[FFmpeg-devel] libavutil simd

[Måns Rullgård]

Rich Felker wrote:
> On Tue, Oct 02, 2007 at 03:27:56PM +0200, Michael Niedermayer wrote:
>> the whole thread seems to be centered around "we must do it at runtime
>> no matter what" i just cant help but keep wondering why that is so
>> important?
>
> I agree totally, this view is unnecessary and causing problems..

We also have no runtime detection for ARM or MIPS extensions.  Alpha
capabilities are cleanly detected, IIRC.

>> PS: let me remind everyone that libavutil is supposed to be LIGHTweight,
>> simple, modular and fast
>> and i really would rather drop SIMD in libavutil completely before we
>
> Agreed. I don't see a point in having SIMD (or any asm or
> non-100%-portable-C) code in libavutil to begin with, but people keep
> proposing optimizations to this stuff with dubious value..

Michael says himself lavu should be fast.  We all know SIMD (and other
CPU-specific things) can sometimes make code much faster.

Personally, I'd be happy if Michael cleaned up the zillions of compiler
warnings from (mainly) aes.c.

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