Distributed dense matrix class. More...

#include <dnsmatrix_mpi.h>

Inheritance diagram for TDenseMatrixMPI< MT >:

Public Member Functions
	TDenseMatrixMPI ()
	Defines a matrix of size 0 x 0.

	TDenseMatrixMPI (int r, int c)
	Defines a matrix of size r x c, and allocates space for a sub-set of rows. More...

	~TDenseMatrixMPI ()

MatrixStorage	StorageType () const
	Matrix storage class. More...

void	New (int r, int c)
	Resize matrix to new dimensions. More...

void	MPIRange (int _r0, int _r1) const
	Returns the row range of the current process. More...

MT	Get (int r, int c) const
	Retrieve a matrix element. More...

MT &	operator() (int r, int c)

TVector< MT >	Row (int r) const
	Returns a row of the matrix. More...

TVector< MT >	Col (int c) const
	Returns a vector containing a copy of column 'c'. More...

int	SparseRow (int r, int ci, MT rv) const
	Returns a row of the matrix in sparse format. More...

void	SetRow (int r, const TVector< MT > &rval)
	Replace entries in row r with the values provided in rval. More...

void	ColScale (const TVector< MT > &scale)

void	RowScale (const TVector< MT > &scale)

TVector< MT >	RowSum () const
	Returns a vector of row sums. More...

TVector< MT >	RowSumSq () const
	Returns a vector of sum of squares for each row. More...

TVector< MT >	ColSum () const
	Returns a vector of column sums. More...

TVector< MT >	ColSumSq () const
	Returns a vector of sum of squares for each column. More...

void	Ax (const TVector< MT > &x, TVector< MT > &b) const
	Matrix x vector product. More...

void	ATx (const TVector< MT > &x, TVector< MT > &b) const
	Matrix-transpose x vector product. More...

MPI_Datatype	MPIType () const
	Returns MPI type compatible with template type.

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 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

Additional Inherited Members
Public Types inherited from TMatrix< MT >
enum	RC { ROW, COL }

Protected Attributes inherited from TMatrix< MT >
int	rows

int	cols

Detailed Description

template<class MT>
class TDenseMatrixMPI< MT >

Distributed dense matrix class.

A dense matrix type that distributes itself across the available processes, by assigning blocks of rows of equal size to each process.

Each process only allocates the memory required to store the associated sub-block. Methods such as matrix x vector (Ax) and transpose(matrix) x vector (ATx) are automaticall performed in parallel, without the need for any special code by the application developer.

Constructor & Destructor Documentation

template<class MT >

TDenseMatrixMPI< MT >::TDenseMatrixMPI	(	int	r,
		int	c
	)

Defines a matrix of size r x c, and allocates space for a sub-set of rows.

Parameters

r	number of rows
c	number of columns

Note: This constructor automatically subdivides the matrix according to the process rank and MPI size.

template<class MT >

TDenseMatrixMPI< MT >::~TDenseMatrixMPI ( )

.

Member Function Documentation

template<class MT >

void TDenseMatrixMPI< MT >::ATx	(	const TVector< MT > &	x,
		TVector< MT > &	b
	)		const

virtual

Matrix-transpose x vector product.

Parameters

[in]	x	Right-hand operand.
[out]	b	Result vector

Note: The size of x must equal the number of rows of the matrix; The size of b must equal the number of columns of the matrix; This is a synchronising method: The result b is collected from all contributions, and re-distributed to all processes.

See Also: Ax

Reimplemented from TMatrix< MT >.

template<class MT >

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

virtual

Matrix x vector product.

Parameters

[in]	x	Right-hand operand.
[out]	b	Result vector

Note: The size of x must equal the number of columns of the matrix.; The size of b must equal the number of rows of the matrix.; This is a synchronising method: The result b is collected from all contributions, and re-distributed to all processes.

See Also: ATx

Reimplemented from TMatrix< MT >.

template<class MT >

TVector<MT> TDenseMatrixMPI< 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 >

TVector<MT> TDenseMatrixMPI< MT >::ColSum ( ) const

Returns a vector of column sums.

Returns: Column sum vector.

Note: The size of the returned vector equals the number of columns of the matrix. Each entry j contains the sum of elements of column j.

See Also: RowSum, RowSumSq, ColSumSq

template<class MT >

TVector<MT> TDenseMatrixMPI< MT >::ColSumSq ( ) const

Returns a vector of sum of squares for each column.

Returns: Vector of sum of squares.

Note: The size of the returned vector equals the number of columns of the matrix. Each entry j contains the sum of the squares of elements of column j.

See Also: ColSum, RowSum, RowSumSq

template<class MT >

MT TDenseMatrixMPI< 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() or operator().

See Also: operator()

Implements TMatrix< MT >.

template<class MT >

void TDenseMatrixMPI< 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.

template<class MT >

void TDenseMatrixMPI< MT >::New	(	int	r,
		int	c
	)

virtual

Resize matrix to new dimensions.

Parameters

r	new row dimension
c	new column dimension

Note: All matrix entries are reset to zero by this method.; This is a blocking function: it must be called for all processes.

Reimplemented from TMatrix< MT >.

template<class MT >

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

virtual

Returns a row of the matrix.

Parameters

r	row index (>= 0)

Returns: Row vector. The size of the returned vector equals the number of columns of the matrix.

Note: This is a blocking method. The process responsible for row r broadcasts its value to all other processes. Each process then returns the same row vector.

Implements TMatrix< MT >.

template<class MT >

TVector<MT> TDenseMatrixMPI< MT >::RowSum ( ) const

Returns a vector of row sums.

Returns: Row sum vector.

Note: The size of the returned vector equals the number of rows of the matrix. Each entry i contains the sum of elements of row i.

See Also: RowSumSq, ColSum, ColSumSq

template<class MT >

TVector<MT> TDenseMatrixMPI< MT >::RowSumSq ( ) const

Returns a vector of sum of squares for each row.

Returns: Vector of sum of squares.

Note: The size of the returned vector equals the number of rows of the matrix. Each entry i contains the sum of the squares of elements of row i.

See Also: RowSum, ColSum, ColSumSq

template<class MT >

void TDenseMatrixMPI< MT >::SetRow	(	int	r,
		const TVector< MT > &	rval
	)

virtual

Replace entries in row r with the values provided in rval.

Parameters

r	row number (>= 0)
rval	replacement row

Note: The length of rval must be equal to the number of columns in the matrix.; This method can be called for all processes, but it affects only the process responsible for row r.

Reimplemented from TMatrix< MT >.

template<class MT >

int TDenseMatrixMPI< 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. This is only really useful for sparse matrix types. For dense matrices this simply returns the complete row, with indices 0, 1, 2, ... n-1

Parameters