[FFmpeg-devel] basic DCT question

[Roman V. Shaposhnik]

On Mon, 2009-02-23 at 19:11 +0100, Michael Niedermayer wrote:
> On Mon, Feb 23, 2009 at 09:23:46AM -0800, Roman V. Shaposhnik wrote:
> > Guys,
> > 
> > the following is probably going to show just how clueless
> > I am (still! :-() in the magic of DCT, but I have to ask it 
> > anyway.
> 
> The dct algrithm is selected by the user though the command line argument
> of matching name.
> the regression tests should/do not use dct algos that differ between
> achitectures, like the auto variant that selects the best, instead they
> use a plain ISO C one.

Got it! But just to clarify: are you saying that ISO C DCT
implementation guarantees even the extra 3 bits to be identical
on all the arcs?

Thanks,
Roman.

P.S. And yes, there's something weird going with how regression
harness sets the dct/idct. Here's the relevant portion of what
gets passed to the ffmpeg (with order of flags preserved):
   -flags +bitexact -dct fastint -idct simple -sws_flags +accurate_rnd+bitexact -dct int

Thanks,
Roman.