cs_blas.h File Reference

#include "cs_defs.h"
#include "cs_base.h"

Include dependency graph for cs_blas.h:

Go to the source code of this file.

Enumerations
enum	cs_blas_reduce_t { CS_BLAS_REDUCE_SUPERBLOCK, CS_BLAS_REDUCE_KAHAN }
	Reduction algorithm type. More...

Functions
void	cs_blas_set_reduce_algorithm (cs_blas_reduce_t mode)
	Set the preferred BLAS reduction algorithm family. More...
void	cs_axpy (cs_lnum_t n, double a, const cs_real_t *x, cs_real_t *restrict y)
	Constant times a vector plus a vector: y <– ax + y. More...
double	cs_sum (cs_lnum_t n, const cs_real_t *x)
double	cs_weighted_sum (cs_lnum_t n, const cs_real_t *w, const cs_real_t *x)
double	cs_dot (cs_lnum_t n, const cs_real_t *x, const cs_real_t *y)
	Return the dot product of 2 vectors: x.y. More...
double	cs_dot_xx (cs_lnum_t n, const cs_real_t *x)
	Return dot products of a vector with itself: x.x. More...
double	cs_dot_wxx (cs_lnum_t n, const cs_real_t *w, const cs_real_t *x)
void	cs_dot_xx_xy (cs_lnum_t n, const cs_real_t *restrict x, const cs_real_t *restrict y, double *xx, double *xy)
	Return 2 dot products of 2 vectors: x.x, and x.y. More...
void	cs_dot_xy_yz (cs_lnum_t n, const cs_real_t *restrict x, const cs_real_t *restrict y, const cs_real_t *restrict z, double *xx, double *xy)
	Return 2 dot products of 3 vectors: x.y, and y.z. More...
void	cs_dot_xx_xy_yz (cs_lnum_t n, const cs_real_t *restrict x, const cs_real_t *restrict y, const cs_real_t *restrict z, double *xx, double *xy, double *yz)
	Return 3 dot products of 3 vectors: x.x, x.y, and y.z. More...
void	cs_dot_xx_yy_xy_xz_yz (cs_lnum_t n, const cs_real_t *restrict x, const cs_real_t *restrict y, const cs_real_t *restrict z, double *xx, double *yy, double *xy, double *xz, double *yz)
	Return 5 dot products of 3 vectors: x.x, y.y, x.y, x.z, and y.z. More...
double	cs_gdot (cs_lnum_t n, const cs_real_t *x, const cs_real_t *y)
	Return the global dot product of 2 vectors: x.y. More...
double	cs_gres (cs_lnum_t n, const cs_real_t *vol, const cs_real_t *x, const cs_real_t *y)
	Return the global residual of 2 extensive vectors: 1/sum(vol) . sum(X.Y/vol) More...

Enumeration Type Documentation

cs_blas_reduce_t

enum cs_blas_reduce_t

Reduction algorithm type.

Enumerator
CS_BLAS_REDUCE_SUPERBLOCK	Reduction based on l3superblock60 algorithm, described in [1]
CS_BLAS_REDUCE_KAHAN	Reduction based on Kahan's compensated summation, described in [4]

Function Documentation

cs_axpy()

void cs_axpy	(	cs_lnum_t	n,
		double	a,
		const cs_real_t *	x,
		cs_real_t *restrict	y
	)

Constant times a vector plus a vector: y <– ax + y.

Parameters

[in]	n	size of arrays x and y
[in]	a	multiplier for x
[in]	x	array of floating-point values
[in,out]	y	array of floating-point values

cs_blas_set_reduce_algorithm()

void cs_blas_set_reduce_algorithm

(

cs_blas_reduce_t

mode

)

Set the preferred BLAS reduction algorithm family.

This may not be enforced for all algorithms, though it should at least be enforced for the most general functions such as cs_dot.

Parameters

[in]

mode

BLAS mode to use

cs_dot()

double cs_dot	(	cs_lnum_t	n,
		const cs_real_t *	x,
		const cs_real_t *	y
	)

Return the dot product of 2 vectors: x.y.

Parameters

[in]	n	size of arrays x and y
[in]	x	array of floating-point values
[in]	y	array of floating-point values

Returns

dot product

cs_dot_wxx()

double cs_dot_wxx	(	cs_lnum_t	n,
		const cs_real_t *	w,
		const cs_real_t *	x
	)

cs_dot_xx()

double cs_dot_xx	(	cs_lnum_t	n,
		const cs_real_t *	x
	)

Return dot products of a vector with itself: x.x.

For better precision, a superblock algorithm is used.

Parameters

[in]	n	size of array x
[in]	x	array of floating-point values

Returns

dot product

cs_dot_xx_xy()

void cs_dot_xx_xy	(	cs_lnum_t	n,
		const cs_real_t *restrict	x,
		const cs_real_t *restrict	y,
		double *	xx,
		double *	xy
	)

Return 2 dot products of 2 vectors: x.x, and x.y.

The products could be computed separately, but computing them simultaneously adds more optimization opportunities and possibly better cache behavior.

Parameters

[in]	n	size of arrays x and y
[in]	x	array of floating-point values
[in]	y	array of floating-point values
[out]	xx	x.x dot product
[out]	xy	x.y dot product

cs_dot_xx_xy_yz()

void cs_dot_xx_xy_yz	(	cs_lnum_t	n,
		const cs_real_t *restrict	x,
		const cs_real_t *restrict	y,
		const cs_real_t *restrict	z,
		double *	xx,
		double *	xy,
		double *	yz
	)

Return 3 dot products of 3 vectors: x.x, x.y, and y.z.

The products could be computed separately, but computing them simultaneously adds more optimization opportunities and possibly better cache behavior.

Parameters

[in]	n	size of arrays x and y, and z
[in]	x	array of floating-point values
[in]	y	array of floating-point values
[in]	z	array of floating-point values
[out]	xx	x.x dot product
[out]	xy	x.y dot product
[out]	yz	y.z dot product

cs_dot_xx_yy_xy_xz_yz()

void cs_dot_xx_yy_xy_xz_yz	(	cs_lnum_t	n,
		const cs_real_t *restrict	x,
		const cs_real_t *restrict	y,
		const cs_real_t *restrict	z,
		double *	xx,
		double *	yy,
		double *	xy,
		double *	xz,
		double *	yz
	)

Return 5 dot products of 3 vectors: x.x, y.y, x.y, x.z, and y.z.

The products could be computed separately, but computing them simultaneously adds more optimization opportunities and possibly better cache behavior.

Parameters

[in]	n	size of arrays x and y, and z
[in]	x	array of floating-point values
[in]	y	array of floating-point values
[in]	z	array of floating-point values
[out]	xx	x.x dot product
[out]	yy	y.y dot product
[out]	xy	x.y dot product
[out]	xz	x.z dot product
[out]	yz	y.z dot product

cs_dot_xy_yz()

void cs_dot_xy_yz	(	cs_lnum_t	n,
		const cs_real_t *restrict	x,
		const cs_real_t *restrict	y,
		const cs_real_t *restrict	z,
		double *	xy,
		double *	yz
	)

Return 2 dot products of 3 vectors: x.y, and y.z.

The products could be computed separately, but computing them simultaneously adds more optimization opportunities and possibly better cache behavior.

Parameters

[in]	n	size of arrays x and y, and z
[in]	x	array of floating-point values
[in]	y	array of floating-point values
[in]	z	array of floating-point values
[out]	xy	x.y dot product
[out]	yz	y.z dot product

cs_gdot()

double cs_gdot	(	cs_lnum_t	n,
		const cs_real_t *	x,
		const cs_real_t *	y
	)

Return the global dot product of 2 vectors: x.y.

In parallel mode, the local results are summed on the default global communicator.

For better precision, a superblock algorithm is used.

Parameters

[in]	n	size of arrays x and y
[in]	x	array of floating-point values
[in]	y	array of floating-point values

Returns

dot product

cs_gres()

double cs_gres	(	cs_lnum_t	n,
		const cs_real_t *	vol,
		const cs_real_t *	x,
		const cs_real_t *	y
	)

Return the global residual of 2 extensive vectors: 1/sum(vol) . sum(X.Y/vol)

In parallel mode, the local results are summed on the default global communicator.

Parameters

[in]	n	size of arrays x and y
[in]	vol	array of floating-point values
[in]	x	array of floating-point values
[in]	y	array of floating-point values

Returns

global residual

cs_sum()

double cs_sum	(	cs_lnum_t	n,
		const cs_real_t *	x
	)

cs_weighted_sum()

double cs_weighted_sum	(	cs_lnum_t	n,
		const cs_real_t *	w,
		const cs_real_t *	x
	)