An class template to encapsulate a MultiVector. A MultiVector is a collection of $N$ vectors belonging to the same finite-dimensional vector space, where usual notion of vector size denotes the dimension of the vector space. Note that this in the mathematical sense and not in the sense of an multi-dimensional array.The MultiVector is stored contiguously with the vector index being the fastest index, or in other words a matrix of size $M \times N$ in row major format with $M $ denoting the dimension of the vector space (size of individual vector). More...

#include <MultiVector.h>

Public Types
using	Storage = dftfe::utils::MemoryStorage<ValueType, memorySpace>

using	value_type = typename Storage::value_type

using	pointer = typename Storage::pointer

using	reference = typename Storage::reference

using	const_reference = typename Storage::const_reference

using	iterator = typename Storage::iterator

using	const_iterator = typename Storage::const_iterator

Public Member Functions
	MultiVector ()=default
	Default Constructor.

	~MultiVector ()=default
	Default Destructor.

	MultiVector (const size_type size, const size_type numVectors, const ValueType initVal=0)
	Constructor for serial MultiVector with vector size, number of vectors and initial value arguments.

	MultiVector (std::unique_ptr< typename MultiVector< ValueType, memorySpace >::Storage > storage, size_type numVectors)
	Constructor for a \serial MultiVector with a predefined MultiVector::Storage (i.e., utils::MemoryStorage). This constructor transfers the ownership of the input Storage to the MultiVector. This is useful when one does not want to allocate new memory and instead use memory allocated in the MultiVector::Storage (i.e., utils::MemoryStorage). The locallyOwnedSize, ghostSize, etc., are automatically set using the size of the input Storage object.

	MultiVector (std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > > mpiPatternP2P, const size_type numVectors, const ValueType initVal=0)
	Constructor for a distributed MultiVector based on an input MPIPatternP2P.

	MultiVector (std::unique_ptr< typename MultiVector< ValueType, memorySpace >::Storage > &storage, std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > > mpiPatternP2P, const size_type numVectors)
	Constructor for a distributed MultiVector with a predefined MultiVector::Storage (i.e., utils::MemoryStorage) and MPIPatternP2P. This constructor transfers the ownership of the input Storage to the MultiVector. This is useful when one does not want to allocate new memory and instead use memory allocated in the input MultiVector::Storage (i.e., utils::MemoryStorage).

	MultiVector (const std::pair< global_size_type, global_size_type > locallyOwnedRange, const std::vector< global_size_type > &ghostIndices, const MPI_Comm &mpiComm, const size_type numVectors, ValueType initVal=0)
	Constructor for a \distributed MultiVector based on locally owned and ghost indices.

	MultiVector (const std::pair< global_size_type, global_size_type > locallyOwnedRange, const MPI_Comm &mpiComm, const size_type numVectors, const ValueType initVal=0)
	Constructor for a special case of distributed MultiVector where none none of the processors have any ghost indices.

	MultiVector (const global_size_type globalSize, const MPI_Comm &mpiComm, const size_type numVectors, const ValueType initVal=0)
	Constructor for a distributed MultiVector based on total number of global indices. The resulting MultiVector will not contain any ghost indices on any of the processors. Internally, the vector is divided to ensure as much equitable distribution across all the processors much as possible.

	MultiVector (const MultiVector &u)
	Copy constructor.

	MultiVector (const MultiVector &u, const ValueType initVal=0)
	Copy constructor with reinitialisation.

	MultiVector (MultiVector &&u) noexcept
	Move constructor.

MultiVector &	operator= (const MultiVector &u)
	Copy assignment operator.

MultiVector &	operator= (MultiVector &&u)
	Move assignment operator.

void	swap (MultiVector &u)
	pointer swap

void	reinit (std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > > mpiPatternP2P, const size_type numVectors, const ValueType initVal=0)
	reinit for a distributed MultiVector based on an input MPIPatternP2P.

void	reinit (const MultiVector &u)
	reinit based on an input distributed MultiVector.

iterator	begin ()
	Return iterator pointing to the beginning of MultiVector data.

const_iterator	begin () const
	Return iterator pointing to the beginning of MultiVector data.

iterator	end ()
	Return iterator pointing to the end of MultiVector data.

const_iterator	end () const
	Return iterator pointing to the end of MultiVector data.

ValueType *	data ()
	Return the raw pointer to the MultiVector data.

const ValueType *	data () const
	Return the constant raw pointer to the MultiVector data.

void	setValue (const ValueType val)
	Set all entries of the MultiVector to a given value.

template<typename ValueBaseType>
void	scale (const ValueBaseType val)

template<typename ValueBaseType>
void	add (const ValueBaseType *valVec, const MultiVector &u)

template<typename ValueBaseType>
void	add (const ValueBaseType val, const MultiVector &u)

template<typename ValueBaseType1, typename ValueBaseType2>
void	addAndScale (const ValueBaseType1 valScale, const ValueBaseType2 valAdd, const MultiVector &u)

template<typename ValueBaseType1, typename ValueBaseType2>
void	scaleAndAdd (const ValueBaseType1 valScale, const ValueBaseType2 valAdd, const MultiVector &u)

void	dot (const MultiVector &u, ValueType *dotVec)

void	zeroOutGhosts ()

void	updateGhostValues (const size_type communicationChannel=0)

void	accumulateAddLocallyOwned (const size_type communicationChannel=0)

void	updateGhostValuesBegin (const size_type communicationChannel=0)

void	updateGhostValuesEnd ()

void	accumulateAddLocallyOwnedBegin (const size_type communicationChannel=0)

void	accumulateAddLocallyOwnedEnd ()

bool	isCompatible (const MultiVector< ValueType, memorySpace > &rhs) const

std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > >	getMPIPatternP2P () const

template<typename ValueBaseType>
void	l2Norm (ValueBaseType *normVec) const

void	setCommunicationPrecision (utils::mpi::communicationPrecision commPrecision)

global_size_type	globalSize () const

size_type	localSize () const

size_type	locallyOwnedSize () const

size_type	ghostSize () const

size_type	numVectors () const

const Storage &	getData () const

Private Attributes
std::unique_ptr< Storage >	d_storage

size_type	d_localSize

global_size_type	d_globalSize

size_type	d_locallyOwnedSize

size_type	d_ghostSize

size_type	d_numVectors

std::unique_ptr< utils::mpi::MPICommunicatorP2P< ValueType, memorySpace > >	d_mpiCommunicatorP2P

std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > >	d_mpiPatternP2P

Detailed Description

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>
class dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >

An class template to encapsulate a MultiVector. A MultiVector is a collection of $N$ vectors belonging to the same finite-dimensional vector space, where usual notion of vector size denotes the dimension of the vector space. Note that this in the mathematical sense and not in the sense of an multi-dimensional array.The MultiVector is stored contiguously with the vector index being the fastest index, or in other words a matrix of size $M \times N$ in row major format with $M $ denoting the dimension of the vector space (size of individual vector).

This class handles both serial and distributed MultiVector in a unfied way. There are different constructors provided for the serial and distributed case.

The serial MultiVector, as the name suggests, resides entirely in a processor.

The distributed MultiVector, on the other hand, is distributed across a set of processors. The storage of each of the $N$ vectors in the distributed MultiVector in a processor follows along similar lines to a distributed Vector object and comprises of two parts:

locally owned part: A part of the distribute MultiVector, defined through a contiguous range of indices ( is included, but is not), for which the current processor is the sole owner. The size of the locally owned part (i.e., ) is termed as locallyOwnedSize. Note that the range of indices that comprises the locally owned part (i.e., ) is same for all the vectors in the MultiVector
ghost part: Part of the MultiVector, defined through a set of ghost indices, that are owned by other processors. The size of ghost indices for each vector is termed as ghostSize. Note that the set of indices that define the ghost indices are same for all the vectors in the MultiVector

The global size of each vector in the distributed MultiVector (i.e., the number of unique indices across all the processors) is simply termed as size. Additionally, we define localSize = locallyOwnedSize + ghostSize.

We handle the serial MultiVector as a special case of the distributed MultiVector, wherein size = locallyOwnedSize and ghostSize = 0.

Note: While typically one would link to an MPI library while compiling this class, care is taken to seamlessly allow usage of this class even while not linking to an MPI library. To do so, we have our own MPI wrappers that redirect to the MPI library's function calls and definitions while linking to an MPI library. While not linking to an MPI library, the MPI wrappers provide equivalent functions and definitions that mimic the MPI functions and definitions, albeit for a single processor. This allows the user of this class to seamlessly switch between linking and de-linking to an MPI library without any change in the code and with the expected behavior.; Note that the case of not linking to an MPI library and the case of creating a serial mult-Vector are two independent things. One can still create a serial MultiVector while linking to an MPI library and running the code across multipe processors. That is, one can create a serial MultiVector in one or more than one of the set of processors used when running in parallel. Internally, we handle this by using MPI_COMM_SELF as our MPI_Comm for the serial MultiVector (i.e., the processor does self communication). However, while not linking to an MPI library (which by definition means running on a single processor), there is no notion of communication (neither with self nor with other processors). In such case, both serial and distributed mult-Vector mean the same thing and the MPI wrappers ensure the expected behavior (i.e., the behavior of a MultiVector while using just one processor)

Template Parameters

template	parameter ValueType defines underlying datatype being stored in the MultiVector (i.e., int, double, complex<double>, etc.)
template	parameter memorySpace defines the MemorySpace (i.e., HOST or DEVICE) in which the MultiVector must reside.

Note

Broadly, there are two ways of constructing a distributed MultiVector.

[Prefered and efficient approach] The first approach takes a pointer to an MPIPatternP2P as an input argument (along with other arguments). The MPIPatternP2P, in turn, contains all the information regarding the locally owned and ghost part of the MultiVector as well as the interaction map between processors. This is the most efficient way of constructing a distributed MultiVector as it allows for reusing of an already constructed MPIPatternP2P.
[ Expensive approach] The second approach takes in the locally owned, ghost indices or the total number of indices across all the processors and internally creates an MPIPatternP2P object. Given that the creation of an MPIPatternP2P is expensive, this route of constructing a distributed MultiVector should be avoided.

Member Typedef Documentation

◆ const_iterator

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::const_iterator = typename Storage::const_iterator

◆ const_reference

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::const_reference = typename Storage::const_reference

◆ iterator

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::iterator = typename Storage::iterator

◆ pointer

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::pointer = typename Storage::pointer

◆ reference

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::reference = typename Storage::reference

◆ Storage

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::Storage = dftfe::utils::MemoryStorage<ValueType, memorySpace>

◆ value_type

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

using dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::value_type = typename Storage::value_type

Constructor & Destructor Documentation

◆ MultiVector() [1/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector ( )

default

Default Constructor.

◆ ~MultiVector()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::~MultiVector ( )

default

Default Destructor.

◆ MultiVector() [2/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	const size_type	size,
		const size_type	numVectors,
		const ValueType	initVal = 0 )

Constructor for serial MultiVector with vector size, number of vectors and initial value arguments.

Constructor for a serial MultiVector using size, numVectors and init value.

Parameters

[in]	size	size of each vector in the MultiVector
[in]	numVectors	number of vectors in the MultiVector
[in]	initVal	initial value of elements of the MultiVector

◆ MultiVector() [3/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	std::unique_ptr< typename MultiVector< ValueType, memorySpace >::Storage >	storage,
		size_type	numVectors )

Constructor for a \serial MultiVector with a predefined MultiVector::Storage (i.e., utils::MemoryStorage). This constructor transfers the ownership of the input Storage to the MultiVector. This is useful when one does not want to allocate new memory and instead use memory allocated in the MultiVector::Storage (i.e., utils::MemoryStorage). The locallyOwnedSize, ghostSize, etc., are automatically set using the size of the input Storage object.

Constructor for a \serial MultiVector with a predefined MultiVector::Storage (i.e., utils::MemoryStorage). This constructor transfers the ownership of the input Storage to the MultiVector. This is useful when one does not want to allocate new memory and instead use memory allocated in the MultiVector::Storage (i.e., utils::MemoryStorage).

Parameters

[in]	storage	unique_ptr to MultiVector::Storage whose ownership is to be transfered to the MultiVector
[in]	numVectors	number of vectors in the MultiVector

Note: This Constructor transfers the ownership from the input unique_ptr storage to the internal data member of the MultiVector. Thus, after the function call storage will point to NULL and any access through storage will lead to undefined behavior.

◆ MultiVector() [4/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > >	mpiPatternP2P,
		const size_type	numVectors,
		const ValueType	initVal = 0 )

Constructor for a distributed MultiVector based on an input MPIPatternP2P.

Parameters

[in]	mpiPatternP2P	A shared_ptr to const MPIPatternP2P based on which the distributed MultiVector will be created.
[in]	numVectors	number of vectors in the MultiVector
[in]	initVal	value with which the MultiVector shoud be initialized

◆ MultiVector() [5/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	std::unique_ptr< typename MultiVector< ValueType, memorySpace >::Storage > &	storage,
		std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > >	mpiPatternP2P,
		const size_type	numVectors )

Constructor for a distributed MultiVector with a predefined MultiVector::Storage (i.e., utils::MemoryStorage) and MPIPatternP2P. This constructor transfers the ownership of the input Storage to the MultiVector. This is useful when one does not want to allocate new memory and instead use memory allocated in the input MultiVector::Storage (i.e., utils::MemoryStorage).

Parameters

[in]	storage	unique_ptr to MultiVector::Storage whose ownership is to be transfered to the MultiVector
[in]	mpiPatternP2P	A shared_ptr to const MPIPatternP2P based on which the distributed MultiVector will be created.
[in]	numVectors	number of vectors in the MultiVector

Note: This Constructor transfers the ownership from the input unique_ptr storage to the internal data member of the MultiVector. Thus, after the function call storage will point to NULL and any access through storage will lead to undefined behavior.

◆ MultiVector() [6/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	const std::pair< global_size_type, global_size_type >	locallyOwnedRange,
		const std::vector< global_size_type > &	ghostIndices,
		const MPI_Comm &	mpiComm,
		const size_type	numVectors,
		ValueType	initVal = 0 )

Constructor for a \distributed MultiVector based on locally owned and ghost indices.

Note: This way of construction is expensive. One should use the other constructor based on an input MPIPatternP2P as far as possible.

Parameters

[in]	locallyOwnedRange	a pair which defines a range of indices (continuous) that are owned by the current processor.
[in]	ghostIndices	vector containing an ordered set of ghost indices (ordered in increasing order and non-repeating)
[in]	mpiComm	MPI_Comm object associated with the group of processors across which the MultiVector is to be distributed
[in]	numVectors	number of vectors in the MultiVector
[in]	initVal	value with which the MultiVector shoud be initialized

Note: The locallyOwnedRange should be an open interval where the start index is included, but the end index is not included.; This way of construction is expensive. One should use the other constructor based on an input MPIPatternP2P as far as possible.

◆ MultiVector() [7/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	const std::pair< global_size_type, global_size_type >	locallyOwnedRange,
		const MPI_Comm &	mpiComm,
		const size_type	numVectors,
		const ValueType	initVal = 0 )

Constructor for a special case of distributed MultiVector where none none of the processors have any ghost indices.

Note: This way of construction is expensive. One should use the other constructor based on an input MPIPatternP2P as far as possible.

Parameters

[in]	locallyOwnedRange	a pair which defines a range of indices (continuous) that are owned by the current processor.
[in]	mpiComm	MPI_Comm object associated with the group of processors across which the MultiVector is to be distributed
[in]	numVectors	number of vectors in the MultiVector
[in]	initVal	value with which the MultiVector shoud be initialized

Note: The locallyOwnedRange should be an open interval where the start index included, but the end index is not included.; This way of construction is expensive. One should use the other constructor based on an input MPIPatternP2P as far as possible.

◆ MultiVector() [8/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	const global_size_type	globalSize,
		const MPI_Comm &	mpiComm,
		const size_type	numVectors,
		const ValueType	initVal = 0 )

Constructor for a distributed MultiVector based on total number of global indices. The resulting MultiVector will not contain any ghost indices on any of the processors. Internally, the vector is divided to ensure as much equitable distribution across all the processors much as possible.

Note: This way of construction is expensive. One should use the other constructor based on an input MPIPatternP2P as far as possible. Further, the decomposition is not compatible with other ways of distributed MultiVector construction.

Parameters

[in]	globalSize	Total number of global indices that is distributed over the processors.
[in]	mpiComm	MPI_Comm object associated with the group of processors across which the MultiVector is to be distributed
[in]	numVectors	number of vectors in the MultiVector
[in]	initVal	value with which the MultiVector shoud be initialized

Note: This way of construction is expensive. One should use the other constructor based on an input MPIPatternP2P as far as possible. Further, the decomposition is not compatible with other ways of distributed MultiVector construction.

◆ MultiVector() [9/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector ( const MultiVector< ValueType, memorySpace > & u )

Copy constructor.

Parameters

[in] u MultiVector object to copy from

◆ MultiVector() [10/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector	(	const MultiVector< ValueType, memorySpace > &	u,
		const ValueType	initVal = 0 )

Copy constructor with reinitialisation.

Parameters

[in]	u	MultiVector object to copy from
[in]	initVal	Initial value of the MultiVector

◆ MultiVector() [11/11]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::MultiVector ( MultiVector< ValueType, memorySpace > && u )

noexcept

Move constructor.

Parameters

[in] u MultiVector object to move from

Member Function Documentation

◆ accumulateAddLocallyOwned()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::accumulateAddLocallyOwned ( const size_type communicationChannel = 0 )

◆ accumulateAddLocallyOwnedBegin()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::accumulateAddLocallyOwnedBegin ( const size_type communicationChannel = 0 )

◆ accumulateAddLocallyOwnedEnd()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::accumulateAddLocallyOwnedEnd ( )

◆ add() [1/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

template<typename ValueBaseType>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::add	(	const ValueBaseType *	valVec,
		const MultiVector< ValueType, memorySpace > &	u )

◆ add() [2/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

template<typename ValueBaseType>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::add	(	const ValueBaseType	val,
		const MultiVector< ValueType, memorySpace > &	u )

◆ addAndScale()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

template<typename ValueBaseType1, typename ValueBaseType2>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::addAndScale	(	const ValueBaseType1	valScale,
		const ValueBaseType2	valAdd,
		const MultiVector< ValueType, memorySpace > &	u )

◆ begin() [1/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

MultiVector< ValueType, memorySpace >::iterator dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::begin ( )

Return iterator pointing to the beginning of MultiVector data.

Returns: Iterator pointing to the beginning of MultiVector.

◆ begin() [2/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

MultiVector< ValueType, memorySpace >::const_iterator dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::begin ( ) const

Return iterator pointing to the beginning of MultiVector data.

Returns: Constant iterator pointing to the beginning of MultiVector.

◆ data() [1/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

ValueType * dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::data ( )

Return the raw pointer to the MultiVector data.

Returns: pointer to data

◆ data() [2/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

const ValueType * dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::data ( ) const

Return the constant raw pointer to the MultiVector data.

Returns: pointer to const data

◆ dot()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::dot	(	const MultiVector< ValueType, memorySpace > &	u,
		ValueType *	dotVec )

◆ end() [1/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

MultiVector< ValueType, memorySpace >::iterator dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::end ( )

Return iterator pointing to the end of MultiVector data.

Returns: Iterator pointing to the end of MultiVector.

◆ end() [2/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

MultiVector< ValueType, memorySpace >::const_iterator dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::end ( ) const

Return iterator pointing to the end of MultiVector data.

Returns: Constant iterator pointing to the end of MultiVector.

◆ getData()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

const dftfe::utils::MemoryStorage< ValueType, memorySpace > & dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::getData ( ) const

◆ getMPIPatternP2P()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > > dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::getMPIPatternP2P ( ) const

◆ ghostSize()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::ghostSize ( ) const

◆ globalSize()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

global_size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::globalSize ( ) const

◆ isCompatible()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

bool dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::isCompatible ( const MultiVector< ValueType, memorySpace > & rhs ) const

◆ l2Norm()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

template<typename ValueBaseType>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::l2Norm ( ValueBaseType * normVec ) const

◆ locallyOwnedSize()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::locallyOwnedSize ( ) const

◆ localSize()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::localSize ( ) const

◆ numVectors()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::numVectors ( ) const

◆ operator=() [1/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

MultiVector< ValueType, memorySpace > & dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::operator= ( const MultiVector< ValueType, memorySpace > & u )

Copy assignment operator.

Parameters

[in] u const reference to MultiVector object to copy from

Returns: reference to this object after copying data from u

◆ operator=() [2/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

MultiVector< ValueType, memorySpace > & dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::operator= ( MultiVector< ValueType, memorySpace > && u )

Move assignment operator.

Parameters

[in] u const reference to MultiVector object to move from

Returns: reference to this object after moving data from u

◆ reinit() [1/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::reinit ( const MultiVector< ValueType, memorySpace > & u )

reinit based on an input distributed MultiVector.

◆ reinit() [2/2]

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::reinit	(	std::shared_ptr< const utils::mpi::MPIPatternP2P< memorySpace > >	mpiPatternP2P,
		const size_type	numVectors,
		const ValueType	initVal = 0 )

reinit for a distributed MultiVector based on an input MPIPatternP2P.

Parameters

[in]	mpiPatternP2P	A shared_ptr to const MPIPatternP2P based on which the distributed MultiVector will be reinitialized.
[in]	numVectors	number of vectors in the MultiVector
[in]	initVal	value with which the MultiVector shoud be reinitialized

◆ scale()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

template<typename ValueBaseType>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::scale ( const ValueBaseType val )

◆ scaleAndAdd()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

template<typename ValueBaseType1, typename ValueBaseType2>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::scaleAndAdd	(	const ValueBaseType1	valScale,
		const ValueBaseType2	valAdd,
		const MultiVector< ValueType, memorySpace > &	u )

◆ setCommunicationPrecision()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::setCommunicationPrecision ( utils::mpi::communicationPrecision commPrecision )

◆ setValue()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::setValue ( const ValueType val )

Set all entries of the MultiVector to a given value.

Parameters

[in] val The value to which the entries are to be set

◆ swap()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::swap ( MultiVector< ValueType, memorySpace > & u )

pointer swap

◆ updateGhostValues()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::updateGhostValues ( const size_type communicationChannel = 0 )

◆ updateGhostValuesBegin()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::updateGhostValuesBegin ( const size_type communicationChannel = 0 )

◆ updateGhostValuesEnd()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::updateGhostValuesEnd ( )

◆ zeroOutGhosts()

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

void dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::zeroOutGhosts ( )

Member Data Documentation

◆ d_ghostSize

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_ghostSize

private

◆ d_globalSize

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

global_size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_globalSize

private

◆ d_locallyOwnedSize

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_locallyOwnedSize

private

◆ d_localSize

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_localSize

private

◆ d_mpiCommunicatorP2P

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

std::unique_ptr<utils::mpi::MPICommunicatorP2P<ValueType, memorySpace> > dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_mpiCommunicatorP2P

private

◆ d_mpiPatternP2P

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

std::shared_ptr<const utils::mpi::MPIPatternP2P<memorySpace> > dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_mpiPatternP2P

private

◆ d_numVectors

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

size_type dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_numVectors

private

◆ d_storage

template<typename ValueType, dftfe::utils::MemorySpace memorySpace>

std::unique_ptr<Storage> dftfe::linearAlgebra::MultiVector< ValueType, memorySpace >::d_storage

private

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

workspace/include/MultiVector.h
workspace/include/MultiVector.t.cc

Public Types

Public Member Functions

Private Attributes

Detailed Description

Member Typedef Documentation

◆ const_iterator

◆ const_reference

◆ iterator

◆ pointer

◆ reference

◆ Storage

◆ value_type

Constructor & Destructor Documentation

◆ MultiVector() [1/11]

◆ ~MultiVector()

◆ MultiVector() [2/11]

◆ MultiVector() [3/11]

◆ MultiVector() [4/11]

◆ MultiVector() [5/11]

◆ MultiVector() [6/11]

◆ MultiVector() [7/11]

◆ MultiVector() [8/11]

◆ MultiVector() [9/11]

◆ MultiVector() [10/11]

◆ MultiVector() [11/11]

Member Function Documentation

◆ accumulateAddLocallyOwned()

◆ accumulateAddLocallyOwnedBegin()

◆ accumulateAddLocallyOwnedEnd()

◆ add() [1/2]

◆ add() [2/2]

◆ addAndScale()

◆ begin() [1/2]

◆ begin() [2/2]

◆ data() [1/2]

◆ data() [2/2]

◆ dot()

◆ end() [1/2]

◆ end() [2/2]

◆ getData()

◆ getMPIPatternP2P()

◆ ghostSize()

◆ globalSize()

◆ isCompatible()

◆ l2Norm()

◆ locallyOwnedSize()

◆ localSize()

◆ numVectors()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ reinit() [1/2]

◆ reinit() [2/2]

◆ scale()

◆ scaleAndAdd()

◆ setCommunicationPrecision()

◆ setValue()

◆ swap()

◆ updateGhostValues()

◆ updateGhostValuesBegin()

◆ updateGhostValuesEnd()

◆ zeroOutGhosts()

Member Data Documentation

◆ d_ghostSize

◆ d_globalSize

◆ d_locallyOwnedSize

◆ d_localSize

◆ d_mpiCommunicatorP2P

◆ d_mpiPatternP2P

◆ d_numVectors

◆ d_storage