[FFmpeg-devel] [PATCH] remove gcc 3.3 workaround in swscale_template.c
Sun Sep 20 21:02:46 CEST 2009
On Sun, Sep 20, 2009 at 08:00:40PM +0200, Diego Biurrun wrote:
> On Sun, Sep 20, 2009 at 07:12:44PM +0200, Reimar D?ffinger wrote:
> > On Sun, Sep 20, 2009 at 06:53:36PM +0200, Diego Biurrun wrote:
> > > There's an ugly preprocessor gcc 3.3 workaround in swscale_template.c:
> > >
> > > /* GCC 3.3 makes MPlayer crash on IA-32 machines when using "g" operand here,
> > > which is needed to support GCC 4.0. */
> > > #if ARCH_X86_64 && ((__GNUC__ > 3) || (__GNUC__ == 3 && __GNUC_MINOR__ >= 4))
> > > :: "m" (src1), "m" (dst), "g" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask),
> > > #else
> > > :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask),
> > > #endif
> > >
> > > At the very least it should be updated to use the AV_GCC_VERSION_AT_LEAST
> > > macro from libavutil. However, I would prefer to get rid of it
> > > completely. If I understand the comment correctly, deleting the whole
> > > #else clause would be the way to achieve this. Since I know little
> > > enough assembler to have no real idea what "g" and "m" operands are all
> > > about I'd like to hear an informed opinion.
> > Huh? No, the upper line should give faster code but causes crashes due
> > to miscompilation with gcc 3.3
> > The lower line does not compile with gcc 4.0 on x86_64 but produces correct code
> > with 3.3.
> > At least that's what the comment says.
> OK, then I propose getting rid of the workaround completely. I don't
> think x86_64 users with gcc 3.3 exist, so it's not worth the ugliness on
> our side.
Hm? Which do you consider a workaround? I think you might still need two cases,
one for x86_32 and one for x86_64 for gcc 4.x, you could only get rid of the
gcc version check.
More information about the ffmpeg-devel