DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize > Struct Template Reference

DeviceElementwiseNormalizationImpl&lt; InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize &gt; Struct Template Reference#

Composable Kernel: ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize > Struct Template Reference
Classes | Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members
ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize > Struct Template Reference

#include <device_elementwise_normalization_impl.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >:
ck::tensor_operation::device::DeviceElementwiseNormalization< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim > ck::tensor_operation::device::BaseOperator

Classes

struct  Argument
struct  Invoker

Public Types

using XDataType = YDataType
using InDataTypePointerTuple = decltype(GenerateInDataTypePointerTuple())
using InGrid2dDescTuple = decltype(GenerateSrcGrid2dDescTuple(Number<NumInput>{}))
using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1))
using GridwiseReduceLayernormGeneric
using GridwiseReduceLayernormSweepOnce

Public Member Functions

bool IsSupportedArgument (const BaseArgument *p_arg) override
std::unique_ptr< BaseArgument > MakeArgumentPointer (const std::vector< index_t > lengths, const std::array< std::vector< index_t >, NumInput > inStridesArray, const std::vector< index_t > gammaStrides, const std::vector< index_t > betaStrides, const std::vector< index_t > yStrides, const std::vector< index_t > reduceDims, double epsilon, const std::array< const void *, NumInput > in_dev_buffers, const void *p_gamma, const void *p_beta, void *p_y, XElementwiseOperation x_elementwise_op, YElementwiseOperation y_elementwise_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 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 GenerateInDataTypePointerTuple ()
static auto MakeSrc2dDescriptor (const std::vector< index_t > &inLengths, const std::vector< index_t > &inStrides, int blkGroupSize, int numBlockTileIteration)
template<index_t TupleSize>
static auto GenerateSrcGrid2dDescTuple (Number< TupleSize >)

Static Public Attributes

static constexpr int NumInput = InDataTypeTuple::Size()
static constexpr index_t M_BlockTileSize
static constexpr index_t K_BlockTileSize
Static Public Attributes inherited from ck::tensor_operation::device::DeviceElementwiseNormalization< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim >
static constexpr int NumInput = InDataTypeTuple::Size()

Member Typedef Documentation

◆ GridDesc_M_K

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
using ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1))

◆ GridwiseReduceLayernormGeneric

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
using ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::GridwiseReduceLayernormGeneric
Initial value:
GridwiseElementwiseLayernormWelfordVariance_mk_to_mk<InDataTypePointerTuple,
XDataType,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
XElementwiseOperation,
YElementwiseOperation,
InGrid2dDescTuple,
GridDesc_M_K,
BlockSize,
MThreadClusterSize,
KThreadClusterSize,
MThreadSliceSize,
KThreadSliceSize,
XYSrcVectorDim,
XSrcVectorSize,
GammaSrcVectorDim,
GammaSrcVectorSize,
BetaSrcVectorDim,
BetaSrcVectorSize,
XYSrcVectorDim,
YDstVectorSize,
false>
ck::GridwiseElementwiseLayernormWelfordVariance_mk_to_mk
Definition gridwise_elementwise_layernorm_welford_variance.hpp:42
ck::tensor_operation::device::DeviceElementwiseNormalizationImpl::XDataType
YDataType XDataType
Definition device_elementwise_normalization_impl.hpp:107
ck::tensor_operation::device::DeviceElementwiseNormalizationImpl::InDataTypePointerTuple
decltype(GenerateInDataTypePointerTuple()) InDataTypePointerTuple
Definition device_elementwise_normalization_impl.hpp:132
ck::tensor_operation::device::DeviceElementwiseNormalizationImpl::GridDesc_M_K
decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1)) GridDesc_M_K
Definition device_elementwise_normalization_impl.hpp:209
ck::tensor_operation::device::DeviceElementwiseNormalizationImpl::InGrid2dDescTuple
decltype(GenerateSrcGrid2dDescTuple(Number< NumInput >{})) InGrid2dDescTuple
Definition device_elementwise_normalization_impl.hpp:207

◆ GridwiseReduceLayernormSweepOnce

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
using ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::GridwiseReduceLayernormSweepOnce
Initial value:
GridwiseElementwiseLayernormWelfordVariance_mk_to_mk<InDataTypePointerTuple,
XDataType,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
XElementwiseOperation,
YElementwiseOperation,
InGrid2dDescTuple,
GridDesc_M_K,
BlockSize,
MThreadClusterSize,
KThreadClusterSize,
MThreadSliceSize,
KThreadSliceSize,
XYSrcVectorDim,
XSrcVectorSize,
GammaSrcVectorDim,
GammaSrcVectorSize,
BetaSrcVectorDim,
BetaSrcVectorSize,
XYSrcVectorDim,
YDstVectorSize,
true>

◆ InDataTypePointerTuple

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
using ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::InDataTypePointerTuple = decltype(GenerateInDataTypePointerTuple())

◆ InGrid2dDescTuple

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
using ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::InGrid2dDescTuple = decltype(GenerateSrcGrid2dDescTuple(Number<NumInput>{}))

◆ XDataType

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
using ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::XDataType = YDataType

Member Function Documentation

◆ GenerateInDataTypePointerTuple()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
auto ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::GenerateInDataTypePointerTuple ( )
inlinestatic

◆ GenerateSrcGrid2dDescTuple()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
template<index_t TupleSize>
auto ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::GenerateSrcGrid2dDescTuple ( Number< TupleSize > )
inlinestatic

◆ GetTypeString()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
std::string ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::GetTypeString ( ) const
inlineoverridevirtual

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

◆ IsSupportedArgument()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
bool ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::IsSupportedArgument ( const BaseArgument * p_arg)
inlineoverridevirtual

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

◆ MakeArgumentPointer()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
std::unique_ptr< BaseArgument > ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::MakeArgumentPointer ( const std::vector< index_t > lengths,
const std::array< std::vector< index_t >, NumInput > inStridesArray,
const std::vector< index_t > gammaStrides,
const std::vector< index_t > betaStrides,
const std::vector< index_t > yStrides,
const std::vector< index_t > reduceDims,
double epsilon,
const std::array< const void *, NumInput > in_dev_buffers,
const void * p_gamma,
const void * p_beta,
void * p_y,
XElementwiseOperation x_elementwise_op,
YElementwiseOperation y_elementwise_op )
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceElementwiseNormalization< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim >.

◆ MakeInvokerPointer()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
std::unique_ptr< BaseInvoker > ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::MakeInvokerPointer ( )
inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceElementwiseNormalization< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim >.

◆ MakeSrc2dDescriptor()

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
auto ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::MakeSrc2dDescriptor ( const std::vector< index_t > & inLengths,
const std::vector< index_t > & inStrides,
int blkGroupSize,
int numBlockTileIteration )
inlinestatic

Member Data Documentation

◆ K_BlockTileSize

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
index_t ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::K_BlockTileSize
staticconstexpr
Initial value:
=
KThreadClusterSize * KThreadSliceSize

◆ M_BlockTileSize

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
index_t ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::M_BlockTileSize
staticconstexpr
Initial value:
=
MThreadClusterSize * MThreadSliceSize

◆ NumInput

template<typename InDataTypeTuple, typename GammaDataType, typename BetaDataType, typename AccDataType, typename YDataType, typename XElementwiseOperation, typename YElementwiseOperation, index_t Rank, index_t NumReduceDim, index_t BlockSize, index_t MThreadClusterSize, index_t KThreadClusterSize, index_t MThreadSliceSize, index_t KThreadSliceSize, index_t XYSrcVectorDim, index_t XSrcVectorSize, index_t GammaSrcVectorDim, index_t GammaSrcVectorSize, index_t BetaSrcVectorDim, index_t BetaSrcVectorSize, index_t YDstVectorSize>
int ck::tensor_operation::device::DeviceElementwiseNormalizationImpl< InDataTypeTuple, GammaDataType, BetaDataType, AccDataType, YDataType, XElementwiseOperation, YElementwiseOperation, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize >::NumInput = InDataTypeTuple::Size()
staticconstexpr
The documentation for this struct was generated from the following file: