[FFmpeg-devel] [PATCH] Add tests for audio encoders: aac, dca, ra144 and nellymoser.

[Reimar Döffinger]

On Fri, Apr 06, 2012 at 08:51:44PM +0200, Reimar Döffinger wrote:
> On Fri, Apr 06, 2012 at 02:42:20PM -0400, Derek Buitenhuis wrote:
> > On 06/04/2012 2:06 PM, Reimar Döffinger wrote:
> > > Those only give a pass/no pass result, not the exact results.
> > 
> > The point is that you cannot rely on exact results to be
> > consistent when floats are involved.
> 
> I believe no floats are involved in the DTS encoder code
> (though I guess in the decoder, so that might break it).
> For nellymoser and ra144 I am not sure that any rounding
> errors can actually cause failures/different output.
> For ra144 at least the decoder is integer-only.
> But even so, running this through FATE once would be an
> easy way to get the right PSNR limit values for the other
> framework.

Hm, it seems upon failing it prints the values, too, so that
should not be a problem.

> Of course that only helps if I can actually use our standard
> audio reference there, too. The wma test use only the "luckynight"
> sample.

A bit messy, but it seems to be possible to test both.

> There is also a bit the problem that when using enc_dec_pcm
> we'll probably never notice when due to decoder improvements
> that CMP_TARGET value should be made more strict.

That remains, so I would prefer testing whether we can use
the strict test for any.