[FFmpeg-devel] [PATCH] samplefmt: change syntax for av_samples_alloc() and av_samples_fill_arrays()

[Stefano Sabatini]

On date Monday 2011-05-23 16:51:25 +0200, Stefano Sabatini encoded:
> On date Monday 2011-05-23 15:15:31 +0200, Reimar Döffinger encoded:
> > On Mon, May 23, 2011 at 12:14:19AM +0200, Stefano Sabatini wrote:
> > > ---
> > >  libavutil/samplefmt.c |   34 ++++++++++++++++++----------------
> > >  libavutil/samplefmt.h |   26 ++++++++++++++++----------
> > >  2 files changed, 34 insertions(+), 26 deletions(-)
> > > 
> > > diff --git a/libavutil/samplefmt.c b/libavutil/samplefmt.c
> > > index ea004d9..346e44e 100644
> > > --- a/libavutil/samplefmt.c
> > > +++ b/libavutil/samplefmt.c
> > > @@ -73,31 +73,29 @@ int av_get_bits_per_sample_fmt(enum AVSampleFormat sample_fmt)
> > >  }
> > >  
> > >  int av_samples_fill_arrays(uint8_t *pointers[8], int linesizes[8],
> > > -                           uint8_t *buf, int nb_channels, int nb_samples,
> > > -                           enum AVSampleFormat sample_fmt, int planar, int align)
> > > +                           uint8_t *buf, int linesize0, int nb_channels,
> > > +                           enum AVSampleFormat sample_fmt, int planar)
> > 
> > The name indicates that av_samples_fill_arrays is a public function, and this
> > would then be an API/ABI change...
> 
> Yes indeed it was prematurely committed, but I have no problem at 
> 
> I retreat the patch for now, since I didn't explain clearly the reason
> for it, sorry for the noise.

Elaborating more: this patches is basically changing the layout of the
samples buffer.

My objective is getting a layout as similiar as possible with that of
the video buffer.

So this was my idea:
pointers  -> contain pointers to each samples plane
linesizes -> the linesize of each plane

This looks simpler and generally easier to manage.

Currently we have:
 * data[c] points to the first sample of channel c.
 * data[c] + linesize[0] points to the second sample of channel c

This has a slight advantage for accessing the samples for each channel
c, but it has the disadvantage that:
* misuse the term "linesize" (with a different semantics depending on
  planar/packed mode)
* creates an asymmetry between audio and video data layout

As for the current signature:
int av_samples_fill_arrays(uint8_t *pointers[8], int linesizes[8],
                           uint8_t *buf, int nb_channels, int nb_samples,
                           enum AVSampleFormat sample_fmt, int planar,
                           int align);

it looks fine and I can't remember why I had this linesize0 idea.
-- 
FFmpeg = Free and Fundamentalist Mournful Peaceless Elaborated God