Opm::MixedMatrixWrapper< Vector > Class Template Reference

Wraps c-implementation of mixed-precision matrix. More...

#include <MatrixWrapper.hpp>

Public Member Functions
	MixedMatrixWrapper (int nrows, int nnz)
	constructor More...
	~MixedMatrixWrapper ()
	destructor More...
void	update (double const *data)
	update matrix entries More...
void	mv (const Vector &x, Vector &y) const
	block-sparse matrix-vector multiplication (y = M.x) More...
void	umv (const Vector &x, Vector &y) const
	block-sparse matrix-vector multiplication with update (y += M.x) More...
void	usmv (double alpha, const Vector &x, Vector &y) const
	block-sparse matrix-vector multiplication with scaled update (y += alpha * M.x) More...
int *	rowptr ()
	access row offset pointer More...
int *	colidx ()
	access column index pointer More...
int	nrows () const

Static Public Attributes
static constexpr auto	block_size = Vector::block_type::dimension

Detailed Description

template<class Vector>
class Opm::MixedMatrixWrapper< Vector >

Wraps c-implementation of mixed-precision matrix.

Note

matrix is stored in single precision, but all operations are performed in double-precision

Template Parameters

Vector

the block-vector used by linear operator

Constructor & Destructor Documentation

MixedMatrixWrapper()

template<class Vector >

Opm::MixedMatrixWrapper< Vector >::MixedMatrixWrapper ( int  nrows, int  nnz  )

inline

constructor

Parameters

nrows	number of block rows
nnz	number of nonzero blocks

References Opm::MixedMatrixWrapper< Vector >::block_size, bsr_alloc(), bsr_init(), and Opm::MixedMatrixWrapper< Vector >::nrows().

~MixedMatrixWrapper()

template<class Vector >

Opm::MixedMatrixWrapper< Vector >::~MixedMatrixWrapper ( )

inline

destructor

References bsr_free().

Member Function Documentation

colidx()

template<class Vector >

int * Opm::MixedMatrixWrapper< Vector >::colidx ( )

inline

access column index pointer

References bsr_matrix::colidx.

mv()

template<class Vector >

void Opm::MixedMatrixWrapper< Vector >::mv	(	const Vector &	x,
		Vector &	y
	)		const

block-sparse matrix-vector multiplication (y = M.x)

mixed-precision block-sparse matrix-vector multiplication (y = M.x)

Parameters

x	input vector
y	output vector

Note

hand-optimized versions are provided for block-sizes 2,3, and 4. A generic implementation is provided for block- sizes > 4

Parameters

x	input vector
y	output vector

References bsr_vmspmv2(), bsr_vmspmv3(), and bsr_vmspmv4().

nrows()

template<class Vector >

int Opm::MixedMatrixWrapper< Vector >::nrows ( ) const

inline

References bsr_matrix::nrows.

Referenced by Opm::MixedMatrixWrapper< Vector >::MixedMatrixWrapper().

rowptr()

template<class Vector >

int * Opm::MixedMatrixWrapper< Vector >::rowptr ( )

inline

access row offset pointer

References bsr_matrix::rowptr.

umv()

template<class Vector >

void Opm::MixedMatrixWrapper< Vector >::umv	(	const Vector &	x,
		Vector &	y
	)		const

block-sparse matrix-vector multiplication with update (y += M.x)

mixed-precision block-sparse matrix-vector multiplication with update (y += M.x)

Parameters

x	input vector
y	output vector

Note

hand-optimized versions are provided for block-sizes 2,3, and 4. A generic implementation for block-sizes > 4 is NOT provided

Parameters

x	input vector
y	output vector

References bsr_vmspumv2(), bsr_vmspumv3(), and bsr_vmspumv4().

update()

template<class Vector >

void Opm::MixedMatrixWrapper< Vector >::update

(

double const *

data

)

update matrix entries

Note

downcasts from double precision and transposes each non-zero block entry

Parameters

data	pointer to double precision data

usmv()

template<class Vector >

void Opm::MixedMatrixWrapper< Vector >::usmv	(	double	alpha,
		const Vector &	x,
		Vector &	y
	)		const

block-sparse matrix-vector multiplication with scaled update (y += alpha * M.x)

mixed-precision block-sparse matrix-vector multiplication with scaled update (y += alpha * M.x)

Parameters

alpha	scaling factor
x	input vector
y	output vector

Note

hand-optimized versions are provided for block-sizes 2,3, and 4. A generic implementation is provided for block- sizes > 4

Parameters

alpha	scaling factor
x	input vector
y	output vector

References bsr_vmspumv2(), bsr_vmspumv3(), and bsr_vmspumv4().

Member Data Documentation

block_size

template<class Vector >

constexpr auto Opm::MixedMatrixWrapper< Vector >::block_size = Vector::block_type::dimension

staticconstexpr

Referenced by Opm::MixedMatrixWrapper< Vector >::MixedMatrixWrapper().

---

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

• MatrixWrapper.hpp