[SciPy-dev] Renaming fftpack to fft, removing backends

David Cournapeau david@ar.media.kyoto-u.ac...
Wed Oct 29 01:59:47 CDT 2008


Charles R Harris wrote:
>
> +1. The fftpack is a pretty good (if old) implementation whose main
> failing is not taking maximum advantage of cache. It does need some
> supplementation, a Bluestein implementation for instance, but that is
> easier to do with a single target.

Bluestein would be useful even for numpy. The cache thing is handled
better by fftw2 and fftw3, but the advantage is largely outweight by the
fact we need to copy the input in a temporary buffer to guarantee SIMD
alignement. In a lot of signal processing tasks, float32 FFT is enough,
and that would give pretty good  speedup, too.

cheers,

David


More information about the Scipy-dev mailing list