[FFmpeg-devel] [flamefest-start] A little something on MMX/SSE intrinsics

[Reimar Döffinger]

Hello,
On Thu, Feb 28, 2008 at 12:39:20AM +0200, ?smail D?nmez wrote:
> On Thu, Feb 28, 2008 at 12:30 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
> >  Thats like giving a painting from Leonardo da Vinci to a wild boar so it can
> >  correct the fine details.
> 
> That assumes every assembly coder is talented like Leonardo da Vinci,
> good luck with that assumption :-)

Well, that is easily compensated by the compiler usually having much
less talent than the wild boar ;-).
But to be serious: I think it is quite possible that PPC/AltiVec is much
easier to handle for a compiler than MMX, but then again it gets not
even a fraction of the attention that x86 gets.
But one issue remains that even in many years the very best compiler
will not be able to fix: There is no way to be sure that the compiler
will not do something very, very stupid in the most critical, innermost
loop - since it does not even remotely understand the code, the compiler
simply has no chance to know what is stupid as soon as there is more
than one way to optimize the code.
But given that our time is limited we might someday consider it worth it
even for places where we would not think of it right now.

Greetings,
Reimar D?ffinger