cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize > Struct Template Reference

Template based on element size (in bits) - defined in terms of pitch-linear memory.

#include <tensor_op_multiplicand_sm70.h>

Public Types
using	Index = int32_t
	Index type used for coordinates. More...
using	LongIndex = int64_t
	Long index type used for offsets. More...
using	TensorCoord = PitchLinearCoord
	Logical coordinate. More...
using	Stride = Coord< kStrideRank, Index, LongIndex >
	Stride vector. More...
using	TileShape = PitchLinearShape< 8, 4 >
	Fundamental tile shape in units of vectors. More...
using	PartitionShape = PitchLinearShape< 4, 4 >
	Fundamental partition shape in units of vectors. More...
using	PartitionCount = PitchLinearShape< TileShape::kContiguous/PartitionShape::kContiguous, TileShape::kStrided/PartitionShape::kStrided >
using	AccessCount = PitchLinearShape< PartitionShape::kContiguous, PartitionShape::kStrided >

Public Member Functions
CUTLASS_HOST_DEVICE	VoltaTensorOpMultiplicandBCongruous (Index ldm=0)
	Ctor. More...
CUTLASS_HOST_DEVICE	VoltaTensorOpMultiplicandBCongruous (Stride stride)
	Ctor. More...
CUTLASS_HOST_DEVICE LongIndex	operator() (TensorCoord const &coord) const
CUTLASS_HOST_DEVICE Stride	stride () const
	Returns the stride of the layout. More...
CUTLASS_HOST_DEVICE Stride &	stride ()
	Returns the stride of the layout. More...
CUTLASS_HOST_DEVICE LongIndex	capacity (TensorCoord const &extent) const
	Compute the number of contiguous elements needed to store a tensor with the given size. More...

Static Public Member Functions
static CUTLASS_HOST_DEVICE VoltaTensorOpMultiplicandBCongruous	packed (TensorCoord const &extent)
	Helper returns a layout to a tightly packed tensor. More...

Static Public Attributes
static int const	kRank = 2
	Logical rank of tensor. More...
static int const	kStrideRank = 1
	Rank of stride vector. More...
static int const	kAccessSize = 128
	This layout is optimized for 128b accesses. More...
static int const	kElementSize = ElementSize
static int const	kElementsPerAccess = kAccessSize / kElementSize

Member Typedef Documentation

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::AccessCount = PitchLinearShape< PartitionShape::kContiguous, PartitionShape::kStrided >

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::Index = int32_t

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::LongIndex = int64_t

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::PartitionCount = PitchLinearShape< TileShape::kContiguous / PartitionShape::kContiguous, TileShape::kStrided / PartitionShape::kStrided >

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::PartitionShape = PitchLinearShape<4, 4>

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::Stride = Coord<kStrideRank, Index, LongIndex>

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::TensorCoord = PitchLinearCoord

template<int ElementSize>

using cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::TileShape = PitchLinearShape<8, 4>

Constructor & Destructor Documentation

template<int ElementSize>

CUTLASS_HOST_DEVICE cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::VoltaTensorOpMultiplicandBCongruous ( Index  ldm = 0 )

inline

template<int ElementSize>

CUTLASS_HOST_DEVICE cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::VoltaTensorOpMultiplicandBCongruous ( Stride  stride )

inline

Member Function Documentation

template<int ElementSize>

CUTLASS_HOST_DEVICE LongIndex cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::capacity ( TensorCoord const &  extent ) const

inline

template<int ElementSize>

CUTLASS_HOST_DEVICE LongIndex cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::operator() ( TensorCoord const &  coord ) const

inline

Returns the offset of a coordinate in linear memory. Assumes coordinate has convention (contiguous, strided)

template<int ElementSize>

static CUTLASS_HOST_DEVICE VoltaTensorOpMultiplicandBCongruous cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::packed ( TensorCoord const &  extent )

inlinestatic

template<int ElementSize>

CUTLASS_HOST_DEVICE Stride cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::stride ( ) const

inline

template<int ElementSize>

CUTLASS_HOST_DEVICE Stride& cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::stride ( )

inline

Member Data Documentation

template<int ElementSize>

int const cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::kAccessSize = 128

static

template<int ElementSize>

int const cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::kElementSize = ElementSize

static

template<int ElementSize>

int const cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::kElementsPerAccess = kAccessSize / kElementSize

static

template<int ElementSize>

int const cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::kRank = 2

static

template<int ElementSize>

int const cutlass::layout::VoltaTensorOpMultiplicandBCongruous< ElementSize >::kStrideRank = 1

static

---

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

• tensor_op_multiplicand_sm70.h