[Ffmpeg-devel] [PATCH] Build failes on i386 if snow is disabled

[Diego Biurrun]

On Sun, Feb 25, 2007 at 10:25:45PM +0200, Martin Storsj? wrote:
> On Thu, 22 Feb 2007, Diego Biurrun wrote:
> 
> >On Thu, Feb 22, 2007 at 07:30:49AM +0200, Martin Storsj? wrote:
> >>On Thu, 22 Feb 2007, Diego Biurrun wrote:
> >>
> >>>Looks wrong, AFAIR snowdsp_mmx is used by the decoder, not the encoder.
> >>>Maybe both, but surely not just the encoder.
> >>
> >>I was able to build a complete ffmpeg using this when enabling only the
> >>decoder.
> >
> >No doubt about this, the optimized routines were just left out..
> >
> >>Additionally, in dsputil_mmx.c, all references to the
> >>ff_snow-routines are wrapped in CONFIG_SNOW_ENCODER only.
> >
> >True, but this looks wrong to me, in libavcodec/ppc/dsputil_ppc.c they
> >are protected by 'if(ENABLE_SNOW_DECODER)' ..
> 
> Yes, it definitively seems to be wrong at the moment. Currently, the 
> generic implementations of the snow dsp functions are set in dsputil.c 
> only if CONFIG_SNOW_ENCODER is defined. Thus, the snow decoder segfaults 
> if the encoder isn't built.
> 
> Attached is a new attempt at fixing this. It changes all occurrances of 
> #ifdef CONFIG_SNOW_ENCODER into #if defined(CONFIG_SNOW_ENCODER) || 
> defined(CONFIG_SNOW_DECODER), and includes the architecture-dependent dsp 
> routines in the build only if the snow decoder/encoder is built (the 
> original problem I had).

Doesn't look correct to me, but I think I have fixed it already by
changing some CONFIG_SNOW_ENCODERs to CONFIG_SNOW_DECODER.  Let me know
if you still have problems.

Diego