CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members
cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock > Struct Template Reference

#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...
 

Public Member Functions

CUTLASS_HOST_DEVICE VoltaTensorOpMultiplicandCrosswise (Index ldm=0)
 Ctor. More...
 
CUTLASS_HOST_DEVICE VoltaTensorOpMultiplicandCrosswise (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 Stridestride ()
 Returns the stride of the layout. More...
 
CUTLASS_HOST_DEVICE LongIndex capacity (TensorCoord const &extent) const
 

Static Public Member Functions

static CUTLASS_HOST_DEVICE VoltaTensorOpMultiplicandCrosswise 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 = 64
 This layout is optimized for 64b accesses. More...
 
static int const kElementSize = ElementSize
 
static int const kElementsPerAccess = kAccessSize / kElementSize
 
static int const kKBlock = KBlock
 

Detailed Description

template<int ElementSize, int KBlock>
struct cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >

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

Member Typedef Documentation

template<int ElementSize, int KBlock>
using cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::Index = int32_t
template<int ElementSize, int KBlock>
using cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::LongIndex = int64_t
template<int ElementSize, int KBlock>
using cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::Stride = Coord<kStrideRank, Index, LongIndex>
template<int ElementSize, int KBlock>
using cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::TensorCoord = PitchLinearCoord

Constructor & Destructor Documentation

template<int ElementSize, int KBlock>
CUTLASS_HOST_DEVICE cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::VoltaTensorOpMultiplicandCrosswise ( Index  ldm = 0)
inline
template<int ElementSize, int KBlock>
CUTLASS_HOST_DEVICE cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::VoltaTensorOpMultiplicandCrosswise ( Stride  stride)
inline

Member Function Documentation

template<int ElementSize, int KBlock>
CUTLASS_HOST_DEVICE LongIndex cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::capacity ( TensorCoord const &  extent) const
inline

Compute the number of contiguous elements needed to store a tensor with the given size

template<int ElementSize, int KBlock>
CUTLASS_HOST_DEVICE LongIndex cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::operator() ( TensorCoord const &  coord) const
inline

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

template<int ElementSize, int KBlock>
static CUTLASS_HOST_DEVICE VoltaTensorOpMultiplicandCrosswise cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::packed ( TensorCoord const &  extent)
inlinestatic
template<int ElementSize, int KBlock>
CUTLASS_HOST_DEVICE Stride cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::stride ( ) const
inline
template<int ElementSize, int KBlock>
CUTLASS_HOST_DEVICE Stride& cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::stride ( )
inline

Member Data Documentation

template<int ElementSize, int KBlock>
int const cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::kAccessSize = 64
static
template<int ElementSize, int KBlock>
int const cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::kElementSize = ElementSize
static
template<int ElementSize, int KBlock>
int const cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::kElementsPerAccess = kAccessSize / kElementSize
static
template<int ElementSize, int KBlock>
int const cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::kKBlock = KBlock
static
template<int ElementSize, int KBlock>
int const cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::kRank = 2
static
template<int ElementSize, int KBlock>
int const cutlass::layout::VoltaTensorOpMultiplicandCrosswise< ElementSize, KBlock >::kStrideRank = 1
static

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