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

Defines the optimal thread map for Wmma TensorOp accumulator layouts.

#include <default_thread_map_wmma_tensor_op.h>

Classes

struct  Detail
 

Public Types

using ThreadblockShape = ThreadblockShape_
 
using WarpShape = WarpShape_
 
using InstructionShape = InstructionShape_
 
using Element = Element_
 
using Type = OutputTileOptimalThreadMap< OutputTileShape< ThreadblockShape::kN, Detail::kTensorOpRows, Detail::WarpCount::kM, 1, 1 >, OutputTileShape< 1, WarpShape::kM/Detail::kTensorOpRows, 1, 1, WarpShape::kM/Detail::kTensorOpRows >, Detail::kThreads, kElementsPerAccess, sizeof_bits< Element >::value >
 ThreadMap to be used by epilogue::PredicatedTileIterator satisfying concept OutputTileThreadMap. More...
 

Static Public Attributes

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

Member Typedef Documentation

template<typename ThreadblockShape_ , typename WarpShape_ , typename InstructionShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess>
using cutlass::epilogue::threadblock::DefaultThreadMapWmmaTensorOp< ThreadblockShape_, WarpShape_, InstructionShape_, PartitionsK, Element_, ElementsPerAccess >::Element = Element_
template<typename ThreadblockShape_ , typename WarpShape_ , typename InstructionShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess>
using cutlass::epilogue::threadblock::DefaultThreadMapWmmaTensorOp< ThreadblockShape_, WarpShape_, InstructionShape_, PartitionsK, Element_, ElementsPerAccess >::InstructionShape = InstructionShape_
template<typename ThreadblockShape_ , typename WarpShape_ , typename InstructionShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess>
using cutlass::epilogue::threadblock::DefaultThreadMapWmmaTensorOp< ThreadblockShape_, WarpShape_, InstructionShape_, PartitionsK, Element_, ElementsPerAccess >::ThreadblockShape = ThreadblockShape_
template<typename ThreadblockShape_ , typename WarpShape_ , typename InstructionShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess>
using cutlass::epilogue::threadblock::DefaultThreadMapWmmaTensorOp< ThreadblockShape_, WarpShape_, InstructionShape_, PartitionsK, Element_, ElementsPerAccess >::Type = OutputTileOptimalThreadMap < OutputTileShape<ThreadblockShape::kN, Detail::kTensorOpRows, Detail::WarpCount::kM, 1, 1>, OutputTileShape<1, WarpShape::kM / Detail::kTensorOpRows, 1, 1, WarpShape::kM / Detail::kTensorOpRows>, Detail::kThreads, kElementsPerAccess, sizeof_bits<Element>::value >
template<typename ThreadblockShape_ , typename WarpShape_ , typename InstructionShape_ , int PartitionsK, typename Element_ , int ElementsPerAccess>
using cutlass::epilogue::threadblock::DefaultThreadMapWmmaTensorOp< ThreadblockShape_, WarpShape_, InstructionShape_, PartitionsK, Element_, ElementsPerAccess >::WarpShape = WarpShape_

Member Data Documentation

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

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