TCompRowMatrixMPI< MT > Class Template Reference

Distributed compressed-row sparse matrix class. More...

#include <crmatrix_mpi.h>

Inheritance diagram for TCompRowMatrixMPI< MT >:

Public Member Functions
	TCompRowMatrixMPI ()
	Creates a matrix of dimension 0 x 0.
	TCompRowMatrixMPI (int rows, int cols)
	Creates a matrix of dimension rows x cols. More...
	TCompRowMatrixMPI (int rows, int cols, const int *_rowptr, const int *_colidx, int proc_nrows, const int *proc_rows)
	TCompRowMatrixMPI (int rows, int cols, const int *_rowptr, const int *_colidx, const MT *_data=0)
	Creates a matrix of dimension rows x cols, and sets up data storage with the specified fill structure. More...
	~TCompRowMatrixMPI ()
	Matrix destructor.
MatrixStorage	StorageType () const
	Storage class identifier.
void	Initialise (const int *_rowptr, const int *_colidx, const MT *_data=0)
	Re-allocate fill structure and assign values. More...
void	MPIRange (int *_r0, int *_r1) const
	Returns the row range of the current process. More...
void	Zero ()
	Zero all elements, but keep fill structure. More...
MT	Get (int r, int c) const
	Retrieve a matrix element. More...
TVector< MT >	Row (int r) const
	Returns a vector containing a copy of row `r'. More...
int	SparseRow (int r, int *colidx, MT *val) const
	Returns a row of the matrix in sparse format. More...
TVector< MT >	Col (int c) const
	Returns a vector containing a copy of column 'c'. More...
virtual void	RowScale (const TVector< MT > &scale)
virtual void	ColScale (const TVector< MT > &scale)
virtual void	Unlink ()
bool	Exists (int r, int c) const
	Checks allocation of a matrix element. More...
MT &	operator() (int r, int c)
int	Get_index (int r, int c) const
MT	GetNext (int &r, int &c) const
void	Ax (const TVector< MT > &x, TVector< MT > &b) const
	Matrix-vector product. More...
void	Ax (const TVector< MT > &x, TVector< MT > &b, int i1, int i2) const
void	ATx (const TVector< MT > &x, TVector< MT > &b) const
void	Add_proc (int r, int c, MT val)
	Add a value to a matrix element (process-specific) More...
void	MPIinit ()
	MPI data initialisation. More...
Public Member Functions inherited from TGenericSparseMatrix< MT >
	TGenericSparseMatrix ()
	Create a sparse matrix of size 0 x 0.
	TGenericSparseMatrix (int rows, int cols, int nv=0, const MT *data=0)
	Create a sparse matrix of logical size rows x cols. More...
	TGenericSparseMatrix (int rows, int cols, int nv, MT *data, CopyMode cmode=DEEP_COPY)
	TGenericSparseMatrix (const TGenericSparseMatrix< MT > &m)
	Constructs a matrix as a copy of 'm'.
virtual	~TGenericSparseMatrix ()
	Destructor.
virtual void	New (int nrows, int ncols)
	Reset the matrix dimensions. More...
void	Initialise (int nv, const MT *data)
void	Initialise (int nv, MT *data, CopyMode cmode)
void	Zero ()
int	nVal () const
MT &	Val (int i)
MT *	ValPtr ()
const MT *	ValPtr () const
virtual void	Add (int r, int c, const MT &val)
TGenericSparseMatrix &	operator*= (const MT &sc)
TVector< MT >	operator* (const TVector< MT > &x) const
TVector< MT >	ATx (const TVector< MT > &x) const
double	FillFraction () const
virtual void	Display (std::ostream &os) const
virtual void	PrintFillinGraph (const char *fname, int maxdim=600, bool binary=true, bool antialias=true)
virtual void	ExportRCV (std::ostream &os)
	Write sparse matrix to ASCII output stream. More...
Public Member Functions inherited from TMatrix< MT >
	TMatrix ()
	Create a matrix of size 0 x 0.
	TMatrix (int nrows, int ncols)
	Create a matrix of logical size nrows x ncols. More...
	TMatrix (const TMatrix< MT > &m)
	Create a matrix as a copy of another matrix. More...
virtual	~TMatrix ()
	Destroy the matrix. More...
int	Dim (RC rc) const
	Return a matrix dimension. More...
int	nRows () const
	Return number of rows of the matrix. More...
int	nCols () const
	Return number of columns of the matrix. More...
bool	isSparse () const
	Return sparse storage flag. More...
bool	isFull () const
	Return dense storage flag. More...
MT	operator() (int r, int c) const
	Matrix element access (read only) More...
virtual void	SetRow (int r, const TVector< MT > &row)
	Substitute a row of the matrix. More...
virtual TVector< MT >	Diag () const
	Returns the matrix diagonal as a vector. More...
virtual TVector< MT >	ColNorm () const
	Returns vector of column norms. More...
TVector< MT >	operator* (const TVector< MT > &x) const
TVector< MT >	ATx (const TVector< MT > &x) const
virtual void	Transpone ()
virtual MT	RowMult (int r, MT *x) const
void	Export (std::ostream &os) const
	Write matrix to ASCII stream. More...
void	Print (std::ostream &os=std::cout, int n=80) const
void	PrintNzeroGraph (char *fname)
virtual int	pcg (const TVector< MT > &b, TVector< MT > &x, double &tol, TPreconditioner< MT > *precon=0, int maxit=0) const
virtual void	pcg (const TVector< MT > *b, TVector< MT > *x, int nrhs, double tol, int maxit=0, TPreconditioner< MT > *precon=0, IterativeSolverResult *res=0) const
virtual int	bicgstab (const TVector< MT > &b, TVector< MT > &x, double &tol, TPreconditioner< MT > *precon=0, int maxit=0) const
virtual void	bicgstab (const TVector< MT > *b, TVector< MT > *x, int nrhs, double tol, int maxit=0, TPreconditioner< MT > *precon=0, IterativeSolverResult *res=0) const
template<>
int	pcg (const FVector &b, FVector &x, double &tol, TPreconditioner< float > *precon, int maxit) const
template<>
void	pcg (const FVector *b, FVector *x, int nrhs, double tol, int maxit, TPreconditioner< float > *precon, IterativeSolverResult *res) const
template<>
int	bicgstab (const FVector &b, FVector &x, double &tol, TPreconditioner< float > *precon, int maxit) const
template<>
void	bicgstab (const FVector *b, FVector *x, int nrhs, double tol, int maxit, TPreconditioner< float > *precon, IterativeSolverResult *res) const

Public Attributes
int	rnk
	MPI process id (0 <= rnk < sze)
int	sze
	Number of MPI processes (>= 1)
int *	rowptr
	Array of row pointers. More...
int *	colidx
	Array of column indices. More...
int	r0
	Low row index for this process - OBSOLETE.
int	r1
	High row index + 1 for this process - OBSOLETE.
int	my_nr
	Number of rows managed by this process.
int *	my_r
	List of rows owned by this process.
int *	mpi_r0
	array of row offsets for all processes
int *	mpi_nr
	array of row numbers for all processes
MPI_Datatype	mpitp
	MPI data type corresponding to template type.

Additional Inherited Members
Public Types inherited from TMatrix< MT >
enum	RC { ROW, COL }
Static Public Member Functions inherited from TGenericSparseMatrix< MT >
static void	GlobalSelectIterativeSolver (IterativeMethod method)
static void	GlobalSelectIterativeSolver_complex (IterativeMethod method)
static IterativeMethod	GetGlobalIterativeSolver ()
static IterativeMethod	GetGlobalIterativeSolver_complex ()
Protected Member Functions inherited from TGenericSparseMatrix< MT >
void	Append (MT v=0)
Protected Attributes inherited from TGenericSparseMatrix< MT >
MT *	val
int	nbuf
int	nval
Protected Attributes inherited from TMatrix< MT >
int	rows
int	cols

Detailed Description

template<class MT>
class TCompRowMatrixMPI< MT >

Distributed compressed-row sparse matrix class.

A sparse matrix type that distributes itself across the available processes. The class interface is designed to be compatible with a subset of the TCompRowMatrix interface, hiding the parallelism within the class implementation.

Upon data allocation, an instance of TCompRowMatrixMPI assigns a block of rows r0 <= r < r1 to each process, such that the number of nonzero elements is approximately evenly distributed. Each block is allocated as a compressed row sub- matrix representing part of the complete matrix. Matrix operations such as Ax or ATx are performed in parallel.

Constructor & Destructor Documentation

template<class MT >

TCompRowMatrixMPI< MT >::TCompRowMatrixMPI	(	int	rows,
		int	cols
	)

Creates a matrix of dimension rows x cols.

Parameters

rows	row dimension (>= 0)
cols	column dimension (>= 0)

Note

No space is allocated for matrix element storage. Use Initialise to allocate data space.

template<class MT >

TCompRowMatrixMPI< MT >::TCompRowMatrixMPI	(	int	rows,
		int	cols,
		const int *	_rowptr,
		const int *	_colidx,
		const MT *	_data = 0
	)

Creates a matrix of dimension rows x cols, and sets up data storage with the specified fill structure.

Parameters

rows	row dimension (>= 0)
cols	column dimension (>= 0)
_rowptr	row pointer array of length rows+1
_colidx	column index array of length nzero
_data	optional data array of length nzero

Note

nzero is the number of allocated entries, and is equal to _rowptr[rows].

If no data array is provided, all allocated entries are initialised to zero.

Member Function Documentation

template<class MT >

void TCompRowMatrixMPI< MT >::Add_proc	(	int	r,
		int	c,
		MT	val
	)

Add a value to a matrix element (process-specific)

Parameters

r	row index (>= 0)
c	column index (>= 0)
val	added value

Note

This method is invoked only for the current process. Therefore, r must be in the row range of the process (r0 <= r < r1)

template<class MT >

void TCompRowMatrixMPI< MT >::Ax ( const TVector< MT > &  x, TVector< MT > &  b  ) const

virtual

Matrix-vector product.

Parameters

[in]	x	vector argument (length ncols)
[out]	b	result vector (length nrows)

Note

Computes Ax = b

Implements TGenericSparseMatrix< MT >.

template<class MT >

TVector<MT> TCompRowMatrixMPI< MT >::Col ( int  c ) const

virtual

Returns a vector containing a copy of column 'c'.

Parameters

c	column index (>= 0)

Returns

vector containing column c

Note

Sparse matrix types expand to a dense column, with missing entries filled with zeros, so this can be used as a "scatter" operation.

See Also

Row

Implements TMatrix< MT >.

template<class MT >

bool TCompRowMatrixMPI< MT >::Exists ( int  r, int  c  ) const

virtual

Checks allocation of a matrix element.

Parameters

r	row index (>= 0)
c	column index (>= 0)

Returns

true if memory space is allocated for the element, false if not.

Implements TGenericSparseMatrix< MT >.

template<class MT >

MT TCompRowMatrixMPI< MT >::Get ( int  r, int  c  ) const

virtual

Retrieve a matrix element.

Parameters

r	matrix row (0 <= r < nRows())
c	matrix column (0 <= c < nCols())

Returns

matrix element (*this)_r,c

Note

This is a read operation and returns the element value. For writing operations, use Put().

Implements TGenericSparseMatrix< MT >.

template<class MT >

void TCompRowMatrixMPI< MT >::Initialise	(	const int *	_rowptr,
		const int *	_colidx,
		const MT *	_data = 0
	)

Re-allocate fill structure and assign values.

Parameters

_rowptr	row pointer array of length rows+1
_colidx	column index array of length nzero
_data	optional data array of length nzero

Note

nzero is the number of allocated entries, and is equal to _rowptr[rows].

If no data array is provided, all allocated entries are initialised to zero.

template<class MT >

void TCompRowMatrixMPI< MT >::MPIinit

(

)

MPI data initialisation.

Note

Sets the sze and rnk parameters.

template<class MT >

void TCompRowMatrixMPI< MT >::MPIRange ( int *  _r0, int *  _r1  ) const

inline

Returns the row range of the current process.

Parameters

[out]	_r0	low row index (>= 0)
[out]	_r1	high row index+1

Note

This method returns the range of rows the current process is operating on.

References TCompRowMatrixMPI< MT >::r0, and TCompRowMatrixMPI< MT >::r1.

template<class MT >

TVector<MT> TCompRowMatrixMPI< MT >::Row ( int  r ) const

virtual

Returns a vector containing a copy of row `r'.

Parameters

r	row index (>= 0)

Returns

vector containing row r

Note

The matrix row is expanded into a full vector, replacing un-allocated elements with zeros.

Implements TMatrix< MT >.

template<class MT >

int TCompRowMatrixMPI< MT >::SparseRow ( int  r, int *  colidx, MT *  val  ) const

virtual

Returns a row of the matrix in sparse format.

Returns a list of column indices and values for all allocated entries of row r.

Parameters

[in]	r	row index (>= 0)
[out]	colidx	pointer to array of column indices
[out]	val	pointer to array of element values

Returns

Actual number of allocated matrix entries in the row.

Note

The arrays must be allocated by the caller and be of sufficient size.

See Also

Row, SetRow

Implements TMatrix< MT >.

template<class MT >

void TCompRowMatrixMPI< MT >::Zero

(

)

Zero all elements, but keep fill structure.

Note

This method resets all allocated elements to zero, but does not deallocate the data and index arrays.

Member Data Documentation

template<class MT >

int* TCompRowMatrixMPI< MT >::colidx

Array of column indices.

Note

The colidx array contains the column indices (0-based) of each allocated data entry for the process sub-block.

The dimension of colidx is nval.

template<class MT >

int* TCompRowMatrixMPI< MT >::rowptr

Array of row pointers.

Note

The rowptr array contains the indices of the first nonzero entry of each row in the data array. Note that for the MPI version, the indices refer to the process data sub-block, not to the global data array.

The dimension of rowptr is m+1, where m = r1-r0.

---

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

• /home/ucacmas/src/toast/toast_sf/toastpp-code/toast/trunk/src/libmath/crmatrix_mpi.h