[FFmpeg-devel] [PATCH] runtime cpu detection for emms

[Alex Converse]

2010/8/30 M?ns Rullg?rd <mans at mansr.com>:
> Alex Converse <alex.converse at gmail.com> writes:
>
>> 2010/8/29 M?ns Rullg?rd <mans at mansr.com>:
>>> Aurelien Jacobs <aurel at gnuage.org> writes:
>>>
>>>> Hi,
>>>>
>>>> Recent change to emms() made it the only MMX dependent function which
>>>> don't support runtime cpu detection.
>>>> Attached patch restore this feature without the need for a global
>>>> mm_flags, and without the overhead of a mm_support() call for each
>>>> emms().
>>>>
>>>> Aurel
>>>>
>>>> diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
>>>> index 3b5ca12..1584b89 100644
>>>> --- a/libavcodec/dsputil.h
>>>> +++ b/libavcodec/dsputil.h
>>>> @@ -627,14 +627,9 @@ void ff_mlp_init_x86(DSPContext* c, AVCodecContext *avctx);
>>>>
>>>> ?#if HAVE_MMX
>>>>
>>>> +extern void (*ff_emms)(void);
>>>> ?#undef emms_c
>>>> -
>>>> -static inline void emms(void)
>>>> -{
>>>> - ? ?__asm__ volatile ("emms;":::"memory");
>>>> -}
>>>> -
>>>> -#define emms_c() emms()
>>>> +#define emms_c() ff_emms()
>>>>
>>>> ?#elif ARCH_ARM
>>>>
>>>> diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
>>>> index a24ed62..0b7de41 100644
>>>> --- a/libavcodec/x86/dsputil_mmx.c
>>>> +++ b/libavcodec/x86/dsputil_mmx.c
>>>> @@ -228,6 +228,15 @@ DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
>>>> ?/***********************************/
>>>> ?/* standard MMX */
>>>>
>>>> +static void emms_null(void) ?{ }
>>>> +static void emms_mmx(void) ? { __asm__ volatile ("emms;":::"memory"); }
>>>> +static void emms_init(void)
>>>> +{
>>>> + ? ?ff_emms = mm_support() & FF_MM_MMX ? emms_mmx : emms_null;
>>>> + ? ?ff_emms();
>>>> +}
>>>> +void (*ff_emms)(void) = emms_init;
>>>
>>> This is worse than the original.
>>>
>>
>> Suppose we put emms() in the DSP context and put the onus of calling
>> it on codecs that actually use MMX accelerated video functions?
>
> Unacceptable.
>

why?

Nothing seems acceptable to you accept breaking x86