[FFmpeg-devel] SoC wmapro project

[Kostya]

On Sun, Jan 11, 2009 at 06:01:25PM +0100, Sascha Sommer wrote:
> Hi,
> 
> On Sonntag, 11. Januar 2009, Diego Biurrun wrote:
> > On Sun, Jan 11, 2009 at 05:08:45PM +0100, Benjamin Larsson wrote:
> > > Hi, the wmapro code is quite usable now, so please test and review the
> > > code so we can start the process of getting the code into main svn.

Congratulations - it works on my PPC (I've had to add 4096-point
sine window table to dsputil.h and mdct.c by hand though) which is the long
awaited thing for me.

> > For starters: How about fixing the issues I mentioned about Sascha's
> > last SoC commit?
> >
> 
> Of course some things are not done yet. I'll try to work on this list next 
> week
> 
> - more sanity checks for the input data, there are a few problematic cases
>   afair
> - the things that Diego mentioned in his mail
> - removing the asserts (they are there to catch decoding errors asap)
>   (seeking won't work atm)

Hmm, I've tried it with the random WMV3+WMA3 file and seeking seems to work
more or less (except for periodic "!! Packet loss detected!", maybe you should also
implement AVCodec.flush() to reset state after seeking)

> - renaming everything to wmapro?
> - a few more comments
> - check if the pow can be replaced with a table

 for inverse quantizing? quite probably

> - renaming some variables
> - reorganizing the context etc.
> - what about the channel layout?

IIRC, channel layout in WMA was the source of inspiration for ffmpeg channel layout,
so mapping will be straightforward. I'm sure Benjamin will give a detailed answer.

> - checking that everything is consistent
> - FIXMEs
> ...
> 
> So I'm mostly interested in:
> - files that do not decode correctly

Have you tried all files from http://samples.mplayerhq.hu/A-codecs/WMA9/ ?
>From my experience you just release it into the wild (ensuring that known MPHQ
samples are decoded fine) and wait for somebody to bugreport :).
 
> - suggestions on how the code sharing between wmapro and the other variants
>   could be done
>   From what I can tell the packet parsing and the imdct and
>   windowing parts could be shared. Most other things look a bit  
>   different
> 
> Regards
> 
> Sascha