default_epilogue_wmma_tensor_op.h File Reference

Epilogue for threadblock scoped GEMMs using Tensor Ops. More...

#include "cutlass/cutlass.h"
#include "cutlass/numeric_types.h"
#include "cutlass/array.h"
#include "cutlass/gemm/gemm.h"
#include "cutlass/epilogue/thread/linear_combination.h"
#include "cutlass/epilogue/thread/conversion_op.h"
#include "cutlass/epilogue/thread/reduction_op.h"
#include "cutlass/transform/threadblock/regular_tile_iterator_pitch_linear.h"
#include "cutlass/epilogue/warp/fragment_iterator_wmma_tensor_op.h"
#include "cutlass/epilogue/warp/tile_iterator_wmma_tensor_op.h"
#include "cutlass/epilogue/threadblock/default_thread_map_wmma_tensor_op.h"
#include "cutlass/epilogue/threadblock/predicated_tile_iterator.h"
#include "cutlass/epilogue/threadblock/shared_load_iterator.h"
#include "cutlass/epilogue/threadblock/epilogue.h"

Include dependency graph for default_epilogue_wmma_tensor_op.h:

Go to the source code of this file.

Classes
struct	cutlass::epilogue::threadblock::DefaultEpilogueWmmaTensorOp< Shape_, WarpMmaTensorOp_, PartitionsK, OutputOp_, ElementsPerAccess >
	Defines sensible defaults for epilogues for WMMA TensorOps. More...

Namespaces
	cutlass
	cutlass::epilogue
	cutlass::epilogue::threadblock

Detailed Description

The epilogue rearranges the result of a matrix product through shared memory to match canonical tensor layouts in global memory. Epilogues support conversion and reduction operations.