[FFmpeg-devel] [PATCH] SSE RDFT

Måns Rullgård mans
Thu Jun 24 23:57:59 CEST 2010

Vitor Sessak <vitor1001 at gmail.com> writes:

> On 06/24/2010 11:37 PM, M?ns Rullg?rd wrote:
>> "Ronald S. Bultje"<rsbultje at gmail.com>  writes:
>>> Hi,
>>> On Thu, Jun 24, 2010 at 5:20 PM, Jason Garrett-Glaser
>>> <darkshikari at gmail.com>  wrote:
>>>> 2010/6/24 M?ns Rullg?rd<mans at mansr.com>:
>>>>> Michael Niedermayer<michaelni at gmx.at>  writes:
>>>>>> btw, speaking of yasm, a not too long while ago i realized that configure
>>>>>> simply compiles without yasm optimizations if yasm isnt installed
>>>>>> i noticed it just because x264 doesnt and warns, we should do the same
>>>>>> this issue alone weighs heavier than any difference between inline and
>>>>>> yasm. Because loosing the optimizations without the user knowing is really
>>>>>> bad.
>>>>> Done.
>>>> This is not sufficient.  From years of experience with x264, people
>>>> will NEVER NOTICE the configure warning, and will proceed to compile
>>>> obliviously.  They will then complain incessantly about slow it is.
>>>> This is doubly important with more and more of ffmpeg's asm being in
>>>> yasm -- particularly the upcoming VP8 decoder, where we are going to
>>>> market ourselves on speed.
>>>> I suggest that configure fail with an error (telling the user to
>>>> install yasm or disable asm) if the following conditions are true:
>>>> 1) architecture is x86 or x86_64
>>>> 2) --disable-mmx is not set
>>>> 3) --disable-yasm is not set
>>> This is probably a good idea.
>>> Note how libvpx has some script that translated from yasm to bla*asm,
>>> similar to mru's iphone script, maybe we need something like that for
>> It's David's script.  I don't touch rotten fruit.
>>> the future? (Wild idea, reject if not practical.)
>> Not needed.  Yasm is available for all targets we support.
> Even PPC? I remember Loren once sent a Altivec-optimized FFT that was
> never committed because of some bikeshed about asm syntax...

Yasm is only for x86.

M?ns Rullg?rd
mans at mansr.com

More information about the ffmpeg-devel mailing list