[FFmpeg-devel] Fwd: Fixpoint FFT optimization, with MDCT and IMDCT wrappers for audio optimization

[Mike Giacomelli]

> well first part you reduce the number of multiplications split radix would do that
> but i guess you can make x operations faster by low level optimizations
> than half of the same operations to which the same low level optimizations
> could be applied
>

Most targets without fpus also have very slow integer multiplications
too.  On ARM for instance, the 32x32->64 multiplies needed to do fix
point ops can be 4 times slower or more then a simple add as I recall.

What I'm getting at is that the fastest algorithm on an x86 or other
modern desktop CPU may not actually be the fastest on most or even any
of the targets where fixed point is desirable (excepting the handful
of formats which use fix point by design).  I'd like to pick a couple
fast FFTs, and then actually test them on arm or coldfire systems and
see which works best on a system with slow multiplies.

Once you've gotten a few well written FFTs done, I can test them on my
Ipod or Sansa and see which works better (though unfortunately I'm not
set up to do timing measurements, but I can give you decode times for
a WMA file using different FFTs).