Defines sensible defaults for epilogues for TensorOps.
#include <default_epilogue_volta_tensor_op.h>
|
| using | Shape = Shape_ |
| |
| using | WarpMmaTensorOp = WarpMmaTensorOp_ |
| |
| using | OutputOp = OutputOp_ |
| |
| using | ElementOutput = typename OutputOp::ElementOutput |
| |
| using | LayoutC = typename WarpMmaTensorOp::LayoutC |
| |
| using | ElementAccumulator = typename WarpMmaTensorOp::ElementC |
| |
| using | OutputTileThreadMap = typename cutlass::epilogue::threadblock::DefaultThreadMapVoltaTensorOp< Shape, typename WarpMmaTensorOp::Shape, kPartitionsK, ElementOutput, kElementsPerAccess, ElementAccumulator >::Type |
| |
| using | OutputTileIterator = cutlass::epilogue::threadblock::PredicatedTileIterator< OutputTileThreadMap, ElementOutput > |
| |
| using | AccumulatorFragmentIterator = cutlass::epilogue::warp::FragmentIteratorVoltaTensorOp< typename WarpMmaTensorOp::Shape, gemm::GemmShape< 32, 32, 4 >, ElementAccumulator, LayoutC > |
| |
| using | WarpTileIterator = cutlass::epilogue::warp::TileIteratorVoltaTensorOp< typename WarpMmaTensorOp::Shape, gemm::GemmShape< 32, 32, 4 >, ElementAccumulator, LayoutC > |
| |
| using | SharedLoadIterator = cutlass::epilogue::threadblock::SharedLoadIterator< typename OutputTileThreadMap::CompactedThreadMap, ElementAccumulator, kSharedMemAlignment > |
| |
| using | Padding = typename WarpTileIterator::Padding |
| | Hard-coded padding elements added. More...
|
| |
| using | Epilogue = cutlass::epilogue::threadblock::Epilogue< Shape, WarpMmaTensorOp, kPartitionsK, OutputTileIterator, AccumulatorFragmentIterator, WarpTileIterator, SharedLoadIterator, OutputOp, Padding > |
| |
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
| using cutlass::epilogue::threadblock::DefaultEpilogueVoltaTensorOp< Shape_, WarpMmaTensorOp_, PartitionsK, OutputOp_, ElementsPerAccess >::Epilogue = cutlass::epilogue::threadblock::Epilogue< Shape, WarpMmaTensorOp, kPartitionsK, OutputTileIterator, AccumulatorFragmentIterator, WarpTileIterator, SharedLoadIterator, OutputOp, Padding > |
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
template<typename Shape_ , typename WarpMmaTensorOp_ , int PartitionsK, typename OutputOp_ , int ElementsPerAccess>
The documentation for this struct was generated from the following file: