strategy_gaussian_conv.h File Reference

Suite of different 1D Gaussian convolution methods. More...

Detailed Description

Suite of different 1D Gaussian convolution methods.

Author

Pascal Getreuer getre.nosp@m.uer@.nosp@m.cmla..nosp@m.ens-.nosp@m.cacha.nosp@m.n.fr

Copyright (c) 2013, Pascal Getreuer All rights reserved.

This program is free software: you can redistribute it and/or modify it under, at your option, the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version, or the terms of the simplified BSD license.

You should have received a copy of these licenses along with this program. If not, see http://www.gnu.org/licenses/ and http://www.opensource.org/licenses/bsd-license.html.

Definition in file strategy_gaussian_conv.h.

#include "num.h"

Go to the source code of this file.

Typedefs
typedef struct gconv_	gconv
	Plan to perform a 1D Gaussian convolution for any algorithm. More...

Functions
gconv *	gconv_plan (num *dest, const num *src, long N, long stride, const char *algo, double sigma, int K, num tol)
	Plan a 1D Gaussian convolution. More...
void	gconv_execute (gconv *g)
	Execute a 1D Gaussian convolution. More...
void	gconv_free (gconv *g)
	Free memory associated with a gconv. More...

Typedef Documentation

typedef struct gconv_ gconv

Plan to perform a 1D Gaussian convolution for any algorithm.

The gconv* object represents a generic plan to perform a 1D Gaussian convolution using any of the implemented algorithms. It is an application of the strategy design pattern to allow the different Gaussian convolution algorithms to be easily interchangeable in the benchmarking code.

Use is similar to FFTW's fftw_plan. A gconv* object is first created by gconv_plan(), then the 1D Gaussian convolution is performed by gconv_execute(), and finally gconv_free() cleans up. The same Gaussian convolution may be executed multiple times by multiple calls to gconv_execute(). This interface allows to separate precomputation and actual execution in timing tests.

Most of the Gaussian convolution algorithm implementations follow this structure of plan, execute, free, but there are variations in the number of algorithm parameters and whether additional workspace memory is required. The implementation of gconv* is simply a set of wrapper functions to unify all algorithms to the same interface.

Example

gconv *g = gconv_plan(dest, src, N, stride,
                      algorithm_name, sigma, K, tol);
gconv_execute(g);
gconv_free(g);

Definition at line 54 of file strategy_gaussian_conv.h.

Function Documentation

void gconv_execute

(

gconv *

g

)

Execute a 1D Gaussian convolution.

Parameters

g	gconv* created by gconv_plan()

Definition at line 125 of file strategy_gaussian_conv.c.

void gconv_free

(

gconv *

g

)

Free memory associated with a gconv.

Parameters

g	gconv* created by gconv_plan()

Definition at line 134 of file strategy_gaussian_conv.c.

gconv* gconv_plan	(	num *	dest,
		const num *	src,
		long	N,
		long	stride,
		const char *	algo,
		double	sigma,
		int	K,
		num	tol
	)

Plan a 1D Gaussian convolution.

Parameters

dest	destination array
src	source array
N	number of samples
stride	stride between successive samples
algo	Gaussian convolution algorithm
sigma	Gaussian standard deviation
K	algorithm steps or filter order parameter
tol	algorithm accuracy parameter

Returns

gconv pointer or NULL on failure

Definition at line 72 of file strategy_gaussian_conv.c.