[FFmpeg-soc] [soc]: r4435 - aacenc/aacenc.c

[Kostya]

On Thu, Jun 11, 2009 at 10:25:46PM +0200, Benjamin Larsson wrote:
> alexc wrote:
> > Author: alexc
> > Date: Thu Jun 11 19:23:53 2009
> > New Revision: 4435
> > 
> > Log:
> > Use scaled MDCT initializer (trunk r18855).
> > 
> > Modified:
> >    aacenc/aacenc.c
> > 
> > Modified: aacenc/aacenc.c
> > ==============================================================================
> > --- aacenc/aacenc.c	Thu Jun 11 18:53:44 2009	(r4434)
> > +++ aacenc/aacenc.c	Thu Jun 11 19:23:53 2009	(r4435)
> > @@ -173,8 +173,8 @@ static av_cold int aac_encode_init(AVCod
> >      s->samplerate_index = i;
> >  
> >      dsputil_init(&s->dsp, avctx);
> > -    ff_mdct_init(&s->mdct1024, 11, 0);
> > -    ff_mdct_init(&s->mdct128,   8, 0);
> > +    ff_mdct_init(&s->mdct1024, 11, 0, 1.0);
> > +    ff_mdct_init(&s->mdct128,   8, 0, 1.0);
> >      // window init
> 
> Hi, at least when doing the inverse transform the sign is flipped
> compared to what most codecs usually use, for example the aac decoder
> is/was adding a - to a scalefactor somewhere. So I just wonder if you
> are sure that it should be 1.0 instead of -1.0? While it shouldn't
> matter to the actual encode it might produce strange results for
> synthetic metrics (psnr, rms, sse).

It's the same with AAC decoder - sign flipping is long incorporated into
other place along with scaling. The right constant is either -512.0 or
-1.0/512.0 but who cares.
 
> MvH
> Benjamin Larsson