[FFmpeg-devel] [PATCH] Unconditionally run emms on calls to emms_c() if HAVE_MMX is defined

Måns Rullgård mans
Sun Jul 4 21:50:14 CEST 2010


Dominik 'Rathann' Mierzejewski <dominik at rangers.eu.org> writes:

> On Sunday, 04 July 2010 at 20:02, M?ns Rullg?rd wrote:
>> Dominik 'Rathann' Mierzejewski <dominik at rangers.eu.org> writes:
>> 
>> > On Sunday, 04 July 2010 at 17:05, M?ns Rullg?rd wrote:
>> >> Eli Friedman <eli.friedman at gmail.com> writes:
>> >> 
>> >> > 2010/7/4 M?ns Rullg?rd <mans at mansr.com>:
>> >> >> Eli Friedman <eli.friedman at gmail.com> writes:
>> >> >>
>> >> >>> On Sat, Jul 3, 2010 at 11:50 PM, Eli Friedman <eli.friedman at gmail.com> wrote:
>> >> >>>> Patch attached. ?We already assume HAVE_MMX implies MMX is available
>> >> >>>> at runtime, so this seems like it should be OK.
>> >> >>>
>> >> >>> Actually, that's wrong... I think this breaks distros that want to
>> >> >>> support Pentium Pro and earlier in the same binary for modern
>> >> >>> computers.
>> >> >>
>> >> >> A binary built for i586 or earlier will be quite crippled even in
>> >> >> plain C code. ?The only sane thing to do as a distro to provide two
>> >> >> packages (or more).
>> >> >
>> >> > So you think this patch is okay?
>> >> 
>> >> _I_ think the idea is OK, but your patch isn't quite what I intended.
>> >> I'm waiting for Reinhard to voice his opinion.  He knows what distros
>> >> want, can, and can't.
>> >
>> > Speaking from Fedora side, the current minimum hw requirement for CPU
>> > is Pentium Pro (i686). MMX is not required, so the basic package should
>> > work even if MMX is not available.
>> 
>> Do you have any mechanism for providing multiple packages of the same
>> software with different optimisations?  Allowing mmx/sse in
>> non-pointered code has real performance advantages.  Do you want to
>> deprive your users of those?
>
> Yes, we have such mechanism and no, I don't want to deprive users
> of the optimizations.

So you wouldn't be screaming bloody murder if we made mmx-enabled
builds use a few mmx instructions unconditionally?

Reinhard?

-- 
M?ns Rullg?rd
mans at mansr.com



More information about the ffmpeg-devel mailing list