The class Matrix is a abstract type that represents a distributed 2D real or complex matrix. More...

#include <Matrix.hpp>

Inheritance diagram for lama::Matrix:

Public Types
enum	SyncKind { ASYNCHRONOUS, SYNCHRONOUS }
	SyncKind describes if the communication and computation should be done synchronously or asynchronously. More...
enum	MatrixKind { DENSE, SPARSE }
	MatrixKind describes if a matrix is dense or sparse. More...
typedef const Matrix &	ExpressionMemberType
	ExpressionMemberType is the type that is used the template Expression to store a Vector.
Public Member Functions
	Matrix (DistributionPtr rowDistribution, DistributionPtr colDistribution)
	Construct a matrix by corresponding distributions.
	Matrix (const IndexType numRows, const IndexType numColumns)
	Construct a matrix of a given size replicated on each partition.
	Matrix (DistributionPtr distribution)
	Construct a square matrix with the given size and the specified distribution.
	Matrix (const IndexType size)
	Construct a square matrix with a replicated distribution.
	Matrix ()
	Default constructor creates a replicated matrix of size 0 x 0.
	Matrix (const Matrix &other, DistributionPtr rowDistribution, DistributionPtr colDistribution)
	Redistribute the given matrix with the specified distributions.
	Matrix (const Matrix &other)
	Copy a matrix to a new matrix with the same distribution.
virtual	~Matrix ()
	Releases all allocated resources.
virtual const char *	getTypeName () const =0
	Virtual method that delivers the class name to which a matrix belongs.
std::auto_ptr< _LAMAArray >	createArray () const
	This routine creates a LAMA array with the same value type as the matrix.
virtual void	clear ()=0
	Clear the full matrix, resets global and local sizes to 0.
virtual void	allocate (const IndexType numRows, const IndexType numColumns)=0
	Reallocate this matrix to a replicated zero-matrix of the given shape.
virtual void	allocate (DistributionPtr rowDistribution, DistributionPtr colDistribution)=0
	Reallocate this matrix to a distributed zero-matrix by the given distributions.
virtual void	setIdentity ()=0
	Operator that sets the matrix to the identity matrix.
virtual void	assign (const Matrix &other)=0
	This method assigns a matrix to this matrix with automatic conversion to the matrix type and the current row / column distribution of *this.
virtual void	assign (const _MatrixStorage &other)=0
	This method assigns replicated/global matrix storage data to this matrix.
virtual void	assign (const _MatrixStorage &storage, DistributionPtr rowDist, DistributionPtr colDist)=0
	Assign local storage for a distributed matrix.
virtual void	buildLocalStorage (_MatrixStorage &storage) const =0
	Get the local part (no splitted columns) of a matrix as if colDist is replicated.
virtual void	redistribute (DistributionPtr rowDistribution, DistributionPtr colDistribution)=0
	This method allows any arbitrary redistribution of the matrix.
virtual void	getRow (Vector &row, const IndexType globalRowIndex) const =0
	This method returns one row of the matrix.
virtual void	getDiagonal (Vector &diagonal) const =0
	This method returns the diagonal.
virtual void	setDiagonal (const Vector &diagonal)=0
	This method replaces the diagonal.
virtual void	setDiagonal (const Scalar scalar)=0
	This method replaces the diagonal by a diagonal value.
virtual void	scale (const Vector &scaling)=0
	This method scales all values with a vector.
virtual void	scale (const Scalar scaling)=0
	This method scales all matrix values with a scalar.
Scalar	operator() (IndexType i, IndexType j) const
	returns a copy of the value at the passed global indexes.
virtual Scalar	getValue (IndexType i, IndexType j) const =0
	returns a copy of the value at the passed global indexes.
virtual void	writeAt (std::ostream &stream) const
	Write some information about this to the passed stream.
IndexType	getNumRows () const
	Returns the number of global rows.
IndexType	getNumColumns () const
	Returns the number of columns.
virtual IndexType	getNumValues () const =0
	Get the total number of non-zero values in the matrix.
double	getSparsityRate () const
virtual void	matrixTimesVector (Vector &result, const Scalar alpha, const Vector &x, const Scalar beta, const Vector &y) const =0
	matrixTimesVector computes result = alpha * this * x + beta * y
virtual void	matrixTimesScalar (const Matrix &other, const Scalar alpha)=0
	matrixTimesScalar computes this = alpha * other
virtual void	matrixTimesMatrix (Matrix &result, const Scalar alpha, const Matrix &B, const Scalar beta, const Matrix &C) const =0
	matrixTimesMatrix computes result = alpha * this * B + beta * C
virtual void	matrix2CSRGraph (IndexType xadj, IndexType adjncy, IndexType vwgt, CommunicatorPtr comm, const IndexType globalRowIndices=NULL, IndexType *vtxdist=NULL) const
	transformation from matrix type to a csr graph
virtual IndexType	getLocalNumValues () const =0
	Getter routine for the local number of stored values.
virtual IndexType	getLocalNumRows () const =0
	Getter routine for the local number of rows.
virtual IndexType	getLocalNumColumns () const =0
	Getter routine for the local number of rows and columns.
const Distribution &	getColDistribution () const
	gets a constant reference to the column distribution.
DistributionPtr	getColDistributionPtr () const
	gets a pointer to the column distribution.
virtual void	setContext (const ContextPtr context)=0
	specifies on which compute back end the matrix operations should take place.
virtual void	setContext (const ContextPtr localContext, const ContextPtr haloContext)
	Only sparse matrices will override this method, others will ignore second argument.
virtual ContextPtr	getContextPtr () const =0
	Getter routine for the context.
const Context &	getContext () const
virtual MatrixKind	getMatrixKind () const =0
	Each derived matrix must give info about its kind (DENSE or SPARSE).
SyncKind	getCommunicationKind () const
	get the communication kind.
SyncKind	getComputeKind () const
	get the compute kind.
void	setCommunicationKind (SyncKind communicationKind)
	set the communication kind.
virtual void	setComputeKind (SyncKind computeKind)
	set the compute kind.
void	inheritAttributes (const Matrix &other)
	Inherit context and kind arguments from another matrix.
virtual void	prefetch () const =0
	Prefetch matrix data to its 'preferred' context location.
virtual void	wait () const =0
	wait for a possibly running prefetch.
Matrix &	operator= (const Matrix &other)
	the assignment operator for matrix.
Matrix &	operator= (Expression< Scalar, Matrix, Times > exp)
	the assignment operator for a scalar matrix multiplication.
Matrix &	operator= (Expression< Matrix, Matrix, Times > exp)
	the assignment operator for a matrix matrix multiplication.
Matrix &	operator= (Expression< Scalar, Expression< Matrix, Matrix, Times >, Times > exp)
	the assignment operator for a scalar matrix matrix multiplication.
Matrix &	operator= (Expression< Expression< Scalar, Expression< Matrix, Matrix, Times >, Times >, Expression< Scalar, Matrix, Times >, Plus > exp)
	the assignment operator for a GEMM expression.
virtual void	invert (const Matrix &other)=0
	Compute the inverse of a matrix.
virtual Scalar	maxDiffNorm (const Matrix &other) const =0
	returns the max norm of ( this - other )
virtual std::auto_ptr< Matrix >	create () const =0
	Constructor function which creates a 'zero' matrix of same type as a given matrix.
std::auto_ptr< Matrix >	create (const IndexType numRows, const IndexType numColumns) const
	Constructor creates a replicated matrix of same type as a given matrix.
std::auto_ptr< Matrix >	create (const IndexType size) const
	Constructor creates a distributed zero matrix of same type as a given matrix.
std::auto_ptr< Matrix >	create (DistributionPtr rowDistribution, DistributionPtr colDistribution) const
	Constructor creates a distributed zero matrix of same type as a given matrix.
std::auto_ptr< Matrix >	create (DistributionPtr distribution) const
	Constructor creates a distributed zero matrix of same type as a given matrix.
VectorPtr	createDenseVector (DistributionPtr distribution, const Scalar value) const
	Constructor creates a distributed dense vector of same type as a given matrix.
virtual std::auto_ptr< Matrix >	copy () const =0
	Constructor function which creates a copy of this matrix.
virtual Scalar::ScalarType	getValueType () const =0
	Query the value type of the matrix elements, e.g.
virtual bool	hasDiagonalProperty () const =0
	hasDiagonalProperty returns the diagonalProperty of the local storage.
virtual void	resetDiagonalProperty ()=0
	resetDiagonalProperty rechecks the storages for their diagonal property
virtual size_t	getMemoryUsage () const =0
	getMemoryUsage returns the global memory that is allocated to hold this matrix.
const Distribution &	getDistribution () const
DistributionPtr	getDistributionPtr () const
Protected Member Functions
void	setReplicatedMatrix (const IndexType numRows, const IndexType numColumns)
	Set the global/local size of replicated matrix.
void	setDistributedMatrix (DistributionPtr distribution, DistributionPtr colDistribution)
	Set the global and local size of distributed matrix.
void	checkSettings ()
void	swapMatrix (Matrix &other)
	LAMA_LOG_DECL_STATIC_LOGGER (logger)
void	swap (Distributed &other)
void	setDistributionPtr (DistributionPtr distributionPtr)
Protected Attributes
DistributionPtr	mColDistribution
IndexType	mNumRows
IndexType	mNumColumns
Private Member Functions
void	setDefaultKind ()
Private Attributes
SyncKind	mCommunicationKind
SyncKind	mComputeKind

Detailed Description

The class Matrix is a abstract type that represents a distributed 2D real or complex matrix.

Matrix is one of the LAMA Base Types and should be used in all situations where it is not necessary to access a single element or to create a new Matrix.

Member Typedef Documentation

typedef const Matrix& lama::Matrix::ExpressionMemberType

ExpressionMemberType is the type that is used the template Expression to store a Vector.

Member Enumeration Documentation

enum lama::Matrix::MatrixKind

MatrixKind describes if a matrix is dense or sparse.

Enumerator:

DENSE	matrix kind for a dense matrix
SPARSE	matrix kind for a sparse matrix

enum lama::Matrix::SyncKind

SyncKind describes if the communication and computation should be done synchronously or asynchronously.

Enumerator:

ASYNCHRONOUS
SYNCHRONOUS

Constructor & Destructor Documentation

lama::Matrix::Matrix	(	DistributionPtr	rowDistribution,
		DistributionPtr	colDistribution
	)

Construct a matrix by corresponding distributions.

Parameters:

[in]	rowDistribution	specifies number of rows and how the rows of the matrix are distributed among the available partitions.
[in]	colDistribution	specifies number of columns and the distribution of a vector when matrix is applied to a matrix-vector multiplication.

The colDstribution is used to setup a communication pattern (e.g. halo) that can be reused for multiple matrix-vector operations. It might also be used to split the matrix data on one partition into a local and non-local part.

The number of rows and columns is given implicitly by the global sizes of the distributions.

References checkSettings(), lama::Distributed::getDistribution(), mNumColumns, mNumRows, setDefaultKind(), and setDistributedMatrix().

lama::Matrix::Matrix	(	const IndexType	numRows,
		const IndexType	numColumns
	)

Construct a matrix of a given size replicated on each partition.

Parameters:

[in]	numRows	is the number of rows
[in]	numColumns	is the number of columns

Same as Matrix( NoDistribution(numRows), NoDistribution(numColumns) )

References mNumColumns, and mNumRows.

lama::Matrix::Matrix ( DistributionPtr distribution )

Construct a square matrix with the given size and the specified distribution.

Parameters:

[in] distribution specifies how the rows of the matrix are distributed among the available partitions.

Same as Matrix( distribution, distribution ); column distribution will be same as that of rows.

References checkSettings(), lama::Distributed::getDistribution(), mNumColumns, mNumRows, and setDistributedMatrix().

lama::Matrix::Matrix ( const IndexType size ) [explicit]

Construct a square matrix with a replicated distribution.

Parameters:

[in] size is the number of rows

Same as Matrix(size, size ).

lama::Matrix::Matrix ( )

Default constructor creates a replicated matrix of size 0 x 0.

lama::Matrix::Matrix	(	const Matrix &	other,
		DistributionPtr	rowDistribution,
		DistributionPtr	colDistribution
	)

Redistribute the given matrix with the specified distributions.

Parameters:

[in]	other	the matrix to redistribute
[in]	rowDistribution	specifies how the rows of the matrix are distributed among the available partitions.
[in]	colDistribution	specifies the distribution of a vector when matrix is applied to a matrix-vector multiplication.

The following relations must hold:

other.getNumRows() == distribution->getGlobalSize()
other.getNumColumns() == colDistribution->getGlobalSize()

References checkSettings(), getColDistribution(), and lama::Distributed::getDistribution().

lama::Matrix::Matrix ( const Matrix & other )

Copy a matrix to a new matrix with the same distribution.

Parameters:

[in] other the matrix to take a copy from

lama::Matrix::~Matrix ( ) [virtual]

Releases all allocated resources.

Member Function Documentation

virtual void lama::Matrix::allocate	(	const IndexType	numRows,
		const IndexType	numColumns
	)		`[pure virtual]`

Reallocate this matrix to a replicated zero-matrix of the given shape.

Parameters:

[in]	numRows	number of rows
[in]	numColumns	number of columns

Remarks:: The call allocate( 0, 0 ) implies a clear on all arrays used internally for the presentation of the matrix data.

Implemented in lama::DenseMatrix< T >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::allocate	(	DistributionPtr	rowDistribution,
		DistributionPtr	colDistribution
	)		`[pure virtual]`

Reallocate this matrix to a distributed zero-matrix by the given distributions.

Parameters:

[in]	distribution	is row distribution, number of rows given by getGlobalSize()
[in]	colDistribution	is col distribution, number of columns given by getGlobalSize()

The distributions specify the new global and the new local sizes for a distributed matrix.

Implemented in lama::DenseMatrix< T >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::assign ( const Matrix & other ) [pure virtual]

This method assigns a matrix to this matrix with automatic conversion to the matrix type and the current row / column distribution of *this.

Note: other.getNumRows() == getNumRows(), other.getNumColumns() == getNumColumns() is mandatory

Implemented in lama::DenseMatrix< T >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by lama::MatrixCreator< T >::fillRandom(), and operator=().

virtual void lama::Matrix::assign ( const _MatrixStorage & other ) [pure virtual]

This method assigns replicated/global matrix storage data to this matrix.

Note: the current matrix data will be overridden.

Implemented in lama::DenseMatrix< T >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::assign	(	const _MatrixStorage &	storage,
		DistributionPtr	rowDist,
		DistributionPtr	colDist
	)		`[pure virtual]`

Assign local storage for a distributed matrix.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::buildLocalStorage ( _MatrixStorage & storage ) const [pure virtual]

Get the local part (no splitted columns) of a matrix as if colDist is replicated.

Parameters:

[out] storage will contain the local part of the matrix with all columns.

As splitting of columns is differently handled for sparse and dense matrices, this method is useful for conversion between them. An alternative solution of copying the whole matrix and replication of columns might be too expensive.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by lama::SparseMatrix< T >::assign().

void lama::Matrix::checkSettings ( ) [protected]

Reimplemented in lama::SparseMatrix< T >, and lama::SparseMatrix< ValueType >.

References getColDistribution(), lama::Distributed::getDistribution(), LAMA_THROWEXCEPTION, mColDistribution, mNumColumns, and mNumRows.

Referenced by Matrix().

virtual void lama::Matrix::clear ( ) [pure virtual]

Clear the full matrix, resets global and local sizes to 0.

     CSRSparseMatrix<double> a ( ... );
     a = CSRSparseMatrix<double> ();     \\ will free all arrays
     a.clear();                          \\ same functionality, clears involved arrays

Implemented in lama::DenseMatrix< T >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual std::auto_ptr<Matrix> lama::Matrix::copy ( ) const [pure virtual]

Constructor function which creates a copy of this matrix.

 std::auto_ptr<Matrix> newMatrix = matrix.copy();
 // More convenient to use, but exactly same as follows: 
 std::auto_ptr<Matrix> newMatrix = matrix.create(); *newMatrix = matrix;

This method is a workaround to call the copy constructor of a derived matrix class where the derived class is not known at compile time.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, and lama::DenseMatrix< T >.

Referenced by lama::DefaultJacobi::initialize().

virtual std::auto_ptr<Matrix> lama::Matrix::create ( ) const [pure virtual]

Constructor function which creates a 'zero' matrix of same type as a given matrix.

 void sub( ..., const Matrix& matrix, ...)
 {
     ...
     // Create a copy of the input matrix

     std::auto_ptr<Matrix> newMatrix = matrix.create();
     *newMatrix = matrix;

     // Create a unity matrix of same type and same row distribution as matrix

     std::auto_ptr<Matrix> newMatrix = matrix.create();
     newMatrix->allocate( matrix.getRowDistributionPtr(), matrix.getRowDistributionPtr() );
     newMatrix->setIdentity();  
     ...
 }

This method is a workaround to call the constructor of a derived matrix class where the derived class is not known at compile time.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::DenseMatrix< T >, and lama::SimpleStorageStrategy< ValueType >.

Referenced by create(), createArray(), lama::SingleGridSetup::initialize(), lama::DefaultJacobi::initialize(), and lama::InverseSolver::initialize().

std::auto_ptr< Matrix > lama::Matrix::create	(	const IndexType	numRows,
		const IndexType	numColumns
	)		const

Constructor creates a replicated matrix of same type as a given matrix.

References create().

std::auto_ptr< Matrix > lama::Matrix::create ( const IndexType size ) const

Constructor creates a distributed zero matrix of same type as a given matrix.

References create().

std::auto_ptr< Matrix > lama::Matrix::create	(	DistributionPtr	rowDistribution,
		DistributionPtr	colDistribution
	)		const

Constructor creates a distributed zero matrix of same type as a given matrix.

References create().

std::auto_ptr< Matrix > lama::Matrix::create ( DistributionPtr distribution ) const

Constructor creates a distributed zero matrix of same type as a given matrix.

References create().

std::auto_ptr< _LAMAArray > lama::Matrix::createArray ( ) const

This routine creates a LAMA array with the same value type as the matrix.

: returns an auto pointer to the LAMA array.

Same as _LAMAArray::create( this.getValueType() )

Value type is known only at runtime, so pointer to the base class is returned. Auto pointer indicates that calling routine takes ownership of the allocated array.

References create(), and getValueType().

VectorPtr lama::Matrix::createDenseVector	(	DistributionPtr	distribution,
		const Scalar	value
	)		const

Constructor creates a distributed dense vector of same type as a given matrix.

References lama::Scalar::DOUBLE, lama::Scalar::FLOAT, lama::Scalar::getValue(), getValueType(), and LAMA_THROWEXCEPTION.

const Distribution & lama::Matrix::getColDistribution ( ) const [inline]

gets a constant reference to the column distribution.

Returns:: a constant reference to the column distribution.

References LAMA_ASSERT_ERROR, and mColDistribution.

Referenced by lama::DenseMatrix< T >::assignSparse(), checkSettings(), lama::LUSolver::computeLUFactorization(), lama::InverseSolver::decompose(), lama::DenseMatrix< T >::DenseMatrix(), lama::DenseMatrixOps::invert(), lama::InverseSolver::invert(), lama::DenseMatrixOps::invertReplicated(), lama::SpecializedJacobi::iterateTyped(), Matrix(), lama::DenseMatrix< T >::matrixTimesMatrix(), lama::SparseMatrix< T >::matrixTimesMatrixImpl(), lama::CRTPMatrix< DenseMatrix< T >, T >::matrixTimesVector(), lama::SparseMatrix< T >::matrixTimesVectorNImpl(), lama::DenseMatrix< T >::maxDiffNorm(), lama::SparseMatrix< T >::maxDiffNorm(), lama::DenseMatrix< T >::maxDiffNormImpl(), lama::SparseMatrix< T >::maxDiffNormImpl(), operator=(), lama::LUSolver::solve(), and lama::Solver::solveInit().

DistributionPtr lama::Matrix::getColDistributionPtr ( ) const [inline]

gets a pointer to the column distribution.

Returns:: a pointer to the column distribution.

References LAMA_ASSERT_ERROR, and mColDistribution.

Referenced by lama::SparseMatrix< T >::assign(), lama::DenseMatrix< T >::assign(), lama::DenseMatrix< T >::assignSparse(), lama::SparseMatrix< T >::assignTransposeImpl(), lama::DenseMatrix< T >::copyDenseMatrix(), lama::MatrixCreator< T >::fillRandom(), lama::DenseMatrixOps::invert(), lama::DenseMatrix< T >::invert(), lama::DenseMatrix< T >::matrixTimesMatrix(), lama::SparseMatrix< T >::matrixTimesMatrixImpl(), and lama::SparseMatrix< T >::matrixTimesVectorNImpl().

Matrix::SyncKind lama::Matrix::getCommunicationKind ( ) const [inline]

get the communication kind.

Returns:: the communication kind.

References mCommunicationKind.

Referenced by inheritAttributes(), lama::SpecializedJacobi::iterateTyped(), and lama::DenseMatrix< T >::matrixTimesVectorImpl().

Matrix::SyncKind lama::Matrix::getComputeKind ( ) const [inline]

get the compute kind.

Returns:: the compute kind.

References mComputeKind.

Referenced by inheritAttributes().

const Context& lama::Matrix::getContext ( ) const [inline]

Reimplemented in lama::SparseMatrix< T >, and lama::SparseMatrix< ValueType >.

virtual ContextPtr lama::Matrix::getContextPtr ( ) const [pure virtual]

Getter routine for the context.

Note: Only for SparseMatrix the context of the halo can be queried.

Implemented in lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, and lama::DenseMatrix< T >.

Referenced by lama::SparseMatrix< T >::assign(), inheritAttributes(), lama::CG::initialize(), lama::DefaultJacobi::initialize(), lama::GMRES::initialize(), lama::InverseSolver::initialize(), lama::InverseSolver::invert(), and lama::LUSolver::piamax_own().

virtual void lama::Matrix::getDiagonal ( Vector & diagonal ) const [pure virtual]

This method returns the diagonal.

Parameters:

[out] diagonal is the destination array

Calculations are dependent to the diagonal property.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by lama::DefaultJacobi::initialize().

const Distribution & lama::Distributed::getDistribution ( ) const [inline, inherited]

References lama::Distributed::mDistribution.

DistributionPtr lama::Distributed::getDistributionPtr ( ) const [inline, inherited]

virtual IndexType lama::Matrix::getLocalNumColumns ( ) const [pure virtual]

Getter routine for the local number of rows and columns.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual IndexType lama::Matrix::getLocalNumRows ( ) const [pure virtual]

Getter routine for the local number of rows.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual IndexType lama::Matrix::getLocalNumValues ( ) const [pure virtual]

Getter routine for the local number of stored values.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual MatrixKind lama::Matrix::getMatrixKind ( ) const [pure virtual]

Each derived matrix must give info about its kind (DENSE or SPARSE).

Implemented in lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, lama::DenseMatrix< T >, and lama::_SparseMatrix.

Referenced by lama::DenseMatrix< T >::assign(), and lama::SparseMatrix< T >::matrixTimesMatrix().

virtual size_t lama::Matrix::getMemoryUsage ( ) const [pure virtual]

getMemoryUsage returns the global memory that is allocated to hold this matrix.

getMemoryUsage returns the global memory that is allocated to hold this matrix. For a distributed matrix all partitions are summed together.

Returns:: the memory consumption of this matrix.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, and lama::DenseMatrix< T >.

IndexType lama::Matrix::getNumColumns ( ) const [inline]

Returns the number of columns.

Returns:: the number of columns.

References mNumColumns.

Referenced by lama::DenseMatrix< T >::assignSparse(), lama::LUSolver::computeLUFactorization(), lama::InverseSolver::decompose(), lama::DenseMatrix< T >::DenseMatrix(), lama::MatrixCreator< T >::fillRandom(), lama::SparseMatrix< T >::getLocalNumColumns(), lama::InverseSolver::invert(), lama::DenseMatrix< T >::invert(), lama::DenseMatrixOps::invertCyclic(), lama::MatrixConfigGrammar::MatrixConfigGrammar(), operator=(), lama::LUSolver::solve(), and lama::Solver::solveInit().

IndexType lama::Matrix::getNumRows ( ) const [inline]

Returns the number of global rows.

Returns:: the number of rows.

References mNumRows.

Referenced by lama::InverseSolver::computeInverse(), lama::LUSolver::computeLUFactorization(), lama::InverseSolver::decompose(), lama::SparseMatrix< T >::getLocalNumRows(), lama::LUSolver::initialize(), lama::DenseMatrixOps::invert(), lama::InverseSolver::invert(), lama::DenseMatrix< T >::invert(), lama::DenseMatrixOps::invertCyclic(), lama::SpecializedJacobi::iterateTyped(), lama::MatrixConfigGrammar::MatrixConfigGrammar(), operator=(), lama::LUSolver::solve(), and lama::Solver::solveInit().

virtual IndexType lama::Matrix::getNumValues ( ) const [pure virtual]

Get the total number of non-zero values in the matrix.

An element is considered to be non-zero if its absolute value is greater equal than mEpsilon. Zero diagonal elements are also counted if this->hasDiagonalProperty() is given.

This routine does not count zero elements even if they are stored (e.g. for dense or dia storage data).

Returns:: the number of non-zero values in this matrix.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by getSparsityRate().

virtual void lama::Matrix::getRow	(	Vector &	row,
		const IndexType	globalRowIndex
	)		const `[pure virtual]`

This method returns one row of the matrix.

Parameters:

[out] row is a replicated vector with all values of the row

Note: the value type of the vector should be the same type as the matrix (otherwise conversion) and it should be a replicated vector (otherwise reallocation)

Unclear: should the distribution always be unchanged ?

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

double lama::Matrix::getSparsityRate ( ) const

References getNumValues(), mNumColumns, and mNumRows.

virtual const char* lama::Matrix::getTypeName ( ) const [pure virtual]

Virtual method that delivers the class name to which a matrix belongs.

Implemented in lama::COOSparseMatrix< T >, lama::CSRSparseMatrix< T >, lama::DIASparseMatrix< T >, lama::ELLSparseMatrix< T >, lama::JDSSparseMatrix< T >, lama::XXXSparseMatrix< T >, lama::DenseMatrix< T >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual Scalar lama::Matrix::getValue	(	IndexType	i,
		IndexType	j
	)		const `[pure virtual]`

returns a copy of the value at the passed global indexes.

Parameters:

[in]	i	the global row index
[in]	j	the global column index

Returns:: a copy of the value at the passed global position.

As this operation requires communication ins SPMD mode it can be very inefficient in some situations.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by operator()().

virtual Scalar::ScalarType lama::Matrix::getValueType ( ) const [pure virtual]

Query the value type of the matrix elements, e.g.

DOUBLE or FLOAT.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, and lama::DenseMatrix< T >.

Referenced by lama::DenseMatrix< T >::assign(), createArray(), createDenseVector(), lama::SingleGridSetup::initialize(), lama::SOR::initialize(), lama::CG::initialize(), lama::DefaultJacobi::initialize(), lama::GMRES::initialize(), lama::GMRES::iterate(), lama::DenseMatrix< T >::maxDiffNorm(), and lama::SparseMatrix< T >::maxDiffNorm().

virtual bool lama::Matrix::hasDiagonalProperty ( ) const [pure virtual]

hasDiagonalProperty returns the diagonalProperty of the local storage.

Returns:: if the diagonal property is full filled.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, and lama::DenseMatrix< T >.

Referenced by lama::SpecializedJacobi::initialize(), and lama::DefaultJacobi::initialize().

void lama::Matrix::inheritAttributes ( const Matrix & other )

Inherit context and kind arguments from another matrix.

This routine will also be used by copy constructors in base classes.

References getCommunicationKind(), getComputeKind(), getContextPtr(), setCommunicationKind(), setComputeKind(), and setContext().

virtual void lama::Matrix::invert ( const Matrix & other ) [pure virtual]

Compute the inverse of a matrix.

Implemented in lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, lama::DenseMatrix< T >, lama::SparseMatrix< T >, and lama::SparseMatrix< ValueType >.

lama::Matrix::LAMA_LOG_DECL_STATIC_LOGGER ( logger ) [protected]

Reimplemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, lama::COOSparseMatrix< T >, lama::CSRSparseMatrix< T >, lama::DIASparseMatrix< T >, lama::ELLSparseMatrix< T >, lama::JDSSparseMatrix< T >, and lama::XXXSparseMatrix< T >.

void lama::Matrix::matrix2CSRGraph	(	IndexType *	xadj,
		IndexType *	adjncy,
		IndexType *	vwgt,
		CommunicatorPtr	comm,
		const IndexType *	globalRowIndices = `NULL`,
		IndexType *	vtxdist = `NULL`
	)		const `[virtual]`

transformation from matrix type to a csr graph

transformation from matrix type to a csr graph, so that it (Par)Metis can work with it.

Parameters:

[out]	xadj	the ia array of the csr graph
[out]	adjncy	the ja array of the csr graph

References LAMA_THROWEXCEPTION.

virtual void lama::Matrix::matrixTimesMatrix	(	Matrix &	result,
		const Scalar	alpha,
		const Matrix &	B,
		const Scalar	beta,
		const Matrix &	C
	)		const `[pure virtual]`

matrixTimesMatrix computes result = alpha * this * B + beta * C

Parameters:

[out]	result	the Matrix to store the result to
[in]	alpha	the Scalar alpha of the expression
[in]	B	the Matrix B of the expression
[in]	beta	the Scalar beta of the expression
[in]	C	the Matrix C of the expression

matrixTimesMatrix computes result = alpha * this * x + beta * y.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by operator=().

virtual void lama::Matrix::matrixTimesScalar	(	const Matrix &	other,
		const Scalar	alpha
	)		`[pure virtual]`

matrixTimesScalar computes this = alpha * other

Parameters:

[out]	other	the Matrix to mutliply
[in]	alpha	the Scalar of the expression

matrixTimesScalar computes this = alpha * other.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by operator=().

virtual void lama::Matrix::matrixTimesVector	(	Vector &	result,
		const Scalar	alpha,
		const Vector &	x,
		const Scalar	beta,
		const Vector &	y
	)		const `[pure virtual]`

matrixTimesVector computes result = alpha * this * x + beta * y

Parameters:

[out]	result	the Vector to store the result to
[in]	alpha	the Scalar alpha of the expression
[in]	x	the Vector x of the expression
[in]	beta	the Scalar beta of the expression
[in]	y	the Vector y of the expression

matrixTimesVector computes result = alpha * this * x + beta * y. If result == x or result == y new storage is allocated to store the result.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, lama::CRTPMatrix< Derived, ValueType >, and lama::CRTPMatrix< DenseMatrix< T >, T >.

Referenced by lama::Vector::operator=().

virtual Scalar lama::Matrix::maxDiffNorm ( const Matrix & other ) const [pure virtual]

returns the max norm of ( this - other )

Parameters:

[in] other another matrix with the same shape as this matrix

Returns:: the max norm of ( this - other )

The maximal value is given by the largest difference between two elements at the same position of the matrices.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, and lama::DenseMatrix< T >.

Scalar lama::Matrix::operator()	(	IndexType	i,
		IndexType	j
	)		const

returns a copy of the value at the passed global indexes.

Parameters:

[in]	i	the global row index
[in]	j	the global column index

Returns:: a copy of the value at the passed global position.

As this operator requires communication ins SPMD mode it can be very inefficient in some situations.

References getValue().

Matrix & lama::Matrix::operator= ( const Matrix & other )

the assignment operator for matrix.

Parameters:

[in] other is the input matrix.

The assignment operator will make a deep copy of the input matrix. Size and distributions are inherited, but there might be implicit conversions regarding storage format and/or value type of the matrix elements.

Reimplemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, and lama::DenseMatrix< T >.

References assign().

Referenced by lama::DenseMatrix< T >::DenseMatrix(), and lama::DenseMatrix< T >::operator=().

Matrix & lama::Matrix::operator= ( Expression< Scalar, Matrix, Times > exp )

the assignment operator for a scalar matrix multiplication.

Reimplemented in lama::DenseMatrix< T >.

References lama::Expression< T1, T2, type >::getArg1(), lama::Expression< T1, T2, type >::getArg2(), and matrixTimesScalar().

Matrix & lama::Matrix::operator= ( Expression< Matrix, Matrix, Times > exp )

the assignment operator for a matrix matrix multiplication.

Reimplemented in lama::DenseMatrix< T >.

References lama::Expression< T1, T2, type >::getArg1(), lama::Plus, lama::Times, and lama::zero.

Matrix & lama::Matrix::operator= ( Expression< Scalar, Expression< Matrix, Matrix, Times >, Times > exp )

the assignment operator for a scalar matrix matrix multiplication.

the assignment operator for a matrix matrix multiplication.

Reimplemented in lama::DenseMatrix< T >.

References lama::Plus, lama::Times, and lama::zero.

Matrix & lama::Matrix::operator= ( Expression< Expression< Scalar, Expression< Matrix, Matrix, Times >, Times >, Expression< Scalar, Matrix, Times >, Plus > exp )

the assignment operator for a GEMM expression.

Reimplemented in lama::DenseMatrix< T >.

References getColDistribution(), lama::Distributed::getDistribution(), getNumColumns(), getNumRows(), LAMA_THROWEXCEPTION, matrixTimesMatrix(), and lama::zero.

virtual void lama::Matrix::prefetch ( ) const [pure virtual]

Prefetch matrix data to its 'preferred' context location.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::DenseMatrix< T >, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::redistribute	(	DistributionPtr	rowDistribution,
		DistributionPtr	colDistribution
	)		`[pure virtual]`

This method allows any arbitrary redistribution of the matrix.

Parameters:

[in]	rowDistribution	is new distribution of rows, global size must be mNumRows
[in]	colDistribution	is new distribution of columns, global size must be mNumColumns

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

Referenced by lama::MatrixConfigGrammar::MatrixConfigGrammar().

virtual void lama::Matrix::resetDiagonalProperty ( ) [pure virtual]

resetDiagonalProperty rechecks the storages for their diagonal property

Returns:: if the diagonal property is full filled.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, and lama::DenseMatrix< T >.

virtual void lama::Matrix::scale ( const Vector & scaling ) [pure virtual]

This method scales all values with a vector.

Parameters:

[in] values is the source array

row wise calculations

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::scale ( const Scalar scaling ) [pure virtual]

This method scales all matrix values with a scalar.

Parameters:

[in] value is the source value

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

void lama::Matrix::setCommunicationKind ( SyncKind communicationKind )

set the communication kind.

Parameters:

[in] communicationKind the communication kind.

References mCommunicationKind.

Referenced by inheritAttributes().

void lama::Matrix::setComputeKind ( SyncKind computeKind ) [virtual]

set the compute kind.

Parameters:

[in] computeKind the compute kind.

Reimplemented in lama::SparseMatrix< T >, and lama::SparseMatrix< ValueType >.

References mComputeKind.

Referenced by inheritAttributes().

virtual void lama::Matrix::setContext ( const ContextPtr context ) [pure virtual]

specifies on which compute back end the matrix operations should take place.

Parameters:

[in] context the compute back to use for calculations with matrix

Note: Only for sparse matrices it is possible to specify separate locations for local and halo computations.

Implemented in lama::MetaMatrix, lama::SimpleStorageStrategy< ValueType >, lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, and lama::DenseMatrix< T >.

Referenced by inheritAttributes(), lama::MatrixConfigGrammar::MatrixConfigGrammar(), and setContext().

void lama::Matrix::setContext	(	const ContextPtr	localContext,
		const ContextPtr	haloContext
	)		`[virtual]`

Only sparse matrices will override this method, others will ignore second argument.

Reimplemented in lama::SparseMatrix< T >, and lama::SparseMatrix< ValueType >.

References setContext().

void lama::Matrix::setDefaultKind ( ) [private]

References mCommunicationKind, mComputeKind, and SYNCHRONOUS.

Referenced by Matrix().

virtual void lama::Matrix::setDiagonal ( const Vector & diagonal ) [pure virtual]

This method replaces the diagonal.

Parameters:

[in] diagonal is the source array

Calculations are dependent to the diagonal property

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

virtual void lama::Matrix::setDiagonal ( const Scalar scalar ) [pure virtual]

This method replaces the diagonal by a diagonal value.

Parameters:

[in] scalar is the source value

Calculations are dependent to the diagonal property

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

void lama::Matrix::setDistributedMatrix	(	DistributionPtr	distribution,
		DistributionPtr	colDistribution
	)		`[protected]`

Set the global and local size of distributed matrix.

Parameters:

[in]	distribution	is distribution for rows
[in]	colDistribution	is distribution for columns

Global and local size is given implicitly by the distributions itself.

References LAMA_ASSERT_ERROR, mColDistribution, mNumColumns, mNumRows, and lama::Distributed::setDistributionPtr().

Referenced by lama::DenseMatrix< T >::allocate(), lama::SimpleStorageStrategy< ValueType >::applyStrategy(), lama::DenseMatrix< T >::assign(), lama::DenseMatrix< T >::assignSparse(), lama::COOSparseMatrix< T >::COOSparseMatrix(), lama::DenseMatrix< T >::copyDenseMatrix(), lama::CSRSparseMatrix< T >::CSRSparseMatrix(), lama::DIASparseMatrix< T >::DIASparseMatrix(), lama::ELLSparseMatrix< T >::ELLSparseMatrix(), lama::JDSSparseMatrix< T >::JDSSparseMatrix(), Matrix(), setReplicatedMatrix(), lama::DenseMatrix< T >::splitColumns(), and lama::XXXSparseMatrix< T >::XXXSparseMatrix().

void lama::Distributed::setDistributionPtr ( DistributionPtr distributionPtr ) [protected, inherited]

References lama::Distributed::mDistribution.

Referenced by lama::Vector::resize(), and setDistributedMatrix().

virtual void lama::Matrix::setIdentity ( ) [pure virtual]

Operator that sets the matrix to the identity matrix.

 void sub( ..., Matrix& a, ... )
 ...
 LAMA_ASSERT_EQUAL_DEBUG( a.getNumRows(), a.getNumColumns() );
 A.setIdentity();

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::DenseMatrix< T >, lama::MetaMatrix, and lama::SimpleStorageStrategy< ValueType >.

void lama::Matrix::setReplicatedMatrix	(	const IndexType	numRows,
		const IndexType	numColumns
	)		`[protected]`

Set the global/local size of replicated matrix.

Parameters:

[in]	numRows	number of rows, must be non-negative.
[in]	numColumns	number of columns, must be non-negative.

References setDistributedMatrix().

Referenced by lama::DenseMatrix< T >::allocate(), lama::DenseMatrix< T >::assign(), lama::DenseMatrix< T >::clear(), and lama::DenseMatrix< T >::DenseMatrix().

void lama::Distributed::swap ( Distributed & other ) [protected, inherited]

References lama::Distributed::mDistribution.

Referenced by swapMatrix().

void lama::Matrix::swapMatrix ( Matrix & other ) [protected]

References mColDistribution, mNumColumns, mNumRows, and lama::Distributed::swap().

Referenced by lama::DenseMatrix< T >::swap().

virtual void lama::Matrix::wait ( ) const [pure virtual]

wait for a possibly running prefetch.

Implemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, lama::MetaMatrix, lama::DenseMatrix< T >, and lama::SimpleStorageStrategy< ValueType >.

void lama::Matrix::writeAt ( std::ostream & stream ) const [virtual]

Write some information about this to the passed stream.

Parameters:

[out] stream the stream to write to.

Reimplemented from Printable.

Reimplemented in lama::SparseMatrix< T >, lama::SparseMatrix< ValueType >, and lama::DenseMatrix< T >.

References mNumColumns, and mNumRows.

Field Documentation

DistributionPtr lama::Matrix::mColDistribution [protected]

Referenced by checkSettings(), getColDistribution(), getColDistributionPtr(), lama::DenseMatrix< T >::redistribute(), setDistributedMatrix(), and swapMatrix().

SyncKind lama::Matrix::mCommunicationKind [private]

Referenced by getCommunicationKind(), setCommunicationKind(), and setDefaultKind().

SyncKind lama::Matrix::mComputeKind [private]

Referenced by getComputeKind(), setComputeKind(), and setDefaultKind().

IndexType lama::Matrix::mNumColumns [protected]

Referenced by checkSettings(), lama::DenseMatrix< T >::DenseMatrix(), getNumColumns(), getSparsityRate(), Matrix(), setDistributedMatrix(), swapMatrix(), and writeAt().

IndexType lama::Matrix::mNumRows [protected]

Referenced by checkSettings(), lama::DenseMatrix< T >::DenseMatrix(), getNumRows(), getSparsityRate(), Matrix(), setDistributedMatrix(), swapMatrix(), and writeAt().

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

/home/brandes/workspace/LAMA/src/lama/matrix/Matrix.hpp
/home/brandes/workspace/LAMA/src/lama/matrix/Matrix.cpp

Public Types

Public Member Functions

Protected Member Functions

Protected Attributes

Private Member Functions

Private Attributes

Detailed Description

Member Typedef Documentation

Member Enumeration Documentation

Constructor & Destructor Documentation

Member Function Documentation

Field Documentation