Re: Implementation of convn

[Søren Hauberg]

tor, 27 03 2008 kl. 09:52 +0100, skrev Fredrik Lingvall:
> I'm the maintainer (and co-author) of an acoustic simulation toolbox 
> that contains two convolution functions  which may be of interest. These 
> functions (conv_p and fftconv_p) are aimed at large problems and have 
> thread support. The (GPLed) toolbox can be found here: 
> http://www.signal.uu.se/Toolbox/dream/

I had a quick look at the web page, but I couldn't find a more detailed
description of the functions you mention. Do they handle N-dimensional
data, or are they limited to 1D and 2D? The implementation I made of
'convn' isn't particular wonderful. It's quite simple, but it could be
faster (don't use increment_idx). If you have fast/robust
implementations of ND convolution, perhaps you could suggest a patch?

Søren