[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?
>   
I don't think it will have much impact, because to use SSE efficiently, 
you need some constraints wrt memory allocation which cannot be met 
easily now in numpy arrays, AND good compiler support (intel compiler, 
basically) for automatic vectorization. Even then, FFT may have specific 
patterns which mean that only hand tuned routines can get most of the 
CPU horsepower: both mkl and fftw use SIMD instructions to get their 
maximum efficiency.

David