Re: [Sc-devel] [commit?] StereoConvolution2L

[Florian Schmidt <mista.tapas@xxxxxxx>]

On Wednesday 05 December 2007, nescivi wrote:

> > Hmm, doesn't BruteFIR just do partitioned convolution in the frequency
> > domain?
>
> no, it does something clever with reordering the coefficients, so you have
> less memory accesses within one period. Memory access is the bottle neck in
> convolution..

http://www.ludd.luth.se/~torger/brutefir.html#bruteconv_4

Ok, makes sense :) I used libDSP once for the complex-multiply-and-add [it had 
support for SIMD loops of different kinds an for different SIMD instruction 
sets). It was quite an improvement, too, over my naive C implementation.

Regards,
Flo

-- 
Palimm Palimm!
http://tapas.affenbande.org