[FFmpeg-devel] [PATCH] x86/emms: run the instruction unconditionally on supported targets

James Almer jamrial at gmail.com
Thu Feb 4 01:15:00 CET 2016


On 2/3/2016 2:21 AM, James Almer wrote:
> Inlined functions like AV_ZERO* and AV_COPY* may use mmx instructions
> regardless of runtime cpuflags.
> 
> Signed-off-by: James Almer <jamrial at gmail.com>
> ---
> On targets where __MMX__ is not defined (like default x86_32 builds) the
> runtime check is a must, and neither of the above functions will use mmx
> instructions anyway.
> 
>  libavutil/x86/emms.h | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/libavutil/x86/emms.h b/libavutil/x86/emms.h
> index a529b6b..0deeb8c 100644
> --- a/libavutil/x86/emms.h
> +++ b/libavutil/x86/emms.h
> @@ -34,7 +34,9 @@ void avpriv_emms_yasm(void);
>   */
>  static av_always_inline void emms_c(void)
>  {
> +#if !defined(__MMX__)
>      if(av_get_cpu_flags() & AV_CPU_FLAG_MMX)
> +#endif
>          __asm__ volatile ("emms" ::: "memory");
>  }
>  #elif HAVE_MMX && HAVE_MM_EMPTY
> 

If you want a simple test case, on x86_32 do:

configure --cpu=pentium-mmx && make fate-dca CPUFLAGS=0

And watch it crash.


More information about the ffmpeg-devel mailing list