A class which defines a sparse matrix distributed across 1D MPI grid. More...

#include <parMatrixSparse.hpp>

Public Member Functions
	parMatrixSparse (parVector< T, S > vec)
	A constructor of `parMatrixSparse`. More...

	parMatrixSparse (parVectorMap< S > map)
	A constructor of `parMatrixSparse`. More...

S	GetNRows ()
	Return the number of rows of local matrix on each MPI proc.

S	GetNCols ()
	Return the number of columns of local matrix on each MPI proc.

S	GetNNzLoc ()
	Return the number of non-zeros entries of local matrix on each MPI proc.

parVectorMap< S >	GetMap ()
	Return the parVectorMap object used to distribute sparse Matrix.

S	GetLowerBound ()
	Return parMatrixSparse::lower_b.

S	GetUpperBound ()
	Return parMatrixSparse::upper_b.

MPI_Comm	GetComm ()
	Return parMatrixSparse::comm.

int	GetProcId ()
	Return parMatrixSparse::ProcID.

int	GetNProcs ()
	Return parMatrixSparse::nProcs.

std::map< S, T > *	GetDynMatLoc ()
	Return parMatrixSparse::dynmat_loc.

void	SetValueLocal (S row, S col, T value)
	Set a value to a distributed sparse matrix with a given local index on each MPI proc. More...

void	SetValuesLocal (S nindex, S rows, S cols, T *values)
	Set multiple values to a distributed sparse matrix with given multiple local indices on each MPI proc. More...

void	SetValue (S row, S col, T value)
	Set a value to a distributed sparse matrix with a given global index on each MPI proc. More...

void	SetDiagonal (parVector< T, S > diag)
	Set the diagonal of a distributed sparse matrix with a given parVector object. More...

void	AddValueLocal (S row, S col, T value)
	Add a value to an entry of a distributed sparse matrix with a given local index on each MPI proc. More...

void	AddValue (S row, S col, T value)
	Add a value to an entry of a distributed sparse matrix with a given global index on each MPI proc. More...

T	GetValueLocal (S row, S col)
	Return a value of an entry of a distributed sparse matrix with a given local index on each MPI proc. More...

T	GetValue (S row, S col)
	Return a value to an entry of a distributed sparse matrix with a given global index on each MPI proc. More...

void	MatScale (T scale)
	Perform an operation `A = scale * A`, in which `scale` is a scalar. More...

void	MatAXPY (parMatrixSparse< T, S > X, T scale)
	Perform `A = A + scale * X`. More...

void	MatAYPX (parMatrixSparse< T, S > X, T scale)
	Perform `A = scale * A + X`. More...

void	initMat (S diag_l, S diag_u, Base< T > scale, T shift, Base< T > sparsity)
	filling the lower part of matrix between diagonal of offset `diag_l` and diagonal of offset `diag_u` with random values `scale * rnd + shift` where `rnd` is a randomly generated value between `0` and `1` More...

void	initMat (S diag_l, S diag_u)
	filling the lower part of matrix between diagonal of offset `diag_l` and diagonal of offset `diag_u` with random values `rnd`, which is a randomly generated value between `0` and `1` More...

void	setSpecNonHerm (parVector< T, S > spectrum)
	Set a given spectrum onto a parMatrixSparse object for constructing a non-Hermitian matrix. More...

void	setSpecNonSymm (parVector< T, S > spectrum)
	Set a given spectrum (all eigenvalues are in real scalars) onto a parMatrixSparse object for constructing a non-Symmetric matrix. More...

void	setSpecNonSymmCmplx (parVector< std::complex< Base< T >>, S > spectrum)
	Set a given spectrum (all eigenvalues can be in real scalars or appear as pairs of conjugate complex scalars) onto a parMatrixSparse object for constructing a non-Symmetric matrix. More...

void	updateNnz ()
	Explicitly re-compute the number of nnz on each MPI proc. More...

void	copy (parMatrixSparse< T, S > X)
	Duplicate from another parMatrixSparse object. More...

void	rmZeros ()
	Explicitly remove all the zeros of parMatrixSparse object. More...

parMatrixSparse< T, S >	MA (Nilpotent< S > nilp)
	Perform `M*A`, in which `A` is a nilpotent matrix. More...

parMatrixSparse< T, S >	AM (Nilpotent< S > nilp)
	Perform `A*M`, in which `A` is a nilpotent matrix. More...

void	ZeroEntries ()
	Make all the entries a sparse matrix to be zeros, but keep the sparsity structure as it is.

MatrixCSR< T, S >	ConvertToCSR ()
	Convert a parMatrixSparse with dynamic memory into a distributed CSR matrix. More...

void	show ()
	Display multiple information of a parMatrixSparse object. More...

void	MatView ()
	Print a parMatrixSparse object in a distributed COO format. More...

void	MatView (std::string matName)
	Print a parMatrixSparse object in a distributed COO format with matrix name indicated at the starting point of each line. More...

void	writeToMatrixMarket (std::string file_name)
	A parallel IO to write a parMatrixSparse object into a file of MatrixMarket format. More...

void	writeToMatrixMarketCmplx (std::string file_name)
	A parallel IO to write a parMatrixSparse object with complex scalar into a file of MatrixMarket format. More...

Private Attributes
std::map< S, T > *	dynmat_loc
	An array of `std::map` in which each map stores the column index and related non-zero entry value of each row.

S	ncols
	number of columns of local matrix on each MPI proc

S	nrows
	number of rows of local matrix on each MPI proc

S	nnz_loc
	number of non-zero entries of local matrix on each MPI proc

parVectorMap< S >	index_map
	the parVectorMap object used to distribute the global matrix across 1D MPI grid

S	lower_b
	the smallest index of row of a distributed matrix on each MPI proc

S	upper_b
	upper_b-1 = the largest index of row of a distributed matrix on each MPI proc

MPI_Comm	comm
	the working MPI communicator

int	ProcID
	rank of each MPI procs within the working MPI communicator

int	nProcs
	number of MPI procs within the working MPI communicator

Detailed Description

template<typename T, typename S>
class parMatrixSparse< T, S >

A class which defines a sparse matrix distributed across 1D MPI grid.

This class can be constructed with a given parVector with the same distribution scheme.
This class can be constructed with a distribution scheme by a given parVectorMap object.

Template Parameters

T	describes the scalar types of the entries of a sparse matrix.
S	type of integer to describes the dimension of vector to be generated.

Constructor & Destructor Documentation

◆ parMatrixSparse() [1/2]

template<typename T , typename S >

parMatrixSparse< T, S >::parMatrixSparse ( parVector< T, S > vec )

A constructor of `parMatrixSparse`.

Parameters

[in] vec a given parVector, a parMatrixSparse object is constructed with the same distribution scheme of `vec`

parMatrixSparse::dynmat_loc is not allocated.

◆ parMatrixSparse() [2/2]

template<typename T , typename S >

parMatrixSparse< T, S >::parMatrixSparse ( parVectorMap< S > map )

A constructor of `parMatrixSparse`.

Parameters

[in] map the distribution scheme determined by this object of type parVectorMap

parMatrixSparse::dynmat_loc is not allocated.

Member Function Documentation

◆ AddValue()

template<typename T , typename S >

void parMatrixSparse< T, S >::AddValue	(	S	row,
		S	col,
		T	value
	)

Add a value to an entry of a distributed sparse matrix with a given global index on each MPI proc.

Parameters

[in]	row	the global index of row
[in]	col	the global index of column
[in]	value	the scalar to be added

◆ AddValueLocal()

template<typename T , typename S >

void parMatrixSparse< T, S >::AddValueLocal	(	S	row,
		S	col,
		T	value
	)

Add a value to an entry of a distributed sparse matrix with a given local index on each MPI proc.

Parameters

[in]	row	the local index of row
[in]	col	the local index of column
[in]	value	the scalar to be added

◆ AM()

template<typename T , typename S >

parMatrixSparse< T, S > parMatrixSparse< T, S >::AM ( Nilpotent< S > nilp )

Perform `A*M`, in which `A` is a nilpotent matrix.

Parameters

[in] nilp a Nilpotent ~~object which determines a nilpotent matrix~~

◆ ConvertToCSR()

template<typename T , typename S >

MatrixCSR< T, S > parMatrixSparse< T, S >::ConvertToCSR

Convert a parMatrixSparse with dynamic memory into a distributed CSR matrix.

Attention, after the converting, any operations on the matrix can be implicity updated on the CSR matrix

◆ copy()

template<typename T , typename S >

void parMatrixSparse< T, S >::copy ( parMatrixSparse< T, S > X )

Duplicate from another parMatrixSparse object.

Parameters

[in] X another parMatrixSparse object to be duplicated with a same distribution scheme

◆ GetValue()

template<typename T , typename S >

T parMatrixSparse< T, S >::GetValue	(	S	row,
		S	col
	)

Return a value to an entry of a distributed sparse matrix with a given global index on each MPI proc.

Parameters

[in]	row	the global index of row
[in]	col	the global index of column

◆ GetValueLocal()

template<typename T , typename S >

T parMatrixSparse< T, S >::GetValueLocal	(	S	row,
		S	col
	)

Return a value of an entry of a distributed sparse matrix with a given local index on each MPI proc.

Parameters

[in]	row	the local index of row
[in]	col	the local index of column

◆ initMat() [1/2]

template<typename T , typename S >

void parMatrixSparse< T, S >::initMat	(	S	diag_l,
		S	diag_u
	)

filling the lower part of matrix between diagonal of offset `diag_l` and diagonal of offset `diag_u` with random values `rnd`, which is a randomly generated value between `0` and `1`

Parameters

[in]	diag_l	the offset of lower diagonal
[in]	diag_u	the offset of lower diagonal

◆ initMat() [2/2]

template<typename T , typename S >

void parMatrixSparse< T, S >::initMat	(	S	diag_l,
		S	diag_u,
		Base< T >	scale,
		T	shift,
		Base< T >	sparsity
	)

filling the lower part of matrix between diagonal of offset `diag_l` and diagonal of offset `diag_u` with random values `scale * rnd + shift` where `rnd` is a randomly generated value between `0` and `1`

Parameters

[in]	diag_l	the offset of lower diagonal
[in]	diag_u	the offset of lower diagonal
[in]	scale	a scalar to be multiplied on the randomly generated value
[in]	shift	a scalar to be added on the randomly generated value
[in]	sparsity	the probability that a entry in the range set to be zero

◆ MA()

template<typename T , typename S >

parMatrixSparse< T, S > parMatrixSparse< T, S >::MA ( Nilpotent< S > nilp )

Perform `M*A`, in which `A` is a nilpotent matrix.

Parameters

[in] nilp a Nilpotent ~~object which determines a nilpotent matrix~~

◆ MatAXPY()

template<typename T , typename S >

void parMatrixSparse< T, S >::MatAXPY	(	parMatrixSparse< T, S >	X,
		T	scale
	)

Perform `A = A + scale * X`.

Parameters

[in]	X	another parMatrixSparse object with a same distribution scheme
[in]	scale	the scalar to be multiplied on `X`

◆ MatAYPX()

template<typename T , typename S >

void parMatrixSparse< T, S >::MatAYPX	(	parMatrixSparse< T, S >	X,
		T	scale
	)

Perform `A = scale * A + X`.

Parameters

[in]	X	another parMatrixSparse object with a same distribution scheme
[in]	scale	the scalar to be multiplied on `A`

◆ MatScale()

template<typename T , typename S >

void parMatrixSparse< T, S >::MatScale ( T scale )

Perform an operation `A = scale * A`, in which `scale` is a scalar.

Parameters

[in] scale the scalar to be multiplied on the distributed sparse matrix

◆ MatView() [1/2]

template<typename T , typename S >

void parMatrixSparse< T, S >::MatView

Print a parMatrixSparse object in a distributed COO format.

This is a distributed function that each MPI proc can only display the piece of local matrix on itself.

◆ MatView() [2/2]

template<typename T , typename S >

void parMatrixSparse< T, S >::MatView ( std::string matName )

Print a parMatrixSparse object in a distributed COO format with matrix name indicated at the starting point of each line.

This is a distributed function that each MPI proc can only display the piece of local matrix on itself.

◆ rmZeros()

template<typename T , typename S >

void parMatrixSparse< T, S >::rmZeros

Explicitly remove all the zeros of parMatrixSparse object.

This member function is introduced in case any matrix operation would introduce some (explicit) zeros.

◆ SetDiagonal()

template<typename T , typename S >

void parMatrixSparse< T, S >::SetDiagonal ( parVector< T, S > diag )

Set the diagonal of a distributed sparse matrix with a given parVector object.

Parameters

[in] diag a given parVector object to be set on the diagonal

◆ setSpecNonHerm()

template<typename T , typename S >

void parMatrixSparse< T, S >::setSpecNonHerm ( parVector< T, S > spectrum )

Set a given spectrum onto a parMatrixSparse object for constructing a non-Hermitian matrix.

Parameters

[in] spectrum a parVector object which stores the given spectrum in complex scalars

This operation is naturally in parallel since the spectrum is stored in a parVector object
Attention, this member function works only with complex scalar

◆ setSpecNonSymm()

template<typename T , typename S >

void parMatrixSparse< T, S >::setSpecNonSymm ( parVector< T, S > spectrum )

Set a given spectrum (all eigenvalues are in real scalars) onto a parMatrixSparse object for constructing a non-Symmetric matrix.

Parameters

[in] spectrum a parVector object which stores the given spectrum in real scalars

This operation is naturally in parallel since the spectrum is stored in a parVector object
Attention, this member function works only with real scalar

◆ setSpecNonSymmCmplx()

template<typename T , typename S >

void parMatrixSparse< T, S >::setSpecNonSymmCmplx ( parVector< std::complex< Base< T >>, S > spectrum )

Set a given spectrum (all eigenvalues can be in real scalars or appear as pairs of conjugate complex scalars) onto a parMatrixSparse object for constructing a non-Symmetric matrix.

Parameters

[in] spectrum a parVector object which stores the given spectrum with conjugate eigenvalues

This operation is naturally in parallel since the spectrum is stored in a parVector object
Attention, for each conjugate pairs of eigenvalues, they should be placed one after another, they cannot be placed in a random order. Before setting the spectrum to the matrix, the spectrum will be checked if it satisfies this condition

◆ SetValue()

template<typename T , typename S >

void parMatrixSparse< T, S >::SetValue	(	S	row,
		S	col,
		T	value
	)

Set a value to a distributed sparse matrix with a given global index on each MPI proc.

Parameters

[in]	row	the global index of row
[in]	col	the global index of column
[in]	value	the scalar to be set

◆ SetValueLocal()

template<typename T , typename S >

void parMatrixSparse< T, S >::SetValueLocal	(	S	row,
		S	col,
		T	value
	)

Set a value to a distributed sparse matrix with a given local index on each MPI proc.

Parameters

[in]	row	the local index of row
[in]	col	the local index of column
[in]	value	the scalar to be set

◆ SetValuesLocal()

template<typename T , typename S >

void parMatrixSparse< T, S >::SetValuesLocal	(	S	nindex,
		S *	rows,
		S *	cols,
		T *	values
	)

Set multiple values to a distributed sparse matrix with given multiple local indices on each MPI proc.

Parameters

[in]	nindex	number of values to be set
[in]	rows	an array storing multiple local indices of row
[in]	cols	an array storing multiple local indices of column
[in]	values	an array storing multiple scalars to be set

◆ show()

template<typename T , typename S >

void parMatrixSparse< T, S >::show

Display multiple information of a parMatrixSparse object.

number of rows of local matrix
number of columns of local matrix
number of nnz of local matrix
parMatrixSparse::lower_b
parMatrixSparse::upper_b

◆ updateNnz()

template<typename T , typename S >

void parMatrixSparse< T, S >::updateNnz

Explicitly re-compute the number of nnz on each MPI proc.

This member function is introduced in case any matrix operation would failed to update the number of nnz

◆ writeToMatrixMarket()

template<typename T , typename S >

void parMatrixSparse< T, S >::writeToMatrixMarket ( std::string file_name )

A parallel IO to write a parMatrixSparse object into a file of MatrixMarket format.

Parameters

[in] file_name the path and file name to write into

Attention, this method works only of sparse matrix with real scalar (`double`, `float`...)

◆ writeToMatrixMarketCmplx()

template<typename T , typename S >

void parMatrixSparse< T, S >::writeToMatrixMarketCmplx ( std::string file_name )

A parallel IO to write a parMatrixSparse object with complex scalar into a file of MatrixMarket format.

Parameters

[in] file_name the path and file name to write into

Attention, this method works only of sparse matrix with scalar scalar (std::complex<double>, std::complex<float>...)

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

include/parMatrix/parMatrixSparse.hpp

Public Member Functions

Private Attributes

Detailed Description

template<typename T, typename S> class parMatrixSparse< T, S >

Constructor & Destructor Documentation

◆ parMatrixSparse() [1/2]

◆ parMatrixSparse() [2/2]

Member Function Documentation

◆ AddValue()

◆ AddValueLocal()

◆ AM()

◆ ConvertToCSR()

◆ copy()

◆ GetValue()

◆ GetValueLocal()

◆ initMat() [1/2]

◆ initMat() [2/2]

◆ MA()

◆ MatAXPY()

◆ MatAYPX()

◆ MatScale()

◆ MatView() [1/2]

◆ MatView() [2/2]

◆ rmZeros()

◆ SetDiagonal()

◆ setSpecNonHerm()

◆ setSpecNonSymm()

◆ setSpecNonSymmCmplx()

◆ SetValue()

◆ SetValueLocal()

◆ SetValuesLocal()

◆ show()

◆ updateNnz()

◆ writeToMatrixMarket()

◆ writeToMatrixMarketCmplx()

template<typename T, typename S>
class parMatrixSparse< T, S >