The class _MatrixStorage is the base class for all matrix storage classes supported by LAMA. More...

#include <MatrixStorage.hpp>

Inheritance diagram for lama::_MatrixStorage:

Public Member Functions
	_MatrixStorage (const IndexType numRows, const IndexType numColumns)
	Constructor for storage of a matrix of size numRows x numColumns.
	_MatrixStorage ()
	Default constructor.
virtual	~_MatrixStorage ()
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 Scalar::ScalarType	getValueType () const =0
	Objects of this class should give info about their value type.
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.
virtual void	assign (const _MatrixStorage &other)=0
	Assign of matrix storage with any format or value type.
virtual void	copyTo (_MatrixStorage &other) const =0
	The opposite routine to assign, for convenience as the other way around is sometimes more efficient.
_MatrixStorage &	operator= (const _MatrixStorage &other)
	Override default assignment operator.
size_t	getMemoryUsage () const
	Returns the number of bytes needed for the current matrix.
virtual void	check (const char *msg) const =0
virtual _MatrixStorage *	create () const =0
	Each matrix storage must provide a routine that creates a new object of the same matrix type (same format and same value type).
virtual _MatrixStorage *	copy () const =0
	Each matrix storage must provide a routine that makes a new copy of the input matrix (same format and same value type).
Static Public Member Functions
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 &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
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
Private Member Functions
virtual bool	checkDiagonalProperty () const =0
	checkDiagonalProperty checks if the diagonal property of this is full filled.

Detailed Description

The class _MatrixStorage is the base class for all matrix storage classes supported by LAMA.

In contratry to a matrix a matrix storage is never distributed. It contains local parts of a distributed matrix that are typically submatrices.

This base class is also common for all supported value types for matrix elements.

Note: Default copy constructor and assignment operator can be used.

Each matrix storage has a context that will be taken as the preferred location where the matrix data resides and where operations on the matrix storage should take place.

Constructor & Destructor Documentation

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

Constructor for storage of a matrix of size numRows x numColumns.

References init(), mNumColumns, and mNumRows.

lama::_MatrixStorage::_MatrixStorage ( )

Default constructor.

References init().

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

Member Function Documentation

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

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

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

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

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

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().

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

Assign of matrix storage with any format or value type.

Format conversion and type conversion is done implicitly. Derived clauses might override this method with more efficient solutions.

Parameters:

[in] other is the matrix storage to be assigned

Implemented in lama::MatrixStorage< T >, lama::MatrixStorage< ValueType >, lama::CSRStorage< T >, and lama::DenseStorageView< T >.

Referenced by operator=().

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

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]

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 getNumValues().

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

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 bool lama::_MatrixStorage::checkDiagonalProperty ( ) const [private, pure virtual]

checkDiagonalProperty checks if the diagonal property of this is full filled.

Returns:: true if the diagonal property is fulfilled for the matrix data

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

Referenced by resetDiagonalProperty().

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

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 >.

virtual _MatrixStorage* lama::_MatrixStorage::copy ( ) const [pure virtual]

Each matrix storage must provide a routine that makes a new copy of the input matrix (same format and same value type).

The implementations in derived classes should use covariant return types to allow polymorphic creation of objects.

Note: as shared_ptr cannot be used for for covariant return types this routine returns directly a new created object that should be wrapped as a shared pointer at calling site.

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

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

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

Implemented in lama::MatrixStorage< T >, lama::MatrixStorage< ValueType >, and lama::CSRStorage< T >.

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

virtual _MatrixStorage* lama::_MatrixStorage::create ( ) const [pure virtual]

Each matrix storage must provide a routine that creates a new object of the same matrix type (same format and same value type).

The implementations in derived classes should use covariant return types to allow polymorphic creation of objects.

Note: as shared_ptr cannot be used for for covariant return types this routine returns directly a new created object that should be wrapped as a shared pointer at calling site.

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

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

Getter for the preferred context of the storage data.

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

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

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]

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]

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

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

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

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 getMemoryUsage().

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

References mNumColumns.

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

References 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(), 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::COOSparseMatrix< T >::swapLocalStorage(), lama::CSRSparseMatrix< T >::swapLocalStorage(), lama::JDSSparseMatrix< T >::swapLocalStorage(), lama::ELLSparseMatrix< T >::swapLocalStorage(), and lama::DIASparseMatrix< T >::swapLocalStorage().

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

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

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

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]

References mRowIndexes.

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

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 >.

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

Objects of this class should give info about their value type.

Implemented in lama::MatrixStorage< T >, lama::MatrixStorage< ValueType >, and lama::DenseStorageView< T >.

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

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

References 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
	)

Initialization of base class due to a resize.

References mCompressThreshold, mDiagonalProperty, mNumColumns, and mNumRows.

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

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

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]`

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(), getNumColumns(), getNumRows(), LAMA_ASSERT_EQUAL_ERROR, and LAMA_THROWEXCEPTION.

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

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

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

Referenced by offsets2sizes().

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

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

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

Override default assignment operator.

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

References assign().

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

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]

Method that prefetches storage data to its preferred location.

References prefetch().

Referenced by prefetch().

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

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 >.

void lama::_MatrixStorage::resetDiagonalProperty ( )

References checkDiagonalProperty(), and 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]

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]

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 )

Allow for additional row compression.

References LAMA_THROWEXCEPTION, and mCompressThreshold.

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

void lama::_MatrixStorage::setContext ( ContextPtr context )

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 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]`

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