[FFmpeg-devel] Question about -fPIC usage for some files
Mon Feb 11 09:40:20 CET 2008
Michael Niedermayer schrieb:
> On Fri, Feb 08, 2008 at 05:30:50PM +0100, Reimar D?ffinger wrote:
>> IIRC that last one does fall in that last category and can be fixed with
>> a __aatribute__((noinline)), the funniest thing about that is that ICC
>> (10.1.008) can not handle that code even without -fPIC, and needs the same hack...
> Why dont you change the asm a little bit to be more easy for our old gcc
> and co?
> "=m"(autoc[j]), "=m"(autoc[j+1]), "=m"(autoc[j+2])
> and access the other 2 by 8+%X 16+%X
> and MANGLE()
> that would be 3 operands less which gcc can stumble over
i offered a patch some monthes ago that did a similar thing and it was
however this would help only for one of the three problematic code
pieces (flacdsp_mmx.c), but i see no chance to fix the -fPIC issue by
similar techniques in cavsdsp_mmx.c and h264dsp_mmx.c.
There some register names are used explicitly together with 1-3 memory
references for different arrays, hard to get that done with -fPIC and
minimum patch size...
More information about the ffmpeg-devel