CUTLASS: cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor > Class Template Reference

CUTLASS

CUDA Templates for Linear Algebra Subroutines and Solvers

Template for reading and writing tiles of accumulators to shared memory.

#include <tile_iterator_wmma_tensor_op.h>

Public Types
using	WarpShape = WarpShape_

using	OperatorShape = OperatorShape_

using	OperatorFragment = OperatorFragment_

using	Layout = layout::RowMajor

using	WmmaDataType = typename OperatorFragment::element_type

using	Element = typename cutlass::arch::WmmaToCutlassDataType< WmmaDataType >::Type
	Data Type of element stored in nvcuda::wmma::frament. More...

using	TensorRef = TensorRef< Element, Layout >
	Tensor Reference object. More...

using	TensorCoord = MatrixCoord
	Logical coordinate in referenced tensor. More...

using	Index = typename TensorRef::Index

using	LongIndex = typename TensorRef::LongIndex

using	Policy = WmmaTensorOpPolicy< WarpShape, OperatorShape, Layout >

using	Shape = MatrixShape< Policy::kRowsPerIteration, WarpShape::kN >
	Shape of the tile in memory. More...

using	Fragment = WmmaFragmentArray< OperatorFragment, Policy::OperatorCount::kColumn *Policy::kWmmaFragmentsPerAccess >
	This is the fragment size produced by one access of the iterator. More...

using	Padding = MatrixShape< 0, 4 *Policy::kElementsPerAccess >
	This is the complete warp-level accumulator tile. More...

Public Member Functions
CUTLASS_HOST_DEVICE	TileIteratorWmmaTensorOp ()
	Default constructor. More...

CUTLASS_HOST_DEVICE	TileIteratorWmmaTensorOp (TensorRef const &ref, unsigned lane_id)
	Constructor from TensorRef. More...

CUTLASS_HOST_DEVICE TileIteratorWmmaTensorOp &	add_pointer_offset (Index pointer_offset)
	Adds a pointer offset. More...

CUTLASS_HOST_DEVICE TileIteratorWmmaTensorOp &	add_tile_offset (TensorCoord const &tile_offset)
	advances in units of whole tiles along the logical coordinate space of the tensor More...

CUTLASS_HOST_DEVICE TileIteratorWmmaTensorOp &	operator+= (TensorCoord const &tile_offset)

CUTLASS_HOST_DEVICE void	store_with_pointer_offset (Fragment const &frag, Index pointer_offset)
	Store. More...

CUTLASS_HOST_DEVICE void	store (Fragment const &frag)
	Store. More...

CUTLASS_HOST_DEVICE void	load_with_pointer_offset (Fragment &frag, Index pointer_offset) const
	Load. More...

CUTLASS_HOST_DEVICE void	load (Fragment &frag) const
	Load. More...

Member Typedef Documentation

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Element = typename cutlass::arch::WmmaToCutlassDataType<WmmaDataType>::Type

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Fragment = WmmaFragmentArray<OperatorFragment, Policy::OperatorCount::kColumn * Policy::kWmmaFragmentsPerAccess>

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Index = typename TensorRef::Index

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Layout = layout::RowMajor

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::LongIndex = typename TensorRef::LongIndex

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::OperatorFragment = OperatorFragment_

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::OperatorShape = OperatorShape_

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Padding = MatrixShape< 0, 4 * Policy::kElementsPerAccess >

Padding quantity

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Policy = WmmaTensorOpPolicy<WarpShape, OperatorShape, Layout>

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::Shape = MatrixShape< Policy::kRowsPerIteration, WarpShape::kN >

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::TensorCoord = MatrixCoord

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::TensorRef = TensorRef<Element, Layout>

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::WarpShape = WarpShape_

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

using cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::WmmaDataType = typename OperatorFragment::element_type

Constructor & Destructor Documentation

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::TileIteratorWmmaTensorOp ( )

inline

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::TileIteratorWmmaTensorOp	(	TensorRef const &	ref,
		unsigned	lane_id
	)

inline

Member Function Documentation

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE TileIteratorWmmaTensorOp& cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::add_pointer_offset ( Index pointer_offset )

inline

advances in units of whole tiles along the logical coordinate space of the tensor

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE TileIteratorWmmaTensorOp& cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::add_tile_offset ( TensorCoord const & tile_offset )

inline

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE void cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::load ( Fragment & frag ) const

inline

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE void cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::load_with_pointer_offset	(	Fragment &	frag,
		Index	pointer_offset
	)		const

inline

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE TileIteratorWmmaTensorOp& cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::operator+= ( TensorCoord const & tile_offset )

inline

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE void cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::store ( Fragment const & frag )

inline

template<typename WarpShape_ , typename OperatorShape_ , typename OperatorFragment_ >

CUTLASS_HOST_DEVICE void cutlass::epilogue::warp::TileIteratorWmmaTensorOp< WarpShape_, OperatorShape_, OperatorFragment_, layout::RowMajor >::store_with_pointer_offset	(	Fragment const &	frag,
		Index	pointer_offset
	)

inline

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

tile_iterator_wmma_tensor_op.h