Base class for low level parallel operations with objects of Solver class. More...

#include <inmost_solver.h>

Public Member Functions
void	Clear ()
	Clear all structures and data.

bool &	HaveMatrix ()
	Return true if Matrix data have already been specified.

	OrderInfo ()
	Default initialize all structures.

	OrderInfo (const OrderInfo &other)
	Copy all structures. More...

OrderInfo &	operator= (OrderInfo const &other)
	Assign all structures. More...

	~OrderInfo ()
	Clears all data.

void	PrepareMatrix (Sparse::Matrix &m, INMOST_DATA_ENUM_TYPE overlap)
	Prepare parallel state of the Matrix with specified overlap size. More...

void	RestoreMatrix (Sparse::Matrix &m)
	Restore initial nonparallel state of the Matrix with no overlap. More...

void	PrepareVector (Sparse::Vector &v) const
	Prepare parallel state of the Vector. More...

void	RestoreVector (Sparse::Vector &v) const
	Restore initial nonparallel state of the Vector. More...

INMOST_DATA_ENUM_TYPE	GetProcessor (INMOST_DATA_ENUM_TYPE gind) const
	Retrieve the processor number by binary search for the specified global index. More...

void	GetOverlapRegion (INMOST_DATA_ENUM_TYPE proc, INMOST_DATA_ENUM_TYPE &mbeg, INMOST_DATA_ENUM_TYPE &mend) const
	Get the overlap index region for the specified processor. More...

void	GetLocalRegion (INMOST_DATA_ENUM_TYPE proc, INMOST_DATA_ENUM_TYPE &mbeg, INMOST_DATA_ENUM_TYPE &mend) const
	Get the local index region for the specified processor. More...

void	GetVectorRegion (INMOST_DATA_ENUM_TYPE &mbeg, INMOST_DATA_ENUM_TYPE &mend) const
	Get the local index region for the current processor. More...

INMOST_DATA_ENUM_TYPE	GetRank () const
	Get the rank of the current communicator, i.e. the current process index.

INMOST_DATA_ENUM_TYPE	GetSize () const
	Get the size of the current communicator, i.e. the total number of processes used.

void	Update (Sparse::Vector &x)
	Update the shared data in parallel vector. More...

void	Accumulate (Sparse::Vector &x)
	Sum shared values in parallel vector. More...

void	Integrate (INMOST_DATA_REAL_TYPE *inout, INMOST_DATA_ENUM_TYPE num) const
	Get the sum of num elements of real array on all processes. More...

INMOST_MPI_Comm	GetComm () const
	Get the communicator which the solver is associated with.

INMOST_MPI_Request *	GetSendRequests ()
	MPI structures that hold information on sent data. More...

INMOST_MPI_Request *	GetRecvRequests ()
	MPI structures that hold information on posted receive requests. More...

INMOST_DATA_ENUM_TYPE	GetSendRequestsSize ()
	Total number of send requests.

INMOST_DATA_ENUM_TYPE	GetRecvRequestsSize ()
	Total number of posted receive requests.

INMOST_DATA_ENUM_TYPE *	GetSendExchangeArray ()
	Request raw access to array used to send data.

INMOST_DATA_ENUM_TYPE	GetSendExchangeSize ()
	Size of the array used to send data.

INMOST_DATA_ENUM_TYPE *	GetRecvExchangeArray ()
	Request raw acces to array used to receive data.

INMOST_DATA_ENUM_TYPE	GetRecvExchangeSize ()
	Size of the array used to receive data.

Detailed Description

Base class for low level parallel operations with objects of Solver class.

This class auguments the matrix with additional layers of overlap for Additive Schwartz method to work. Also allows to expand vectors according to new matrix size and to organize vector data exchange in both directions.

Definition at line 87 of file inmost_solver.h.

Constructor & Destructor Documentation

◆ OrderInfo()

INMOST::Solver::OrderInfo::OrderInfo ( const OrderInfo & other )

Copy all structures.

Parameters

other Right hand side.

Member Function Documentation

◆ Accumulate()

void INMOST::Solver::OrderInfo::Accumulate ( Sparse::Vector & x )

Sum shared values in parallel vector.

Parameters

x	Vector for which the shared data should be accumulated.

◆ GetLocalRegion()

void INMOST::Solver::OrderInfo::GetLocalRegion	(	INMOST_DATA_ENUM_TYPE	proc,
		INMOST_DATA_ENUM_TYPE &	mbeg,
		INMOST_DATA_ENUM_TYPE &	mend
	)		const

Get the local index region for the specified processor.

Parameters

proc	Processor number.
mbeg	Record the first index here.
mbeg	Record the last index here.

◆ GetOverlapRegion()

void INMOST::Solver::OrderInfo::GetOverlapRegion	(	INMOST_DATA_ENUM_TYPE	proc,
		INMOST_DATA_ENUM_TYPE &	mbeg,
		INMOST_DATA_ENUM_TYPE &	mend
	)		const

Get the overlap index region for the specified processor.

Parameters

proc	Processor number.
mbeg	Record the first index here.
mbeg	Record the last index here.

◆ GetProcessor()

INMOST_DATA_ENUM_TYPE INMOST::Solver::OrderInfo::GetProcessor ( INMOST_DATA_ENUM_TYPE gind ) const

Retrieve the processor number by binary search for the specified global index.

Finds the processor that contains provided index in it's interval of indices.

Parameters

gind	Global index in the matrix.

◆ GetRecvRequests()

INMOST_MPI_Request* INMOST::Solver::OrderInfo::GetRecvRequests ( )

inline

MPI structures that hold information on posted receive requests.

Access to MPI structures that allow for asynchronous exchange of data.

Definition at line 199 of file inmost_solver.h.

◆ GetSendRequests()

INMOST_MPI_Request* INMOST::Solver::OrderInfo::GetSendRequests ( )

inline

MPI structures that hold information on sent data.

Access to MPI structures that allow for asynchronous exchange of data.

Definition at line 193 of file inmost_solver.h.

◆ GetVectorRegion()

void INMOST::Solver::OrderInfo::GetVectorRegion	(	INMOST_DATA_ENUM_TYPE &	mbeg,
		INMOST_DATA_ENUM_TYPE &	mend
	)		const

inline

Get the local index region for the current processor.

Parameters

mbeg	Record the first index here.
mbeg	Record the last index here.

Definition at line 168 of file inmost_solver.h.

◆ Integrate()

void INMOST::Solver::OrderInfo::Integrate	(	INMOST_DATA_REAL_TYPE *	inout,
		INMOST_DATA_ENUM_TYPE	num
	)		const

Get the sum of num elements of real array on all processes.

Parameters

inout	Data that should be integrated.
num	Number of entries in inout array.

◆ operator=()

OrderInfo& INMOST::Solver::OrderInfo::operator= ( OrderInfo const & other )

Assign all structures.

Parameters

other Right hand side.

◆ PrepareMatrix()

void INMOST::Solver::OrderInfo::PrepareMatrix	(	Sparse::Matrix &	m,
		INMOST_DATA_ENUM_TYPE	overlap
	)

Prepare parallel state of the Matrix with specified overlap size.

This state of the matrix can be used, for instance, to construct the preconditioner for Additive Swartz method.

Parameters

m	Matrix to be expanded.
overlap	Overlap size, viz. the number of overlap layers.

◆ PrepareVector()

void INMOST::Solver::OrderInfo::PrepareVector ( Sparse::Vector & v ) const

Prepare parallel state of the Vector.

Parameters

v	Vector to be expanded.

◆ RestoreMatrix()

void INMOST::Solver::OrderInfo::RestoreMatrix ( Sparse::Matrix & m )

Restore initial nonparallel state of the Matrix with no overlap.

Parameters

m	Matrix to be restored.

◆ RestoreVector()

void INMOST::Solver::OrderInfo::RestoreVector ( Sparse::Vector & v ) const

Restore initial nonparallel state of the Vector.

Parameters

v	Vector to be restored.

◆ Update()

void INMOST::Solver::OrderInfo::Update ( Sparse::Vector & x )

Update the shared data in parallel vector.

Parameters

x	Vector for which the shared data should be updated.

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

inmost_solver.h

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ OrderInfo()

Member Function Documentation

◆ Accumulate()

◆ GetLocalRegion()

◆ GetOverlapRegion()

◆ GetProcessor()

◆ GetRecvRequests()

◆ GetSendRequests()

◆ GetVectorRegion()

◆ Integrate()

◆ operator=()

◆ PrepareMatrix()

◆ PrepareVector()

◆ RestoreMatrix()

◆ RestoreVector()

◆ Update()