[FFmpeg-devel] [flamefest-start] A little something on MMX/SSE intrinsics
Sat Mar 1 16:04:01 CET 2008
On Sat, Mar 01, 2008 at 09:26:34PM +0800, Zuxy Meng wrote:
> 2008/3/1, Michael Niedermayer <michaelni at gmx.at>:
> > On Sat, Mar 01, 2008 at 02:09:14PM +0200, Ivan Kalvachev wrote:
> > [...]
> > > Of course there are architectures like VLIW where you have fixed
> > > pre-determined number of instructions that could be grouped together.
> > Not that VLIW is the subject but ...
> > The compiler generally does not know (gcc even less so) which instructions
> > are independant thus it is likely going to mess up worse than a human ...
> > A human plain knows what a write over a given pointer will change and
> > with what such an instruction can be grouped for a compiler this is not so
> > easy to find out (restrict helps but still ...)
> > Also i dont know any VLIW arch which has intrinsics ...
> IA-64 supports MMX and SSE intrinsics, if you'd call it VLIW.
Those are not VLIW intrinsics though, and I seriously doubt you can even
get anywhere close to optimal speed with MMX/SSE intrinsics on IA-64.
Not that it matters much, since for almost all applications nobody is
going to write IA-64 optimizations anyway...
More information about the ffmpeg-devel