[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-multimedia
Subject:    Re: FFT (the djbfft)
From:       Hans Meine <hans_meine () gmx ! net>
Date:       2001-03-05 9:40:03
[Download RAW message or body]

Charles Samuels <charles@kde.org> writes:

> > - same speed (djbfft claims to be faster, though the fftw people
> >  are working on importing the needed loc to gain the last missing
> >  percent)
> The thing I'm really worrying about on the fftw is how it has to "learn".  
> How long does this take?  They say several seconds.  That's too long.
Not noticeable. I don't know why and where they say so.

In fftw, you got three options when you "create a plan" (prepare an fft):
- let fftw measure the speed of different algorithms to use the fastest
  for this specific application/use/size of data and so on (this takes
  "some time", but not in the magnitude of seconds!)
- let fftw guess the fastest method (takes virtually no time, but might
  not choose the fastest method - should be the fastest option if you
  only want to do _one_ fft)
- use "wisdom" ;-), that is - load the above information/plan from a file
  combines both advantages but needs more programming

-- 
Ciao,  /  /
      /--/
     /  / ANS                          .,* Hamburg, Germany *,.
_______________________________________________
Kde-multimedia mailing list
Kde-multimedia@master.kde.org
http://master.kde.org/mailman/listinfo/kde-multimedia

[prev in list] [next in list] [prev in thread] [next in thread]