[Numpy-discussion] fourier with single precision

[David Cournapeau]

Lars Friedrich wrote:
> Hello,
>
> thanks for your comments. If I got you right, I should look for a 
> FFT-code that uses SSE (what does this actually stand for?), which means 
> that it vectorizes 32bit-single-operations into larger chunks that make 
> efficient use of recent CPUs.
>
> You mentioned FFTW and MKL. Is this www.fftw.org and the 'intel math 
> kernel library'? If I would like to use one of them, is numpy the right 
> place to put it in?
>
> Does anyone know, if it is possible to switch on SSE support (at compile 
> time) in the fftpack.c that numpy uses?
>
MKL is from Intel (free as in beer on Linux and for academic purpose I 
think, but of course, you should check whether this applies to you). 
FFTW is GPL, and AFAIK is considered to be the fastest general purpose 
open source FFT.

Here are your options as far as I understand:

    - if you care about speed (that is, faster than numpy), then use 
scipy.fftpack with fftw3: there are wrappers in scipy for it. There is 
no float support (yet), but it is planned. Even with double, it will be 
faster (how much is really platform dependent). There is also MKL 
support, which may be faster (never used it).
    - if you care also about memory, then maybe you will have no choice 
but using your own routines for float support. FFTW support both single 
and double precision, but only double is available in scipy.

    cheers,

    David
> Thanks
>
> Lars
> _______________________________________________
> Numpy-discussion mailing list
> Numpy-discussion at scipy.org
> http://projects.scipy.org/mailman/listinfo/numpy-discussion
>
>