CUTLASS
CUDA Templates for Linear Algebra Subroutines and Solvers
Classes | Public Types | Static Public Attributes | List of all members
cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK > Struct Template Reference

Defines the optimal thread map for TensorOp accumulator layouts.

#include <default_thread_map_tensor_op.h>

Classes

struct  Detail
 

Public Types

using ThreadblockShape = ThreadblockShape_
 
using WarpShape = WarpShape_
 
using Element = Element_
 
using Type = InterleavedOutputTileThreadMap< layout::PitchLinearShape< Detail::WarpCount::kM, Detail::WarpCount::kN >, layout::PitchLinearShape< WarpShape::kM/Detail::kTensorOpRows, WarpShape::kN/InterleavedK >, Detail::kThreads, kElementsPerAccess, sizeof_bits< Element >::value >
 

Static Public Attributes

static int const kPartitionsK = PartitionsK
 
static int const kElementsPerAccess = ElementsPerAccess
 
static int const kInterleavedK = InterleavedK
 

Member Typedef Documentation

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::Element = Element_
template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::ThreadblockShape = ThreadblockShape_
template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::Type = InterleavedOutputTileThreadMap< layout::PitchLinearShape<Detail::WarpCount::kM, Detail::WarpCount::kN>, layout::PitchLinearShape<WarpShape::kM / Detail::kTensorOpRows, WarpShape::kN / InterleavedK>, Detail::kThreads, kElementsPerAccess, sizeof_bits<Element>::value>

ThreadMap to be used by epilogue::PredicatedTileIterator satisfying concept InterleavedOutputTileThreadMap

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
using cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::WarpShape = WarpShape_

Member Data Documentation

template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
int const cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::kElementsPerAccess = ElementsPerAccess
static
template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
int const cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::kInterleavedK = InterleavedK
static
template<typename ThreadblockShape_ , typename WarpShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess, int InterleavedK>
int const cutlass::epilogue::threadblock::DefaultInterleavedThreadMapTensorOp< ThreadblockShape_, WarpShape_, PartitionsK, Element_, ElementsPerAccess, InterleavedK >::kPartitionsK = PartitionsK
static

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