Bug#567725: mplayer: gets SIGILL on Pentium II when playing an MPEG video with -vo caca

[Diego Biurrun]

On Sun, Jan 31, 2010 at 04:47:13PM +0100, Reinhard Tartler wrote:
> On So, Jan 31, 2010 at 12:18:29 (CET), Reimar Döffinger wrote:
> 
> > On Sun, Jan 31, 2010 at 09:32:05AM +0100, Reinhard Tartler wrote:
> >> libswscale uses MMX2:
> >> 
> >> > -- The GDB backtrace
> >> > #0  0xb620657d in yuv420_rgb24_MMX2 (c=0x8601ec0, src=0xbfffcb70, srcStride=0xbfffcb40,
> >> >     srcSliceY=0, srcSliceH=16, dst=0x862c944, dstStride=0xbfffcb50)
> >> >     at /build/buildd-ffmpeg_0.5+svn20090706-5-i386-gCmK4F/ffmpeg-0.5+svn20090706/libswscale/yuv2rgb_template.c:292
> >> 
> >> This is the same bug as
> >> https://bugs.launchpad.net/ubuntu/+source/ffmpeg/+bug/386397
> >> 
> >> As easy workaround, you can move /usr/lib/i686/cmov/libswscale.so.0 out
> >> of the way.
> >> 
> >> Not sure how to fix this properly.
> >
> > By doing what the first comment in that bug report says?
> > Compile FFmpeg with --enable-runtime-cpudetection
> 
> upon closer inspection, this is not possible in ffmpeg 0.5 yet. Unless
> someone wants to work on backporting this change from current trunk, I
> fear this will remain unfixed until ffmpeg 0.6.

This sounds like a candidate for backporting, i.e. it is likely a change
that I would approve.

Diego