The template class MatrixStorage<ValueType> is the base class for all matrix storage classes of a given ValueType. More...

#include <MatrixStorage.hpp>

Inheritance diagram for lama::MatrixStorage< T >:

[legend]

Public Types
typedef T	ValueType
	ValueType stands for type of matrix elements, is given by the template argument.
Public Member Functions
	MatrixStorage (const IndexType numRows, const IndexType numColumns)
	Constructor of matrix storage contains dimensions of the matrix.
	MatrixStorage ()
	Overwrite default constructor, same as MatrixStorage(0,0)
virtual	~MatrixStorage ()
	Destructor.
virtual MatrixStorage *	create () const =0
	Override _MatrixStorage::create with routine that uses covariant return type.
virtual MatrixStorage *	copy () const =0
	Override _MatrixStorage::copy with routine that uses covariant return type.
virtual Scalar::ScalarType	getValueType () const
	Implementation of pure method.
template<typename OtherValueType >
void	setRawDenseData (const IndexType numRows, const IndexType numColumns, const OtherValueType values[], const ValueType eps=0.0)
	Construct a matrix from a dense matrix in row-major order (C-style).
template<typename OtherValueType >
void	setRawCSRData (const IndexType numRows, const IndexType numColumns, const IndexType numValues, const IndexType const ia, const IndexType const ja, const OtherValueType *const values)
	fills matrix storage by csr sparse data.
virtual void	joinHalo (const _MatrixStorage &localData, const _MatrixStorage &haloData, const class Halo &halo, const class Distribution &colDist)
	Join local and halo storage back into one storage as needed for NoDistribution.
virtual void	splitHalo (MatrixStorage< ValueType > &localData, MatrixStorage< ValueType > &haloData, class Halo &halo, const class Distribution &colDist, const class Distribution *rowDist) const
	Splitting of matrix storage for halo.
virtual void	buildHalo (class Halo &halo, const class Distribution &colDist)
	Special version of splitHalo where this matrix contains no local data and where haloData is aliased to this matrix.
virtual void	localize (const _MatrixStorage &globalData, const class Distribution &rowDist)
	This method build for this matrix the local part of a global matrix.
virtual void	replicate (const _MatrixStorage &localData, const class Distribution &rowDist)
	This routine builds the full matrix storage for a distributed matrix.
virtual ValueType	getValue (IndexType i, IndexType j) const =0
	Get a value of the matrix.
virtual void	buildCSCData (LAMAArray< IndexType > &cscIA, LAMAArray< IndexType > &cscJA, LAMAArray< ValueType > &cscValues) const
	This method builds CSC sparse data (column sizes, row indexes and data values) for a matrix storage.
virtual void	assign (const _MatrixStorage &other)
	Format conversion of matrix storage.
virtual void	copyTo (_MatrixStorage &other) const
	The opposite routine to assign, for convenience as the other way around is sometimes more efficient.
virtual void	assignTranspose (const MatrixStorage< ValueType > &other)
	Transpose of matrix storage.
virtual void	redistribute (const _MatrixStorage &other, const class Redistributor &redistributor)
	This methods assigns a redistributed matrix.
virtual void	exchangeHalo (const class Halo &halo, const MatrixStorage< ValueType > &matrix, const class Communicator &comm)
	Build a halo matrix with all rows of required indexes.
MatrixStorage &	operator= (const _MatrixStorage &other)
	Assignment operator for matrix storage.
virtual void	writeToFile (const std::string &fileName, const File::FileType fileType=File::BINARY, const File::DataType dataType=File::INTERNAL, const File::IndexDataType indexDataTypeIA=File::INT, const File::IndexDataType indexDataTypeJA=File::INT) const
	get the number of elements left/right of the diagonal for the given row
virtual void	writeToFile (const PartitionId size, const PartitionId rank, const std::string &fileName, const File::FileType fileType=File::BINARY, const File::DataType dataType=File::INTERNAL, const File::IndexDataType indexDataTypeIA=File::INT, const File::IndexDataType indexDataTypeJA=File::INT) const
virtual void	readFromFile (const std::string &fileName)
virtual void	invert (const MatrixStorage< ValueType > &other)
	assign this storage with the inverse of another matrix.
virtual void	matrixTimesVector (LAMAArrayView< ValueType > result, const ValueType alpha, const LAMAArrayConstView< ValueType > x, const ValueType beta, const LAMAArrayConstView< ValueType > y) const
	This method implements result = alpha * thisMatrix * x + beta * y.
virtual void	matrixTimesVectorN (LAMAArrayView< ValueType > result, const IndexType n, const ValueType alpha, const LAMAArrayConstView< ValueType > x, const ValueType beta, const LAMAArrayConstView< ValueType > y) const
virtual std::auto_ptr< SyncToken >	matrixTimesVectorAsync (LAMAArrayView< ValueType > result, const ValueType alpha, const LAMAArrayConstView< ValueType > x, const ValueType beta, const LAMAArrayConstView< ValueType > y) const
	This method implements result = alpha * thisMatrix * x + beta * y that is executed asynchronously.
virtual void	matrixTimesScalar (const ValueType alpha, const MatrixStorage< ValueType > &a)
	Assign this = alpha * a.
virtual void	matrixTimesMatrix (const ValueType alpha, const MatrixStorage< ValueType > &a, const MatrixStorage< ValueType > &b, const ValueType beta, const MatrixStorage< ValueType > &c)
	assign alpha * a * b + beta * c
virtual ValueType	maxDiffNorm (const MatrixStorage< ValueType > &other)
	Get the maximal absolue element-wise difference between two matrix storages.
virtual void	jacobiIterate (LAMAArrayView< ValueType > solution, const LAMAArrayConstView< ValueType > oldSolution, const LAMAArrayConstView< ValueType > rhs, const ValueType omega) const
	Jacobi iteration step on a local storage.
virtual std::auto_ptr< SyncToken >	jacobiIterateAsync (LAMAArrayView< ValueType > solution, const LAMAArrayConstView< ValueType > oldSolution, const LAMAArrayConstView< ValueType > rhs, const ValueType omega) const
	Asynchrounous version of jacobiIterate.
virtual void	jacobiIterateHalo (LAMAArrayView< ValueType > localSolution, const MatrixStorage< ValueType > &localStorage, const LAMAArrayConstView< ValueType > haloOldSolution, const ValueType omega) const
	Jacobi iteration step on a halo storage.
virtual const char *	getTypeName () const =0
	This method returns the type name of a matrix storage, e.g.
void	init (const IndexType numRows, const IndexType numColumns)
	Initialization of base class due to a resize.
virtual void	clear ()=0
	Clear the matrix storage, resets size to 0 x 0.
virtual void	purge ()=0
	Purge clears the matrix and even frees the allocated memory.
virtual void	allocate (const IndexType numRows, const IndexType numColumns)=0
	This method allocates new matrix storage for the matrix.
void	setContext (ContextPtr context)
	Set the preferred context for the matrix storage.
ContextPtr	getContextPtr () const
	Getter for the preferred context of the storage data, returns pointer.
const Context &	getContext () const
	Getter for the preferred context of the storage data.
virtual void	prefetch (const ContextPtr context) const =0
	Pure method that prefetches storage data into a given context.
void	prefetch () const
	Method that prefetches storage data to its preferred location.
virtual void	wait () const =0
	Will wait for all outstanding asynchronous data transfers.
void	setCompressThreshold (float ratio)
	Allow for additional row compression.
IndexType	getNumRows () const
IndexType	getNumColumns () const
virtual void	writeAt (std::ostream &stream) const
	Writes some Information about this to the passed stream.
void	resetDiagonalProperty ()
bool	hasDiagonalProperty () const
virtual MatrixStorageFormat	getFormat () const =0
virtual void	setIdentity (const IndexType n)=0
	This method sets storage for the identity matrix.
virtual void	getRow (_LAMAArray &row, const IndexType i) const =0
	This method returns the i-th row of the matrix.
virtual void	getDiagonal (_LAMAArray &diagonal) const =0
	This method returns the diagonal of the matrix.
virtual void	setDiagonal (const _LAMAArray &diagonal)=0
	This method sets the diagonal of a matrix storage.
virtual void	setDiagonal (const Scalar value)=0
virtual void	scale (const Scalar value)=0
	This method scales all matrix values with a scalar.
virtual void	scale (const _LAMAArray &values)=0
	This method scales each row of a matrix with a separate value.
virtual void	localize (const _MatrixStorage &global, const Distribution &rowDist)
	This operation localizes the matrix storage of a full matrix to the part that is owned by this processor.
virtual IndexType	getNumValues () const
	Get the total number of non-zero values in the matrix.
const LAMAArray< IndexType > &	getRowIndexes () const
virtual void	buildCSRSizes (LAMAArray< IndexType > &csrIA) const =0
	Get the number of entries in each row.
virtual void	buildCSRData (LAMAArray< IndexType > &csrIA, LAMAArray< IndexType > &csrJA, _LAMAArray &csrValues) const =0
	Get the matrix data of the storage in CSR format.
virtual void	setCSRData (const IndexType numRows, const IndexType numColumns, const IndexType numValues, const LAMAArray< IndexType > &csrIA, const LAMAArray< IndexType > &csrJA, const _LAMAArray &csrValues)=0
	Each storage class must provide a routine to set CSR storage data.
size_t	getMemoryUsage () const
	Returns the number of bytes needed for the current matrix.
virtual void	check (const char *msg) const =0
Static Public Member Functions
static void	convertCSR2CSC (LAMAArray< IndexType > &colIA, LAMAArray< IndexType > &colJA, LAMAArray< ValueType > &colValues, const IndexType numColumns, const LAMAArray< IndexType > &rowIA, const LAMAArray< IndexType > &rowJA, const LAMAArray< ValueType > &rowValues, const ContextPtr loc)
	Conversion routine of Compressed Sparse Row data to Compressed Sparse Column.
static void	joinRows (LAMAArray< IndexType > &outIA, LAMAArray< IndexType > &outJA, LAMAArray< ValueType > &outValues, const IndexType numRows, const LAMAArray< IndexType > &rowIndexes, const LAMAArray< IndexType > &inIA, const LAMAArray< IndexType > &inJA, const LAMAArray< ValueType > &inValues)
	Method that joins rows of another matrix storage.
static void	offsets2sizes (LAMAArray< IndexType > &offsets)
	Help routines to convert arrays with sizes to offsets and vice versa.
static void	offsets2sizes (LAMAArray< IndexType > &sizes, const LAMAArray< IndexType > &offsets)
static IndexType	sizes2offsets (LAMAArray< IndexType > &sizes)
	Utitily to compute the offset array from a sizes array.
Protected Member Functions
void	swap (MatrixStorage< ValueType > &other)
	Swap member variables of base class MatrixStorage<T>
void	swap (_MatrixStorage &other)
	Swaps this with other.
virtual void	_assignTranspose (const _MatrixStorage &other)
virtual void	_assign (const _MatrixStorage &other)
virtual size_t	getMemoryUsageImpl () const =0
	Returns the number of bytes needed for the current matrix.
	LAMA_LOG_DECL_STATIC_LOGGER (logger)
	logger for this matrix format
Protected Attributes
ValueType	mEpsilon
	The value mEpsilon is an individual value for each matrix storage that specifies a threshold when a matrix values can be considered as zero.
IndexType	mNumRows
IndexType	mNumColumns
LAMAArray< IndexType >	mRowIndexes
	used in case of sparse representation of ia
float	mCompressThreshold
	ratio at which compression is done, 0 for never, 1 for always
bool	mDiagonalProperty
	if true, diagonal elements are always stored at first position in each row
ContextPtr	mContext
	preferred context for the storage

Detailed Description

template<typename T>
class lama::MatrixStorage< T >

The template class MatrixStorage<ValueType> is the base class for all matrix storage classes of a given ValueType.

Template Parameters:

T	is the value type of the matrix elements.

Member Typedef Documentation

template<typename T>

typedef T lama::MatrixStorage< T >::ValueType

ValueType stands for type of matrix elements, is given by the template argument.

Reimplemented in lama::DenseStorage< T >, lama::CSRStorage< T >, lama::JDSStorage< T >, lama::COOStorage< T >, lama::ELLStorage< T >, lama::DenseStorageView< T >, lama::SparseAssemblyStorage< T >, and lama::DIAStorage< T >.

Constructor & Destructor Documentation

template<typename ValueType >

lama::MatrixStorage< ValueType >::MatrixStorage	(	const IndexType	numRows,
		const IndexType	numColumns
	)

Constructor of matrix storage contains dimensions of the matrix.

template<typename ValueType >

lama::MatrixStorage< ValueType >::MatrixStorage ( )

Overwrite default constructor, same as MatrixStorage(0,0)

template<typename ValueType >

lama::MatrixStorage< ValueType >::~MatrixStorage ( ) [virtual]

Destructor.

Member Function Documentation

void lama::_MatrixStorage::_assign ( const _MatrixStorage & other ) [protected, virtual, inherited]

References lama::LAMAArray< T >::clear(), lama::_MatrixStorage::mCompressThreshold, lama::_MatrixStorage::mDiagonalProperty, lama::_MatrixStorage::mNumColumns, lama::_MatrixStorage::mNumRows, and lama::_MatrixStorage::mRowIndexes.

Referenced by lama::CSRStorage< T >::assign(), lama::DenseStorageView< T >::assignDenseStorageImpl(), and lama::JDSStorage< T >::assignJDS().

void lama::_MatrixStorage::_assignTranspose ( const _MatrixStorage & other ) [protected, virtual, inherited]

References lama::LAMAArray< T >::clear(), lama::_MatrixStorage::mCompressThreshold, lama::_MatrixStorage::mDiagonalProperty, lama::_MatrixStorage::mNumColumns, lama::_MatrixStorage::mNumRows, and lama::_MatrixStorage::mRowIndexes.

Referenced by lama::CSRStorage< T >::assignTranspose().

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

This method allocates new matrix storage for the matrix.

The matrix contains only zero elements.

Note: This method contains an implicit clear, but not a purge. So allocation of a much smaller matrix compared to the existing one might result in some waste of memory.

Implemented in lama::CSRStorage< T >, lama::DenseStorageView< T >, lama::JDSStorage< T >, lama::DIAStorage< T >, lama::COOStorage< T >, lama::ELLStorage< T >, and lama::SparseAssemblyStorage< T >.

Referenced by lama::MatrixStorage< T >::splitHalo().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::assign ( const _MatrixStorage< T > & other ) [virtual]

Format conversion of matrix storage.

A default implementation is provided using CSR data. Derived clauses might override this method with more efficient solutions.

Implements lama::_MatrixStorage.

Reimplemented in lama::CSRStorage< T >, and lama::DenseStorageView< T >.

References lama::_MatrixStorage::buildCSRData(), lama::_MatrixStorage::copyTo(), lama::CSR, lama::File::DOUBLE, lama::File::FLOAT, lama::_MatrixStorage::getFormat(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), lama::_MatrixStorage::getValueType(), and lama::_LAMAArray::size().

Referenced by lama::DenseStorageView< T >::assign(), lama::SparseMatrix< T >::assign(), lama::COOStorage< T >::COOStorage(), lama::DIAStorage< T >::DIAStorage(), lama::ELLStorage< T >::ELLStorage(), lama::JDSStorage< T >::JDSStorage(), lama::SparseAssemblyStorage< T >::SparseAssemblyStorage(), and lama::MatrixStorage< T >::splitHalo().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::assignTranspose ( const MatrixStorage< ValueType > & other ) [virtual]

Transpose of matrix storage.

A default implementation is provided using CSR data. Derived clauses might override this method with more efficient solutions.

Note: type conversion is not supported here

Reimplemented in lama::CSRStorage< T >.

References lama::MatrixStorage< T >::buildCSCData(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), and lama::_LAMAArray::size().

template<typename T >

void lama::MatrixStorage< T >::buildCSCData	(	LAMAArray< IndexType > &	cscIA,
		LAMAArray< IndexType > &	cscJA,
		LAMAArray< ValueType > &	cscValues
	)		const `[virtual]`

This method builds CSC sparse data (column sizes, row indexes and data values) for a matrix storage.

For efficiency a derived class might override this method, e.g. for CSR storage.

Reimplemented in lama::CSRStorage< T >.

Referenced by lama::CSRStorage< T >::assignTranspose(), lama::MatrixStorage< T >::assignTranspose(), and lama::SparseMatrix< T >::assignTransposeImpl().

virtual void lama::_MatrixStorage::buildCSRData	(	LAMAArray< IndexType > &	csrIA,
		LAMAArray< IndexType > &	csrJA,
		_LAMAArray &	csrValues
	)		const `[pure virtual, inherited]`

Get the matrix data of the storage in CSR format.

Parameters:

[out]	csrIA	offset array for rows, csrIA.size() == numRows + 1
[out]	csrJA	column indexes, csrJA.size() == csrIA[ csrIA.size() ]
[out]	csrValues	are the non-zero matrix values, csrJA.size() == csrValues.size()

The csr data will have the diagonal property if this->hasDiagonalProperty() is true.

Note: This routine supports also type conversion between different value types.

     DenseStorage<double> dense( ... )
     LAMAArray<IndexType> ia, ja;
     LAMAArray<float> values;
     // get non-zero values of dense matrix as float values
     dense.buildCSRData( ia, ja, values );

This routine must be provided by each matrix storage format.

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

Referenced by lama::CSRStorage< T >::assign(), lama::MatrixStorage< T >::assign(), lama::DenseMatrix< T >::assignLocal(), lama::MatrixStorage< T >::exchangeHalo(), lama::MatrixStorage< T >::joinHalo(), lama::MatrixStorage< T >::localize(), and lama::MatrixStorage< T >::redistribute().

virtual void lama::_MatrixStorage::buildCSRSizes ( LAMAArray< IndexType > & csrIA ) const [pure virtual, inherited]

Get the number of entries in each row.

Parameters:

[out] csrIA size array for rows, csrIA.size() == numRows

If this->hasDiagonalProperty() is true, diagonal elements are also counted.

     DenseStorage<double> dense( ... )
     LAMAArray<IndexType> ia;
     dense.buildCSRSizes( ia );

This routine must be provided by each matrix storage format.

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

Referenced by lama::_MatrixStorage::getNumValues().

template<typename T>

void lama::MatrixStorage< ValueType >::buildHalo	(	class Halo &	halo,
		const class Distribution &	colDist
	)		`[virtual]`

Special version of splitHalo where this matrix contains no local data and where haloData is aliased to this matrix.

This routine is used to translate the non-local column indexes (required values) to local halo indexes and to set up the exchange schedule.

References lama::Distribution::getGlobalSize(), LAMA_ASSERT_EQUAL, and lama::_LAMAArray::size().

virtual void lama::_MatrixStorage::check ( const char * msg ) const [pure virtual, inherited]

Implemented in lama::SparseAssemblyStorage< T >, lama::CSRStorage< T >, lama::JDSStorage< T >, lama::ELLStorage< T >, lama::DIAStorage< T >, lama::COOStorage< T >, and lama::DenseStorageView< T >.

Referenced by lama::MatrixStorage< T >::splitHalo().

virtual void lama::_MatrixStorage::clear ( ) [pure virtual, inherited]

Clear the matrix storage, resets size to 0 x 0.

Clearing of matrix storage will not free allocated data, so resize to the original size is a cheap operation.

Implemented in lama::DenseStorageView< T >, lama::JDSStorage< T >, lama::ELLStorage< T >, lama::CSRStorage< T >, lama::SparseAssemblyStorage< T >, lama::DIAStorage< T >, and lama::COOStorage< T >.

template<typename ValueType >

void lama::MatrixStorage< ValueType >::convertCSR2CSC	(	LAMAArray< IndexType > &	colIA,
		LAMAArray< IndexType > &	colJA,
		LAMAArray< ValueType > &	colValues,
		const IndexType	numColumns,
		const LAMAArray< IndexType > &	rowIA,
		const LAMAArray< IndexType > &	rowJA,
		const LAMAArray< ValueType > &	rowValues,
		const ContextPtr	loc
	)		`[static]`

Conversion routine of Compressed Sparse Row data to Compressed Sparse Column.

References lama::ReadAccess< T >::get(), lama::WriteAccess< T >::get(), LAMA_ASSERT_EQUAL_DEBUG, LAMA_CONTEXT_ACCESS, LAMA_INTERFACE_FN_T, LAMA_REGION, and lama::_LAMAArray::size().

template<typename T>

virtual MatrixStorage* lama::MatrixStorage< T >::copy ( ) const [pure virtual]

Override _MatrixStorage::copy with routine that uses covariant return type.

Implements lama::_MatrixStorage.

Implemented in lama::SparseAssemblyStorage< T >, lama::COOStorage< T >, lama::CSRStorage< T >, lama::JDSStorage< T >, lama::ELLStorage< T >, lama::DIAStorage< T >, and lama::DenseStorageView< T >.

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

template<typename ValueType >

void lama::MatrixStorage< ValueType >::copyTo ( _MatrixStorage< T > & other ) const [virtual]

The opposite routine to assign, for convenience as the other way around is sometimes more efficient.

Implements lama::_MatrixStorage.

Reimplemented in lama::CSRStorage< T >.

References LAMA_ASSERT_EQUAL_DEBUG, lama::_MatrixStorage::setCSRData(), and lama::_LAMAArray::size().

template<typename T>

virtual MatrixStorage* lama::MatrixStorage< T >::create ( ) const [pure virtual]

Override _MatrixStorage::create with routine that uses covariant return type.

Implements lama::_MatrixStorage.

Implemented in lama::SparseAssemblyStorage< T >, lama::COOStorage< T >, lama::CSRStorage< T >, lama::JDSStorage< T >, lama::ELLStorage< T >, lama::DIAStorage< T >, and lama::DenseStorageView< T >.

template<typename T>

void lama::MatrixStorage< ValueType >::exchangeHalo	(	const class Halo &	halo,
		const MatrixStorage< ValueType > &	matrix,
		const class Communicator &	comm
	)		`[virtual]`

Build a halo matrix with all rows of required indexes.

References lama::_MatrixStorage::buildCSRData(), lama::_MatrixStorage::getNumColumns(), and lama::_LAMAArray::size().

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

const Context& lama::_MatrixStorage::getContext ( ) const [inline, inherited]

Getter for the preferred context of the storage data.

Referenced by lama::ELLStorage< T >::ELLStorage().

ContextPtr lama::_MatrixStorage::getContextPtr ( ) const [inline, inherited]

Getter for the preferred context of the storage data, returns pointer.

Referenced by lama::SpecializedJacobi::iterateTyped(), and lama::JDSStorage< T >::JDSStorage().

virtual void lama::_MatrixStorage::getDiagonal ( _LAMAArray & diagonal ) const [pure virtual, inherited]

This method returns the diagonal of the matrix.

Parameters:

[out] diagonal is the destination array

The values will be stored in diagonal that can be of any type for which implicit conversion is available.

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

Referenced by lama::JDSStorage< T >::jacobiIterateHalo(), and lama::ELLStorage< T >::jacobiIterateHalo().

virtual MatrixStorageFormat lama::_MatrixStorage::getFormat ( ) const [pure virtual, inherited]

Implemented in lama::SparseAssemblyStorage< T >, lama::CSRStorage< T >, lama::JDSStorage< T >, lama::ELLStorage< T >, lama::DIAStorage< T >, lama::COOStorage< T >, and lama::DenseStorageView< T >.

Referenced by lama::DenseStorageView< T >::assign(), lama::DenseMatrix< T >::assign(), lama::MatrixStorage< T >::assign(), lama::SOR::initialize(), lama::DenseStorageView< T >::invert(), lama::ELLStorage< T >::jacobiIterateHalo(), lama::CSRStorage< T >::jacobiIterateHalo(), lama::DenseStorageView< T >::matrixTimesMatrix(), lama::ELLStorage< T >::matrixTimesMatrix(), lama::CSRStorage< T >::matrixTimesMatrix(), lama::DenseStorageView< T >::maxDiffNorm(), and lama::CSRStorage< T >::maxDiffNorm().

size_t lama::_MatrixStorage::getMemoryUsage ( ) const [inherited]

Returns the number of bytes needed for the current matrix.

Note: This routine does not tell how many memory is really allocated. Storage data might be allocated on more than one device. Furthermore, it is possible that arrays have more memory reserved than needed for its current size.

References lama::_MatrixStorage::getMemoryUsageImpl(), lama::_MatrixStorage::mRowIndexes, and lama::_LAMAArray::size().

virtual size_t lama::_MatrixStorage::getMemoryUsageImpl ( ) const [protected, pure virtual, inherited]

Returns the number of bytes needed for the current matrix.

This pure function must be implemented by each derived class. Relevant for the number of bytes is the current size of the used (LAMA) arrays.

Note: This routine does not tell how many memory is really allocated. Storage data might be allocated on more than one device. Furthermore, it is possible that arrays have more memory reserved than needed for its current size.

Implemented in lama::CSRStorage< T >, lama::ELLStorage< T >, lama::DIAStorage< T >, lama::COOStorage< T >, lama::JDSStorage< T >, lama::DenseStorageView< T >, and lama::SparseAssemblyStorage< T >.

Referenced by lama::_MatrixStorage::getMemoryUsage().

IndexType lama::_MatrixStorage::getNumColumns ( ) const [inline, inherited]

References lama::_MatrixStorage::mNumColumns.

IndexType lama::_MatrixStorage::getNumRows ( ) const [inline, inherited]

References lama::_MatrixStorage::mNumRows.

Referenced by lama::SparseMatrix< T >::assign(), lama::DenseMatrix< T >::assign(), lama::MatrixStorage< T >::assign(), lama::MatrixStorage< T >::assignTranspose(), lama::InverseSolver::decompose(), lama::InverseSolver::invert(), lama::DenseMatrix< T >::invertCyclic(), lama::SOR::iterateImpl(), lama::JDSStorage< T >::jacobiIterateHalo(), lama::ELLStorage< T >::jacobiIterateHalo(), lama::CSRStorage< T >::jacobiIterateHalo(), lama::DenseMatrix< T >::joinColumnData(), lama::MatrixStorage< T >::joinHalo(), lama::_MatrixStorage::localize(), lama::DenseMatrix< T >::localize(), lama::MatrixStorage< T >::localize(), lama::CSRStorage< T >::matrixAddMatrixCSR(), lama::ELLStorage< T >::matrixAddMatrixELL(), lama::CSRStorage< T >::matrixTimesMatrixCSR(), lama::DenseStorageView< T >::matrixTimesMatrixDense(), lama::ELLStorage< T >::matrixTimesMatrixELL(), lama::DenseStorageView< T >::maxDiffNorm(), lama::CSRStorage< T >::maxDiffNorm(), lama::MatrixStorage< T >::maxDiffNorm(), lama::MatrixStorage< T >::redistribute(), lama::MatrixStorage< T >::replicate(), lama::replicate(), lama::SparseMatrix< T >::set(), lama::DenseMatrix< T >::splitColumnData(), lama::ELLSparseMatrix< T >::swapLocalStorage(), lama::CSRSparseMatrix< T >::swapLocalStorage(), lama::DIASparseMatrix< T >::swapLocalStorage(), lama::JDSSparseMatrix< T >::swapLocalStorage(), and lama::COOSparseMatrix< T >::swapLocalStorage().

IndexType lama::_MatrixStorage::getNumValues ( ) const [virtual, inherited]

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

Returns:: total number of non-zero values

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).

This routine gives exactly the same number of elements that will be the size of the ja and values array when calling buildCSRData.

The default implementation uses the routine buildCSRSizes, but derived classes should override it if they can do it more efficiently.

Reimplemented in lama::SparseAssemblyStorage< T >, lama::JDSStorage< T >, lama::CSRStorage< T >, lama::COOStorage< T >, lama::ELLStorage< T >, and lama::DenseStorageView< T >.

References lama::_MatrixStorage::buildCSRSizes(), lama::ReadAccess< T >::get(), lama::_MatrixStorage::mNumRows, and lama::OpenMPUtils::sum().

virtual void lama::_MatrixStorage::getRow	(	_LAMAArray &	row,
		const IndexType	i
	)		const `[pure virtual, inherited]`

This method returns the i-th row of the matrix.

Parameters:

[out]	row	is the destination array that will contain the row
[in]	i	is the row index

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

const LAMAArray< IndexType > & lama::_MatrixStorage::getRowIndexes ( ) const [inline, inherited]

References lama::_MatrixStorage::mRowIndexes.

virtual const char* lama::_MatrixStorage::getTypeName ( ) const [pure virtual, inherited]

This method returns the type name of a matrix storage, e.g.

CSRStorage<double>.

This routine is more convenient for error messages than typeid(*this).name()

Implemented in lama::DenseStorage< T >, lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

template<typename T>

virtual ValueType lama::MatrixStorage< T >::getValue	(	IndexType	i,
		IndexType	j
	)		const `[pure virtual]`

Get a value of the matrix.

Parameters:

[in]	i	is the row index, 0 <= i < mNumRows
[in]	j	is the colum index, 0 <= j < mNumRows

Exceptions:

Exception out-of-range is enabled for ASSERT_LEVEL=DEBUG.

Implemented in lama::DIAStorage< T >, lama::COOStorage< T >, lama::JDSStorage< T >, lama::CSRStorage< T >, lama::ELLStorage< T >, lama::SparseAssemblyStorage< T >, and lama::DenseStorageView< T >.

template<typename ValueType >

Scalar::ScalarType lama::MatrixStorage< ValueType >::getValueType ( ) const [virtual]

Implementation of pure method.

Implements lama::_MatrixStorage.

Reimplemented in lama::DenseStorageView< T >.

Referenced by lama::DenseStorageView< T >::maxDiffNorm(), and lama::CSRStorage< T >::maxDiffNorm().

bool lama::_MatrixStorage::hasDiagonalProperty ( ) const [inline, inherited]

References lama::_MatrixStorage::mDiagonalProperty.

Referenced by lama::SOR::iterateImpl(), lama::JDSStorage< T >::jacobiIterateHalo(), lama::ELLStorage< T >::jacobiIterateHalo(), lama::CSRStorage< T >::jacobiIterateHalo(), and lama::MatrixStorage< T >::joinHalo().

void lama::_MatrixStorage::init	(	const IndexType	numRows,
		const IndexType	numColumns
	)		`[inherited]`

Initialization of base class due to a resize.

References lama::_MatrixStorage::mCompressThreshold, lama::_MatrixStorage::mDiagonalProperty, lama::_MatrixStorage::mNumColumns, and lama::_MatrixStorage::mNumRows.

Referenced by lama::_MatrixStorage::_MatrixStorage(), lama::SparseAssemblyStorage< T >::allocate(), lama::CSRStorage< T >::allocate(), lama::ELLStorage< T >::setCSRDataImpl(), lama::JDSStorage< T >::setCSRDataImpl(), lama::DIAStorage< T >::setCSRDataImpl(), and lama::CSRStorage< T >::setIdentity().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::invert ( const MatrixStorage< ValueType > & other ) [virtual]

assign this storage with the inverse of another matrix.

Parameters:

[in] other is matrix to invert, must be square

Note: other matrix storage can be aliased with this storage.

Reimplemented in lama::DenseStorageView< T >.

References lama::DenseStorageView< T >::invert().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::jacobiIterate	(	LAMAArrayView< ValueType >	solution,
		const LAMAArrayConstView< ValueType >	oldSolution,
		const LAMAArrayConstView< ValueType >	rhs,
		const ValueType	omega
	)		const `[virtual]`

Jacobi iteration step on a local storage.

solution = omega * ( rhs + B * oldSolution) * dinv + ( 1 - omega ) * oldSolution

where B is this storage without diagonal, dinv the inverse diagonal

Parameters:

[out]	solution	is the solution vector to be computed
[in]	oldSolution	is the old solution vector
[in]	rhs	is the right hand side of equation system to solve
[in]	omega	is the scaling factor

Reimplemented in lama::CSRStorage< T >, lama::ELLStorage< T >, lama::COOStorage< T >, lama::DIAStorage< T >, and lama::JDSStorage< T >.

References lama::CSRStorage< T >::jacobiIterate(), and LAMA_UNSUPPORTED.

Referenced by lama::SpecializedJacobi::iterateTyped().

template<typename ValueType >

std::auto_ptr< SyncToken > lama::MatrixStorage< ValueType >::jacobiIterateAsync	(	LAMAArrayView< ValueType >	solution,
		const LAMAArrayConstView< ValueType >	oldSolution,
		const LAMAArrayConstView< ValueType >	rhs,
		const ValueType	omega
	)		const `[virtual]`

Asynchrounous version of jacobiIterate.

Referenced by lama::SpecializedJacobi::iterateTyped().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::jacobiIterateHalo	(	LAMAArrayView< ValueType >	localSolution,
		const MatrixStorage< ValueType > &	localStorage,
		const LAMAArrayConstView< ValueType >	haloOldSolution,
		const ValueType	omega
	)		const `[virtual]`

Jacobi iteration step on a halo storage.

solution -= omega * ( B(halo) * oldSolution) * dinv

Parameters:

[in,out]	localSolution	is the solution vector that is updated
[in]	localStorage	is needed to get the diagonal
[in]	haloOldSolution	is the old solution vector of halo part
[in]	omega	is the scaling factor.

While local storage must be square and have the diagonal property, this matrix storage does not have to be square.

Reimplemented in lama::CSRStorage< T >, lama::ELLStorage< T >, and lama::JDSStorage< T >.

References lama::CSRStorage< T >::jacobiIterateHalo(), and LAMA_UNSUPPORTED.

Referenced by lama::SpecializedJacobi::iterateTyped().

template<typename T>

void lama::MatrixStorage< ValueType >::joinHalo	(	const _MatrixStorage< T > &	localData,
		const _MatrixStorage< T > &	haloData,
		const class Halo &	halo,
		const class Distribution &	colDist
	)		`[virtual]`

Join local and halo storage back into one storage as needed for NoDistribution.

This matrix storage is used as output matrix.

Parameters:

[in]	localData	is matrix storage with local data
[in]	haloData	is the matrix storage with halo data
[in]	halo	is the communicaton halo, contains also mapping halo to global indexes
[in]	colDist	is the distribution that has be used for getting the local data

Attention: localData might be aliased with this matrix storage, while haloData must not

Derived classes might use a default implementation that is based on joining CSR data with corresponding conversions.

References lama::_MatrixStorage::buildCSRData(), lama::Distribution::getGlobalSize(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), lama::Halo::getRequiredIndexes(), lama::_MatrixStorage::hasDiagonalProperty(), lama::Distribution::local2global(), lama::min(), and lama::_LAMAArray::size().

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

template<typename ValueType >

void lama::MatrixStorage< ValueType >::joinRows	(	LAMAArray< IndexType > &	outIA,
		LAMAArray< IndexType > &	outJA,
		LAMAArray< ValueType > &	outValues,
		const IndexType	numRows,
		const LAMAArray< IndexType > &	rowIndexes,
		const LAMAArray< IndexType > &	inIA,
		const LAMAArray< IndexType > &	inJA,
		const LAMAArray< ValueType > &	inValues
	)		`[static]`

Method that joins rows of another matrix storage.

row(i) of out contains all elements of in(k) with rowIndexes[k] == i

References lama::ReadAccess< T >::get(), lama::WriteAccess< T >::get(), lama::WriteAccess< T >::resize(), lama::ReadAccess< T >::size(), and lama::WriteAccess< T >::size().

lama::_MatrixStorage::LAMA_LOG_DECL_STATIC_LOGGER ( logger ) [protected, inherited]

logger for this matrix format

Reimplemented in lama::CSRStorage< T >, lama::SparseAssemblyStorage< T >, lama::ELLStorage< T >, lama::DIAStorage< T >, lama::JDSStorage< T >, lama::COOStorage< T >, and lama::DenseStorageView< T >.

void lama::_MatrixStorage::localize	(	const _MatrixStorage &	global,
		const Distribution &	rowDist
	)		`[virtual, inherited]`

This operation localizes the matrix storage of a full matrix to the part that is owned by this processor.

This means that only the owned rows of the matrix will be kept.

Notes:

* routine can also be used if global is aliased with this matrix. * this routine is the same as an assign in case of a replicated distribution

References lama::Distribution::getGlobalSize(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), LAMA_ASSERT_EQUAL_ERROR, and LAMA_THROWEXCEPTION.

template<typename T>

void lama::MatrixStorage< ValueType >::localize	(	const _MatrixStorage< T > &	globalData,
		const class Distribution &	rowDist
	)		`[virtual]`

This method build for this matrix the local part of a global matrix.

The row distribution specifies which rows of the global matrix will be used for this storage locally.

Attention: globalMatrix might be aliased to this storage.

References lama::_MatrixStorage::buildCSRData(), lama::Distribution::getGlobalSize(), lama::Distribution::getLocalSize(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), lama::Distribution::isReplicated(), LAMA_ASSERT_EQUAL_ERROR, and lama::_LAMAArray::size().

Referenced by lama::DenseMatrix< T >::assign(), and lama::MatrixStorage< T >::splitHalo().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::matrixTimesMatrix	(	const ValueType	alpha,
		const MatrixStorage< ValueType > &	a,
		const MatrixStorage< ValueType > &	b,
		const ValueType	beta,
		const MatrixStorage< ValueType > &	c
	)		`[virtual]`

assign alpha * a * b + beta * c

Reimplemented in lama::CSRStorage< T >, lama::ELLStorage< T >, and lama::DenseStorageView< T >.

References lama::CSRStorage< T >::check(), LAMA_UNSUPPORTED, and lama::CSRStorage< T >::matrixTimesMatrix().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::matrixTimesScalar	(	const ValueType	alpha,
		const MatrixStorage< ValueType > &	a
	)		`[virtual]`

Assign this = alpha * a.

The default implementation assigns the matrix a and scales it afterwards.

template<typename ValueType >

void lama::MatrixStorage< ValueType >::matrixTimesVector	(	LAMAArrayView< ValueType >	result,
		const ValueType	alpha,
		const LAMAArrayConstView< ValueType >	x,
		const ValueType	beta,
		const LAMAArrayConstView< ValueType >	y
	)		const `[virtual]`

This method implements result = alpha * thisMatrix * x + beta * y.

Each matrix storage must provide this kind of matrix-vector multiplication. Default implementation throws exception for non-availability in derived class.

Reimplemented in lama::CSRStorage< T >, lama::ELLStorage< T >, lama::COOStorage< T >, lama::DenseStorageView< T >, lama::DIAStorage< T >, and lama::JDSStorage< T >.

References LAMA_UNSUPPORTED, and lama::CSRStorage< T >::matrixTimesVector().

template<typename ValueType >

std::auto_ptr< SyncToken > lama::MatrixStorage< ValueType >::matrixTimesVectorAsync	(	LAMAArrayView< ValueType >	result,
		const ValueType	alpha,
		const LAMAArrayConstView< ValueType >	x,
		const ValueType	beta,
		const LAMAArrayConstView< ValueType >	y
	)		const `[virtual]`

This method implements result = alpha * thisMatrix * x + beta * y that is executed asynchronously.

Each matrix storage should provide this kind of matrix-vector multiplication.

A default implementation is provided that does the operation synchronsly and returns a NoSyncToken.

Reimplemented in lama::CSRStorage< T >, and lama::ELLStorage< T >.

template<typename ValueType >

void lama::MatrixStorage< ValueType >::matrixTimesVectorN	(	LAMAArrayView< ValueType >	result,
		const IndexType	n,
		const ValueType	alpha,
		const LAMAArrayConstView< ValueType >	x,
		const ValueType	beta,
		const LAMAArrayConstView< ValueType >	y
	)		const `[virtual]`

Reimplemented in lama::CSRStorage< T >.

References LAMA_UNSUPPORTED, and lama::CSRStorage< T >::matrixTimesVectorN().

template<typename ValueType >

ValueType lama::MatrixStorage< ValueType >::maxDiffNorm ( const MatrixStorage< ValueType > & other ) [virtual]

Get the maximal absolue element-wise difference between two matrix storages.

Reimplemented in lama::CSRStorage< T >, and lama::DenseStorageView< T >.

References lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), LAMA_ASSERT_EQUAL_ERROR, LAMA_UNSUPPORTED, and lama::DenseStorageView< T >::maxDiffNorm().

void lama::_MatrixStorage::offsets2sizes ( LAMAArray< IndexType > & offsets ) [static, inherited]

Help routines to convert arrays with sizes to offsets and vice versa.

References lama::WriteAccess< T >::resize(), and lama::_LAMAArray::size().

Referenced by lama::_MatrixStorage::offsets2sizes().

void lama::_MatrixStorage::offsets2sizes	(	LAMAArray< IndexType > &	sizes,
		const LAMAArray< IndexType > &	offsets
	)		`[static, inherited]`

References lama::WriteAccess< T >::clear(), lama::_MatrixStorage::offsets2sizes(), lama::WriteAccess< T >::resize(), and lama::_LAMAArray::size().

template<typename ValueType >

MatrixStorage< ValueType > & lama::MatrixStorage< ValueType >::operator= ( const _MatrixStorage< T > & other )

Assignment operator for matrix storage.

Using 'assign'

Reimplemented from lama::_MatrixStorage.

Reimplemented in lama::DenseStorage< T >, lama::SparseAssemblyStorage< T >, lama::CSRStorage< T >, lama::JDSStorage< T >, lama::COOStorage< T >, lama::ELLStorage< T >, and lama::DIAStorage< T >.

virtual void lama::_MatrixStorage::prefetch ( const ContextPtr context ) const [pure virtual, inherited]

Pure method that prefetches storage data into a given context.

Implemented in lama::DIAStorage< T >, lama::COOStorage< T >, lama::JDSStorage< T >, lama::CSRStorage< T >, lama::ELLStorage< T >, lama::SparseAssemblyStorage< T >, and lama::DenseStorageView< T >.

Referenced by lama::SpecializedJacobi::iterateTyped().

void lama::_MatrixStorage::prefetch ( ) const [inline, inherited]

Method that prefetches storage data to its preferred location.

References lama::_MatrixStorage::prefetch().

Referenced by lama::_MatrixStorage::prefetch().

virtual void lama::_MatrixStorage::purge ( ) [pure virtual, inherited]

Purge clears the matrix and even frees the allocated memory.

Implemented in lama::DenseStorageView< T >, lama::JDSStorage< T >, lama::DIAStorage< T >, lama::ELLStorage< T >, lama::COOStorage< T >, lama::CSRStorage< T >, and lama::SparseAssemblyStorage< T >.

template<typename ValueType >

void lama::MatrixStorage< ValueType >::readFromFile ( const std::string & fileName ) [virtual]

References lama::_LAMAArray::size().

template<typename T>

void lama::MatrixStorage< ValueType >::redistribute	(	const _MatrixStorage< T > &	other,
		const class Redistributor &	redistributor
	)		`[virtual]`

This methods assigns a redistributed matrix.

Parameters:

[in]	other	is the local part of the matrix data to redistribute
[in]	redistributor	contains source distribution of other and target distribution of this

References lama::_MatrixStorage::buildCSRData(), lama::Distribution::getLocalSize(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), lama::Redistributor::getSourceDistributionPtr(), lama::Redistributor::getTargetDistributionPtr(), lama::Distribution::isReplicated(), LAMA_ASSERT_EQUAL_ERROR, and lama::_LAMAArray::size().

template<typename T>

void lama::MatrixStorage< ValueType >::replicate	(	const _MatrixStorage< T > &	localData,
		const class Distribution &	rowDist
	)		`[virtual]`

This routine builds the full matrix storage for a distributed matrix.

This routine is exactly the inverse routine to the localize routine. After this operation this matrix storage will contain on each processor the global matrix data.

References lama::Distribution::getGlobalSize(), lama::Distribution::getLocalSize(), lama::_MatrixStorage::getNumColumns(), lama::_MatrixStorage::getNumRows(), lama::Distribution::isReplicated(), LAMA_ASSERT_EQUAL_ERROR, and lama::_LAMAArray::size().

void lama::_MatrixStorage::resetDiagonalProperty ( ) [inherited]

References lama::_MatrixStorage::checkDiagonalProperty(), and lama::_MatrixStorage::mDiagonalProperty.

Referenced by lama::COOStorage< T >::COOStorage(), lama::ELLStorage< T >::ELLStorage(), and lama::JDSStorage< T >::JDSStorage().

virtual void lama::_MatrixStorage::scale ( const Scalar value ) [pure virtual, inherited]

This method scales all matrix values with a scalar.

Parameters:

[in] value is the source value

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

void lama::_MatrixStorage::scale ( const _LAMAArray & values ) [pure virtual, inherited]

This method scales each row of a matrix with a separate value.

Parameters:

[in] values contains one scale factor for each row.

Each row of the matrix is scaled with the corresponding value.

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

References LAMA_THROWEXCEPTION.

void lama::_MatrixStorage::setCompressThreshold ( float ratio ) [inherited]

Allow for additional row compression.

References LAMA_THROWEXCEPTION, and lama::_MatrixStorage::mCompressThreshold.

Referenced by lama::MatrixStorage< T >::splitHalo().

void lama::_MatrixStorage::setContext ( ContextPtr context ) [inherited]

Set the preferred context for the matrix storage.

Parameters:

[in] context specifies where the storage should be allocated and where operations should be done

References lama::_MatrixStorage::mContext.

Referenced by lama::CSRStorage< T >::CSRStorage(), lama::DenseStorage< T >::DenseStorage(), and lama::ELLStorage< T >::ELLStorage().

virtual void lama::_MatrixStorage::setCSRData	(	const IndexType	numRows,
		const IndexType	numColumns,
		const IndexType	numValues,
		const LAMAArray< IndexType > &	csrIA,
		const LAMAArray< IndexType > &	csrJA,
		const _LAMAArray &	csrValues
	)		`[pure virtual, inherited]`

Each storage class must provide a routine to set CSR storage data.

Parameters:

numRows	number of rows
numColumns	number of columns
numValues	number of non-zero values
csrIA	offset array ia for column indexes, size is numRows + 1
csrJA	are the column indexes of matrix entries, size is numValues
csrValues	are the values of matrix entries, size is numValues

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

Referenced by lama::CSRStorage< T >::copyTo(), lama::MatrixStorage< T >::copyTo(), and lama::MatrixStorage< T >::splitHalo().

void lama::_MatrixStorage::setDiagonal ( const _LAMAArray & diagonal ) [pure virtual, inherited]

This method sets the diagonal of a matrix storage.

Implementation of this routine must be provided by all derived classes.

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

References LAMA_THROWEXCEPTION.

virtual void lama::_MatrixStorage::setDiagonal ( const Scalar value ) [pure virtual, inherited]

Implemented in lama::CRTPMatrixStorage< Derived, ValueType >, lama::CRTPMatrixStorage< COOStorage< T >, T >, lama::CRTPMatrixStorage< ELLStorage< T >, T >, lama::CRTPMatrixStorage< JDSStorage< T >, T >, lama::CRTPMatrixStorage< SparseAssemblyStorage< T >, T >, lama::CRTPMatrixStorage< DenseStorageView< T >, T >, lama::CRTPMatrixStorage< DIAStorage< T >, T >, and lama::CRTPMatrixStorage< CSRStorage< T >, T >.

virtual void lama::_MatrixStorage::setIdentity ( const IndexType n ) [pure virtual, inherited]

This method sets storage for the identity matrix.

Parameters:

in)	n is the size of the square matrix

Implemented in lama::SparseAssemblyStorage< T >, lama::DIAStorage< T >, lama::JDSStorage< T >, lama::CSRStorage< T >, lama::ELLStorage< T >, lama::COOStorage< T >, and lama::DenseStorageView< T >.

template<typename ValueType >

template<typename OtherValueType >

void lama::MatrixStorage< ValueType >::setRawCSRData	(	const IndexType	numRows,
		const IndexType	numColumns,
		const IndexType	numValues,
		const IndexType *const	ia,
		const IndexType *const	ja,
		const OtherValueType *const	values
	)

fills matrix storage by csr sparse data.

Parameters:

[in]	numRows	number of rows
[in]	numColumns	number of columns
[in]	numValues	the number of stored elements in the matrix
[in]	ia	row pointer of the input csr sparse matrix
[in]	ja	column indexes of the input csr sparse matrix
[in]	values	the data values of the input csr sparse matrix

Referenced by lama::MatrixCreator< T >::fillRandom().

template<typename ValueType >

template<typename OtherValueType >

template LAMA_DLL_IMPORTEXPORT void lama::MatrixStorage< T >::setRawDenseData	(	const IndexType	numRows,
		const IndexType	numColumns,
		const OtherValueType	values[],
		const ValueType	eps = `0.0`
	)

Construct a matrix from a dense matrix in row-major order (C-style).

Values of the matrix will be considered as zero if their absolute value is smaller than eps.

Parameters:

[in]	numRows	number of rows
[in]	numColumns	number of columns
[in]	values	the dense matrix values in row-major order (C-style)
[in]	eps	threshold value for non-zero elements

Sparse matrix formats will have the diagonal property if numRows == numColums

References LAMA_ASSERT_ERROR, and lama::_LAMAArray::size().

IndexType lama::_MatrixStorage::sizes2offsets ( LAMAArray< IndexType > & sizes ) [static, inherited]

Utitily to compute the offset array from a sizes array.

References lama::WriteAccess< T >::resize(), and lama::_LAMAArray::size().

Referenced by lama::SparseAssemblyStorage< T >::buildCSR().

template<typename T>

void lama::MatrixStorage< ValueType >::splitHalo	(	MatrixStorage< ValueType > &	localData,
		MatrixStorage< ValueType > &	haloData,
		class Halo &	halo,
		const class Distribution &	colDist,
		const class Distribution *	rowDist
	)		const `[virtual]`

Splitting of matrix storage for halo.

Parameters:

[out]	localData	will contain all columns owned by this processor
[out]	haloData	will contain all columns not owned by this processor
[out]	halo	will be the exchange schedule
[in]	colDist	specifies the distribution of the columns
[in]	rowDist	optional, localizes the rows before splitting

References lama::_MatrixStorage::allocate(), lama::MatrixStorage< T >::assign(), lama::_MatrixStorage::check(), lama::Distribution::getGlobalSize(), lama::Distribution::getLocalSize(), lama::Distribution::isReplicated(), LAMA_ASSERT_EQUAL, LAMA_ASSERT_EQUAL_DEBUG, lama::MatrixStorage< T >::localize(), lama::_MatrixStorage::setCompressThreshold(), lama::_MatrixStorage::setCSRData(), and lama::_LAMAArray::size().

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

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

Swaps this with other.

swap is protected to avoid accidently wrong swaps of base classes which do not implement their own swap.

Parameters:

[in,out] other the _MatrixStorage to swap this with

References lama::_MatrixStorage::mCompressThreshold, lama::_MatrixStorage::mContext, lama::_MatrixStorage::mDiagonalProperty, lama::_MatrixStorage::mNumColumns, lama::_MatrixStorage::mNumRows, lama::_MatrixStorage::mRowIndexes, and lama::LAMAArray< T >::swap().

template<typename ValueType >

void lama::MatrixStorage< ValueType >::swap ( MatrixStorage< ValueType > & other ) [protected]

Swap member variables of base class MatrixStorage<T>

Parameters:

[in,out] other the MatrixStorage to swap this with

swap is protected to avoid accidently wrong swaps of derived classes which do not implement their own swap or where storages of different format or types are involved.

 CSRStorage<float> csr;
 CSRStorage<double> csr1;
 ELLStorage<float> ell;
 csr.swap( ell ); // NOT ALLOWED, different format
 csr.swap( csr1 ); // NOT ALLOWED, different type

References lama::MatrixStorage< T >::mEpsilon.

Referenced by lama::SparseAssemblyStorage< T >::swap(), lama::DenseStorageView< T >::swap(), lama::JDSStorage< T >::swap(), lama::COOStorage< T >::swap(), lama::DIAStorage< T >::swap(), lama::CSRStorage< T >::swap(), and lama::ELLStorage< T >::swap().

virtual void lama::_MatrixStorage::wait ( ) const [pure virtual, inherited]

Will wait for all outstanding asynchronous data transfers.

Implemented in lama::DIAStorage< T >, lama::COOStorage< T >, lama::JDSStorage< T >, lama::CSRStorage< T >, lama::ELLStorage< T >, lama::SparseAssemblyStorage< T >, and lama::DenseStorageView< T >.

void lama::_MatrixStorage::writeAt ( std::ostream & stream ) const [virtual, inherited]

Writes some Information about this to the passed stream.

If a deriving class does not overrides writeAt, typeid(this).name() is written to stream.

Parameters:

[out] stream the stream to write to.

Reimplemented from Printable.

Reimplemented in lama::DIAStorage< T >, lama::JDSStorage< T >, lama::CSRStorage< T >, lama::ELLStorage< T >, lama::COOStorage< T >, lama::DenseStorageView< T >, and lama::SparseAssemblyStorage< T >.

References lama::_MatrixStorage::mNumColumns, and lama::_MatrixStorage::mNumRows.

template<typename ValueType >

void lama::MatrixStorage< ValueType >::writeToFile	(	const std::string &	fileName,
		const File::FileType	fileType = `File::BINARY`,
		const File::DataType	dataType = `File::INTERNAL`,
		const File::IndexDataType	indexDataTypeIA = `File::INT`,
		const File::IndexDataType	indexDataTypeJA = `File::INT`
	)		const `[virtual]`

get the number of elements left/right of the diagonal for the given row

Parameters:

[in]	row	global Index of row
[out]	left	maximal distance of non-zero element to diagonal from the left
[out]	right	maximal distance of non-zero element to diagonal from the right

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

template<typename ValueType >

void lama::MatrixStorage< ValueType >::writeToFile	(	const PartitionId	size,
		const PartitionId	rank,
		const std::string &	fileName,
		const File::FileType	fileType = `File::BINARY`,
		const File::DataType	dataType = `File::INTERNAL`,
		const File::IndexDataType	indexDataTypeIA = `File::INT`,
		const File::IndexDataType	indexDataTypeJA = `File::INT`
	)		const `[virtual]`