[Bug 696770] Memory leak when processing h.264 field pictures

[GStreamer (bugzilla.gnome.org)]

https://bugzilla.gnome.org/show_bug.cgi?id=696770
  GStreamer | gst-libav | 1.0.5

--- Comment #14 from Sebastian Dröge <slomo at circular-chaos.org> 2013-06-05 11:06:55 UTC ---
(In reply to comment #12)
> (In reply to comment #8)
> > Review of attachment 245998 [details] [details]:
> > 
> > ::: ext/libav/gstavviddec.c
> > @@ +1400,3 @@
> > +     ffmpeg is allocating the output buffer only for complete frames and not
> > +     for the individual fields. */
> > +  if (frame && !frame->output_buffer) {
> > 
> > I don't think this is correct. libav is not necessarily allocating an output
> > buffer for the current input frame immediately
> 
> Does the following code path in get_buffer() making sure the latest frame
> buffer allocation?
> 
> picture->reordered_opaque = context->reordered_opaque;
>   frame =
>       gst_video_decoder_get_frame (GST_VIDEO_DECODER (ffmpegdec),
>       picture->reordered_opaque);

No, it only makes sure that some frames gets a buffer allocation. Might be the
latest, might be any other frame.

> > Also does this mean that every GstVideoCodecFrame was a field here, not a
> > frame? In that case it would be something that the parser should've fixed,
> > GstVideoCodecFrame is supposed to be one complete frame with everything related
> > to it.
> 
> Yes, for field-encoded stream, every GstVideoCodecFrame is a field. The issue
> is exactly what I mentioned before.  I don't think it is the duty of parser.
> Because if I understood correctly some decoders can send fields as frames to
> renderer directly.  For eg, vaapi has a feature for this. vaPutsurface() api
> has flags to specify whether the surface has top fields or bottom fields. But
> AFAIK this is not yet supported in driver. So in current gstreamer-vaapi we are
> using the same surface to render both top and bottom fields.
> 
> I have a fix for this to gstreamer-vaapi also:
> https://bugzilla.gnome.org/show_bug.cgi?id=701257
> 
> We need similar fixes to mpeg2dec also.

Then this somehow needs to be handled in the GstVideoDecoder::parse() function
to accumulate all fields before claiming something to be a frame. The detection
of this could be based on buffer flags, that are put on there by the parser
(http://gstreamer.freedesktop.org/data/doc/gstreamer/head/gst-plugins-base-libs/html/gst-plugins-base-libs-gstvideo.html#GstVideoBufferFlags
 ONEFIELD | TFF and ONEFIELD | RFF)

-- 
Configure bugmail: https://bugzilla.gnome.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug.
You are the assignee for the bug.