Template Numerical Library: TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal

Implementation of sparse matrix view. More...

#include <TNL/Matrices/SparseMatrixView.h>

Inheritance diagram for TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >:

[legend]

Collaboration diagram for TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >:

[legend]

Public Types
using	ConstViewType = SparseMatrixView< std::add_const_t< Real >, Device, Index, MatrixType, SegmentsViewTemplate >
	Matrix view type for constant instances.

template<typename Device_ , typename Index_ >
using	SegmentsViewTemplate = SegmentsView< Device_, Index_ >
	Templated type of segments view, i.e. sparse matrix format.

template<typename _Real = Real, typename _Device = Device, typename _Index = Index, typename _MatrixType = MatrixType, template< typename, typename > class _SegmentsView = SegmentsViewTemplate, typename _ComputeReal = ComputeReal>
using	Self = SparseMatrixView< _Real, _Device, _Index, _MatrixType, _SegmentsView, _ComputeReal >
	Helper type for getting self type or its modifications.

using	ViewType = SparseMatrixView< Real, Device, Index, MatrixType, SegmentsViewTemplate >
	Type of related matrix view.

Public Types inherited from TNL::Matrices::SparseMatrixBase< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >
using	ColumnIndexesViewType

using	ComputeRealType = ComputeReal

using	ConstRowView = typename RowView::ConstRowView
	Type for accessing constant matrix rows.

using	DefaultSegmentsReductionKernel = typename Algorithms::SegmentsReductionKernels::DefaultKernel< SegmentsView >::type
	Type of the kernel used for parallel reductions on segments.

using	DeviceType = Device
	The device where the matrix is allocated.

using	IndexType = Index
	The type used for matrix elements indexing.

using	RealType = typename Base::RealType
	The type of matrix elements.

using	RowView
	Type for accessing matrix rows.

using	SegmentsViewType = SegmentsView
	Type of segments view used by this matrix. It represents the sparse matrix format.

Public Types inherited from TNL::Matrices::MatrixBase< Real, Device, Index, MatrixType, SegmentsView::getOrganization() >
using	ConstValuesViewType
	Type of constant vector view holding values of matrix elements.

using	DeviceType
	The device where the matrix is allocated.

using	IndexType
	The type used for matrix elements indexing.

using	RealType
	The type of matrix elements.

using	RowCapacitiesType

using	ValuesViewType
	Type of vector view holding values of matrix elements.

Public Member Functions
__cuda_callable__	SparseMatrixView ()=default
	Constructor with no parameters.

__cuda_callable__	SparseMatrixView (const SparseMatrixView &matrix)=default
	Copy constructor.

__cuda_callable__	SparseMatrixView (Index rows, Index columns, typename Base::ValuesViewType values, typename Base::ColumnIndexesViewType columnIndexes, typename Base::SegmentsViewType segments)
	Constructor with all necessary data and views.

__cuda_callable__	SparseMatrixView (SparseMatrixView &&matrix) noexcept=default
	Move constructor.

__cuda_callable__ void	bind (SparseMatrixView &&view)
	Method for rebinding (reinitialization) using another sparse matrix view.

__cuda_callable__ void	bind (SparseMatrixView &view)
	Method for rebinding (reinitialization) using another sparse matrix view.

__cuda_callable__ ConstViewType	getConstView () const
	Returns a non-modifiable view of the sparse matrix.

__cuda_callable__ ViewType	getView ()
	Returns a modifiable view of the sparse matrix.

SparseMatrixView &	operator= (const SparseMatrixView &)=delete
	Copy-assignment operator.

void	save (File &file) const
	Method for saving the matrix to a file.

Public Member Functions inherited from TNL::Matrices::SparseMatrixBase< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >
__cuda_callable__	SparseMatrixBase ()=default
	Constructor with no parameters.

__cuda_callable__	SparseMatrixBase (const SparseMatrixBase &matrix)=default
	Copy constructor.

__cuda_callable__	SparseMatrixBase (IndexType rows, IndexType columns, typename Base::ValuesViewType values, ColumnIndexesViewType columnIndexes, SegmentsViewType segments)
	Constructor with all necessary data and views.

__cuda_callable__	SparseMatrixBase (SparseMatrixBase &&matrix) noexcept=default
	Move constructor.

__cuda_callable__ void	addElement (IndexType row, IndexType column, const RealType &value, const RealType &thisElementMultiplicator=1.0)
	Add element at given row and column to given value.

__cuda_callable__ IndexType	findElement (IndexType row, IndexType column) const
	Finds element in the matrix and returns its position in the arrays with values and columnIndexes.

template<typename Function >
void	forAllElements (Function &&function)
	This method calls forElements for all matrix rows.

template<typename Function >
void	forAllElements (Function &&function) const
	This method calls forElements for all matrix rows (for constant instances).

template<typename Function >
void	forAllRows (Function &&function)
	Method for parallel iteration over all matrix rows.

template<typename Function >
void	forAllRows (Function &&function) const
	Method for parallel iteration over all matrix rows for constant instances.

template<typename Function >
void	forElements (IndexType begin, IndexType end, Function &&function)
	Method for iteration over all matrix rows for non-constant instances.

template<typename Function >
void	forElements (IndexType begin, IndexType end, Function &&function) const
	Method for iteration over all matrix rows for constant instances.

template<typename Function >
void	forRows (IndexType begin, IndexType end, Function &&function)
	Method for parallel iteration over matrix rows from interval `[begin, end)`.

template<typename Function >
void	forRows (IndexType begin, IndexType end, Function &&function) const
	Method for parallel iteration over matrix rows from interval `[begin, end)` for constant instances.

ColumnIndexesViewType &	getColumnIndexes ()
	Getter of column indexes for nonconstant instances.

const ColumnIndexesViewType &	getColumnIndexes () const
	Getter of column indexes for constant instances.

template<typename Vector >
void	getCompressedRowLengths (Vector &rowLengths) const
	Computes number of non-zeros in each row.

__cuda_callable__ RealType	getElement (IndexType row, IndexType column) const
	Returns value of matrix element at position given by its row and column index.

IndexType	getNonzeroElementsCount () const override
	Returns number of non-zero matrix elements.

__cuda_callable__ RowView	getRow (IndexType rowIdx)
	Non-constant getter of simple structure for accessing given matrix row.

__cuda_callable__ ConstRowView	getRow (IndexType rowIdx) const
	Constant getter of simple structure for accessing given matrix row.

template<typename Vector >
void	getRowCapacities (Vector &rowCapacities) const
	Compute capacities of all rows.

__cuda_callable__ IndexType	getRowCapacity (IndexType row) const
	Returns capacity of given matrix row.

SegmentsViewType &	getSegments ()
	Getter of segments for non-constant instances.

const SegmentsViewType &	getSegments () const
	Getter of segments for constant instances.

template<typename Matrix >
bool	operator!= (const Matrix &matrix) const
	Comparison operator with another arbitrary matrix type.

SparseMatrixBase &	operator= (const SparseMatrixBase &)=delete
	Copy-assignment operator.

SparseMatrixBase &	operator= (SparseMatrixBase &&)=delete
	Move-assignment operator.

template<typename Matrix >
bool	operator== (const Matrix &matrix) const
	Comparison operator with another arbitrary matrix type.

void	print (std::ostream &str) const
	Method for printing the matrix to output stream.

template<typename Fetch , typename Reduce , typename Keep , typename FetchValue , typename SegmentsReductionKernel = DefaultSegmentsReductionKernel>
void	reduceAllRows (Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchValue &identity, const SegmentsReductionKernel &kernel=SegmentsReductionKernel{}) const
	Method for performing general reduction on all matrix rows for constant instances.

template<typename Fetch , typename Reduce , typename Keep , typename FetchValue , typename SegmentsReductionKernel = DefaultSegmentsReductionKernel>
void	reduceRows (IndexType begin, IndexType end, Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchValue &identity, const SegmentsReductionKernel &kernel=SegmentsReductionKernel{}) const
	Method for performing general reduction on matrix rows for constant instances.

template<typename Function >
void	sequentialForAllRows (Function &&function)
	This method calls sequentialForRows for all matrix rows.

template<typename Function >
void	sequentialForAllRows (Function &&function) const
	This method calls sequentialForRows for all matrix rows (for constant instances).

template<typename Function >
void	sequentialForRows (IndexType begin, IndexType end, Function &&function)
	Method for sequential iteration over all matrix rows for non-constant instances.

template<typename Function >
void	sequentialForRows (IndexType begin, IndexType end, Function &&function) const
	Method for sequential iteration over all matrix rows for constant instances.

__cuda_callable__ void	setElement (IndexType row, IndexType column, const RealType &value)
	Sets element at given row and column to given value.

void	sortColumnIndexes ()
	Sort matrix elements in each row by column indexes in ascending order.

template<typename InVector , typename OutVector >
void	transposedVectorProduct (const InVector &inVector, OutVector &outVector, ComputeReal matrixMultiplicator=1.0, ComputeReal outVectorMultiplicator=0.0, Index begin=0, Index end=0) const
	Computes product of transposed matrix and vector.

template<typename InVector , typename OutVector , typename SegmentsReductionKernel = DefaultSegmentsReductionKernel>
void	vectorProduct (const InVector &inVector, OutVector &outVector, ComputeRealType matrixMultiplicator=1.0, ComputeRealType outVectorMultiplicator=0.0, IndexType begin=0, IndexType end=0, const SegmentsReductionKernel &kernel=SegmentsReductionKernel{}) const
	Computes product of matrix and vector.

template<typename InVector , typename OutVector , typename SegmentsReductionKernel , typename... , std::enable_if_t< ! std::is_convertible_v< SegmentsReductionKernel, ComputeRealType >, bool > = true>
void	vectorProduct (const InVector &inVector, OutVector &outVector, const SegmentsReductionKernel &kernel) const

Public Member Functions inherited from TNL::Matrices::MatrixBase< Real, Device, Index, MatrixType, SegmentsView::getOrganization() >
__cuda_callable__	MatrixBase ()=default
	Basic constructor with no parameters.

__cuda_callable__	MatrixBase (const MatrixBase &view)=default
	Shallow copy constructor.

__cuda_callable__	MatrixBase (IndexType rows, IndexType columns, ValuesViewType values)
	Constructor with matrix dimensions and matrix elements values.

__cuda_callable__	MatrixBase (MatrixBase &&view) noexcept=default
	Move constructor.

IndexType	getAllocatedElementsCount () const
	Tells the number of allocated matrix elements.

__cuda_callable__ IndexType	getColumns () const
	Returns number of matrix columns.

__cuda_callable__ IndexType	getRows () const
	Returns number of matrix rows.

__cuda_callable__ ValuesViewType &	getValues ()
	Returns a reference to a vector with the matrix elements values.

__cuda_callable__ const ValuesViewType &	getValues () const
	Returns a constant reference to a vector with the matrix elements values.

bool	operator!= (const Matrix &matrix) const
	Comparison operator with another arbitrary matrix view type.

bool	operator!= (const MatrixT &matrix) const

__cuda_callable__ MatrixBase &	operator= (const MatrixBase &)=delete
	Copy-assignment operator.

__cuda_callable__ MatrixBase &	operator= (MatrixBase &&)=delete
	Move-assignment operator.

bool	operator== (const Matrix &matrix) const
	Comparison operator with another arbitrary matrix view type.

bool	operator== (const MatrixT &matrix) const

Additional Inherited Members
Static Public Member Functions inherited from TNL::Matrices::SparseMatrixBase< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >
static std::string	getSerializationType ()
	Returns string with serialization type.

Static Public Member Functions inherited from TNL::Matrices::MatrixBase< Real, Device, Index, MatrixType, SegmentsView::getOrganization() >
static constexpr ElementsOrganization	getOrganization ()
	Matrix elements organization getter.

static constexpr bool	isBinary ()
	Test of binary matrix type.

static constexpr bool	isMatrix ()
	Test of matrix type.

static constexpr bool	isSymmetric ()
	Test of symmetric matrix type.

Protected Member Functions inherited from TNL::Matrices::SparseMatrixBase< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >
__cuda_callable__ void	bind (IndexType rows, IndexType columns, typename Base::ValuesViewType values, ColumnIndexesViewType columnIndexes, SegmentsViewType segments)
	Re-initializes the internal attributes of the base class.

Protected Member Functions inherited from TNL::Matrices::MatrixBase< Real, Device, Index, MatrixType, SegmentsView::getOrganization() >
__cuda_callable__ void	bind (IndexType rows, IndexType columns, ValuesViewType values)
	Re-initializes the internal attributes of the base class.

Protected Attributes inherited from TNL::Matrices::SparseMatrixBase< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >
ColumnIndexesViewType	columnIndexes

SegmentsViewType	segments

Protected Attributes inherited from TNL::Matrices::MatrixBase< Real, Device, Index, MatrixType, SegmentsView::getOrganization() >
IndexType	columns

IndexType	rows

ValuesViewType	values

Detailed Description

template<typename Real, typename Device = Devices::Host, typename Index = int, typename MatrixType = GeneralMatrix, template< typename Device_, typename Index_ > class SegmentsView = Algorithms::Segments::CSRView, typename ComputeReal = typename ChooseSparseMatrixComputeReal< Real, Index >::type>
class TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >

Implementation of sparse matrix view.

This is to prevent from appearing in Doxygen documentation.

It serves as an accessor to SparseMatrix for example when passing the matrix to lambda functions. SparseMatrix view can be also created in CUDA kernels.

Template Parameters

Real	is a type of matrix elements. If Real equals bool the matrix is treated as binary and so the matrix elements values are not stored in the memory since we need to remember only coordinates of non-zero elements (which equal one).
Device	is a device where the matrix is allocated.
Index	is a type for indexing of the matrix elements.
MatrixType	specifies a symmetry of matrix. See MatrixType. Symmetric matrices store only lower part of the matrix and its diagonal. The upper part is reconstructed on the fly. GeneralMatrix with no symmetry is used by default.
Segments	is a structure representing the sparse matrix format. Depending on the pattern of the non-zero elements different matrix formats can perform differently especially on GPUs. By default Algorithms::Segments::CSR format is used. See also Algorithms::Segments::Ellpack, Algorithms::Segments::SlicedEllpack, Algorithms::Segments::ChunkedEllpack, and Algorithms::Segments::BiEllpack.
ComputeReal	is the same as Real mostly but for binary matrices it is set to Index type. This can be changed by the user, of course.

Constructor & Destructor Documentation

◆ SparseMatrixView() [1/3]

template<typename Real , typename Device , typename Index , typename MatrixType , template< typename, typename > class SegmentsView, typename ComputeReal >

__cuda_callable__ TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::SparseMatrixView	(	Index	rows,
		Index	columns,
		typename Base::ValuesViewType	values,
		typename Base::ColumnIndexesViewType	columnIndexes,
		typename Base::SegmentsViewType	segments )

Constructor with all necessary data and views.

Parameters

rows	is a number of matrix rows.
columns	is a number of matrix columns.
values	is a vector view with matrix elements values.
columnIndexes	is a vector view with matrix elements column indexes.
segments	is a segments view representing the sparse matrix format.

◆ SparseMatrixView() [2/3]

template<typename Real , typename Device = Devices::Host, typename Index = int, typename MatrixType = GeneralMatrix, template< typename Device_, typename Index_ > class SegmentsView = Algorithms::Segments::CSRView, typename ComputeReal = typename ChooseSparseMatrixComputeReal< Real, Index >::type>

__cuda_callable__ TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::SparseMatrixView ( const SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal > & matrix )

default

Copy constructor.

Parameters

matrix is an input sparse matrix view.

◆ SparseMatrixView() [3/3]

template<typename Real , typename Device = Devices::Host, typename Index = int, typename MatrixType = GeneralMatrix, template< typename Device_, typename Index_ > class SegmentsView = Algorithms::Segments::CSRView, typename ComputeReal = typename ChooseSparseMatrixComputeReal< Real, Index >::type>

__cuda_callable__ TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::SparseMatrixView ( SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal > && matrix )

defaultnoexcept

Move constructor.

Parameters

matrix is an input sparse matrix view.

Member Function Documentation

◆ bind() [1/2]

template<typename Real , typename Device , typename Index , typename MatrixType , template< typename, typename > class SegmentsView, typename ComputeReal >

__cuda_callable__ void TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::bind ( SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal > && view )

Method for rebinding (reinitialization) using another sparse matrix view.

Parameters

view	The sparse matrix view to be bound.

◆ bind() [2/2]

template<typename Real , typename Device , typename Index , typename MatrixType , template< typename, typename > class SegmentsView, typename ComputeReal >

Method for rebinding (reinitialization) using another sparse matrix view.

Parameters

view	The sparse matrix view to be bound.

◆ getConstView()

template<typename Real , typename Device , typename Index , typename MatrixType , template< typename, typename > class SegmentsView, typename ComputeReal >

__cuda_callable__ auto TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::getConstView ( ) const

Returns a non-modifiable view of the sparse matrix.

Returns: sparse matrix view.

◆ getView()

template<typename Real , typename Device , typename Index , typename MatrixType , template< typename, typename > class SegmentsView, typename ComputeReal >

__cuda_callable__ auto TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::getView ( )

Returns a modifiable view of the sparse matrix.

Returns: sparse matrix view.

◆ operator=()

template<typename Real , typename Device = Devices::Host, typename Index = int, typename MatrixType = GeneralMatrix, template< typename Device_, typename Index_ > class SegmentsView = Algorithms::Segments::CSRView, typename ComputeReal = typename ChooseSparseMatrixComputeReal< Real, Index >::type>

SparseMatrixView & TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::operator= ( const SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal > & )

delete

Copy-assignment operator.

It is a deleted function, because sparse matrix assignment in general requires reallocation.

◆ save()

template<typename Real , typename Device , typename Index , typename MatrixType , template< typename, typename > class SegmentsView, typename ComputeReal >

void TNL::Matrices::SparseMatrixView< Real, Device, Index, MatrixType, SegmentsView, ComputeReal >::save ( File & file ) const

Method for saving the matrix to a file.

Parameters

file	is the output file.

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

src/TNL/Matrices/SparseMatrixView.h
src/TNL/Matrices/SparseMatrixView.hpp

Public Types

Public Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ SparseMatrixView() [1/3]

◆ SparseMatrixView() [2/3]

◆ SparseMatrixView() [3/3]

Member Function Documentation

◆ bind() [1/2]

◆ bind() [2/2]

◆ getConstView()

◆ getView()

◆ operator=()

◆ save()