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

[Marc Hoffman]

On 7/28/07, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Fri, Jul 27, 2007 at 09:38:53PM -0400, Marc Hoffman wrote:
> > On 7/27/07, Michael Niedermayer <michaelni at gmx.at> wrote:
> > > Hi
> > >
> > > On Fri, Jul 27, 2007 at 05:40:14PM -0400, mmh wrote:
> > > [...]
> >
> ff_fft_init() should call the fixed point init
> also
> the mmx code in it should be split out into a i386/whatever.c
> and allso called from ff_fft_init()
>
> i think that would solve it?
>

So we should do something like this to create the tables?  We should
probably allow something like the arch dependent dsp init routine to
override the mmx stuff.

yoda:~/r2/ffmpeg/libavcodec mmh$ svn diff fft.c
Index: fft.c
===================================================================
--- fft.c       (revision 9807)
+++ fft.c       (working copy)
@@ -59,6 +59,8 @@
     s->imdct_calc = ff_imdct_calc;
     s->exptab1 = NULL;

+    ff_fixedpoint_fft_init (s, nbits, inverse);
+
     /* compute constant table for HAVE_SSE version */
 #if defined(HAVE_MMX) \
     || (defined(HAVE_ALTIVEC) && !defined(ALTIVEC_USE_REFERENCE_C_CODE))
yoda:~/r2/ffmpeg/libavcodec mmh$