[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Speed concerns with fft
From: |
David Bateman |
Subject: |
Re: Speed concerns with fft |
Date: |
Fri, 8 Oct 2004 15:52:45 +0200 |
User-agent: |
Mutt/1.4.1i |
The speed of FFT is closely related to the largest prime factors of the
size of the fft. For example
octave:25> a = rand(1024,100); tic; b = fft(a); toc
ans = 0.024215
octave:26> a = rand(1024,100); tic; b = fft(a); toc
ans = 0.025095
octave:27> a = rand(1024,100); tic; b = fft(a); toc
ans = 0.024253
octave:28> a = rand(1021,100); tic; b = fft(a); toc
ans = 0.10520
octave:29> a = rand(1021,100); tic; b = fft(a); toc
ans = 0.10243
octave:30> a = rand(1021,100); tic; b = fft(a); toc
ans = 0.10138
Since 1024 is 2^10 it is faster to calculate than the fft of 1021 which
is prime.
If you have to do the same calculation main times, you might want to
look at the command fftw_wisdom, and the command line program fftw-wisdom,
which can be used to save to the system-wide FFTW wisdom file in
/etc/fftw/wisdom, to make things even a bit faster..
As for the fink version, I'm not sure what libraries it links to, but I
hope it links to FFTW v3.
Regards
David
According to Sébastien Maerten <address@hidden> (on 10/08/04):
> Hello,
>
> I need to perform fft on "large" sets of data, for that I've used
> octave binary installed by fink on macosx 10.3.5 on an ibook G4 1GHz (
> 256 Mo RAM). It turns out that computing fft( my_data ) takes about 15
> sec when my_data is a 500K floats vector (timed by tic() ; fft () ;
> toc() ). With a 1M floats vector, the time jumps to 6000 sec !
> Do anybody know if this is the intended behavior ? ( I don't know a
> word about fft algorythms )
>
> I'm sorry I can't provide more info about octave's build options than
> it's current fink version.
>
> Do anybody know how I can get a "faster" result ? Building octave with
> proper math libraries ? Which one ? It seems to me that building octave
> from source on osx is not trivial (I've failed with fink, darwinports
> and manually with versions 2.1.57 , 2.1.59 and 2.1.60), btw, is it
> worth reporting such build failures ?
>
> I'm in trouble because I *need* those fft's but I can't give the
> computer around 100 x 6000 sec just for tha :( .
>
> Any idea / pointer / help would be greatly apreciated. Thanks.
>
>
>
> -------------------------------------------------------------
> Octave is freely available under the terms of the GNU GPL.
>
> Octave's home on the web: http://www.octave.org
> How to fund new projects: http://www.octave.org/funding.html
> Subscription information: http://www.octave.org/archive.html
> -------------------------------------------------------------
--
David Bateman address@hidden
Motorola CRM +33 1 69 35 48 04 (Ph)
Parc Les Algorithmes, Commune de St Aubin +33 1 69 35 77 01 (Fax)
91193 Gif-Sur-Yvette FRANCE
The information contained in this communication has been classified as:
[x] General Business Information
[ ] Motorola Internal Use Only
[ ] Motorola Confidential Proprietary
-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.
Octave's home on the web: http://www.octave.org
How to fund new projects: http://www.octave.org/funding.html
Subscription information: http://www.octave.org/archive.html
-------------------------------------------------------------