#include <mma_tensor_op_tile_iterator_sm70.h>

Public Types
using	Shape = Shape_
	Shape of tile to load (concept: PitchLinearShape) More...

using	Element = Element_
	Element type. More...

using	Layout = cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, kKBlock >
	Layout of source tile. More...

using	InstructionShape = InstructionShape_
	Shape of one matrix product operation (concept: MatrixShape) More...

using	TensorRef = TensorRef< Element, Layout >
	TensorRef type for loading element from a tensor. More...

using	Index = typename TensorRef::Index
	Index type. More...

using	LongIndex = typename TensorRef::LongIndex
	Long Index type. More...

using	TensorCoord = typename TensorRef::TensorCoord
	Coordinate for an element in the tensor. More...

using	Base = MmaVoltaTensorOpMultiplicandTileIterator< layout::PitchLinearShape< Shape::kColumn, Shape::kRow >, kOperand, Element, layout::VoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, kKBlock >, layout::PitchLinearShape< InstructionShape::kColumn, InstructionShape::kRow >, kOpDelta, kThreads >
	Underlying tile iterator implementation. More...

using	Fragment = Array< Element, Shape::kCount/kThreads *2 >
	Fragment object holding a thread's part of a tile. More...

Public Member Functions
CUTLASS_HOST_DEVICE	MmaVoltaTensorOpMultiplicandTileIterator ()
	Default ctor constructs null iterator. More...

CUTLASS_HOST_DEVICE	MmaVoltaTensorOpMultiplicandTileIterator (TensorRef const &ref, int lane_id)
	Constructor from TensorRef. More...

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator &	add_pointer_offset (LongIndex offset)
	Adds a pointer offset to internal pointer(s) to advance through memory. More...

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator &	add_tile_offset (TensorCoord const &tile_offset)

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator &	operator++ ()
	Advances the iterator along the advance dimension. More...

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator &	operator-- ()
	Advances the iterator along the advance dimension. More...

CUTLASS_DEVICE MmaVoltaTensorOpMultiplicandTileIterator &	operator+= (TensorCoord const &tile_offset)

CUTLASS_DEVICE MmaVoltaTensorOpMultiplicandTileIterator &	operator-= (TensorCoord const &tile_offset)

CUTLASS_HOST_DEVICE void	load (Fragment &frag) const
	Loads a fragment from memory at the location pointed to by the iterator. More...

CUTLASS_DEVICE void	load_with_pointer_offset (Fragment &frag, Index pointer_offset) const
	Loads a fragment from memory with additional logical offset. More...

CUTLASS_DEVICE void	load_with_byte_offset (Fragment &frag, Index byte_offset) const
	Loads a fragment from memory with additional logical offset. More...

CUTLASS_DEVICE void	load (Fragment &frag, TensorCoord const &tile_offset) const
	Loads a fragment from memory with logical offset in units of whole tiles. More...

CUTLASS_DEVICE void	load (Fragment &frag, TensorCoord const &tile_offset, Index pointer_offset) const
	Loads a fragment from memory with logical offset in units of whole tiles. More...

CUTLASS_DEVICE void	load_with_byte_offset (Fragment &frag, TensorCoord const &tile_offset, Index byte_offset) const
	Loads a fragment from memory with logical offset in units of whole tiles. More...

CUTLASS_DEVICE void	set_kgroup_index (int k_group)

Static Public Attributes
static Operand const	kOperand = Operand_
	Operand tag. More...

static int const	kKBlock = KBlock
	KBlock size. More...

static int const	kOpDelta = OpDelta_

static int const	kThreads = 32
	Number of participating threads. More...

Detailed Description

template<typename Shape_, Operand Operand_, typename Element_, typename InstructionShape_, int OpDelta_, int KBlock>
class cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >

This tile iterator is specialized for 32-thread TensorOps. It uses LDS to load from shared memory and therefore must be initialized with a TensorRef to shared memory.

Satisfies: ReadableRandomAccessContiguousTileIteratorConcept

Member Typedef Documentation

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::Base = MmaVoltaTensorOpMultiplicandTileIterator< layout::PitchLinearShape<Shape::kColumn, Shape::kRow>, kOperand, Element, layout::VoltaTensorOpMultiplicandCrosswise<sizeof_bits<Element_>::value, kKBlock>, layout::PitchLinearShape<InstructionShape::kColumn, InstructionShape::kRow>, kOpDelta, kThreads>

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::Element = Element_

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::Fragment = Array<Element, Shape::kCount / kThreads * 2>

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::Index = typename TensorRef::Index

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::InstructionShape = InstructionShape_

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::Layout = cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits<Element_>::value, kKBlock>

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::LongIndex = typename TensorRef::LongIndex

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::Shape = Shape_

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::TensorCoord = typename TensorRef::TensorCoord

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

using cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::TensorRef = TensorRef<Element, Layout>

Constructor & Destructor Documentation

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::MmaVoltaTensorOpMultiplicandTileIterator ( )

inline

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::MmaVoltaTensorOpMultiplicandTileIterator	(	TensorRef const &	ref,
		int	lane_id
	)

inline

Member Function Documentation

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator& cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::add_pointer_offset ( LongIndex offset )

inline

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator& cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::add_tile_offset ( TensorCoord const & tile_offset )

inline

Advances an iterator along logical dimensions of matrix in units of whole tiles

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::load ( Fragment & frag ) const

inline

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::load	(	Fragment &	frag,
		TensorCoord const &	tile_offset
	)		const

inline

Parameters

frag	fragment to load from the tensor
tile_offset	loads a tile with a logical offset in units of whole tiles

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::load	(	Fragment &	frag,
		TensorCoord const &	tile_offset,
		Index	pointer_offset
	)		const

inline

Parameters

frag	fragment to load from the tensor
tile_offset	loads a tile with a logical offset in units of whole tiles
pointer_offset	loads a tile with a logical offset AND a pointer offset

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::load_with_byte_offset	(	Fragment &	frag,
		Index	byte_offset
	)		const

inline

Parameters

frag	fragment to load from the tensor
byte_offset	loads a tile with a linear offset

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::load_with_byte_offset	(	Fragment &	frag,
		TensorCoord const &	tile_offset,
		Index	byte_offset
	)		const

inline

Parameters

frag	fragment to load from the tensor
tile_offset	loads a tile with a logical offset in units of whole tiles
byte_offset	loads a tile with a logical offset AND a pointer offset

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::load_with_pointer_offset	(	Fragment &	frag,
		Index	pointer_offset
	)		const

inline

Parameters

frag	fragment to load from the tensor
pointer_offset	loads a tile with a linear offset

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator& cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::operator++ ( )

inline

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE MmaVoltaTensorOpMultiplicandTileIterator& cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::operator+= ( TensorCoord const & tile_offset )

inline

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

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_HOST_DEVICE MmaVoltaTensorOpMultiplicandTileIterator& cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::operator-- ( )

inline

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

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE MmaVoltaTensorOpMultiplicandTileIterator& cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::operator-= ( TensorCoord const & tile_offset )

inline

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

CUTLASS_DEVICE void cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::set_kgroup_index ( int k_group )

inline

Notify the iterator which k-group it is currently pointing to.

This does not advance the iterator. Rather, it overrides its internal tracking with constant-valued k-group index to enable the compiler to fold constants and achieve more efficient code.

This is used by some nontrivial permuted layouts.

Member Data Documentation

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

int const cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::kKBlock = KBlock

static

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

int const cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::kOpDelta = OpDelta_

static

Delta between *MMA operations (in units of *MMA operations, concept: MatrixShape)

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

Operand const cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::kOperand = Operand_

static

template<typename Shape_ , Operand Operand_, typename Element_ , typename InstructionShape_ , int OpDelta_, int KBlock>

int const cutlass::gemm::warp::MmaVoltaTensorOpMultiplicandTileIterator< Shape_, Operand_, Element_, cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise< sizeof_bits< Element_ >::value, KBlock >, InstructionShape_, OpDelta_, 32 >::kThreads = 32

static

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

mma_tensor_op_tile_iterator_sm70.h

Public Types

Public Member Functions

Static Public Attributes

Detailed Description

Member Typedef Documentation

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation