[Ffmpeg-devel] VP3/Theora Perfection

[Mike Melanson]

Rich Felker wrote:
> Somehow I expect their specialized code is slower than our general
> idct..

	I doubt that, especially for the 1-element (non-zero DC) IDCT case. The 
other one handles 10 elements.

>>>* using a 2*width*height array to store dct coefficients, which is 
>>>memset(0) for every frame
> 
> 
> very bad..

	I wrote most of this 2 years ago, probably with the thinking that I 
would optimize it later. I suppose now is later so it's time to get back 
to work on it.

> then someone else should write it.. really, though, it's easy and
> logical...

	I think the x86 ASM routines should be written in Intel/NASM syntax so 
they could be compiled under Microsoft compilers as well (and NASM under 
Linux). Or perhaps an intermediate macro language that could be 
converted to Intel or AT&T syntax. But that's just me.

-- 
	-Mike Melanson