrcDstScalarPerVector_NPerBlock, index_t CReduceThreadVgpr2GlobalCopySrcDstScalarPerVector_MPerBlock, LoopScheduler LoopSched = make_default_loop_scheduler()>
|
staticconstexpr |
The documentation for this struct was generated from the following file: