[SciPy-User] Single precision FFT insufficiently accurate.

Sturla Molden sturla@molden...
Mon Jun 28 08:58:46 CDT 2010


Den 28.06.2010 14:37, skrev Ralf Gommers:
> Anne's argument sounded convincing, so I think the way it is now in 
> the 0.8.x branch is fine. Probably good to add a warning in the 
> docstrings like:
>
>     .. note:: In scipy 0.8.0 `fft` in single precision is available, 
> but *only*
>         for input array sizes which can be factorized into 
> (combinations of) 2,
>         3 and 5. For other sizes the computation will be done in double
>         precision.

This is actually quite diagnostic. 2, 3, and 5 are the only primes 
FFTPACK support. Larger primes has an O(N**2) fallback.

The source of the (rounding?) error must therefore be in the O(N**2) 
fallback code.

Sturla


More information about the SciPy-User mailing list