no.uib.cipr.matrix.distributed
Class DistRowMatrix

java.lang.Object
  no.uib.cipr.matrix.AbstractMatrix
      no.uib.cipr.matrix.distributed.DistRowMatrix

All Implemented Interfaces:

java.lang.Iterable<MatrixEntry>, Matrix

public class DistRowMatrix
extends AbstractMatrix

Distributed matrix with row major blocks

Nested Class Summary

Nested classes/interfaces inherited from interface no.uib.cipr.matrix.Matrix

Matrix.Norm

Field Summary

Fields inherited from class no.uib.cipr.matrix.AbstractMatrix

numColumns, numRows

Constructor Summary

DistRowMatrix(int numRows, int numColumns, Communicator comm, Matrix A, Matrix B)
Constructor for DistRowMatrix

Method Summary

void	add(int row, int column, double value) A(row,column) += value
DistRowMatrix	copy() Creates a deep copy of the matrix
double	get(int row, int column) Returns A(row,column)
Matrix	getBlock() Returns the diagonal block matrix
int[]	getColumnOwnerships() Returns which columns are owned by which ranks.
Communicator	getCommunicator() Gets the communicator associated with this matrix
Matrix	getOff() Returns the off-diagonal matrix
int[]	getRowOwnerships() Returns which rows are owned by which ranks.
java.util.Iterator<MatrixEntry>	iterator()
boolean	local(int row, int column) Returns true if the insertion indices are local to this rank, and no communication is required afterwards.
protected double	max() Returns the largest absolute value
Vector	multAdd(double alpha, Vector x, Vector y) y = alpha*A*x + y
protected double	norm1() Computes the 1 norm
protected double	normF() Computes the Frobenius norm.
protected double	normInf() Computes the infinity norm
Matrix	rank1(double alpha, Vector x, Vector y) A = alpha*x*yT + A.
Matrix	rank2(double alpha, Vector x, Vector y) A = alpha*x*yT + alpha*y*xT + A.
void	set(int row, int column, double value) A(row,column) = value
Vector	transMultAdd(double alpha, Vector x, Vector y) y = alpha*AT*x + y
DistRowMatrix	zero() Zeros all the entries in the matrix, while preserving any underlying structure.

Methods inherited from class no.uib.cipr.matrix.AbstractMatrix

add, add, check, checkMultAdd, checkMultAdd, checkRank1, checkRank1, checkRank2, checkRank2, checkSize, checkSolve, checkSolve, checkTransABmultAdd, checkTransAmultAdd, checkTransBmultAdd, checkTransMultAdd, checkTranspose, checkTranspose, checkTransRank1, checkTransRank2, isSquare, max, mult, mult, mult, mult, multAdd, multAdd, multAdd, norm, numColumns, numRows, rank1, rank1, rank1, rank1, rank1, rank2, rank2, rank2, scale, set, set, solve, solve, toString, transABmult, transABmult, transABmultAdd, transABmultAdd, transAmult, transAmult, transAmultAdd, transAmultAdd, transBmult, transBmult, transBmultAdd, transBmultAdd, transMult, transMult, transMultAdd, transpose, transpose, transRank1, transRank1, transRank2, transRank2, transSolve, transSolve

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

DistRowMatrix

public DistRowMatrix(int numRows,
                     int numColumns,
                     Communicator comm,
                     Matrix A,
                     Matrix B)

Constructor for DistRowMatrix

Parameters:

numRows - Global number of rows

numColumns - Global number of columns

comm - Communicator to use

A - Block diagonal matrix. The sum of the local row sizes of A must equal the global number, and likewise with the column sizes.

B - Off-diagonal matrix part. Its number of columns must equal the global number of columns, and its number of rows must equal that of A

Method Detail

add

public void add(int row,
                int column,
                double value)

Description copied from interface: Matrix

A(row,column) += value

Specified by:

add in interface Matrix

Overrides:

add in class AbstractMatrix

set

public void set(int row,
                int column,
                double value)

Description copied from interface: Matrix

A(row,column) = value

Specified by:

set in interface Matrix

Overrides:

set in class AbstractMatrix

get

public double get(int row,
                  int column)

Description copied from interface: Matrix

Returns A(row,column)

Specified by:

get in interface Matrix

Overrides:

get in class AbstractMatrix

copy

public DistRowMatrix copy()

Description copied from interface: Matrix

Creates a deep copy of the matrix

Specified by:

copy in interface Matrix

Overrides:

copy in class AbstractMatrix

Returns:

A

iterator

public java.util.Iterator<MatrixEntry> iterator()

Specified by:

iterator in interface java.lang.Iterable<MatrixEntry>

Overrides:

iterator in class AbstractMatrix

multAdd

public Vector multAdd(double alpha,
                      Vector x,
                      Vector y)

Description copied from interface: Matrix

y = alpha*A*x + y

Specified by:

multAdd in interface Matrix

Overrides:

multAdd in class AbstractMatrix

x - Vector of size A.numColumns()

y - Vector of size A.numRows()

Returns:

y

transMultAdd

public Vector transMultAdd(double alpha,
                           Vector x,
                           Vector y)

Description copied from interface: Matrix

y = alpha*AT*x + y

Specified by:

transMultAdd in interface Matrix

Overrides:

transMultAdd in class AbstractMatrix

x - Vector of size A.numRows()

y - Vector of size A.numColumns()

Returns:

y

local

public boolean local(int row,
                     int column)

Returns true if the insertion indices are local to this rank, and no communication is required afterwards. However, you still need to call flushAssembly to set up things like matrix/vector multiplication

norm1

protected double norm1()

Description copied from class: AbstractMatrix

Computes the 1 norm

Overrides:

norm1 in class AbstractMatrix

normInf

protected double normInf()

Description copied from class: AbstractMatrix

Computes the infinity norm

Overrides:

normInf in class AbstractMatrix

zero

public DistRowMatrix zero()

Description copied from interface: Matrix

Zeros all the entries in the matrix, while preserving any underlying structure. Useful for general, unstructured matrices.

Specified by:

zero in interface Matrix

Returns:

A

getRowOwnerships

public int[] getRowOwnerships()

Returns which rows are owned by which ranks. The current rank owns the rows n[comm.rank()] (inclusive) to n[comm.rank()+1] (exclusive)

getColumnOwnerships

public int[] getColumnOwnerships()

Returns which columns are owned by which ranks. The current rank owns the columns m[comm.rank()] (inclusive) to m[comm.rank()+1] (exclusive)

getBlock

public Matrix getBlock()

Returns the diagonal block matrix

getOff

public Matrix getOff()

Returns the off-diagonal matrix

max

protected double max()

Description copied from class: AbstractMatrix

Returns the largest absolute value

Overrides:

max in class AbstractMatrix

normF

protected double normF()

Description copied from class: AbstractMatrix

Computes the Frobenius norm. This implementation is overflow resistant

Overrides:

normF in class AbstractMatrix

rank1

public Matrix rank1(double alpha,
                    Vector x,
                    Vector y)

Description copied from interface: Matrix

A = alpha*x*yT + A. The matrix must be square, and the vectors of the same length

Specified by:

rank1 in interface Matrix

Overrides:

rank1 in class AbstractMatrix

Returns:

A

rank2

public Matrix rank2(double alpha,
                    Vector x,
                    Vector y)

Description copied from interface: Matrix

A = alpha*x*yT + alpha*y*xT + A. The matrix must be square, and the vectors of the same length

Specified by:

rank2 in interface Matrix

Overrides:

rank2 in class AbstractMatrix

Returns:

A

getCommunicator

public Communicator getCommunicator()

Gets the communicator associated with this matrix