TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization > Class Template Reference

Implementation of dense matrix view. More...

#include <TNL/Matrices/DenseMatrixView.h>

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

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

Public Types
using	ConstViewType = DenseMatrixView< std::add_const_t< Real >, Device, Index, Organization >
	Matrix view type for constant instances.
template<typename _Real = Real, typename _Device = Device, typename _Index = Index>
using	Self = DenseMatrixView< _Real, _Device, _Index >
	Helper type for getting self type or its modifications.
using	ViewType = DenseMatrixView< Real, Device, Index, Organization >
	Matrix view type.
Public Types inherited from TNL::Matrices::DenseMatrixBase< double, Devices::Host, int, Algorithms::Segments::DefaultElementsOrganization< Devices::Host >::getOrganization() >
using	ConstRowView
	Type for accessing immutable matrix row.
using	DeviceType
	The device where the matrix is allocated.
using	IndexType
	The type used for matrix elements indexing.
using	RealType
	The type of matrix elements.
using	RowView
	Type for accessing matrix row.
Public Types inherited from TNL::Matrices::MatrixBase< double, Devices::Host, int, GeneralMatrix, Organization >
using	ConstValuesViewType
	Type of constant vector view holding values of matrix elements.
using	DeviceType
	The device where the matrix is allocated.
using	IndexType
	The type used for matrix elements indexing.
using	RealType
	The type of matrix elements.
using	RowCapacitiesType
using	ValuesViewType
	Type of vector view holding values of matrix elements.

Public Member Functions
__cuda_callable__	DenseMatrixView ()=default
	Constructor without parameters.
__cuda_callable__	DenseMatrixView (const DenseMatrixView &matrix)=default
	Copy constructor.
__cuda_callable__	DenseMatrixView (DenseMatrixView &&matrix) noexcept=default
	Move constructor.
__cuda_callable__	DenseMatrixView (Index rows, Index columns, typename Base::ValuesViewType values)
	Constructor with matrix dimensions and values.
__cuda_callable__ void	bind (DenseMatrixView &&view)
	Method for rebinding (reinitialization) using another dense matrix view.
__cuda_callable__ void	bind (DenseMatrixView &view)
	Method for rebinding (reinitialization) using another dense matrix view.
__cuda_callable__ ConstViewType	getConstView () const
	Returns a non-modifiable dense matrix view.
template<int tileDim = 16>
void	getInPlaceTransposition (Real matrixMultiplicator=1.0)
	Performs an in-place transposition of this matrix.
__cuda_callable__ ViewType	getView ()
	Returns a modifiable dense matrix view.
DenseMatrixView &	operator= (const DenseMatrixView &)=delete
	Copy-assignment operator.
DenseMatrixView &	operator= (DenseMatrixView &&)=delete
	Move-assignment operator.
Public Member Functions inherited from TNL::Matrices::DenseMatrixBase< double, Devices::Host, int, Algorithms::Segments::DefaultElementsOrganization< Devices::Host >::getOrganization() >
__cuda_callable__	DenseMatrixBase ()=default
	Constructor without parameters.
__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.
void	addMatrix (const Matrix &matrix, const RealType &matrixMultiplicator=1.0, const RealType &thisMatrixMultiplicator=1.0, TransposeState transpose=TransposeState::None)
	Computes matrix addition.
void	forAllElements (Function &&function) const
	This method calls forElements for all matrix rows.
void	forAllRows (Function &&function)
	Method for parallel iteration over all matrix rows.
void	forElements (IndexType begin, IndexType end, Function &&function) const
	Method for iteration over all matrix rows for constant instances.
void	forRows (IndexType begin, IndexType end, Function &&function)
	Method for parallel iteration over matrix rows from interval [begin, end).
void	getCompressedRowLengths (Vector &rowLengths) const
	Computes number of non-zeros in each row.
__cuda_callable__ double	getElement (IndexType row, IndexType column) const
	Returns value of matrix element at position given by its row and column index.
IndexType	getNonzeroElementsCount () const
	Computes a current number of nonzero matrix elements.
__cuda_callable__ ConstRowView	getRow (IndexType rowIdx) const
	Constant getter of simple structure for accessing given matrix row.
void	getRowCapacities (Vector &rowCapacities) const
	Compute capacities of all rows.
__cuda_callable__ IndexType	getRowCapacity (IndexType row) const
	Returns capacity of given matrix row.
bool	operator!= (const DenseMatrixBase< Real_, Device_, Index_, Organization > &matrix) const
	Comparison operator with another dense matrix view.
__cuda_callable__ double &	operator() (IndexType row, IndexType column)
	Returns non-constant reference to element at row row and column column.
DenseMatrixBase &	operator= (const DenseMatrixBase &)=delete
	Copy-assignment operator.
bool	operator== (const DenseMatrixBase< Real_, Device_, Index_, Organization > &matrix) const
	Comparison operator with another dense matrix view.
void	print (std::ostream &str) const
	Method for printing the matrix to output stream.
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.
void	reduceRows (IndexType begin, IndexType end, Fetch &&fetch, const Reduce &reduce, Keep &&keep, const FetchValue &identity) const
void	sequentialForAllRows (Function &&function) const
	This method calls sequentialForRows for all matrix rows (for constant instances).
void	sequentialForRows (IndexType begin, IndexType end, Function &&function) const
	Method for sequential iteration over all matrix rows for constant instances.
__cuda_callable__ void	setElement (IndexType row, IndexType column, const RealType &value)
	Sets element at given row and column to given value.
void	setValue (const RealType &v)
	Sets all matrix elements to value v.
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.
Public Member Functions inherited from TNL::Matrices::MatrixBase< double, Devices::Host, int, GeneralMatrix, Organization >
__cuda_callable__	MatrixBase ()=default
	Basic constructor with no parameters.
IndexType	getAllocatedElementsCount () const
	Tells the number of allocated matrix elements.
__cuda_callable__ IndexType	getColumns () const
	Returns number of matrix columns.
__cuda_callable__ IndexType	getRows () const
	Returns number of matrix rows.
__cuda_callable__ const ValuesViewType &	getValues () const
	Returns a constant reference to a vector with the matrix elements values.
__cuda_callable__ MatrixBase &	operator= (const MatrixBase &)=delete
	Copy-assignment operator.

Additional Inherited Members
Static Public Member Functions inherited from TNL::Matrices::DenseMatrixBase< double, Devices::Host, int, Algorithms::Segments::DefaultElementsOrganization< Devices::Host >::getOrganization() >
static std::string	getSerializationType ()
	Returns string with serialization type.
Static Public Member Functions inherited from TNL::Matrices::MatrixBase< double, Devices::Host, int, GeneralMatrix, Organization >
static constexpr ElementsOrganization	getOrganization ()
	Matrix elements organization getter.
static constexpr bool	isBinary ()
	Test of binary matrix type.
static constexpr bool	isSymmetric ()
	Test of symmetric matrix type.
Protected Types inherited from TNL::Matrices::DenseMatrixBase< double, Devices::Host, int, Algorithms::Segments::DefaultElementsOrganization< Devices::Host >::getOrganization() >
using	Base
using	SegmentsReductionKernel
using	SegmentsType
using	SegmentsViewType
using	SegmentViewType
Protected Member Functions inherited from TNL::Matrices::DenseMatrixBase< double, Devices::Host, int, Algorithms::Segments::DefaultElementsOrganization< Devices::Host >::getOrganization() >
__cuda_callable__ void	bind (IndexType rows, IndexType columns, typename Base::ValuesViewType values, SegmentsViewType segments)
	Re-initializes the internal attributes of the base class.
__cuda_callable__ IndexType	getElementIndex (IndexType row, IndexType column) const
Protected Member Functions inherited from TNL::Matrices::MatrixBase< double, Devices::Host, int, GeneralMatrix, Organization >
__cuda_callable__ void	bind (IndexType rows, IndexType columns, ValuesViewType values)
	Re-initializes the internal attributes of the base class.
Protected Attributes inherited from TNL::Matrices::DenseMatrixBase< double, Devices::Host, int, Algorithms::Segments::DefaultElementsOrganization< Devices::Host >::getOrganization() >
SegmentsViewType	segments
Protected Attributes inherited from TNL::Matrices::MatrixBase< double, Devices::Host, int, GeneralMatrix, Organization >
IndexType	columns
IndexType	rows
ValuesViewType	values

Detailed Description

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

Implementation of dense matrix view.

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

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.
MatrixElementsOrganization	tells the ordering of matrix elements in memory. It is either TNL::Algorithms::Segments::RowMajorOrder or TNL::Algorithms::Segments::ColumnMajorOrder.

See DenseMatrix.

Member Typedef Documentation

ConstViewType

template<typename Real = double, typename Device = Devices::Host, typename Index = int, ElementsOrganization Organization = Algorithms::Segments::DefaultElementsOrganization< Device >::getOrganization()>

using TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::ConstViewType = DenseMatrixView< std::add_const_t< Real >, Device, Index, Organization >

Matrix view type for constant instances.

See DenseMatrixView.

ViewType

template<typename Real = double, typename Device = Devices::Host, typename Index = int, ElementsOrganization Organization = Algorithms::Segments::DefaultElementsOrganization< Device >::getOrganization()>

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

Matrix view type.

See DenseMatrixView.

Constructor & Destructor Documentation

DenseMatrixView() [1/3]

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

__cuda_callable__ TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::DenseMatrixView	(	Index	rows,
		Index	columns,
		typename Base::ValuesViewType	values )

Constructor with matrix dimensions and values.

Organization of matrix elements values in

Parameters

rows	number of matrix rows.
columns	number of matrix columns.
values	is vector view with matrix elements values.

Example

#include <iostream>
#include <TNL/Matrices/DenseMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void
createMatrixView()
{
   TNL::Containers::Vector< double, Device > values{
      // clang-format off
      1,  2,  3,  4,
      5,  6,  7,  8,
      9, 10, 11, 12
      // clang-format on
   };
 
   /***
    * Create dense matrix view with row major order
    */
   TNL::Matrices::DenseMatrixView< double, Device, int, TNL::Algorithms::Segments::RowMajorOrder > rowMajorMatrix(
      3, 4, values.getView() );
   std::cout << "Row major order matrix:" << std::endl;
   std::cout << rowMajorMatrix << std::endl;
 
   /***
    * Create dense matrix view with column major order
    */
   TNL::Matrices::DenseMatrixView< double, Device, int, TNL::Algorithms::Segments::RowMajorOrder > columnMajorMatrix(
      4, 3, values.getView() );
   std::cout << "Column major order matrix:" << std::endl;
   std::cout << columnMajorMatrix << std::endl;
}
 
int
main( int argc, char* argv[] )
{
   std::cout << "Creating matrix view on host: " << std::endl;
   createMatrixView< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix view on CUDA device: " << std::endl;
   createMatrixView< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix view on host: 
Row major order matrix:
Row: 0 ->     0:1       1:2       2:3       3:4   
Row: 1 ->     0:5       1:6       2:7       3:8   
Row: 2 ->     0:9       1:10      2:11      3:12  
Column major order matrix:
Row: 0 ->     0:1       1:2       2:3   
Row: 1 ->     0:4       1:5       2:6   
Row: 2 ->     0:7       1:8       2:9   
Row: 3 ->     0:10      1:11      2:12  
Creating matrix view on CUDA device: 
Row major order matrix:
Row: 0 ->     0:1       1:2       2:3       3:4   
Row: 1 ->     0:5       1:6       2:7       3:8   
Row: 2 ->     0:9       1:10      2:11      3:12  
Column major order matrix:
Row: 0 ->     0:1       1:2       2:3   
Row: 1 ->     0:4       1:5       2:6   
Row: 2 ->     0:7       1:8       2:9   
Row: 3 ->     0:10      1:11      2:12  

DenseMatrixView() [2/3]

template<typename Real = double, typename Device = Devices::Host, typename Index = int, ElementsOrganization Organization = Algorithms::Segments::DefaultElementsOrganization< Device >::getOrganization()>

__cuda_callable__ TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::DenseMatrixView ( const DenseMatrixView< Real, Device, Index, Organization > & matrix )

default

Copy constructor.

Parameters

matrix

is the source matrix view.

DenseMatrixView() [3/3]

template<typename Real = double, typename Device = Devices::Host, typename Index = int, ElementsOrganization Organization = Algorithms::Segments::DefaultElementsOrganization< Device >::getOrganization()>

__cuda_callable__ TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::DenseMatrixView ( DenseMatrixView< Real, Device, Index, Organization > && matrix )

defaultnoexcept

Move constructor.

Parameters

matrix

is the source matrix view.

Member Function Documentation

bind() [1/2]

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

__cuda_callable__ void TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::bind

(

DenseMatrixView< Real, Device, Index, Organization > &&

view

)

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

Parameters

view	The dense matrix view to be bound.

bind() [2/2]

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

__cuda_callable__ void TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::bind

(

DenseMatrixView< Real, Device, Index, Organization > &

view

)

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

Parameters

view	The dense matrix view to be bound.

getConstView()

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

__cuda_callable__ auto TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::getConstView ( ) const

nodiscard

Returns a non-modifiable dense matrix view.

Returns

dense matrix view.

getInPlaceTransposition()

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

template<int tileDim>

void TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::getInPlaceTransposition

(

Real

matrixMultiplicator = 1.0

)

Performs an in-place transposition of this matrix.

This method transposes this matrix in place, modifying the original matrix. The operation can optionally scale the matrix by a specified factor.

Template Parameters

tileDim

Tile dimension for GPU computation optimization. Default is 16.

Parameters

matrixMultiplicator

A scalar value by which the matrix is scaled during the transposition. Default is 1.0.

Example

#include <iostream>
#include <TNL/Matrices/DenseMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void
getInPlaceTranspositionExample()
{
   // clang-format off
   TNL::Matrices::DenseMatrix< double, Device > matrix{
      { 1, 2, 3, 4 },
      { 6, 7, 8, 9 },
      { 10, 11, 12, 13 },
      { 14, 15, 16, 17 }
   };
   // clang-format on
 
   std::cout << "Dense matrix: " << std::endl << matrix << std::endl;
 
   matrix.getInPlaceTransposition();
 
   std::cout << "Transposed dense matrix: " << std::endl << matrix << std::endl;
}
 
int
main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on CPU ... " << std::endl;
   getInPlaceTranspositionExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA GPU ... " << std::endl;
   getInPlaceTranspositionExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on CPU ... 
Dense matrix: 
Row: 0 ->     0:1       1:2       2:3       3:4   
Row: 1 ->     0:6       1:7       2:8       3:9   
Row: 2 ->     0:10      1:11      2:12      3:13  
Row: 3 ->     0:14      1:15      2:16      3:17  
Transposed dense matrix: 
Row: 0 ->     0:1       1:6       2:10      3:14  
Row: 1 ->     0:2       1:7       2:11      3:15  
Row: 2 ->     0:3       1:8       2:12      3:16  
Row: 3 ->     0:4       1:9       2:13      3:17  
Creating matrix on CUDA GPU ... 
Dense matrix: 
Row: 0 ->     0:1       1:2       2:3       3:4   
Row: 1 ->     0:6       1:7       2:8       3:9   
Row: 2 ->     0:10      1:11      2:12      3:13  
Row: 3 ->     0:14      1:15      2:16      3:17  
Transposed dense matrix: 
Row: 0 ->     0:1       1:6       2:10      3:14  
Row: 1 ->     0:2       1:7       2:11      3:15  
Row: 2 ->     0:3       1:8       2:12      3:16  
Row: 3 ->     0:4       1:9       2:13      3:17  

getView()

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

__cuda_callable__ auto TNL::Matrices::DenseMatrixView< Real, Device, Index, Organization >::getView ( )

nodiscard

Returns a modifiable dense matrix view.

Returns

dense matrix view.

operator=()

template<typename Real = double, typename Device = Devices::Host, typename Index = int, ElementsOrganization Organization = Algorithms::Segments::DefaultElementsOrganization< Device >::getOrganization()>

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

delete

Copy-assignment operator.

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

---

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

• src/TNL/Matrices/DenseMatrixView.h
• src/TNL/Matrices/DenseMatrixView.hpp