[FFmpeg-soc] libavfilter audio work - qualification task

[Stefano Sabatini]

On date Monday 2010-04-05 08:44:15 -0700, S.N. Hemanth Meenakshisundaram encoded:
> On 04/05/2010 02:43 AM, Stefano Sabatini wrote:
> >On date Thursday 2010-04-01 11:53:42 -0700, Baptiste Coudurier encoded:
> >>On 03/31/2010 10:44 PM, S.N. Hemanth Meenakshisundaram wrote:
> >>>Hi All,
> >>>
> >>>Based on the posts above, here's what I was thinking of doing as a
> >>>qualification task :
> >>>
> >>>1. Fix the configure_filters function in ffmpeg.c to work for both video
> >>>and audio filters.
> >>>
> >>>2. Bring the af_null and af_vol filters from soc/afilters to
> >>>soc/avfilter and generalize avfilter framework to work with the af_*
> >>>filters. This should involve a minimal design for the avfilter audio
> >>>changes.
> >>>
> >>>3. af_vol currently just scales input by two. Modify af_vol to have some
> >>>basic volume filter functionality.
> >>>
> >>>Please let me know if I have understood the tasks right and if this will
> >>>work as a qualification task.
> >I believe that is even too burden as a qualification task, much of the
> >libavfilter audio design has yet to be done.
> >
> >>A qualification task could be to add AVCodecContext get_buffer
> >>functionality to audio decoders.
> >As alternative to this, any video libavfilter task should be OK
> >(e.g. port a filter from MPlayer/libmpcodecs or write a new one), note
> >that vf_imlib.c is *not* a valid qualification task (see
> >https://roundup.ffmpeg.org/issue302), this should help you to get
> >confident with the FFmpeg/libavfilter API.
> >
> >I especially recommend: a deinterlace filter (check the code in
> >libavcodec/imgconvert.c), a libfreetype wrapper (check as reference
> >vhook/drawtext.c in ffmpeg-0.5).
> >
> >Ask here or on IRC if you have questions.
> >
> Hi,
> 
> For the deinterlace filter, do we want just a vf_* wrapper for
> imgconvert functions or should the code be ported to the
> vf_deinterlace filter?

It should be ported, lavfi should not depend on lavc (that is it
should be usable without any libavcodec installed).
 
> I have been working on getting the audio filters from soc/afilters
> to work more like the current video filters and wanted to check that
> am on the right track :
> 
> 1. Client will use a 'asrc_buffer_add_samples' function to give
> buffers to the source input filter or define an input filter that
> can fetch audio data.
> 2. Client then calls request_buffer on output filter,
> request_buffers propagate all the way to source (input filter).
> 3. The buffer is then passed along the filter chain and processed.

Looks fine, BTW also check this thread:
http://thread.gmane.org/gmane.comp.video.ffmpeg.devel/107363

> Also have a couple of questions :
> 
> 1. Should the formats for audio be from the SampleFormat enum
> (libavcodec/avcodec.h) or the PCM formats in the CodecID enum. The
> old audio filter seems to be wrongly using both in two different
> places.

That has to do with how we choose to represent samples. Currently we
should use enum SampleFormat, but as you can see the CODEC_ID_ provide
more information (e.g. CODEC_ID_PCM_F32BE, CODEC_ID_PCM_F32LE,
CODEC_ID_PCM_F64BE, CODEC_ID_PCM_F64LE while we have just a
SAMPLE_FMT_DBL SampleFormat).

Preferably use enum SampleFormat, but be aware that it will need to be
extended (also maybe we will need some sort of sample format
descriptor of the kind of that defined in pixdesc.h, doesn't look
strictly necessary by now though).

> 2. While video filters work a single frame at a time, audio would
> need to work on a buffer. Is it ok to use a single call similar to
> the current 'filter_buffer' to get this done instead of the
> start_frame, draw_slice, end_frame sequence in video?

Yes, check also the abovementioned thread.

> I am planning to do the deinterlace, libfreetype first and then
> continue with this.

Good. Feel free to join the discussion on ffmpeg-devel regarding the
libavfilter API, consider that as part of the qual task.

Regards.