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

[Michael Niedermayer]

Hi

On Tue, Aug 21, 2007 at 05:02:18PM -0400, Marc Hoffman wrote:
> Hi,
> 
> On 8/19/07, Justin Ruggles <justinruggles at bellsouth.net> wrote:
> >
> >                      256    512   1024   2048    4096
> > -------------------------------------------------------
> > (1) fft32           46.1   101.0  205.0  494.4  1169.3
> > (2) fftr2            7.6    16.0   34.1  101.3   258.5
> > (3) fft ffmpeg       7.4    16.4   33.8   82.4   187.7
> > (4) fftr4/2          6.7    13.6   28.2   74.0   169.8
> >
> > (1) fixedpoint 32bit simple RAD2 fft
> > (2) simple rad2 first stage rad4
> > (3) current lavc RAD2, first 2 stages, and middle pulled
> > (4) rad4/2 with last stage rad2 if needed
> 
> I'm adding a split radix to the list.  Its a lot harder to optimize
> because of the extra pointer manipulation and control structures.  I
> did manage to unroll the W0 multipliers out but getting a clean
> iteration space has not happened yet.

could you also add the split radix fft from 
liba52-cvs/a52dec/liba52/imdct.c
as comparission

also djbfft and fftw2 would be interresting
IIRC they all use split radix ffts

and looking at them might help you with optimizing your code

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Opposition brings concord. Out of discord comes the fairest harmony.
-- Heraclitus
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20070822/4e5ee709/attachment.pgp>