Template Numerical Library: TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator

Implementation of dense matrix, i.e. matrix storing explicitly all of its elements including zeros. More...

#include <TNL/Matrices/DenseMatrix.h>

Inheritance diagram for TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >:

Collaboration diagram for TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >:

[legend]

Public Types
using	ConstRowView = typename ConstViewType::ConstRowView
	Type for accessing immutable matrix rows.

using	ConstViewType = typename DenseMatrixView< Real, Device, Index, Organization >::ConstViewType
	Matrix view type for constant instances. More...

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

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

using	RealAllocatorType = RealAllocator
	The allocator for matrix elements.

using	RealType = Real
	The type of matrix elements.

using	RowView = typename ViewType::RowView
	Type for accessing matrix rows.

template<typename _Real = Real, typename _Device = Device, typename _Index = Index, ElementsOrganization _Organization = Algorithms::Segments::DefaultElementsOrganization< _Device >::getOrganization(), typename _RealAllocator = typename Allocators::Default< _Device >::template Allocator< _Real >>
using	Self = DenseMatrix< _Real, _Device, _Index, _Organization, _RealAllocator >
	Helper type for getting self type or its modifications.

using	ViewType = DenseMatrixView< Real, Device, Index, Organization >
	Type of related matrix view. More...

Public Types inherited from TNL::Matrices::Matrix< Real, Device, Index, RealAllocator >
using	ConstRowsCapacitiesView = typename RowsCapacitiesView::ConstViewType

using	ConstValuesType = Containers::Vector< std::add_const_t< Real >, Device, Index, RealAllocator >
	Type of constant vector holding values of matrix elements.

using	ConstValuesView = typename ViewType::ConstValuesView
	Type of constant vector view holding values of matrix elements.

using	ConstViewType = typename MatrixView< Real, Device, Index >::ConstViewType
	Type of base matrix view for constant instances. More...

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

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

using	RealAllocatorType = RealAllocator

using	RealType = Real
	The type of matrix elements.

using	RowsCapacitiesType = Containers::Vector< Index, Device, Index >

using	RowsCapacitiesView = Containers::VectorView< Index, Device, Index >

using	ValuesType = Containers::Vector< Real, Device, Index, RealAllocator >
	Type of vector holding values of matrix elements.

using	ValuesView = typename ViewType::ValuesView
	Type of vector view holding values of matrix elements.

using	ViewType = MatrixView< Real, Device, Index >
	Type of base matrix view. More...

Public Member Functions
	DenseMatrix (const DenseMatrix &matrix)=default
	Copy constructor. More...

	DenseMatrix (const RealAllocatorType &allocator=RealAllocatorType())
	Constructor only with values allocator. More...

	DenseMatrix (DenseMatrix &&matrix) noexcept=default
	Move constructor. More...

	DenseMatrix (IndexType rows, IndexType columns, const RealAllocatorType &allocator=RealAllocatorType())
	Constructor with matrix dimensions. More...

template<typename Value >
	DenseMatrix (std::initializer_list< std::initializer_list< Value > > data, const RealAllocatorType &allocator=RealAllocatorType())
	Constructor with 2D initializer list. More...

__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. More...

template<typename Matrix >
void	addMatrix (const Matrix &matrix, const RealType &matrixMultiplicator=1.0, const RealType &thisMatrixMultiplicator=1.0)

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

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

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

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

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

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

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

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. More...

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

ConstViewType	getConstView () const
	Returns a non-modifiable view of the dense matrix. More...

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

template<typename Matrix1 , typename Matrix2 , int tileDim = 32>
void	getMatrixProduct (const Matrix1 &matrix1, const Matrix2 &matrix2, const RealType &matrixMultiplicator=1.0)

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

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

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

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

std::string	getSerializationTypeVirtual () const override
	Returns string with serialization type. More...

template<typename Matrix , int tileDim = 32>
void	getTransposition (const Matrix &matrix, const RealType &matrixMultiplicator=1.0)

ViewType	getView ()
	Returns a modifiable view of the dense matrix. More...

void	load (const String &fileName)
	Method for loading the matrix from the file with given filename. More...

void	load (File &file) override
	Method for loading the matrix from a file. More...

template<typename Real_ , typename Device_ , typename Index_ , typename RealAllocator_ >
bool	operator!= (const DenseMatrix< Real_, Device_, Index_, Organization, RealAllocator_ > &matrix) const
	Comparison operator with another dense matrix. More...

template<typename Real_ , typename Device_ , typename Index_ >
bool	operator!= (const DenseMatrixView< Real_, Device_, Index_, Organization > &matrix) const
	Comparison operator with another dense matrix view. More...

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

__cuda_callable__ Real &	operator() (IndexType row, IndexType column)
	Returns non-constant reference to element at row row and column column. More...

__cuda_callable__ const Real &	operator() (IndexType row, IndexType column) const
	Returns constant reference to element at row row and column column. More...

DenseMatrix &	operator= (const DenseMatrix &matrix)
	Assignment operator with exactly the same type of the dense matrix. More...

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >
DenseMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const DenseMatrix< Real, Device, Index, Organization, RealAllocator > &matrix)

template<typename RHSReal , typename RHSDevice , typename RHSIndex , typename RHSRealAllocator >
DenseMatrix &	operator= (const DenseMatrix< RHSReal, RHSDevice, RHSIndex, Organization, RHSRealAllocator > &matrix)
	Assignment operator with the same organization. More...

template<typename RHSReal , typename RHSDevice , typename RHSIndex , typename RHSRealAllocator >
DenseMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const DenseMatrix< RHSReal, RHSDevice, RHSIndex, Organization, RHSRealAllocator > &matrix)

template<typename RHSReal , typename RHSDevice , typename RHSIndex , ElementsOrganization RHSOrganization, typename RHSRealAllocator >
DenseMatrix &	operator= (const DenseMatrix< RHSReal, RHSDevice, RHSIndex, RHSOrganization, RHSRealAllocator > &matrix)
	Assignment operator with other dense matrices. More...

template<typename RHSReal , typename RHSDevice , typename RHSIndex , ElementsOrganization RHSOrganization, typename RHSRealAllocator >
DenseMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const DenseMatrix< RHSReal, RHSDevice, RHSIndex, RHSOrganization, RHSRealAllocator > &matrix)

template<typename RHSReal , typename RHSDevice , typename RHSIndex >
DenseMatrix &	operator= (const DenseMatrixView< RHSReal, RHSDevice, RHSIndex, Organization > &matrix)
	Assignment operator with matrix view having the same elements organization. More...

template<typename RHSReal , typename RHSDevice , typename RHSIndex >
DenseMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const DenseMatrixView< RHSReal, RHSDevice, RHSIndex, Organization > &matrix)

template<typename RHSReal , typename RHSDevice , typename RHSIndex , ElementsOrganization RHSOrganization>
DenseMatrix &	operator= (const DenseMatrixView< RHSReal, RHSDevice, RHSIndex, RHSOrganization > &matrix)
	Assignment operator with other dense matrices. More...

template<typename RHSReal , typename RHSDevice , typename RHSIndex , ElementsOrganization RHSOrganization>
DenseMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const DenseMatrixView< RHSReal, RHSDevice, RHSIndex, RHSOrganization > &matrix)

template<typename RHSMatrix >
DenseMatrix &	operator= (const RHSMatrix &matrix)
	Assignment operator with other (sparse) types of matrices. More...

template<typename RHSMatrix >
DenseMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const RHSMatrix &matrix)

template<typename Real_ , typename Device_ , typename Index_ , typename RealAllocator_ >
bool	operator== (const DenseMatrix< Real_, Device_, Index_, Organization, RealAllocator_ > &matrix) const
	Comparison operator with another dense matrix. More...

template<typename Real_ , typename Device_ , typename Index_ >
bool	operator== (const DenseMatrixView< Real_, Device_, Index_, Organization > &matrix) const
	Comparison operator with another dense matrix view. More...

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

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

template<typename Fetch , typename Reduce , typename Keep , typename FetchReal >
void	reduceAllRows (Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchReal &identity)
	Method for performing general reduction on ALL matrix rows. More...

template<typename Fetch , typename Reduce , typename Keep , typename FetchReal >
void	reduceAllRows (Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchReal &identity) const
	Method for performing general reduction on ALL matrix rows for constant instances. More...

template<typename Fetch , typename Reduce , typename Keep , typename FetchValue >
void	reduceRows (IndexType begin, IndexType end, Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchValue &identity)
	Method for performing general reduction on matrix rows. More...

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

void	reset ()
	Resets the matrix to zero dimensions.

void	save (const String &fileName) const
	Method for saving the matrix to the file with given filename. More...

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

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

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

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

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

void	setDimensions (IndexType rows, IndexType columns) override
	Set number of rows and columns of this matrix. More...

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

template<typename Value >
void	setElements (std::initializer_list< std::initializer_list< Value > > data)
	This method recreates the dense matrix from 2D initializer list. More...

template<typename Matrix >
void	setLike (const Matrix &matrix)
	Set the number of matrix rows and columns by the given matrix. More...

template<typename Matrix_ >
void	setLike (const Matrix_ &matrix)

template<typename RowCapacitiesVector >
void	setRowCapacities (const RowCapacitiesVector &rowCapacities)
	This method is only for the compatibility with the sparse matrices. More...

void	setValue (const RealType &v)
	Sets all matrix elements to value v. More...

template<typename InVector , typename OutVector >
void	vectorProduct (const InVector &inVector, OutVector &outVector, const RealType &matrixMultiplicator=1.0, const RealType &outVectorMultiplicator=0.0, IndexType begin=0, IndexType end=0) const
	Computes product of matrix and vector. More...

Public Member Functions inherited from TNL::Matrices::Matrix< Real, Device, Index, RealAllocator >
	Matrix (const RealAllocatorType &allocator=RealAllocatorType())
	Construct a new Matrix object possibly with user defined allocator of the matrix values. More...

	Matrix (IndexType rows, IndexType columns, const RealAllocatorType &allocator=RealAllocatorType())
	Construct a new Matrix object with given dimensions and possibly user defined allocator of the matrix values. More...

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

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

virtual IndexType	getNonzeroElementsCount () const
	Computes a current number of nonzero matrix elements. More...

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

ValuesType &	getValues ()
	Returns a reference to a vector with the matrix elements values. More...

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

void	load (File &file) override
	Method for loading the matrix from a file. More...

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

template<typename MatrixT >
bool	operator!= (const MatrixT &matrix) const

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

template<typename MatrixT >
bool	operator== (const MatrixT &matrix) const

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

void	reset ()
	Reset the matrix. More...

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

virtual void	setDimensions (IndexType rows, IndexType columns)
	Method for setting or changing of the matrix dimensions. More...

template<typename Matrix_ >
void	setLike (const Matrix_ &matrix)
	Set the matrix dimensions to be equal to those of the input matrix. More...

Public Member Functions inherited from TNL::Object
virtual	~Object ()=default
	Destructor. More...

virtual std::string	getSerializationTypeVirtual () const

void	load (const String &fileName)
	Method for restoring the object from a file. More...

virtual void	load (File &file)
	Method for restoring the object from a file. More...

void	save (const String &fileName) const
	Method for saving the object to a file as a binary data. More...

virtual void	save (File &file) const
	Method for saving the object to a file as a binary data. More...

Static Public Member Functions
static constexpr ElementsOrganization	getOrganization ()
	Matrix elements organization getter. More...

static std::string	getSerializationType ()
	Returns string with serialization type. More...

static constexpr bool	isSymmetric ()
	This is only for compatibility with sparse matrices. More...

Static Public Member Functions inherited from TNL::Object
static std::string	getSerializationType ()
	Static serialization type getter. More...

Protected Types
using	BaseType = Matrix< Real, Device, Index, RealAllocator >

using	SegmentsType = Algorithms::Segments::Ellpack< Device, Index, typename Allocators::Default< Device >::template Allocator< Index >, Organization, 1 >

using	SegmentViewType = typename SegmentsType::SegmentViewType

using	ValuesVectorType = typename BaseType::ValuesType

using	ValuesViewType = typename ValuesVectorType::ViewType

Protected Member Functions
__cuda_callable__ IndexType	getElementIndex (IndexType row, IndexType column) const

Protected Attributes
SegmentsType	segments

ViewType	view

Protected Attributes inherited from TNL::Matrices::Matrix< Real, Device, Index, RealAllocator >
IndexType	columns

IndexType	rows

ValuesType	values
	Array containing the allocated matrix elements.

Detailed Description

template<typename Real = double, typename Device = Devices::Host, typename Index = int, ElementsOrganization Organization = Algorithms::Segments::DefaultElementsOrganization< Device >::getOrganization(), typename RealAllocator = typename Allocators::Default< Device >::template Allocator< Real >>
class TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >

Implementation of dense matrix, i.e. matrix storing explicitly all of its elements including zeros.

Template Parameters

Real	is a type of matrix elements.
Device	is a device where the matrix is allocated.
Index	is a type for indexing of the matrix elements.
Organization	tells the ordering of matrix elements. It is either RowMajorOrder or ColumnMajorOrder.
RealAllocator	is allocator for the matrix elements.

Member Typedef Documentation

◆ ConstViewType

using TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::ConstViewType = typename DenseMatrixView< Real, Device, Index, Organization >::ConstViewType

Matrix view type for constant instances.

See DenseMatrixView.

◆ ViewType

using TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::ViewType = DenseMatrixView< Real, Device, Index, Organization >

Type of related matrix view.

See DenseMatrixView.

Constructor & Destructor Documentation

◆ DenseMatrix() [1/5]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::DenseMatrix ( const RealAllocatorType & allocator = RealAllocatorType() )

Constructor only with values allocator.

Parameters

allocator is used for allocation of matrix elements values.

◆ DenseMatrix() [2/5]

TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::DenseMatrix ( const DenseMatrix< Real, Device, Index, Organization, RealAllocator > & matrix )

default

Copy constructor.

Parameters

matrix is the source matrix

◆ DenseMatrix() [3/5]

TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::DenseMatrix ( DenseMatrix< Real, Device, Index, Organization, RealAllocator > && matrix )

defaultnoexcept

Move constructor.

Parameters

matrix is the source matrix

◆ DenseMatrix() [4/5]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::DenseMatrix	(	IndexType	rows,
		IndexType	columns,
		const RealAllocatorType &	allocator = `RealAllocatorType()`
	)

Constructor with matrix dimensions.

Parameters

rows	is number of matrix rows.
columns	is number of matrix columns.
allocator	is used for allocation of matrix elements values.

◆ DenseMatrix() [5/5]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Value >

TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::DenseMatrix	(	std::initializer_list< std::initializer_list< Value > >	data,
		const RealAllocatorType &	allocator = `RealAllocatorType()`
	)

Constructor with 2D initializer list.

The number of matrix rows is set to the outer list size and the number of matrix columns is set to maximum size of inner lists. Missing elements are filled in with zeros.

Parameters

data	is a initializer list of initializer lists representing list of matrix rows.
allocator	is used for allocation of matrix elements values.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

template< typename Device >

void initializerListExample()

{

TNL::Matrices::DenseMatrix< double, Device > matrix {

{ 1, 2, 3, 4, 5, 6 },

{ 7, 8, 9, 10, 11, 12 },

{ 13, 14, 15, 16, 17, 18 }

};

std::cout << "General dense matrix: " << std::endl << matrix << std::endl;

TNL::Matrices::DenseMatrix< double, Device > triangularMatrix {

{ 1 },

{ 2, 3 },

{ 4, 5, 6 },

{ 7, 8, 9, 10 },

{ 11, 12, 13, 14, 15 }

};

std::cout << "Triangular dense matrix: " << std::endl << triangularMatrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Creating matrices on CPU ... " << std::endl;

initializerListExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Creating matrices on CUDA GPU ... " << std::endl;

initializerListExample< TNL::Devices::Cuda >();

#endif

}

std::cout

TNL::Matrices::DenseMatrix
Implementation of dense matrix, i.e. matrix storing explicitly all of its elements including zeros.
Definition: DenseMatrix.h:35

std::endl
T endl(T... args)

Output: Creating matrices on CPU ...

General dense matrix:

Row: 0 -> 0:1 1:2 2:3 3:4 4:5 5:6

Row: 1 -> 0:7 1:8 2:9 3:10 4:11 5:12

Row: 2 -> 0:13 1:14 2:15 3:16 4:17 5:18

Triangular dense matrix:

Row: 0 -> 0:1 1:0 2:0 3:0 4:0

Row: 1 -> 0:2 1:3 2:0 3:0 4:0

Row: 2 -> 0:4 1:5 2:6 3:0 4:0

Row: 3 -> 0:7 1:8 2:9 3:10 4:0

Row: 4 -> 0:11 1:12 2:13 3:14 4:15

Creating matrices on CUDA GPU ...

General dense matrix:

Row: 0 -> 0:1 1:2 2:3 3:4 4:5 5:6

Row: 1 -> 0:7 1:8 2:9 3:10 4:11 5:12

Row: 2 -> 0:13 1:14 2:15 3:16 4:17 5:18

Triangular dense matrix:

Row: 0 -> 0:1 1:0 2:0 3:0 4:0

Row: 1 -> 0:2 1:3 2:0 3:0 4:0

Row: 2 -> 0:4 1:5 2:6 3:0 4:0

Row: 3 -> 0:7 1:8 2:9 3:10 4:0

Row: 4 -> 0:11 1:12 2:13 3:14 4:15

Member Function Documentation

◆ addElement()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::addElement	(	IndexType	row,
		IndexType	column,
		const RealType &	value,
		const RealType &	thisElementMultiplicator = `1.0`
	)

Add element at given row and column to given value.

This method can be called from the host system (CPU) no matter where the matrix is allocated. If the matrix is allocated on GPU this method can be called even from device kernels. If the matrix is allocated in GPU device this method is called from CPU, it transfers values of each matrix element separately and so the performance is very low. For higher performance see. DenseMatrix::getRow or DenseMatrix::forElements and DenseMatrix::forAllElements.

Parameters

row	is row index of the element.
column	is columns index of the element.
value	is the value the element will be set to.
thisElementMultiplicator	is multiplicator the original matrix element value is multiplied by before addition of given value.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

template< typename Device >

void addElements()

{

TNL::Matrices::DenseMatrix< double, Device > matrix( 5, 5 );

for( int i = 0; i < 5; i++ )

matrix.setElement( i, i, i );

std::cout << "Initial matrix is: " << std::endl << matrix << std::endl;

for( int i = 0; i < 5; i++ )

for( int j = 0; j < 5; j++ )

matrix.addElement( i, j, 1.0, 5.0 );

std::cout << "Matrix after addition is: " << std::endl << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Add elements on host:" << std::endl;

addElements< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Add elements on CUDA device:" << std::endl;

addElements< TNL::Devices::Cuda >();

#endif

}

Output: Add elements on host:

Initial matrix is:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:0 1:1 2:0 3:0 4:0

Row: 2 -> 0:0 1:0 2:2 3:0 4:0

Row: 3 -> 0:0 1:0 2:0 3:3 4:0

Row: 4 -> 0:0 1:0 2:0 3:0 4:4

Matrix after addition is:

Row: 0 -> 0:1 1:1 2:1 3:1 4:1

Row: 1 -> 0:1 1:6 2:1 3:1 4:1

Row: 2 -> 0:1 1:1 2:11 3:1 4:1

Row: 3 -> 0:1 1:1 2:1 3:16 4:1

Row: 4 -> 0:1 1:1 2:1 3:1 4:21

Add elements on CUDA device:

Initial matrix is:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:0 1:1 2:0 3:0 4:0

Row: 2 -> 0:0 1:0 2:2 3:0 4:0

Row: 3 -> 0:0 1:0 2:0 3:3 4:0

Row: 4 -> 0:0 1:0 2:0 3:0 4:4

Matrix after addition is:

Row: 0 -> 0:1 1:1 2:1 3:1 4:1

Row: 1 -> 0:1 1:6 2:1 3:1 4:1

Row: 2 -> 0:1 1:1 2:11 3:1 4:1

Row: 3 -> 0:1 1:1 2:1 3:16 4:1

Row: 4 -> 0:1 1:1 2:1 3:1 4:21

◆ forAllElements() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forAllElements ( Function && function )

This method calls forElements for all matrix rows.

See DenseMatrix::forAllElements.

Template Parameters

Function is a type of lambda function that will operate on matrix elements.

Parameters

function is an instance of the lambda function to be called in each row.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forAllElementsExample()

{

TNL::Matrices::DenseMatrix< double, Device > matrix( 5, 5 );

auto f = [=] __cuda_callable__ ( int rowIdx, int columnIdx, int columnIdx_, double& value ) {

if( rowIdx >= columnIdx )

value = rowIdx + columnIdx;

};

matrix.forAllElements( f );

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Creating matrix on host: " << std::endl;

forAllElementsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Creating matrix on CUDA device: " << std::endl;

forAllElementsExample< TNL::Devices::Cuda >();

#endif

}

__cuda_callable__
#define __cuda_callable__
Definition: CudaCallable.h:22

Output: Creating matrix on host:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:1 1:2 2:0 3:0 4:0

Row: 2 -> 0:2 1:3 2:4 3:0 4:0

Row: 3 -> 0:3 1:4 2:5 3:6 4:0

Row: 4 -> 0:4 1:5 2:6 3:7 4:8

Creating matrix on CUDA device:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:1 1:2 2:0 3:0 4:0

Row: 2 -> 0:2 1:3 2:4 3:0 4:0

Row: 3 -> 0:3 1:4 2:5 3:6 4:0

Row: 4 -> 0:4 1:5 2:6 3:7 4:8

◆ forAllElements() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forAllElements ( Function && function ) const

This method calls forElements for all matrix rows (for constant instances).

See DenseMatrix::forElements.

Template Parameters

Function is a type of lambda function that will operate on matrix elements.

Parameters

function is an instance of the lambda function to be called in each row.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forAllElementsExample()

{

TNL::Matrices::DenseMatrix< double, Device > matrix( 5, 5 );

auto f = [=] __cuda_callable__ ( int rowIdx, int columnIdx, int columnIdx_, double& value ) {

if( rowIdx >= columnIdx )

value = rowIdx + columnIdx;

};

matrix.forAllElements( f );

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Creating matrix on host: " << std::endl;

forAllElementsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Creating matrix on CUDA device: " << std::endl;

forAllElementsExample< TNL::Devices::Cuda >();

#endif

}

Output: Creating matrix on host:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:1 1:2 2:0 3:0 4:0

Row: 2 -> 0:2 1:3 2:4 3:0 4:0

Row: 3 -> 0:3 1:4 2:5 3:6 4:0

Row: 4 -> 0:4 1:5 2:6 3:7 4:8

Creating matrix on CUDA device:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:1 1:2 2:0 3:0 4:0

Row: 2 -> 0:2 1:3 2:4 3:0 4:0

Row: 3 -> 0:3 1:4 2:5 3:6 4:0

Row: 4 -> 0:4 1:5 2:6 3:7 4:8

◆ forAllRows() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forAllRows ( Function && function )

Method for parallel iteration over all matrix rows.

In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method DenseMatrix::forAllElements where more than one thread can be mapped to each row.

Template Parameters

Function is type of the lambda function.

Parameters

function is an instance of the lambda function to be called for each row.

auto function = [] __cuda_callable__ ( RowView& row ) mutable { ... };

TNL::Matrices::DenseMatrix::RowView

typename ViewType::RowView RowView

Type for accessing matrix rows.

Definition: DenseMatrix.h:104

RowView represents matrix row - see TNL::Matrices::DenseMatrix::RowView.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forRowsExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

using RowView = typename MatrixType::RowView;

const int size = 5;

MatrixType matrix( size, size );

/***

* Set the matrix elements.

*/

auto f = [=] __cuda_callable__ ( RowView& row ) mutable {

const int& rowIdx = row.getRowIndex();

if( rowIdx > 0 )

row.setValue( rowIdx - 1, -1.0 );

row.setValue( rowIdx, rowIdx + 1.0 );

if( rowIdx < size - 1 )

row.setValue( rowIdx + 1, -1.0 );

};

matrix.forAllRows( f );

std::cout << matrix << std::endl;

/***

* Now divide each matrix row by its largest element with use of iterators.

*/

matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {

double largest = std::numeric_limits< double >::lowest();

for( auto element : row )

largest = TNL::max( largest, element.value() );

for( auto element : row )

element.value() /= largest;

} );

std::cout << "Divide each matrix row by its largest element... " << std::endl;

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

forRowsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

forRowsExample< TNL::Devices::Cuda >();

#endif

}

std::numeric_limits::lowest
T lowest(T... args)

TNL::max
constexpr ResultType max(const T1 &a, const T2 &b)
This function returns maximum of two numbers.
Definition: Math.h:65

TNL::Matrices::MatrixType
Structure for specifying type of sparse matrix.
Definition: MatrixType.h:21

Output: Getting matrix rows on host:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

Getting matrix rows on CUDA device:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

◆ forAllRows() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forAllRows ( Function && function ) const

Method for parallel iteration over all matrix rows for constant instances.

In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method DenseMatrix::forAllElements where more than one thread can be mapped to each row.

Template Parameters

Function is type of the lambda function.

Parameters

function is an instance of the lambda function to be called for each row.

auto function = [] __cuda_callable__ ( RowView& row ) { ... };

RowView represents matrix row - see TNL::Matrices::DenseMatrix::RowView.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forRowsExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

using RowView = typename MatrixType::RowView;

const int size = 5;

MatrixType matrix( size, size );

/***

* Set the matrix elements.

*/

auto f = [=] __cuda_callable__ ( RowView& row ) mutable {

const int& rowIdx = row.getRowIndex();

if( rowIdx > 0 )

row.setValue( rowIdx - 1, -1.0 );

row.setValue( rowIdx, rowIdx + 1.0 );

if( rowIdx < size - 1 )

row.setValue( rowIdx + 1, -1.0 );

};

matrix.forAllRows( f );

std::cout << matrix << std::endl;

/***

* Now divide each matrix row by its largest element with use of iterators.

*/

matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {

double largest = std::numeric_limits< double >::lowest();

for( auto element : row )

largest = TNL::max( largest, element.value() );

for( auto element : row )

element.value() /= largest;

} );

std::cout << "Divide each matrix row by its largest element... " << std::endl;

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

forRowsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

forRowsExample< TNL::Devices::Cuda >();

#endif

}

Output: Getting matrix rows on host:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

Getting matrix rows on CUDA device:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

◆ forElements() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forElements	(	IndexType	begin,
		IndexType	end,
		Function &&	function
	)

Method for iteration over all matrix rows for non-constant instances.

Template Parameters

Function is type of lambda function that will operate on matrix elements. It is should have form like

auto function = [=] __cuda_callable__
     ( IndexType rowIdx, IndexType columnIdx, IndexType columnIdx_, RealType& value )
     { ... };

The column index repeats twice only for compatibility with sparse matrices.

Parameters

begin	defines beginning of the range [begin,end) of rows to be processed.
end	defines ending of the range [begin,end) of rows to be processed.
function	is an instance of the lambda function to be called in each row.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forRowsExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

using RowView = typename MatrixType::RowView;

const int size = 5;

MatrixType matrix( size, size );

/***

* Set the matrix elements.

*/

auto f = [=] __cuda_callable__ ( RowView& row ) mutable {

const int& rowIdx = row.getRowIndex();

if( rowIdx > 0 )

row.setValue( rowIdx - 1, -1.0 );

row.setValue( rowIdx, rowIdx + 1.0 );

if( rowIdx < size - 1 )

row.setValue( rowIdx + 1, -1.0 );

};

matrix.forAllRows( f );

std::cout << matrix << std::endl;

/***

* Now divide each matrix row by its largest element with use of iterators.

*/

matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {

double largest = std::numeric_limits< double >::lowest();

for( auto element : row )

largest = TNL::max( largest, element.value() );

for( auto element : row )

element.value() /= largest;

} );

std::cout << "Divide each matrix row by its largest element... " << std::endl;

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

forRowsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

forRowsExample< TNL::Devices::Cuda >();

#endif

}

Output: Getting matrix rows on host:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

Getting matrix rows on CUDA device:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

◆ forElements() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forElements	(	IndexType	begin,
		IndexType	end,
		Function &&	function
	)		const

Method for iteration over all matrix rows for constant instances.

Template Parameters

Function is type of lambda function that will operate on matrix elements. It is should have form like

auto function = [=] __cuda_callable__
     ( IndexType rowIdx, IndexType columnIdx, IndexType columnIdx_, const RealType& value )
     { ... };

The column index repeats twice only for compatibility with sparse matrices.

Parameters

begin	defines beginning of the range [begin,end) of rows to be processed.
end	defines ending of the range [begin,end) of rows to be processed.
function	is an instance of the lambda function to be called in each row.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forRowsExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

using RowView = typename MatrixType::RowView;

const int size = 5;

MatrixType matrix( size, size );

/***

* Set the matrix elements.

*/

auto f = [=] __cuda_callable__ ( RowView& row ) mutable {

const int& rowIdx = row.getRowIndex();

if( rowIdx > 0 )

row.setValue( rowIdx - 1, -1.0 );

row.setValue( rowIdx, rowIdx + 1.0 );

if( rowIdx < size - 1 )

row.setValue( rowIdx + 1, -1.0 );

};

matrix.forAllRows( f );

std::cout << matrix << std::endl;

/***

* Now divide each matrix row by its largest element with use of iterators.

*/

matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {

double largest = std::numeric_limits< double >::lowest();

for( auto element : row )

largest = TNL::max( largest, element.value() );

for( auto element : row )

element.value() /= largest;

} );

std::cout << "Divide each matrix row by its largest element... " << std::endl;

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

forRowsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

forRowsExample< TNL::Devices::Cuda >();

#endif

}

Output: Getting matrix rows on host:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

Getting matrix rows on CUDA device:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

◆ forRows() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forRows	(	IndexType	begin,
		IndexType	end,
		Function &&	function
	)

Method for parallel iteration over matrix rows from interval [ begin, end).

In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method DenseMatrix::forElements where more than one thread can be mapped to each row.

Template Parameters

Function is type of the lambda function.

Parameters

begin	defines beginning of the range [ begin,end ) of rows to be processed.
end	defines ending of the range [ begin, end ) of rows to be processed.
function	is an instance of the lambda function to be called for each row.

auto function = [] __cuda_callable__ ( RowView& row ) mutable { ... };

RowView represents matrix row - see TNL::Matrices::DenseMatrix::RowView.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forRowsExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

using RowView = typename MatrixType::RowView;

const int size = 5;

MatrixType matrix( size, size );

/***

* Set the matrix elements.

*/

auto f = [=] __cuda_callable__ ( RowView& row ) mutable {

const int& rowIdx = row.getRowIndex();

if( rowIdx > 0 )

row.setValue( rowIdx - 1, -1.0 );

row.setValue( rowIdx, rowIdx + 1.0 );

if( rowIdx < size - 1 )

row.setValue( rowIdx + 1, -1.0 );

};

matrix.forAllRows( f );

std::cout << matrix << std::endl;

/***

* Now divide each matrix row by its largest element with use of iterators.

*/

matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {

double largest = std::numeric_limits< double >::lowest();

for( auto element : row )

largest = TNL::max( largest, element.value() );

for( auto element : row )

element.value() /= largest;

} );

std::cout << "Divide each matrix row by its largest element... " << std::endl;

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

forRowsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

forRowsExample< TNL::Devices::Cuda >();

#endif

}

Output: Getting matrix rows on host:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

Getting matrix rows on CUDA device:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

◆ forRows() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::forRows	(	IndexType	begin,
		IndexType	end,
		Function &&	function
	)		const

Method for parallel iteration over matrix rows from interval [ begin, end) for constant instances.

In each row, given lambda function is performed. Each row is processed by at most one thread unlike the method DenseMatrix::forElements where more than one thread can be mapped to each row.

Template Parameters

Function is type of the lambda function.

Parameters

begin	defines beginning of the range [ begin,end ) of rows to be processed.
end	defines ending of the range [ begin, end ) of rows to be processed.
function	is an instance of the lambda function to be called for each row.

auto function = [] __cuda_callable__ ( RowView& row ) { ... };

RowView represents matrix row - see TNL::Matrices::DenseMatrix::RowView.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void forRowsExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

using RowView = typename MatrixType::RowView;

const int size = 5;

MatrixType matrix( size, size );

/***

* Set the matrix elements.

*/

auto f = [=] __cuda_callable__ ( RowView& row ) mutable {

const int& rowIdx = row.getRowIndex();

if( rowIdx > 0 )

row.setValue( rowIdx - 1, -1.0 );

row.setValue( rowIdx, rowIdx + 1.0 );

if( rowIdx < size - 1 )

row.setValue( rowIdx + 1, -1.0 );

};

matrix.forAllRows( f );

std::cout << matrix << std::endl;

/***

* Now divide each matrix row by its largest element with use of iterators.

*/

matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {

double largest = std::numeric_limits< double >::lowest();

for( auto element : row )

largest = TNL::max( largest, element.value() );

for( auto element : row )

element.value() /= largest;

} );

std::cout << "Divide each matrix row by its largest element... " << std::endl;

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

forRowsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

forRowsExample< TNL::Devices::Cuda >();

#endif

}

Output: Getting matrix rows on host:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

Getting matrix rows on CUDA device:

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-1 1:2 2:-1 3:0 4:0

Row: 2 -> 0:0 1:-1 2:3 3:-1 4:0

Row: 3 -> 0:0 1:0 2:-1 3:4 4:-1

Row: 4 -> 0:0 1:0 2:0 3:-1 4:5

Divide each matrix row by its largest element...

Row: 0 -> 0:1 1:-1 2:0 3:0 4:0

Row: 1 -> 0:-0.5 1:1 2:-0.5 3:0 4:0

Row: 2 -> 0:0 1:-0.333333 2:1 3:-0.333333 4:0

Row: 3 -> 0:0 1:0 2:-0.25 3:1 4:-0.25

Row: 4 -> 0:0 1:0 2:0 3:-0.2 4:1

◆ getCompressedRowLengths()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename RowLengthsVector >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getCompressedRowLengths ( RowLengthsVector & rowLengths ) const

Computes number of non-zeros in each row.

Parameters

rowLengths is a vector into which the number of non-zeros in each row will be stored.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void getCompressedRowLengthsExample()

{

TNL::Matrices::DenseMatrix< double, Device > triangularMatrix {

{ 1 },

{ 2, 3 },

{ 4, 5, 6 },

{ 7, 8, 9, 10 },

{ 11, 12, 13, 14, 15 }

};

std::cout << triangularMatrix << std::endl;

TNL::Containers::Vector< int, Device > rowLengths;

triangularMatrix.getCompressedRowLengths( rowLengths );

std::cout << "Compressed row lengths are: " << rowLengths << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting compressed row lengths on host: " << std::endl;

getCompressedRowLengthsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting compressed row lengths on CUDA device: " << std::endl;

getCompressedRowLengthsExample< TNL::Devices::Cuda >();

#endif

}

TNL::Containers::Vector
Vector extends Array with algebraic operations.
Definition: Vector.h:40

Output: Getting compressed row lengths on host:

Row: 0 -> 0:1 1:0 2:0 3:0 4:0

Row: 1 -> 0:2 1:3 2:0 3:0 4:0

Row: 2 -> 0:4 1:5 2:6 3:0 4:0

Row: 3 -> 0:7 1:8 2:9 3:10 4:0

Row: 4 -> 0:11 1:12 2:13 3:14 4:15

Compressed row lengths are: [ 1, 2, 3, 4, 5 ]

Getting compressed row lengths on CUDA device:

Row: 0 -> 0:1 1:0 2:0 3:0 4:0

Row: 1 -> 0:2 1:3 2:0 3:0 4:0

Row: 2 -> 0:4 1:5 2:6 3:0 4:0

Row: 3 -> 0:7 1:8 2:9 3:10 4:0

Row: 4 -> 0:11 1:12 2:13 3:14 4:15

Compressed row lengths are: [ 1, 2, 3, 4, 5 ]

◆ getConstView()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

auto TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getConstView

Returns a non-modifiable view of the dense matrix.

See DenseMatrixView.

Returns: dense matrix view.

◆ getElement()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ Real TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getElement	(	IndexType	row,
		IndexType	column
	)		const

Returns value of matrix element at position given by its row and column index.

Parameters

row	is a row index of the matrix element.
column	i a column index of the matrix element.

Returns: value of given matrix element.

Example: #include <iostream>

#include <iomanip>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

template< typename Device >

void getElements()

{

TNL::Matrices::DenseMatrix< double, Device > matrix {

{ 1, 0, 0, 0, 0 },

{ -1, 2, -1, 0, 0 },

{ 0, -1, 2, -1, 0 },

{ 0, 0, -1, 2, -1 },

{ 0, 0, 0, 0, 1 } };

for( int i = 0; i < 5; i++ )

{

for( int j = 0; j < 5; j++ )

std::cout << std::setw( 5 ) << matrix.getElement( i, j );

std::cout << std::endl;

}

}

int main( int argc, char* argv[] )

{

std::cout << "Get elements on host:" << std::endl;

getElements< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Get elements on CUDA device:" << std::endl;

getElements< TNL::Devices::Cuda >();

#endif

}

std::setw
T setw(T... args)

Output: Get elements on host:

1 0 0 0 0

-1 2 -1 0 0

0 -1 2 -1 0

0 0 -1 2 -1

0 0 0 0 1

Get elements on CUDA device:

1 0 0 0 0

-1 2 -1 0 0

0 -1 2 -1 0

0 0 -1 2 -1

0 0 0 0 1

◆ getNonzeroElementsCount()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

Index TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getNonzeroElementsCount

overridevirtual

Returns number of non-zero matrix elements.

Returns: number of all non-zero matrix elements.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void getElementsCountExample()

{

TNL::Matrices::DenseMatrix< double, Device > triangularMatrix {

{ 1 },

{ 2, 3 },

{ 4, 5, 6 },

{ 7, 8, 9, 10 },

{ 11, 12, 13, 14, 15 }

};

std::cout << "Matrix elements count is " << triangularMatrix.getAllocatedElementsCount() << "." << std::endl;

std::cout << "Non-zero matrix elements count is " << triangularMatrix.getNonzeroElementsCount() << "." << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Computing matrix elements on host: " << std::endl;

getElementsCountExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Computing matrix elements on CUDA device: " << std::endl;

getElementsCountExample< TNL::Devices::Cuda >();

#endif

}

Output: Computing matrix elements on host:

Matrix elements count is 25.

Non-zero matrix elements count is 15.

Computing matrix elements on CUDA device:

Matrix elements count is 25.

Non-zero matrix elements count is 15.

Reimplemented from TNL::Matrices::Matrix< Real, Device, Index, RealAllocator >.

◆ getOrganization()

static constexpr ElementsOrganization TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getOrganization ( )

inlinestaticconstexpr

Matrix elements organization getter.

Returns: matrix elements organization - RowMajorOrder of ColumnMajorOrder.

◆ getRow() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ auto TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getRow ( const IndexType & rowIdx )

Non-constant getter of simple structure for accessing given matrix row.

Parameters

rowIdx is matrix row index.

Returns: RowView for accessing given matrix row.

Example: #include <iostream>

#include <TNL/Algorithms/parallelFor.h>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Pointers/SharedPointer.h>

template< typename Device >

void getRowExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

TNL::Pointers::SharedPointer< MatrixType > matrix( 5, 5 );

MatrixType* matrix_device = &matrix.template modifyData< Device >();

auto f = [=] __cuda_callable__ ( int rowIdx ) mutable {

auto row = matrix_device->getRow( rowIdx );

row.setValue( rowIdx, 10 * ( rowIdx + 1 ) );

};

/***

* For the case when Device is CUDA device we need to synchronize smart

* pointers. To avoid this you may use DenseMatrixView. See

* DenseMatrixView::getRow example for details.

*/

TNL::Pointers::synchronizeSmartPointersOnDevice< Device >();

/***

* Set the matrix elements.

*/

TNL::Algorithms::parallelFor< Device >( 0, matrix->getRows(), f );

std::cout << matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

getRowExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

getRowExample< TNL::Devices::Cuda >();

#endif

}

TNL::Pointers::SharedPointer
Cross-device shared smart pointer.
Definition: SharedPointer.h:49

Output: Getting matrix rows on host:

1

Getting matrix rows on CUDA device:

1

See DenseMatrixRowView.

◆ getRow() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ auto TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getRow ( const IndexType & rowIdx ) const

Constant getter of simple structure for accessing given matrix row.

Parameters

rowIdx is matrix row index.

Returns: RowView for accessing given matrix row.

Example: #include <iostream>

#include <functional>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Pointers/SharedPointer.h>

template< typename Device >

void getRowExample()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

TNL::Pointers::SharedPointer< MatrixType > matrix {

{ 1, 0, 0, 0, 0 },

{ 1, 2, 0, 0, 0 },

{ 1, 2, 3, 0, 0 },

{ 1, 2, 3, 4, 0 },

{ 1, 2, 3, 4, 5 }

};

/***

* Fetch lambda function returns diagonal element in each row.

*/

const MatrixType* matrix_device = &matrix.template getData< Device >();

auto fetch = [=] __cuda_callable__ ( int rowIdx ) mutable -> double {

auto row = matrix_device->getRow( rowIdx );

return row.getValue( rowIdx );

};

/***

* For the case when Device is CUDA device we need to synchronize smart

* pointers. To avoid this you may use DenseMatrixView. See

* DenseMatrixView::getConstRow example for details.

*/

TNL::Pointers::synchronizeSmartPointersOnDevice< Device >();

/***

* Compute the matrix trace.

*/

int trace = TNL::Algorithms::reduce< Device >( 0, matrix->getRows(), fetch, std::plus<>{}, 0 );

std::cout << "Matrix trace is " << trace << "." << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Getting matrix rows on host: " << std::endl;

getRowExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Getting matrix rows on CUDA device: " << std::endl;

getRowExample< TNL::Devices::Cuda >();

#endif

}

std::plus

Output: Getting matrix rows on host:

Matrix trace is 15.

Getting matrix rows on CUDA device:

Matrix trace is 15.

See DenseMatrixRowView.

◆ getRowCapacities()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Vector >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getRowCapacities ( Vector & rowCapacities ) const

Compute capacities of all rows.

The row capacities are not stored explicitly and must be computed.

Parameters

rowCapacities is a vector where the row capacities will be stored.

◆ getSerializationType()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

std::string TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getSerializationType

static

Returns string with serialization type.

The string has a form `MatricesDenseMatrix< RealType, [any_device], IndexType, [any_allocator], true/false >`.

Returns: String with the serialization type.

◆ getSerializationTypeVirtual()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

std::string TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getSerializationTypeVirtual

overridevirtual

Returns string with serialization type.

See DenseMatrix::getSerializationType.

Returns: String with the serialization type.

Reimplemented from TNL::Object.

◆ getView()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

auto TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::getView

Returns a modifiable view of the dense matrix.

See DenseMatrixView.

Returns: dense matrix view.

◆ isSymmetric()

static constexpr bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::isSymmetric ( )

inlinestaticconstexpr

This is only for compatibility with sparse matrices.

Returns: false.

◆ load() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::load ( const String & fileName )

Method for loading the matrix from the file with given filename.

Parameters

fileName is name of the file.

◆ load() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::load ( File & file )

overridevirtual

Method for loading the matrix from a file.

Parameters

file	is the file from which the matrix will be loaded.

Reimplemented from TNL::Object.

◆ operator!=() [1/3]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Real_ , typename Device_ , typename Index_ , typename RealAllocator_ >

bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator!= ( const DenseMatrix< Real_, Device_, Index_, Organization, RealAllocator_ > & matrix ) const

Comparison operator with another dense matrix.

Parameters

matrix is the right-hand side matrix.

Returns: false if the RHS matrix is equal, true otherwise.

◆ operator!=() [2/3]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Real_ , typename Device_ , typename Index_ >

bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator!= ( const DenseMatrixView< Real_, Device_, Index_, Organization > & matrix ) const

Comparison operator with another dense matrix view.

Parameters

matrix is the right-hand side matrix view.

Returns: false if the RHS matrix view is equal, true otherwise.

◆ operator!=() [3/3]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Matrix >

bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator!= ( const Matrix & matrix ) const

Comparison operator with another arbitrary matrix type.

Parameters

matrix is the right-hand side matrix.

Returns: true if the RHS matrix is equal, false otherwise.

◆ operator()() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ Real & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator()	(	IndexType	row,
		IndexType	column
	)

Returns non-constant reference to element at row row and column column.

Since this method returns reference to the element, it cannot be called across different address spaces. It means that it can be called only form CPU if the matrix is allocated on CPU or only from GPU kernels if the matrix is allocated on GPU.

Parameters

row	is a row index of the element.
column	is a columns index of the element.

Returns: reference to given matrix element.

◆ operator()() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ const Real & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator()	(	IndexType	row,
		IndexType	column
	)		const

Returns constant reference to element at row row and column column.

Parameters

row	is a row index of the element.
column	is a columns index of the element.

Returns: reference to given matrix element.

◆ operator=() [1/6]

DenseMatrix & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator= ( const DenseMatrix< Real, Device, Index, Organization, RealAllocator > & matrix )

Assignment operator with exactly the same type of the dense matrix.

Parameters

matrix is the right-hand side matrix.

Returns: reference to this matrix.

◆ operator=() [2/6]

template<typename RHSReal , typename RHSDevice , typename RHSIndex , typename RHSRealAllocator >

DenseMatrix & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator= ( const DenseMatrix< RHSReal, RHSDevice, RHSIndex, Organization, RHSRealAllocator > & matrix )

Assignment operator with the same organization.

Parameters

matrix is the right-hand side matrix.

Returns: reference to this matrix.

◆ operator=() [3/6]

template<typename RHSReal , typename RHSDevice , typename RHSIndex , ElementsOrganization RHSOrganization, typename RHSRealAllocator >

DenseMatrix & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator= ( const DenseMatrix< RHSReal, RHSDevice, RHSIndex, RHSOrganization, RHSRealAllocator > & matrix )

Assignment operator with other dense matrices.

Parameters

matrix is the right-hand side matrix.

Returns: reference to this matrix.

◆ operator=() [4/6]

template<typename RHSReal , typename RHSDevice , typename RHSIndex >

DenseMatrix & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator= ( const DenseMatrixView< RHSReal, RHSDevice, RHSIndex, Organization > & matrix )

Assignment operator with matrix view having the same elements organization.

Parameters

matrix is the right-hand side matrix.

Returns: reference to this matrix.

◆ operator=() [5/6]

template<typename RHSReal , typename RHSDevice , typename RHSIndex , ElementsOrganization RHSOrganization>

DenseMatrix & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator= ( const DenseMatrixView< RHSReal, RHSDevice, RHSIndex, RHSOrganization > & matrix )

Assignment operator with other dense matrices.

Parameters

matrix is the right-hand side matrix.

Returns: reference to this matrix.

◆ operator=() [6/6]

template<typename RHSMatrix >

DenseMatrix & TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator= ( const RHSMatrix & matrix )

Assignment operator with other (sparse) types of matrices.

Parameters

matrix is the right-hand side matrix.

Returns: reference to this matrix.

◆ operator==() [1/3]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Real_ , typename Device_ , typename Index_ , typename RealAllocator_ >

bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator== ( const DenseMatrix< Real_, Device_, Index_, Organization, RealAllocator_ > & matrix ) const

Comparison operator with another dense matrix.

Parameters

matrix is the right-hand side matrix.

Returns: true if the RHS matrix is equal, false otherwise.

◆ operator==() [2/3]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Real_ , typename Device_ , typename Index_ >

bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator== ( const DenseMatrixView< Real_, Device_, Index_, Organization > & matrix ) const

Comparison operator with another dense matrix view.

Parameters

matrix is the right-hand side matrix view.

Returns: true if the RHS matrix view is equal, false otherwise.

◆ operator==() [3/3]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Matrix >

bool TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::operator== ( const Matrix & matrix ) const

Comparison operator with another arbitrary matrix type.

Parameters

matrix is the right-hand side matrix.

Returns: true if the RHS matrix is equal, false otherwise.

◆ print()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::print ( std::ostream & str ) const

overridevirtual

Method for printing the matrix to output stream.

Parameters

str	is the output stream.

Reimplemented from TNL::Matrices::Matrix< Real, Device, Index, RealAllocator >.

◆ reduceAllRows() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Fetch , typename Reduce , typename Keep , typename FetchReal >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::reduceAllRows	(	Fetch &	fetch,
		const Reduce &	reduce,
		Keep &	keep,
		const FetchReal &	identity
	)

Method for performing general reduction on ALL matrix rows.

Template Parameters

Fetch is a type of lambda function for data fetch declared as

auto fetch = [=] __cuda_callable__ ( IndexType rowIdx, IndexType columnIdx, RealType elementValue ) -> FetchValue { ... };

The return type of this lambda can be any non void.

Template Parameters

Reduce is a type of lambda function for reduction declared as

auto reduce = [=] __cuda_callable__ ( const FetchValue& v1, const FetchValue& v2 ) -> FetchValue { ... };

Template Parameters

Keep	is a type of lambda function for storing results of reduction in each row. It is declared as auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const double& value ) { ... };
FetchValue	is type returned by the Fetch lambda function.

Parameters

fetch	is an instance of lambda function for data fetch.
reduce	is an instance of lambda function for reduction.
keep	in an instance of lambda function for storing results.
identity	is the identity element for the reduction operation, i.e. element which does not change the result of the reduction.

Example: #include <iostream>

#include <iomanip>

#include <functional>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void reduceAllRows()

{

TNL::Matrices::DenseMatrix< double, Device > matrix {

{ 1, 0, 0, 0, 0 },

{ 1, 2, 0, 0, 0 },

{ 0, 1, 8, 0, 0 },

{ 0, 0, 1, 9, 0 },

{ 0, 0, 0, 0, 1 } };

/***

* Find largest element in each row.

*/

TNL::Containers::Vector< double, Device > rowMax( matrix.getRows() );

/***

* Prepare vector view and matrix view for lambdas.

*/

auto rowMaxView = rowMax.getView();

/***

* Fetch lambda just returns absolute value of matrix elements.

*/

auto fetch = [=] __cuda_callable__ ( int rowIdx, int columnIdx, const double& value ) -> double {

return TNL::abs( value );

};

/***

* Reduce lambda return maximum of given values.

*/

auto reduce = [=] __cuda_callable__ ( double& a, const double& b ) -> double {

return TNL::max( a, b );

};

/***

* Keep lambda store the largest value in each row to the vector rowMax.

*/

auto keep = [=] __cuda_callable__ ( int rowIdx, const double& value ) mutable {

rowMaxView[ rowIdx ] = value;

};

/***

* Compute the largest values in each row.

*/

matrix.reduceAllRows( fetch, reduce, keep, std::numeric_limits< double >::lowest() );

std::cout << "Max. elements in rows are: " << rowMax << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "All rows reduction on host:" << std::endl;

reduceAllRows< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "All rows reduction on CUDA device:" << std::endl;

reduceAllRows< TNL::Devices::Cuda >();

#endif

}

TNL::Matrices::DenseMatrix::reduceAllRows
void reduceAllRows(Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchReal &identity)
Method for performing general reduction on ALL matrix rows.
Definition: DenseMatrix.hpp:261

TNL::Algorithms::reduce
Result reduce(Index begin, Index end, Fetch &&fetch, Reduction &&reduction, const Result &identity)
reduce implements (parallel) reduction for vectors and arrays.
Definition: reduce.h:71

TNL::abs
__cuda_callable__ T abs(const T &n)
This function returns absolute value of given number n.
Definition: Math.h:91

std::numeric_limits

Output: All rows reduction on host:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

All rows reduction on CUDA device:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

◆ reduceAllRows() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Fetch , typename Reduce , typename Keep , typename FetchReal >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::reduceAllRows	(	Fetch &	fetch,
		const Reduce &	reduce,
		Keep &	keep,
		const FetchReal &	identity
	)		const

Method for performing general reduction on ALL matrix rows for constant instances.

Template Parameters

Fetch is a type of lambda function for data fetch declared as

auto fetch = [=] __cuda_callable__ ( IndexType rowIdx, IndexType columnIdx, RealType elementValue ) -> FetchValue { ... };

     The return type of this lambda can be any non void.

Template Parameters

Reduce is a type of lambda function for reduction declared as

auto reduce = [=] __cuda_callable__ ( const FetchValue& v1, const FetchValue& v2 ) -> FetchValue { ... };

Template Parameters

Keep	is a type of lambda function for storing results of reduction in each row. It is declared as

auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const double& value ) { ... };

Template Parameters

FetchValue is type returned by the Fetch lambda function.

Parameters

fetch	is an instance of lambda function for data fetch.
reduce	is an instance of lambda function for reduction.
keep	in an instance of lambda function for storing results.
identity	is the identity element for the reduction operation, i.e. element which does not change the result of the reduction.

Example: #include <iostream>

#include <iomanip>

#include <functional>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void reduceAllRows()

{

TNL::Matrices::DenseMatrix< double, Device > matrix {

{ 1, 0, 0, 0, 0 },

{ 1, 2, 0, 0, 0 },

{ 0, 1, 8, 0, 0 },

{ 0, 0, 1, 9, 0 },

{ 0, 0, 0, 0, 1 } };

/***

* Find largest element in each row.

*/

TNL::Containers::Vector< double, Device > rowMax( matrix.getRows() );

/***

* Prepare vector view and matrix view for lambdas.

*/

auto rowMaxView = rowMax.getView();

/***

* Fetch lambda just returns absolute value of matrix elements.

*/

auto fetch = [=] __cuda_callable__ ( int rowIdx, int columnIdx, const double& value ) -> double {

return TNL::abs( value );

};

/***

* Reduce lambda return maximum of given values.

*/

auto reduce = [=] __cuda_callable__ ( double& a, const double& b ) -> double {

return TNL::max( a, b );

};

/***

* Keep lambda store the largest value in each row to the vector rowMax.

*/

auto keep = [=] __cuda_callable__ ( int rowIdx, const double& value ) mutable {

rowMaxView[ rowIdx ] = value;

};

/***

* Compute the largest values in each row.

*/

matrix.reduceAllRows( fetch, reduce, keep, std::numeric_limits< double >::lowest() );

std::cout << "Max. elements in rows are: " << rowMax << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "All rows reduction on host:" << std::endl;

reduceAllRows< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "All rows reduction on CUDA device:" << std::endl;

reduceAllRows< TNL::Devices::Cuda >();

#endif

}

Output: All rows reduction on host:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

All rows reduction on CUDA device:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

◆ reduceRows() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Fetch , typename Reduce , typename Keep , typename FetchValue >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::reduceRows	(	IndexType	begin,
		IndexType	end,
		Fetch &	fetch,
		const Reduce &	reduce,
		Keep &	keep,
		const FetchValue &	identity
	)

Method for performing general reduction on matrix rows.

Template Parameters

Fetch is a type of lambda function for data fetch declared as

auto fetch = [=] __cuda_callable__ ( IndexType rowIdx, IndexType columnIdx, RealType elementValue ) -> FetchValue { ... };

The return type of this lambda can be any non void.

Template Parameters

Reduce is a type of lambda function for reduction declared as

auto reduce = [=] __cuda_callable__ ( const FetchValue& v1, const FetchValue& v2 ) -> FetchValue { ... };

Template Parameters

Keep	is a type of lambda function for storing results of reduction in each row. It is declared as

auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const double& value ) { ... };

Template Parameters

FetchValue is type returned by the Fetch lambda function.

Parameters

begin	defines beginning of the range [begin,end) of rows to be processed.
end	defines ending of the range [begin,end) of rows to be processed.
fetch	is an instance of lambda function for data fetch.
reduce	is an instance of lambda function for reduction.
keep	in an instance of lambda function for storing results.
identity	is the identity element for the reduction operation, i.e. element which does not change the result of the reduction.

Example: #include <iostream>

#include <iomanip>

#include <functional>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

template< typename Device >

void reduceRows()

{

TNL::Matrices::DenseMatrix< double, Device > matrix {

{ 1, 0, 0, 0, 0 },

{ 1, 2, 0, 0, 0 },

{ 0, 1, 8, 0, 0 },

{ 0, 0, 1, 9, 0 },

{ 0, 0, 0, 0, 1 } };

/***

* Find largest element in each row.

*/

TNL::Containers::Vector< double, Device > rowMax( matrix.getRows() );

/***

* Prepare vector view for lambdas.

*/

auto rowMaxView = rowMax.getView();

/***

* Fetch lambda just returns absolute value of matrix elements.

*/

auto fetch = [=] __cuda_callable__ ( int rowIdx, int columnIdx, const double& value ) -> double {

return TNL::abs( value );

};

/***

* Reduce lambda return maximum of given values.

*/

auto reduce = [=] __cuda_callable__ ( double& a, const double& b ) -> double {

return TNL::max( a, b );

};

/***

* Keep lambda store the largest value in each row to the vector rowMax.

*/

auto keep = [=] __cuda_callable__ ( int rowIdx, const double& value ) mutable {

rowMaxView[ rowIdx ] = value;

};

/***

* Compute the largest values in each row.

*/

matrix.reduceRows( 0, matrix.getRows(), fetch, reduce, keep, std::numeric_limits< double >::lowest() );

std::cout << "Max. elements in rows are: " << rowMax << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Rows reduction on host:" << std::endl;

reduceRows< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Rows reduction on CUDA device:" << std::endl;

reduceRows< TNL::Devices::Cuda >();

#endif

}

TNL::Matrices::DenseMatrix::reduceRows
void reduceRows(IndexType begin, IndexType end, Fetch &fetch, const Reduce &reduce, Keep &keep, const FetchValue &identity)
Method for performing general reduction on matrix rows.
Definition: DenseMatrix.hpp:235

Output: Rows reduction on host:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

Rows reduction on CUDA device:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

◆ reduceRows() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Fetch , typename Reduce , typename Keep , typename FetchValue >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::reduceRows	(	IndexType	begin,
		IndexType	end,
		Fetch &	fetch,
		const Reduce &	reduce,
		Keep &	keep,
		const FetchValue &	identity
	)		const

Method for performing general reduction on matrix rows for constant instances.

Template Parameters

Fetch is a type of lambda function for data fetch declared as

auto fetch = [=] __cuda_callable__ ( IndexType rowIdx, IndexType columnIdx, RealType elementValue ) -> FetchValue { ... };

The return type of this lambda can be any non void.

Template Parameters

Reduce is a type of lambda function for reduction declared as

auto reduce = [=] __cuda_callable__ ( const FetchValue& v1, const FetchValue& v2 ) -> FetchValue { ... };

Template Parameters

Keep	is a type of lambda function for storing results of reduction in each row. It is declared as

auto keep = [=] __cuda_callable__ ( const IndexType rowIdx, const double& value ) { ... };

Template Parameters

FetchValue is type returned by the Fetch lambda function.

Parameters

begin	defines beginning of the range [begin,end) of rows to be processed.
end	defines ending of the range [begin,end) of rows to be processed.
fetch	is an instance of lambda function for data fetch.
reduce	is an instance of lambda function for reduction.
keep	in an instance of lambda function for storing results.
identity	is the identity element for the reduction operation, i.e. element which does not change the result of the reduction.

Example: #include <iostream>

#include <iomanip>

#include <functional>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

template< typename Device >

void reduceRows()

{

TNL::Matrices::DenseMatrix< double, Device > matrix {

{ 1, 0, 0, 0, 0 },

{ 1, 2, 0, 0, 0 },

{ 0, 1, 8, 0, 0 },

{ 0, 0, 1, 9, 0 },

{ 0, 0, 0, 0, 1 } };

/***

* Find largest element in each row.

*/

TNL::Containers::Vector< double, Device > rowMax( matrix.getRows() );

/***

* Prepare vector view for lambdas.

*/

auto rowMaxView = rowMax.getView();

/***

* Fetch lambda just returns absolute value of matrix elements.

*/

auto fetch = [=] __cuda_callable__ ( int rowIdx, int columnIdx, const double& value ) -> double {

return TNL::abs( value );

};

/***

* Reduce lambda return maximum of given values.

*/

auto reduce = [=] __cuda_callable__ ( double& a, const double& b ) -> double {

return TNL::max( a, b );

};

/***

* Keep lambda store the largest value in each row to the vector rowMax.

*/

auto keep = [=] __cuda_callable__ ( int rowIdx, const double& value ) mutable {

rowMaxView[ rowIdx ] = value;

};

/***

* Compute the largest values in each row.

*/

matrix.reduceRows( 0, matrix.getRows(), fetch, reduce, keep, std::numeric_limits< double >::lowest() );

std::cout << "Max. elements in rows are: " << rowMax << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Rows reduction on host:" << std::endl;

reduceRows< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Rows reduction on CUDA device:" << std::endl;

reduceRows< TNL::Devices::Cuda >();

#endif

}

Output: Rows reduction on host:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

Rows reduction on CUDA device:

Max. elements in rows are: [ 1, 2, 8, 9, 1 ]

◆ save() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::save ( const String & fileName ) const

Method for saving the matrix to the file with given filename.

Parameters

fileName is name of the file.

◆ save() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::save ( File & file ) const

overridevirtual

Method for saving the matrix to a file.

Parameters

file	is the file where the matrix will be saved.

Reimplemented from TNL::Object.

◆ sequentialForAllRows() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::sequentialForAllRows ( Function && function )

This method calls sequentialForRows for all matrix rows.

See DenseMatrix::sequentialForAllRows.

Template Parameters

Function is a type of lambda function that will operate on matrix elements.

Parameters

function is an instance of the lambda function to be called in each row.

◆ sequentialForAllRows() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::sequentialForAllRows ( Function && function ) const

This method calls sequentialForRows for all matrix rows (for constant instances).

See DenseMatrix::sequentialForRows.

Template Parameters

Function is a type of lambda function that will operate on matrix elements.

Parameters

function is an instance of the lambda function to be called in each row.

◆ sequentialForRows() [1/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::sequentialForRows	(	IndexType	begin,
		IndexType	end,
		Function &&	function
	)

Method for sequential iteration over all matrix rows for non-constant instances.

Template Parameters

Function is type of lambda function that will operate on matrix elements. It is should have form like

auto function = [] __cuda_callable__ ( RowView& row ) { ... };

RowView represents matrix row - see TNL::Matrices::DenseMatrix::RowView.

Parameters

begin	defines beginning of the range [begin,end) of rows to be processed.
end	defines ending of the range [begin,end) of rows to be processed.
function	is an instance of the lambda function to be called in each row.

◆ sequentialForRows() [2/2]

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Function >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::sequentialForRows	(	IndexType	begin,
		IndexType	end,
		Function &&	function
	)		const

Method for sequential iteration over all matrix rows for constant instances.

Template Parameters

Function is type of lambda function that will operate on matrix elements. It is should have form like

auto function = [] __cuda_callable__ ( RowView& row ) { ... };

RowView represents matrix row - see TNL::Matrices::DenseMatrix::RowView.

Parameters

begin	defines beginning of the range [begin,end) of rows to be processed.
end	defines ending of the range [begin,end) of rows to be processed.
function	is an instance of the lambda function to be called in each row.

◆ setDimensions()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::setDimensions	(	IndexType	rows,
		IndexType	columns
	)

overridevirtual

Set number of rows and columns of this matrix.

Parameters

rows	is the number of matrix rows.
columns	is the number of matrix columns.

Reimplemented from TNL::Matrices::Matrix< Real, Device, Index, RealAllocator >.

◆ setElement()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

__cuda_callable__ void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::setElement	(	IndexType	row,
		IndexType	column,
		const RealType &	value
	)

Sets element at given row and column to given value.

Parameters

row	is row index of the element.
column	is columns index of the element.
value	is the value the element will be set to.

Example: #include <iostream>

#include <TNL/Algorithms/parallelFor.h>

#include <TNL/Containers/StaticArray.h>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Pointers/SharedPointer.h>

#include <TNL/Pointers/SmartPointersRegister.h>

template< typename Device >

void setElements()

{

using MatrixType = TNL::Matrices::DenseMatrix< double, Device >;

TNL::Pointers::SharedPointer< MatrixType > matrix( 5, 5 );

for( int i = 0; i < 5; i++ )

matrix->setElement( i, i, i );

std::cout << "Matrix set from the host:" << std::endl;

std::cout << *matrix << std::endl;

MatrixType* matrix_device = &matrix.template modifyData< Device >();

auto f = [=] __cuda_callable__ ( const TNL::Containers::StaticArray< 2, int >& i ) mutable {

matrix_device->addElement( i[ 0 ], i[ 1 ], 5.0 );

};

/***

* For the case when Device is CUDA device we need to synchronize smart

* pointers. To avoid this you may use DenseMatrixView. See

* DenseMatrixView::getRow example for details.

*/

TNL::Pointers::synchronizeSmartPointersOnDevice< Device >();

TNL::Containers::StaticArray< 2, int > begin = { 0, 0 };

TNL::Containers::StaticArray< 2, int > end = { 5, 5 };

TNL::Algorithms::parallelFor< Device >( begin, end, f );

std::cout << "Matrix set from its native device:" << std::endl;

std::cout << *matrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Set elements on host:" << std::endl;

setElements< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Set elements on CUDA device:" << std::endl;

setElements< TNL::Devices::Cuda >();

#endif

}

TNL::Containers::StaticArray
Array with constant size.
Definition: StaticArray.h:24

TNL::Matrices::DenseMatrix::setElements
void setElements(std::initializer_list< std::initializer_list< Value > > data)
This method recreates the dense matrix from 2D initializer list.
Definition: DenseMatrix.hpp:45

Output: Set elements on host:

Matrix set from the host:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:0 1:1 2:0 3:0 4:0

Row: 2 -> 0:0 1:0 2:2 3:0 4:0

Row: 3 -> 0:0 1:0 2:0 3:3 4:0

Row: 4 -> 0:0 1:0 2:0 3:0 4:4

Matrix set from its native device:

Row: 0 -> 0:5 1:5 2:5 3:5 4:5

Row: 1 -> 0:5 1:6 2:5 3:5 4:5

Row: 2 -> 0:5 1:5 2:7 3:5 4:5

Row: 3 -> 0:5 1:5 2:5 3:8 4:5

Row: 4 -> 0:5 1:5 2:5 3:5 4:9

Set elements on CUDA device:

Matrix set from the host:

Row: 0 -> 0:0 1:0 2:0 3:0 4:0

Row: 1 -> 0:0 1:1 2:0 3:0 4:0

Row: 2 -> 0:0 1:0 2:2 3:0 4:0

Row: 3 -> 0:0 1:0 2:0 3:3 4:0

Row: 4 -> 0:0 1:0 2:0 3:0 4:4

Matrix set from its native device:

Row: 0 -> 0:5 1:5 2:5 3:5 4:5

Row: 1 -> 0:5 1:6 2:5 3:5 4:5

Row: 2 -> 0:5 1:5 2:7 3:5 4:5

Row: 3 -> 0:5 1:5 2:5 3:8 4:5

Row: 4 -> 0:5 1:5 2:5 3:5 4:9

◆ setElements()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename Value >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::setElements ( std::initializer_list< std::initializer_list< Value > > data )

This method recreates the dense matrix from 2D initializer list.

The number of matrix rows is set to the outer list size and the number of matrix columns is set to maximum size of inner lists. Missing elements are filled in with zeros.

Parameters

data	is a initializer list of initializer lists representing list of matrix rows.

Example: #include <iostream>

#include <TNL/Matrices/DenseMatrix.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

template< typename Device >

void setElementsExample()

{

TNL::Matrices::DenseMatrix< double, Device > matrix;

matrix.setElements( {

{ 1, 2, 3, 4, 5, 6 },

{ 7, 8, 9, 10, 11, 12 },

{ 13, 14, 15, 16, 17, 18 }

} );

std::cout << matrix << std::endl;

TNL::Matrices::DenseMatrix< double, Device > triangularMatrix;

triangularMatrix.setElements( {

{ 1 },

{ 2, 3 },

{ 4, 5, 6 },

{ 7, 8, 9, 10 },

{ 11, 12, 13, 14, 15 }

} );

std::cout << triangularMatrix << std::endl;

}

int main( int argc, char* argv[] )

{

std::cout << "Setting matrix elements on host: " << std::endl;

setElementsExample< TNL::Devices::Host >();

#ifdef __CUDACC__

std::cout << "Setting matrix elements on CUDA device: " << std::endl;

setElementsExample< TNL::Devices::Cuda >();

#endif

}

Output: Setting matrix elements on host:

Row: 0 -> 0:1 1:2 2:3 3:4 4:5 5:6

Row: 1 -> 0:7 1:8 2:9 3:10 4:11 5:12

Row: 2 -> 0:13 1:14 2:15 3:16 4:17 5:18

Row: 0 -> 0:1 1:0 2:0 3:0 4:0

Row: 1 -> 0:2 1:3 2:0 3:0 4:0

Row: 2 -> 0:4 1:5 2:6 3:0 4:0

Row: 3 -> 0:7 1:8 2:9 3:10 4:0

Row: 4 -> 0:11 1:12 2:13 3:14 4:15

Setting matrix elements on CUDA device:

Row: 0 -> 0:1 1:2 2:3 3:4 4:5 5:6

Row: 1 -> 0:7 1:8 2:9 3:10 4:11 5:12

Row: 2 -> 0:13 1:14 2:15 3:16 4:17 5:18

Row: 0 -> 0:1 1:0 2:0 3:0 4:0

Row: 1 -> 0:2 1:3 2:0 3:0 4:0

Row: 2 -> 0:4 1:5 2:6 3:0 4:0

Row: 3 -> 0:7 1:8 2:9 3:10 4:0

Row: 4 -> 0:11 1:12 2:13 3:14 4:15

◆ setLike()

template<typename Matrix >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::setLike ( const Matrix & matrix )

Set the number of matrix rows and columns by the given matrix.

Template Parameters

Matrix is matrix type. This can be any matrix having methods getRows and getColumns.

Parameters

matrix in the input matrix dimensions of which are to be adopted.

◆ setRowCapacities()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename RowCapacitiesVector >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::setRowCapacities ( const RowCapacitiesVector & rowCapacities )

This method is only for the compatibility with the sparse matrices.

This method does nothing. In debug mode it contains assertions checking that given rowCapacities are compatible with the current matrix dimensions.

◆ setValue()

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::setValue ( const RealType & v )

Sets all matrix elements to value v.

Parameters

v	is value all matrix elements will be set to.

◆ vectorProduct()

template<typename Real , typename Device , typename Index , ElementsOrganization Organization, typename RealAllocator >

template<typename InVector , typename OutVector >

void TNL::Matrices::DenseMatrix< Real, Device, Index, Organization, RealAllocator >::vectorProduct	(	const InVector &	inVector,
		OutVector &	outVector,
		const RealType &	matrixMultiplicator = `1.0`,
		const RealType &	outVectorMultiplicator = `0.0`,
		IndexType	begin = `0`,
		IndexType	end = `0`
	)		const

Computes product of matrix and vector.

More precisely, it computes:

outVector = matrixMultiplicator * ( *this ) * inVector + outVectorMultiplicator * outVector

Template Parameters

InVector	is type of input vector. It can be TNL::Containers::Vector, TNL::Containers::VectorView, TNL::Containers::Array, TNL::Containers::ArrayView, or similar container.
OutVector	is type of output vector. It can be TNL::Containers::Vector, TNL::Containers::VectorView, TNL::Containers::Array, TNL::Containers::ArrayView, or similar container.

Parameters

inVector	is input vector.
outVector	is output vector.
matrixMultiplicator	is a factor by which the matrix is multiplied. It is one by default.
outVectorMultiplicator	is a factor by which the outVector is multiplied before added to the result of matrix-vector product. It is zero by default.
begin	is the beginning of the rows range for which the vector product is computed. It is zero by default.
end	is the end of the rows range for which the vector product is computed. It is number if the matrix rows by default.

Note that the ouput vector dimension must be the same as the number of matrix rows no matter how we set begin and end parameters. These parameters just say that some matrix rows and the output vector elements are omitted.

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

src/TNL/Matrices/DenseMatrix.h
src/TNL/Matrices/DenseMatrix.hpp

Public Types

Public Member Functions

Static Public Member Functions

Protected Types

Protected Member Functions

Protected Attributes

Detailed Description

Member Typedef Documentation

◆ ConstViewType

◆ ViewType

Constructor & Destructor Documentation

◆ DenseMatrix() [1/5]

◆ DenseMatrix() [2/5]

◆ DenseMatrix() [3/5]

◆ DenseMatrix() [4/5]

◆ DenseMatrix() [5/5]

Member Function Documentation

◆ addElement()

◆ forAllElements() [1/2]

◆ forAllElements() [2/2]

◆ forAllRows() [1/2]

◆ forAllRows() [2/2]

◆ forElements() [1/2]

◆ forElements() [2/2]

◆ forRows() [1/2]

◆ forRows() [2/2]

◆ getCompressedRowLengths()

◆ getConstView()

◆ getElement()

◆ getNonzeroElementsCount()

◆ getOrganization()

◆ getRow() [1/2]

◆ getRow() [2/2]

◆ getRowCapacities()

◆ getSerializationType()

◆ getSerializationTypeVirtual()

◆ getView()

◆ isSymmetric()

◆ load() [1/2]

◆ load() [2/2]

◆ operator!=() [1/3]

◆ operator!=() [2/3]

◆ operator!=() [3/3]

◆ operator()() [1/2]

◆ operator()() [2/2]

◆ operator=() [1/6]

◆ operator=() [2/6]

◆ operator=() [3/6]

◆ operator=() [4/6]

◆ operator=() [5/6]

◆ operator=() [6/6]

◆ operator==() [1/3]

◆ operator==() [2/3]

◆ operator==() [3/3]

◆ print()

◆ reduceAllRows() [1/2]

◆ reduceAllRows() [2/2]

◆ reduceRows() [1/2]

◆ reduceRows() [2/2]

◆ save() [1/2]

◆ save() [2/2]

◆ sequentialForAllRows() [1/2]

◆ sequentialForAllRows() [2/2]

◆ sequentialForRows() [1/2]

◆ sequentialForRows() [2/2]

◆ setDimensions()

◆ setElement()

◆ setElements()

◆ setLike()

◆ setRowCapacities()

◆ setValue()

◆ vectorProduct()