On 2014-02-18 14:35:35 +0100, Niels Möller wrote:
> Christophe Gisquet <[email protected]> writes:
> 
> > This scale is constant so it is simpler to just scale the
> > FIR coefficients in the tables instead.
> > ---
> >  libavcodec/dcadata.h | 384 
> > +++++++++++++++++++++++++--------------------------
> >  libavcodec/dcadec.c  |   7 +-
> >  libavcodec/dcadsp.c  |  16 +--
> >  libavcodec/dcadsp.h  |   3 +-
> >  4 files changed, 203 insertions(+), 207 deletions(-)
> >
> > diff --git a/libavcodec/dcadata.h b/libavcodec/dcadata.h
> > index dc4e8b9..2e4ac5f 100644
> > --- a/libavcodec/dcadata.h
> > +++ b/libavcodec/dcadata.h
> > @@ -7307,202 +7307,202 @@ DECLARE_ALIGNED(16, static const float, 
> > fir_32bands_nonperfect)[] =
> >  
> >  DECLARE_ALIGNED(16, static const float, lfe_fir_64)[] =
> >  {
> > -    2.658434386830777e-4, 9.029330685734748e-3,
> [...]
> > +    3.1691007457e-11, 1.0763800962e-09,
> 
> It looks like you're not only scaling, you also drop a couple of digits
> of precision here. Do you really want to?

it still has more precision than float can hold so I think that should
be ok but Christophe said he is going redo the scaling with a macro.
I assume that will preserve the current constants.

Janne 
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to