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

Implementation of sparse tridiagonal matrix. More...

#include <TNL/Matrices/TridiagonalMatrix.h>

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

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

Public Types
using	BaseType = Matrix< Real, Device, Index, RealAllocator >
using	ConstRowView = typename ViewType::ConstRowView
	Type for accessing constant matrix rows.
using	ConstViewType = TridiagonalMatrixView< typename std::add_const< Real >::type, Device, Index, Organization >
	Matrix view type for constant instances. More...
using	DeviceType = Device
	The device where the matrix is allocated.
using	IndexerType = details::TridiagonalMatrixIndexer< Index, Organization >
using	IndexType = Index
	The type used for matrix elements indexing.
using	RealAllocatorType = RealAllocator
	The allocator for matrix elements values.
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 = TridiagonalMatrix< _Real, _Device, _Index, _Organization, _RealAllocator >
	Helper type for getting self type or its modifications.
using	ValuesVectorType = typename BaseType::ValuesType
using	ValuesViewType = typename ValuesVectorType::ViewType
using	ViewType = TridiagonalMatrixView< 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
	TridiagonalMatrix ()
	Constructor with no parameters.
	TridiagonalMatrix (const TridiagonalMatrix &matrix)=default
	Copy constructor. More...
template<typename ListReal >
	TridiagonalMatrix (IndexType columns, const std::initializer_list< std::initializer_list< ListReal > > &data)
	Constructor with matrix dimensions, diagonals offsets and matrix elements. More...
	TridiagonalMatrix (IndexType rows, IndexType columns)
	Constructor with matrix dimensions. More...
	TridiagonalMatrix (TridiagonalMatrix &&matrix) noexcept=default
	Move constructor. 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 Real_ , typename Device_ , typename Index_ , ElementsOrganization Organization_, typename RealAllocator_ >
void	addMatrix (const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &matrix, const RealType &matrixMultiplicator=1.0, const RealType &thisMatrixMultiplicator=1.0)
template<typename Function >
void	forAllElements (Function &function)
	Method for iteration over all matrix rows for non-constant instances. More...
template<typename Function >
void	forAllElements (Function &function) const
	Method for iteration over 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 matrix rows for non-constant instances. More...
template<typename Function >
void	forElements (IndexType begin, IndexType end, Function &function) const
	Method for iteration over 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 Vector >
void	getCompressedRowLengths (Vector &rowLengths) const
	Computes number of non-zeros in each row. More...
ConstViewType	getConstView () const
	Returns a non-modifiable view of the tridiagonal matrix. More...
__cuda_callable__ RealType	getElement (IndexType row, IndexType column) const
	Returns value of matrix element at position given by its row and column index. More...
IndexerType &	getIndexer ()
	This method returns matrix elements indexer used by this matrix. More...
const IndexerType &	getIndexer () const
	This method returns matrix elements indexer used by this matrix. More...
IndexType	getNonzeroElementsCount () const override
	Returns number of non-zero matrix elements. More...
__cuda_callable__ IndexType	getPaddingIndex () const
	Returns padding index denoting padding zero elements. More...
__cuda_callable__ RowView	getRow (IndexType rowIdx)
	Non-constant getter of simple structure for accessing given matrix row. More...
__cuda_callable__ ConstRowView	getRow (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 Real2 , typename Index2 >
void	getTransposition (const TridiagonalMatrix< Real2, Device, Index2 > &matrix, const RealType &matrixMultiplicator=1.0)
ViewType	getView ()
	Returns a modifiable view of the tridiagonal 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_ , ElementsOrganization Organization_, typename RealAllocator_ >
bool	operator!= (const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &matrix) const
	Comparison operator with another tridiagonal matrix. More...
TridiagonalMatrix &	operator= (const TridiagonalMatrix &matrix)
	Assignment of exactly the same matrix type. More...
template<typename Real_ , typename Device_ , typename Index_ , ElementsOrganization Organization_, typename RealAllocator_ >
TridiagonalMatrix &	operator= (const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &matrix)
	Assignment of another tridiagonal matrix. More...
template<typename Real_ , typename Device_ , typename Index_ , ElementsOrganization Organization_, typename RealAllocator_ >
TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator > &	operator= (const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &matrix)
template<typename Real_ , typename Device_ , typename Index_ , ElementsOrganization Organization_, typename RealAllocator_ >
bool	operator== (const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &matrix) const
	Comparison operator with another tridiagonal matrix. 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, 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, Reduce &reduce, Keep &keep, const FetchReal &identity) const
	Method for performing general reduction on all matrix rows of constant matrix instances. More...
template<typename Fetch , typename Reduce , typename Keep , typename FetchReal >
void	reduceRows (IndexType begin, IndexType end, Fetch &fetch, Reduce &reduce, Keep &keep, const FetchReal &identity)
	Method for performing general reduction on matrix rows. More...
template<typename Fetch , typename Reduce , typename Keep , typename FetchReal >
void	reduceRows (IndexType begin, IndexType end, Fetch &fetch, Reduce &reduce, Keep &keep, const FetchReal &identity) const
	Method for performing general reduction on matrix rows of constant matrix 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 matrix dimensions. 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 ListReal >
void	setElements (const std::initializer_list< std::initializer_list< ListReal > > &data)
	Set matrix elements from an initializer list. More...
template<typename Real_ , typename Device_ , typename Index_ , ElementsOrganization Organization_, typename RealAllocator_ >
void	setLike (const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &matrix)
	Setup the matrix dimensions and diagonals offsets based on another tridiagonal matrix. More...
template<typename RowCapacitiesVector >
void	setRowCapacities (const RowCapacitiesVector &rowCapacities)
	This method is for compatibility with SparseMatrix. More...
void	setValue (const RealType &value)
	Set all matrix elements to given value. More...
template<typename InVector , typename OutVector >
void	vectorProduct (const InVector &inVector, OutVector &outVector, RealType matrixMultiplicator=1.0, 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 ()
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 Member Functions
__cuda_callable__ IndexType	getElementIndex (IndexType row, IndexType column) const

Protected Attributes
IndexerType	indexer
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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >

Implementation of sparse tridiagonal matrix.

Use this matrix type for storing of tridiagonal matrices i.e., matrices having non-zero matrix elements only on its diagonal and immediately above and bellow the diagonal. This is an example:

\[ \left( \begin{array}{ccccccc} 4 & -1 & . & . & . & . \\ -1 & 4 & -1 & . & . & . \\ . & -1 & 4 & -1 & . & . \\ . & . & -1 & 4 & -1 & . \\ . & . & . & -1 & 4 & -1 \\ . & . & . & . & -1 & 4 \end{array} \right) \]

Advantage is that we do not store the column indexes explicitly as it is in SparseMatrix. This can reduce significantly the memory requirements which also means better performance. See the following table for the storage requirements comparison between TridiagonalMatrix and SparseMatrix.

Real	Index	SparseMatrix	TridiagonalMatrix	Ratio
float	32-bit int	8 bytes per element	4 bytes per element	50%
double	32-bit int	12 bytes per element	8 bytes per element	75%
float	64-bit int	12 bytes per element	4 bytes per element	30%
double	64-bit int	16 bytes per element	8 bytes per element	50%

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

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 >>

using TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::ConstViewType = TridiagonalMatrixView< typename std::add_const< Real >::type, Device, Index, Organization >

Matrix view type for constant instances.

See TridiagonalMatrixView.

ViewType

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 >>

using TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::ViewType = TridiagonalMatrixView< Real, Device, Index, Organization >

Type of related matrix view.

See TridiagonalMatrixView.

Constructor & Destructor Documentation

TridiagonalMatrix() [1/4]

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

TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::TridiagonalMatrix	(	IndexType	rows,
		IndexType	columns
	)

Constructor with matrix dimensions.

Parameters

rows	is number of matrix rows.
columns	is number of matrix columns.

TridiagonalMatrix() [2/4]

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

template<typename ListReal >

TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::TridiagonalMatrix	(	IndexType	columns,
		const std::initializer_list< std::initializer_list< ListReal > > &	data
	)

Constructor with matrix dimensions, diagonals offsets and matrix elements.

The number of matrix rows is deduced from the size of the initializer list data.

Template Parameters

ListReal

is type used in the initializer list defining matrix elements values.

Parameters

columns	is number of matrix columns.
data	is initializer list holding matrix elements. The size of the outer list defines the number of matrix rows. Each inner list defines values of each sub-diagonal and so its size should be lower or equal to three. Values of sub-diagonals which do not fit to given row are omitted.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
 
template< typename Device >
void createTridiagonalMatrix()
{
   const int matrixSize = 6;
 
   /***
    * Setup the following matrix (dots represent zeros):
    *
    * /  2 -1 .   .  .  . \
    * | -1  2 -1  .  .  . |
    * |  . -1  2 -1  .  . |
    * |  .  . -1  2 -1  . |
    * |  .  .  . -1  2 -1 |
    * \  .  .  .  . -1  2 /
    *
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix(
      matrixSize, {
   /***
    * To set the matrix elements we first extend the diagonals to their full
    * lengths even outside the matrix (dots represent zeros and zeros are
    * artificial zeros used for memory alignment):
    *
    * 0 /  2 -1 .   .  .  . \    -> {  0,  2, -1 }
    *   | -1  2 -1  .  .  . |    -> { -1,  2, -1 }
    *   |  . -1  2 -1  .  . |    -> { -1,  2, -1 }
    *   |  .  . -1  2 -1  . |    -> { -1,  2, -1 }
    *   |  .  .  . -1  2 -1 |    -> { -1,  2, -1 }
    *   \  .  .  .  . -1  2 / 0  -> { -1,  2,  0 }
    *
    */
      {  0,  2, -1 },
      { -1,  2, -1 },
      { -1,  2, -1 },
      { -1,  2, -1 },
      { -1,  2, -1 },
      { -1,  2,  0 }
      } );
   std::cout << "The matrix reads as: " << std::endl << matrix << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating tridiagonal matrix on CPU ... " << std::endl;
   createTridiagonalMatrix< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating tridiagonal matrix on CUDA GPU ... " << std::endl;
   createTridiagonalMatrix< TNL::Devices::Cuda >();
#endif
}

Output

Creating tridiagonal matrix on CPU ... 
The matrix reads as: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2       5:-1  
Row: 5 ->     4:-1      5:2   
 
Creating tridiagonal matrix on CUDA GPU ... 
The matrix reads as: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2       5:-1  
Row: 5 ->     4:-1      5:2   
 

TridiagonalMatrix() [3/4]

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 >>

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

default

Copy constructor.

Parameters

matrix

is an input matrix.

TridiagonalMatrix() [4/4]

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 >>

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

defaultnoexcept

Move constructor.

Parameters

matrix

is an input matrix.

Member Function Documentation

addElement()

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

__cuda_callable__ void TNL::Matrices::TridiagonalMatrix< 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. TridiagonalMatrix::getRow or TridiagonalMatrix::forElements and TridiagonalMatrix::forAllElements. The call may fail if the matrix row capacity is exhausted.

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void addElements()
{
   const int matrixSize( 5 );
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix(
      matrixSize,    // number of rows
      matrixSize     // number of columns
   );
   for( int i = 0; i < matrixSize; i++ )
      matrix.setElement( i, i, i );
 
   std::cout << "Initial matrix is: " << std::endl << matrix << std::endl;
 
   for( int i = 0; i < matrixSize; i++ )
   {
      if( i > 0 )
         matrix.addElement( i, i - 1, 1.0, 5.0 );
      matrix.addElement( i, i, 1.0, 5.0 );
      if( i < matrixSize - 1 )
         matrix.addElement( i, i + 1, 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 -> 
Row: 1 ->     1:1   
Row: 2 ->     2:2   
Row: 3 ->     3:3   
Row: 4 ->     4:4   
 
Matrix after addition is: 
Row: 0 ->     0:1       1:1   
Row: 1 ->     0:1       1:6       2:1   
Row: 2 ->     1:1       2:11      3:1   
Row: 3 ->     2:1       3:16      4:1   
Row: 4 ->     3:1       4:21  
 
Add elements on CUDA device:
Initial matrix is: 
Row: 0 -> 
Row: 1 ->     1:1   
Row: 2 ->     2:2   
Row: 3 ->     3:3   
Row: 4 ->     4:4   
 
Matrix after addition is: 
Row: 0 ->     0:1       1:1   
Row: 1 ->     0:1       1:6       2:1   
Row: 2 ->     1:1       2:11      3:1   
Row: 3 ->     2:1       3:16      4:1   
Row: 4 ->     3:1       4:21  
 

forAllElements() [1/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::forAllElements

(

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 localIdx, IndexType columnIdx, const RealType& value )
{ ... };

The localIdx parameter is a rank of the non-zero element in given row.

Parameters

function

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

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forAllElementsExample()
{
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    * 0 / 1  3  .  .  . \   -> { 0, 1, 3 }
    *   | 2  1  3  .  . |   -> { 2, 1, 3 }
    *   | .  2  1  3  . |   -> { 2, 1, 3 }
    *   | .  .  2  1  3 |   -> { 2, 1, 3 }
    *   \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix(
      5,      // number of matrix rows
      5 );    // number of matrix columns
 
   auto f = [=] __cuda_callable__ ( int rowIdx, int localIdx, int columnIdx, double& value ) {
      /***
       * 'forElements' method iterates only over matrix elements lying on given subdiagonals
       * and so we do not need to check anything. The element value can be expressed
       * by the 'localIdx' variable, see the following figure:
       *
       *                           0  1  2  <- localIdx values
       *                           -------
       * 0 / 1  3  .  .  . \   -> { 0, 1, 3 }
       *   | 2  1  3  .  . |   -> { 2, 1, 3 }
       *   | .  2  1  3  . |   -> { 2, 1, 3 }
       *   | .  .  2  1  3 |   -> { 2, 1, 3 }
       *   \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
       *
       */
      value = 3 - localIdx;
   };
   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 ->     0:2       1:1   
Row: 1 ->     0:3       1:2       2:1   
Row: 2 ->     1:3       2:2       3:1   
Row: 3 ->     2:3       3:2       4:1   
Row: 4 ->     3:3       4:2   
 
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:1   
Row: 1 ->     0:3       1:2       2:1   
Row: 2 ->     1:3       2:2       3:1   
Row: 3 ->     2:3       3:2       4:1   
Row: 4 ->     3:3       4:2   
 

forAllElements() [2/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::forAllElements

(

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 localIdx, IndexType columnIdx, const RealType& value )
{ ... };

The localIdx parameter is a rank of the non-zero element in given row.

Parameters

function

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

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forAllElementsExample()
{
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    * 0 / 1  3  .  .  . \   -> { 0, 1, 3 }
    *   | 2  1  3  .  . |   -> { 2, 1, 3 }
    *   | .  2  1  3  . |   -> { 2, 1, 3 }
    *   | .  .  2  1  3 |   -> { 2, 1, 3 }
    *   \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix(
      5,      // number of matrix rows
      5 );    // number of matrix columns
 
   auto f = [=] __cuda_callable__ ( int rowIdx, int localIdx, int columnIdx, double& value ) {
      /***
       * 'forElements' method iterates only over matrix elements lying on given subdiagonals
       * and so we do not need to check anything. The element value can be expressed
       * by the 'localIdx' variable, see the following figure:
       *
       *                           0  1  2  <- localIdx values
       *                           -------
       * 0 / 1  3  .  .  . \   -> { 0, 1, 3 }
       *   | 2  1  3  .  . |   -> { 2, 1, 3 }
       *   | .  2  1  3  . |   -> { 2, 1, 3 }
       *   | .  .  2  1  3 |   -> { 2, 1, 3 }
       *   \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
       *
       */
      value = 3 - localIdx;
   };
   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 ->     0:2       1:1   
Row: 1 ->     0:3       1:2       2:1   
Row: 2 ->     1:3       2:2       3:1   
Row: 3 ->     2:3       3:2       4:1   
Row: 4 ->     3:3       4:2   
 
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:1   
Row: 1 ->     0:3       1:2       2:1   
Row: 2 ->     1:3       2:2       3:1   
Row: 3 ->     2:3       3:2       4:1   
Row: 4 ->     3:3       4:2   
 

forAllRows() [1/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< 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 TridiagonalMatrix::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 { ... };

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

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forRowsExample()
{
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   using RowView = typename MatrixType::RowView;
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *    0 /  2  .  .  .  . \  -> { 0, 0, 1 }
    *      | -1  2 -1  .  . |  -> { 0, 2, 1 }
    *      |  . -1  2 -1. . |  -> { 3, 2, 1 }
    *      |  .  . -1  2 -1 |  -> { 3, 2, 1 }
    *      \  .  .  .  .  2 /  -> { 3, 2, 1 }
    *
    */
    const int size = 5;
    MatrixType matrix( size, size );
 
   auto f = [=] __cuda_callable__ ( RowView& row ) {
      const int& rowIdx = row.getRowIndex();
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < size - 1 )        // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
   matrix.forAllRows( f );
   std::cout << matrix << std::endl;
 
   /***
    * Compute sum of elements in each row and store it into a vector.
    */
   TNL::Containers::Vector< double, Device > sum_vector( size );
   auto sum_view = sum_vector.getView();
   matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {
      double sum( 0.0 );
      for( auto element : row )
         sum += TNL::abs( element.value() );
      sum_view[ row.getRowIndex() ] = sum;
   } );
 
   std::cout << "Sums in matrix rows = " << sum_vector << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on host: " << std::endl;
   forRowsExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA device: " << std::endl;
   forRowsExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on host: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]

forAllRows() [2/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< 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 TridiagonalMatrix::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::TridiagonalMatrix::RowView.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forRowsExample()
{
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   using RowView = typename MatrixType::RowView;
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *    0 /  2  .  .  .  . \  -> { 0, 0, 1 }
    *      | -1  2 -1  .  . |  -> { 0, 2, 1 }
    *      |  . -1  2 -1. . |  -> { 3, 2, 1 }
    *      |  .  . -1  2 -1 |  -> { 3, 2, 1 }
    *      \  .  .  .  .  2 /  -> { 3, 2, 1 }
    *
    */
    const int size = 5;
    MatrixType matrix( size, size );
 
   auto f = [=] __cuda_callable__ ( RowView& row ) {
      const int& rowIdx = row.getRowIndex();
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < size - 1 )        // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
   matrix.forAllRows( f );
   std::cout << matrix << std::endl;
 
   /***
    * Compute sum of elements in each row and store it into a vector.
    */
   TNL::Containers::Vector< double, Device > sum_vector( size );
   auto sum_view = sum_vector.getView();
   matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {
      double sum( 0.0 );
      for( auto element : row )
         sum += TNL::abs( element.value() );
      sum_view[ row.getRowIndex() ] = sum;
   } );
 
   std::cout << "Sums in matrix rows = " << sum_vector << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on host: " << std::endl;
   forRowsExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA device: " << std::endl;
   forRowsExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on host: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]

forElements() [1/2]

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

template<typename Function >

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

Method for iteration over 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 localIdx, IndexType columnIdx, const RealType& value )
{ ... };

The localIdx parameter is a rank of the non-zero element in given row.

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forRowsExample()
{
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   using RowView = typename MatrixType::RowView;
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *    0 /  2  .  .  .  . \  -> { 0, 0, 1 }
    *      | -1  2 -1  .  . |  -> { 0, 2, 1 }
    *      |  . -1  2 -1. . |  -> { 3, 2, 1 }
    *      |  .  . -1  2 -1 |  -> { 3, 2, 1 }
    *      \  .  .  .  .  2 /  -> { 3, 2, 1 }
    *
    */
    const int size = 5;
    MatrixType matrix( size, size );
 
   auto f = [=] __cuda_callable__ ( RowView& row ) {
      const int& rowIdx = row.getRowIndex();
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < size - 1 )        // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
   matrix.forAllRows( f );
   std::cout << matrix << std::endl;
 
   /***
    * Compute sum of elements in each row and store it into a vector.
    */
   TNL::Containers::Vector< double, Device > sum_vector( size );
   auto sum_view = sum_vector.getView();
   matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {
      double sum( 0.0 );
      for( auto element : row )
         sum += TNL::abs( element.value() );
      sum_view[ row.getRowIndex() ] = sum;
   } );
 
   std::cout << "Sums in matrix rows = " << sum_vector << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on host: " << std::endl;
   forRowsExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA device: " << std::endl;
   forRowsExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on host: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]

forElements() [2/2]

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

template<typename Function >

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

Method for iteration over 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 localIdx, IndexType columnIdx, const RealType& value )
{ ... };

The localIdx parameter is a rank of the non-zero element in given row.

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forRowsExample()
{
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   using RowView = typename MatrixType::RowView;
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *    0 /  2  .  .  .  . \  -> { 0, 0, 1 }
    *      | -1  2 -1  .  . |  -> { 0, 2, 1 }
    *      |  . -1  2 -1. . |  -> { 3, 2, 1 }
    *      |  .  . -1  2 -1 |  -> { 3, 2, 1 }
    *      \  .  .  .  .  2 /  -> { 3, 2, 1 }
    *
    */
    const int size = 5;
    MatrixType matrix( size, size );
 
   auto f = [=] __cuda_callable__ ( RowView& row ) {
      const int& rowIdx = row.getRowIndex();
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < size - 1 )        // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
   matrix.forAllRows( f );
   std::cout << matrix << std::endl;
 
   /***
    * Compute sum of elements in each row and store it into a vector.
    */
   TNL::Containers::Vector< double, Device > sum_vector( size );
   auto sum_view = sum_vector.getView();
   matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {
      double sum( 0.0 );
      for( auto element : row )
         sum += TNL::abs( element.value() );
      sum_view[ row.getRowIndex() ] = sum;
   } );
 
   std::cout << "Sums in matrix rows = " << sum_vector << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on host: " << std::endl;
   forRowsExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA device: " << std::endl;
   forRowsExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on host: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]

forRows() [1/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< 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 TridiagonalMatrix::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::TridiagonalMatrix::RowView.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forRowsExample()
{
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   using RowView = typename MatrixType::RowView;
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *    0 /  2  .  .  .  . \  -> { 0, 0, 1 }
    *      | -1  2 -1  .  . |  -> { 0, 2, 1 }
    *      |  . -1  2 -1. . |  -> { 3, 2, 1 }
    *      |  .  . -1  2 -1 |  -> { 3, 2, 1 }
    *      \  .  .  .  .  2 /  -> { 3, 2, 1 }
    *
    */
    const int size = 5;
    MatrixType matrix( size, size );
 
   auto f = [=] __cuda_callable__ ( RowView& row ) {
      const int& rowIdx = row.getRowIndex();
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < size - 1 )        // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
   matrix.forAllRows( f );
   std::cout << matrix << std::endl;
 
   /***
    * Compute sum of elements in each row and store it into a vector.
    */
   TNL::Containers::Vector< double, Device > sum_vector( size );
   auto sum_view = sum_vector.getView();
   matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {
      double sum( 0.0 );
      for( auto element : row )
         sum += TNL::abs( element.value() );
      sum_view[ row.getRowIndex() ] = sum;
   } );
 
   std::cout << "Sums in matrix rows = " << sum_vector << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on host: " << std::endl;
   forRowsExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA device: " << std::endl;
   forRowsExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on host: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]

forRows() [2/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< 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 TridiagonalMatrix::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::TridiagonalMatrix::RowView.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
template< typename Device >
void forRowsExample()
{
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   using RowView = typename MatrixType::RowView;
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *    0 /  2  .  .  .  . \  -> { 0, 0, 1 }
    *      | -1  2 -1  .  . |  -> { 0, 2, 1 }
    *      |  . -1  2 -1. . |  -> { 3, 2, 1 }
    *      |  .  . -1  2 -1 |  -> { 3, 2, 1 }
    *      \  .  .  .  .  2 /  -> { 3, 2, 1 }
    *
    */
    const int size = 5;
    MatrixType matrix( size, size );
 
   auto f = [=] __cuda_callable__ ( RowView& row ) {
      const int& rowIdx = row.getRowIndex();
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < size - 1 )        // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
   matrix.forAllRows( f );
   std::cout << matrix << std::endl;
 
   /***
    * Compute sum of elements in each row and store it into a vector.
    */
   TNL::Containers::Vector< double, Device > sum_vector( size );
   auto sum_view = sum_vector.getView();
   matrix.forAllRows( [=] __cuda_callable__ ( RowView& row ) mutable {
      double sum( 0.0 );
      for( auto element : row )
         sum += TNL::abs( element.value() );
      sum_view[ row.getRowIndex() ] = sum;
   } );
 
   std::cout << "Sums in matrix rows = " << sum_vector << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating matrix on host: " << std::endl;
   forRowsExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating matrix on CUDA device: " << std::endl;
   forRowsExample< TNL::Devices::Cuda >();
#endif
}

Output

Creating matrix on host: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]
Creating matrix on CUDA device: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Sums in matrix rows = [ 3, 4, 4, 4, 3 ]

getCompressedRowLengths()

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

template<typename Vector >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getCompressedRowLengths

(

Vector &

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
 
template< typename Device >
void laplaceOperatorMatrix()
{
   const int gridSize( 6 );
   const int matrixSize = gridSize;
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix(
      matrixSize, // number of rows
      matrixSize  // number of columns
   );
   matrix.setElements( {
         {  0.0, 1.0 },
         { -1.0, 2.0, -1.0 },
         { -1.0, 2.0, -1.0 },
         { -1.0, 2.0, -1.0 },
         { -1.0, 2.0, -1.0 },
         {  0.0, 1.0 }
      } );
   TNL::Containers::Vector< int, Device > rowLengths;
   matrix.getCompressedRowLengths( rowLengths );
   std::cout << "Laplace operator matrix: " << std::endl << matrix << std::endl;
   std::cout << "Compressed row lengths: " << rowLengths << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating Laplace operator matrix on CPU ... " << std::endl;
   laplaceOperatorMatrix< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating Laplace operator matrix on CUDA GPU ... " << std::endl;
   laplaceOperatorMatrix< TNL::Devices::Cuda >();
#endif
}

Output

Creating Laplace operator matrix on CPU ... 
Laplace operator matrix: 
Row: 0 ->     0:1   
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2       5:-1  
Row: 5 ->     5:1   
 
Compressed row lengths: [ 1, 3, 3, 3, 3, 1 ]
Creating Laplace operator matrix on CUDA GPU ... 
Laplace operator matrix: 
Row: 0 ->     0:1   
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2       5:-1  
Row: 5 ->     5:1   
 
Compressed row lengths: [ 1, 3, 3, 3, 3, 1 ]

getConstView()

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

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

Returns a non-modifiable view of the tridiagonal matrix.

See TridiagonalMatrixView.

Returns

tridiagonal matrix view.

getElement()

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

__cuda_callable__ Real TNL::Matrices::TridiagonalMatrix< 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.

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. TridiagonalMatrix::getRow or TridiagonalMatrix::forElements and TridiagonalMatrix::forAllElements.

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void getElements()
{
   const int matrixSize( 5 );
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix (
      matrixSize,   // number of matrix columns
      {             // matrix elements definition
         {  0.0, 2.0, -1.0 },
         { -1.0, 2.0, -1.0 },
         { -1.0, 2.0, -1.0 },
         { -1.0, 2.0, -1.0 },
         { -1.0, 2.0,  0.0 }
      } );
 
 
   for( int i = 0; i < matrixSize; i++ )
   {
      for( int j = 0; j < matrixSize; 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
}

Output

Get elements on host:
    2   -1    0    0    0
   -1    2   -1    0    0
    0   -1    2   -1    0
    0    0   -1    2   -1
    0    0    0   -1    2
Get elements on CUDA device:
    2   -1    0    0    0
   -1    2   -1    0    0
    0   -1    2   -1    0
    0    0   -1    2   -1
    0    0    0   -1    2

getIndexer() [1/2]

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

auto TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getIndexer

This method returns matrix elements indexer used by this matrix.

Returns

non-constant reference to the indexer.

getIndexer() [2/2]

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

auto TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getIndexer

This method returns matrix elements indexer used by this matrix.

Returns

constant reference to the indexer.

getNonzeroElementsCount()

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

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

overridevirtual

Returns number of non-zero matrix elements.

This method really counts the non-zero matrix elements and so it returns zero for matrix having all allocated elements set to zero.

Returns

number of non-zero matrix elements.

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

getPaddingIndex()

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

__cuda_callable__ Index TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getPaddingIndex

Returns padding index denoting padding zero elements.

These elements are used for efficient data alignment in memory.

Returns

value of the padding index.

getRow() [1/2]

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

__cuda_callable__ auto TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getRow

(

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
#include <TNL/Pointers/SharedPointer.h>
 
template< typename Device >
void getRowExample()
{
   const int matrixSize( 5 );
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   TNL::Pointers::SharedPointer< MatrixType > matrix(
      matrixSize,  // number of matrix rows
      matrixSize  // number of matrix columns
   );
 
   MatrixType* matrix_device = &matrix.template modifyData< Device >();
   auto f = [=] __cuda_callable__ ( int rowIdx ) mutable {
      auto row = matrix_device->getRow( rowIdx );
      if( rowIdx > 0 )
         row.setElement( 0, -1.0 );  // elements below the diagonal
      row.setElement( 1, 2.0 );      // elements on the diagonal
      if( rowIdx < matrixSize - 1 )  // elements above the diagonal
         row.setElement( 2, -1.0 );
   };
 
   /***
    * For the case when Device is CUDA device we need to synchronize smart
    * pointers. To avoid this you may use TridiagonalMatrixView. See
    * TridiagonalMatrixView::getRow example for details.
    */
   TNL::Pointers::synchronizeSmartPointersOnDevice< Device >();
 
   /***
    * Set the matrix elements.
    */
   TNL::Algorithms::parallelFor< Device >( 0, matrix->getRows(), f );
   std::cout << std::endl << *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
}

Output

Getting matrix rows on host: 
 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 
Getting matrix rows on CUDA device: 
 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2   
 

See TridiagonalMatrixRowView.

getRow() [2/2]

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

__cuda_callable__ auto TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getRow

(

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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
#include <TNL/Pointers/SharedPointer.h>
 
template< typename Device >
void getRowExample()
{
   const int matrixSize = 5;
   using MatrixType = TNL::Matrices::TridiagonalMatrix< double, Device >;
   TNL::Pointers::SharedPointer< MatrixType > matrix (
      matrixSize,  // number of matrix rows
      matrixSize   // number of matrix columns
    );
   matrix->setElements(
      {  { 0.0, 2.0, 1.0 },
         { 0.0, 2.0, 1.0 },
         { 3.0, 2.0, 1.0 },
         { 3.0, 2.0, 1.0 },
         { 0.0, 2.0, 1.0 } } );
 
   /***
    * Fetch lambda function returns diagonal element in each row.
    */
   const MatrixType* matrix_device = &matrix.template modifyData< Device >();
   auto fetch = [=] __cuda_callable__ ( int rowIdx ) mutable -> double {
      auto row = matrix_device->getRow( rowIdx );
      return row.getValue( 2 ); // get value from subdiagonal with index 2, i.e. the main diagonal
   };
 
   /***
    * For the case when Device is CUDA device we need to synchronize smart
    * pointers. To avoid this you may use TridiagonalMatrixView. See
    * TridiagonalMatrixView::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 reads as: " << std::endl << *matrix << std::endl;
   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
}

Output

Getting matrix rows on host: 
Matrix reads as: 
Row: 0 ->     0:2       1:1   
Row: 1 ->     1:2       2:1   
Row: 2 ->     1:3       2:2       3:1   
Row: 3 ->     2:3       3:2       4:1   
Row: 4 ->     4:2   
 
Matrix trace is: 5.
Getting matrix rows on CUDA device: 
Matrix reads as: 
Row: 0 ->     0:2       1:1   
Row: 1 ->     1:2       2:1   
Row: 2 ->     1:3       2:2       3:1   
Row: 3 ->     2:3       3:2       4:1   
Row: 4 ->     4:2   
 
Matrix trace is: 5.

See TridiagonalMatrixRowView.

getRowCapacities()

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

template<typename Vector >

void TNL::Matrices::TridiagonalMatrix< 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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::getSerializationType

static

Returns string with serialization type.

The string has a form Matrices::TridiagonalMatrix< RealType, [any_device], IndexType, ElementsOrganization, [any_allocator] >.

Returns

String with the serialization type.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
 
template< typename Device >
void getSerializationTypeExample()
{
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix;
 
   std::cout << "Matrix type is: " << matrix.getSerializationType();
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Get serialization type on CPU ... " << std::endl;
   getSerializationTypeExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Get serialization type on CUDA GPU ... " << std::endl;
   getSerializationTypeExample< TNL::Devices::Cuda >();
#endif
}

Output

Get serialization type on CPU ... 
Matrix type is: Matrices::TridiagonalMatrix< double, [any_device], int, RowMajorOrder, [any_allocator] >Get serialization type on CUDA GPU ... 
Matrix type is: Matrices::TridiagonalMatrix< double, [any_device], int, ColumnMajorOrder, [any_allocator] >

getSerializationTypeVirtual()

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

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

overridevirtual

Returns string with serialization type.

See TridiagonalMatrix::getSerializationType.

Returns

String with the serialization type.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
 
template< typename Device >
void getSerializationTypeExample()
{
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix;
 
   std::cout << "Matrix type is: " << matrix.getSerializationType();
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Get serialization type on CPU ... " << std::endl;
   getSerializationTypeExample< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Get serialization type on CUDA GPU ... " << std::endl;
   getSerializationTypeExample< TNL::Devices::Cuda >();
#endif
}

Output

Get serialization type on CPU ... 
Matrix type is: Matrices::TridiagonalMatrix< double, [any_device], int, RowMajorOrder, [any_allocator] >Get serialization type on CUDA GPU ... 
Matrix type is: Matrices::TridiagonalMatrix< double, [any_device], int, ColumnMajorOrder, [any_allocator] >

Reimplemented from TNL::Object.

getView()

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

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

Returns a modifiable view of the tridiagonal matrix.

See TridiagonalMatrixView.

Returns

tridiagonal matrix view.

isSymmetric()

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 >>

static constexpr bool TNL::Matrices::TridiagonalMatrix< 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::TridiagonalMatrix< 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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::load ( File &  file )

overridevirtual

Method for loading the matrix from a file.

Parameters

file	is the input file.

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

operator!=()

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

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

bool TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::operator!=

(

const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &

matrix

)

const

Comparison operator with another tridiagonal matrix.

Template Parameters

Real_	is Real type of the source matrix.
Device_	is Device type of the source matrix.
Index_	is Index type of the source matrix.
Organization_	is Organization of the source matrix.
RealAllocator_	is RealAllocator of the source matrix.

Parameters

matrix

is the source matrix.

Returns

true if both matrices are NOT identical and false otherwise.

operator=() [1/2]

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

TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator > & TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::operator=

(

const TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator > &

matrix

)

Assignment of exactly the same matrix type.

Parameters

matrix

is input matrix for the assignment.

Returns

reference to this matrix.

operator=() [2/2]

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 >>

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

TridiagonalMatrix & TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::operator=

(

const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &

matrix

)

Assignment of another tridiagonal matrix.

Parameters

matrix

is input matrix for the assignment.

Returns

reference to this matrix.

operator==()

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

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

bool TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::operator==

(

const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &

matrix

)

const

Comparison operator with another tridiagonal matrix.

Template Parameters

Real_	is Real type of the source matrix.
Device_	is Device type of the source matrix.
Index_	is Index type of the source matrix.
Organization_	is Organization of the source matrix.
RealAllocator_	is RealAllocator of the source matrix.

Returns

true if both matrices are identical and false otherwise.

print()

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

void TNL::Matrices::TridiagonalMatrix< 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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::reduceAllRows	(	Fetch &	fetch,
		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__ ( IndexType rowIdx, const RealType& 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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void reduceRows()
{
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *  0 / 1  3  .  .  . \   -> { 0, 1, 3 }
    *    | 2  1  3  .  . |   -> { 2, 1, 3 }
    *    | .  2  1  3  . |   -> { 2, 1, 3 }
    *    | .  .  2  1  3 |   -> { 2, 1, 3 }
    *    \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
    *
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix (
      5,              // number of matrix columns
      { { 0, 1, 3 },  // matrix elements
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 } } );
 
   /***
    * 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.reduceAllRows( fetch, reduce, keep, std::numeric_limits< double >::lowest() );
 
   std::cout << "The matrix reads as: " << std::endl << matrix << std::endl;
   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:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]
Rows reduction on CUDA device:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]

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

Method for performing general reduction on all matrix rows of constant matrix 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__ ( IndexType rowIdx, const RealType& 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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void reduceRows()
{
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *  0 / 1  3  .  .  . \   -> { 0, 1, 3 }
    *    | 2  1  3  .  . |   -> { 2, 1, 3 }
    *    | .  2  1  3  . |   -> { 2, 1, 3 }
    *    | .  .  2  1  3 |   -> { 2, 1, 3 }
    *    \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
    *
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix (
      5,              // number of matrix columns
      { { 0, 1, 3 },  // matrix elements
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 } } );
 
   /***
    * 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.reduceAllRows( fetch, reduce, keep, std::numeric_limits< double >::lowest() );
 
   std::cout << "The matrix reads as: " << std::endl << matrix << std::endl;
   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:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]
Rows reduction on CUDA device:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]

reduceRows() [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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::reduceRows	(	IndexType	begin,
		IndexType	end,
		Fetch &	fetch,
		Reduce &	reduce,
		Keep &	keep,
		const FetchReal &	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__ ( IndexType rowIdx, const RealType& 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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void reduceRows()
{
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *  0 / 1  3  .  .  . \   -> { 0, 1, 3 }
    *    | 2  1  3  .  . |   -> { 2, 1, 3 }
    *    | .  2  1  3  . |   -> { 2, 1, 3 }
    *    | .  .  2  1  3 |   -> { 2, 1, 3 }
    *    \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
    *
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix (
      5,              // number of matrix columns
      { { 0, 1, 3 },  // matrix elements
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 } } );
 
   /***
    * 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__ ( const 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 << "The matrix reads as: " << std::endl << matrix << std::endl;
   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:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]
Rows reduction on CUDA device:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]

reduceRows() [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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::reduceRows	(	IndexType	begin,
		IndexType	end,
		Fetch &	fetch,
		Reduce &	reduce,
		Keep &	keep,
		const FetchReal &	identity
	)		const

Method for performing general reduction on matrix rows of constant matrix 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__ ( IndexType rowIdx, const RealType& 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/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
 
template< typename Device >
void reduceRows()
{
   /***
    * Set the following matrix (dots represent zero matrix elements and zeros are
    * padding zeros for memory alignment):
    *
    *  0 / 1  3  .  .  . \   -> { 0, 1, 3 }
    *    | 2  1  3  .  . |   -> { 2, 1, 3 }
    *    | .  2  1  3  . |   -> { 2, 1, 3 }
    *    | .  .  2  1  3 |   -> { 2, 1, 3 }
    *    \ .  .  .  2  1 / 0 -> { 2, 1, 0 }
    *
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix (
      5,              // number of matrix columns
      { { 0, 1, 3 },  // matrix elements
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 },
        { 2, 1, 3 } } );
 
   /***
    * 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__ ( const 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 << "The matrix reads as: " << std::endl << matrix << std::endl;
   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:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]
Rows reduction on CUDA device:
The matrix reads as: 
Row: 0 ->     0:1       1:3   
Row: 1 ->     0:2       1:1       2:3   
Row: 2 ->     1:2       2:1       3:3   
Row: 3 ->     2:2       3:1       4:3   
Row: 4 ->     3:2       4:1   
 
Max. elements in rows are: [ 3, 3, 3, 3, 2 ]

save() [1/2]

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

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

overridevirtual

Method for saving the matrix to a file.

Parameters

file	is the output file.

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

sequentialForAllRows() [1/2]

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

template<typename Function >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::sequentialForAllRows

(

Function &

function

)

This method calls sequentialForRows for all matrix rows.

See TridiagonalMatrix::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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::sequentialForAllRows

(

Function &

function

)

const

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

See TridiagonalMatrix::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::TridiagonalMatrix< 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::TridiagonalMatrix::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::TridiagonalMatrix< 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::TridiagonalMatrix::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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::setDimensions ( IndexType  rows, IndexType  columns  )

overridevirtual

Set matrix dimensions.

Parameters

rows	is number of matrix rows.
columns	is 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::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::setElement	(	IndexType	row,
		IndexType	column,
		const RealType &	value
	)

Sets 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. TridiagonalMatrix::getRow or TridiagonalMatrix::forElements and TridiagonalMatrix::forAllElements. The call may fail if the matrix row capacity is exhausted.

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/Matrices/TridiagonalMatrix.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()
{
   const int matrixSize( 5 );
   using Matrix = TNL::Matrices::TridiagonalMatrix< double, Device >;
   TNL::Pointers::SharedPointer< Matrix > matrix( matrixSize, matrixSize );
   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;
 
   Matrix* matrix_device = &matrix.template modifyData< Device >();
   auto f = [=] __cuda_callable__ ( int i ) mutable {
      if( i > 0 )
         matrix_device->setElement( i, i - 1, 1.0 );
      matrix_device->setElement( i, i, -i );
      if( i < matrixSize - 1 )
         matrix_device->setElement( i, i + 1, 1.0 );
   };
 
   /***
    * For the case when Device is CUDA device we need to synchronize smart
    * pointers. To avoid this you may use TridiagonalMatrixView. See
    * TridiagonalMatrixView::getRow example for details.
    */
   TNL::Pointers::synchronizeSmartPointersOnDevice< Device >();
   TNL::Algorithms::parallelFor< Device >( 0, matrixSize, 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
}

Output

Set elements on host:
Matrix set from the host:
Row: 0 -> 
Row: 1 ->     1:1   
Row: 2 ->     2:2   
Row: 3 ->     3:3   
Row: 4 ->     4:4   
 
Matrix set from its native device:
Row: 0 ->     1:1   
Row: 1 ->     0:1       1:-1      2:1   
Row: 2 ->     1:1       2:-2      3:1   
Row: 3 ->     2:1       3:-3      4:1   
Row: 4 ->     3:1       4:-4  
 
Set elements on CUDA device:
Matrix set from the host:
Row: 0 -> 
Row: 1 ->     1:1   
Row: 2 ->     2:2   
Row: 3 ->     3:3   
Row: 4 ->     4:4   
 
Matrix set from its native device:
Row: 0 ->     1:1   
Row: 1 ->     0:1       1:-1      2:1   
Row: 2 ->     1:1       2:-2      3:1   
Row: 3 ->     2:1       3:-3      4:1   
Row: 4 ->     3:1       4:-4  
 

setElements()

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

template<typename ListReal >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::setElements

(

const std::initializer_list< std::initializer_list< ListReal > > &

data

)

Set matrix elements from an initializer list.

Template Parameters

ListReal

is data type of the initializer list.

Parameters

data	is initializer list holding matrix elements. The size of the outer list defines the number of matrix rows. Each inner list defines values of each sub-diagonal and so its size should be lower or equal to three. Values of sub-diagonals which do not fit to given row are omitted.

Example

#include <iostream>
#include <TNL/Matrices/TridiagonalMatrix.h>
#include <TNL/Devices/Host.h>
#include <TNL/Devices/Cuda.h>
 
 
template< typename Device >
void createTridiagonalMatrix()
{
   const int matrixSize = 6;
 
   /***
    * Setup the following matrix (dots represent zeros):
    *
    * /  2 -1 .   .  .  . \
    * | -1  2 -1  .  .  . |
    * |  . -1  2 -1  .  . |
    * |  .  . -1  2 -1  . |
    * |  .  .  . -1  2 -1 |
    * \  .  .  .  . -1  2 /
    *
    */
   TNL::Matrices::TridiagonalMatrix< double, Device > matrix( matrixSize, matrixSize );
   matrix.setElements( {
   /***
    * To set the matrix elements we first extend the diagonals to their full
    * lengths even outside the matrix (dots represent zeros and zeros are
    * artificial zeros used for memory alignment):
    *
    * 0 /  2 -1 .   .  .  . \    -> {  0,  2, -1 }
    *   | -1  2 -1  .  .  . |    -> { -1,  2, -1 }
    *   |  . -1  2 -1  .  . |    -> { -1,  2, -1 }
    *   |  .  . -1  2 -1  . |    -> { -1,  2, -1 }
    *   |  .  .  . -1  2 -1 |    -> { -1,  2, -1 }
    *   \  .  .  .  . -1  2 / 0  -> { -1,  2,  0 }
    *
    */
      {  0,  2, -1 },
      { -1,  2, -1 },
      { -1,  2, -1 },
      { -1,  2, -1 },
      { -1,  2, -1 },
      { -1,  2,  0 }
      } );
   std::cout << "The matrix reads as: " << std::endl << matrix << std::endl;
}
 
int main( int argc, char* argv[] )
{
   std::cout << "Creating tridiagonal matrix on CPU ... " << std::endl;
   createTridiagonalMatrix< TNL::Devices::Host >();
 
#ifdef __CUDACC__
   std::cout << "Creating tridiagonal matrix on CUDA GPU ... " << std::endl;
   createTridiagonalMatrix< TNL::Devices::Cuda >();
#endif
}

Output

Creating tridiagonal matrix on CPU ... 
The matrix reads as: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2       5:-1  
Row: 5 ->     4:-1      5:2   
 
Creating tridiagonal matrix on CUDA GPU ... 
The matrix reads as: 
Row: 0 ->     0:2       1:-1  
Row: 1 ->     0:-1      1:2       2:-1  
Row: 2 ->     1:-1      2:2       3:-1  
Row: 3 ->     2:-1      3:2       4:-1  
Row: 4 ->     3:-1      4:2       5:-1  
Row: 5 ->     4:-1      5:2   
 

setLike()

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

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

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::setLike

(

const TridiagonalMatrix< Real_, Device_, Index_, Organization_, RealAllocator_ > &

matrix

)

Setup the matrix dimensions and diagonals offsets based on another tridiagonal matrix.

Template Parameters

Real_	is Real type of the source matrix.
Device_	is Device type of the source matrix.
Index_	is Index type of the source matrix.
Organization_	is Organization of the source matrix.
RealAllocator_	is RealAllocator of the source matrix.

Parameters

matrix

is the source matrix.

setRowCapacities()

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

template<typename RowCapacitiesVector >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::setRowCapacities

(

const RowCapacitiesVector &

rowCapacities

)

This method is for compatibility with SparseMatrix.

It checks if the number of matrix diagonals is compatible with required number of non-zero matrix elements in each row. If not exception is thrown.

Template Parameters

RowCapacitiesVector

is vector-like container type for holding required row capacities.

Parameters

rowCapacities

is vector-like container holding required row capacities.

setValue()

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

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::setValue

(

const RealType &

value

)

Set all matrix elements to given value.

Parameters

value

is the new value of all matrix elements.

vectorProduct()

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

template<typename InVector , typename OutVector >

void TNL::Matrices::TridiagonalMatrix< Real, Device, Index, Organization, RealAllocator >::vectorProduct	(	const InVector &	inVector,
		OutVector &	outVector,
		RealType	matrixMultiplicator = 1.0,
		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.

---

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

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