[Bug 732266] vaapidecode: h264: add support for decoding SVC base layers only

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=732266

--- Comment #25 from sreerenj <bsreerenj at gmail.com> ---
(In reply to Orestis Floros from comment #24)
> (In reply to sreerenj from comment #23)
> > (In reply to Orestis Floros from comment #22)
> > > Created attachment 358028 [details] [review] [review] [review]
> > > (In reply to sreerenj from comment #21)
> > > 
> > > > One access unit can have multiple view components, and yes you don't have to
> > > > add the slice_ext in the code for checking the complete_au. Sorry about the
> > > > confusion.
> > > > But ideally, we should add other checkings like "compare the view_ids of
> > > > nals" for vcl boundary detection (even though current code can work without
> > > > that).
> > > 
> > > I think this requires the previous slice / vlc nalu right? I think h264parse
> > > doesn't keep that.
> > 
> > Yup, It requires some extra parsing code. But no need to include that one
> > here I believe (unless it is necessary to fix any known issue). 
> > 
> > > 
> > > > BTW I hit a segfault issue while testing your branch:
> > > > 
> > > > Generate an mvc sample like this: 
> > > > gst-launch-1.0 -v videotestsrc num-buffers=30 blocksize=115200 !
> > > > vaapih264enc num-views=2 ! video/x-h264, profile=stereo-high ! filesink
> > > > location=sample.mvc
> > > > 
> > > > playback the stream with only base view
> > > > gst-launch-1.0 filesrc location=sample.mvc ! h264parse ! vaapih264dec
> > > > base-only=true ! vaapisink
> > > 
> > > Thanks, fixed.
> > 
> > K great.
> > So I guess the only know issue is
> > https://bugzilla.gnome.org/show_bug.cgi?id=732266#c18 . right?
> 
> yes.
> 
> The errors:
> 
> SVCHS-2.264:
> lt-gst-launch-1.0: gen75_mfd.c:556: gen75_mfd_avc_img_state: Assertion
> `pic_param->pic_fields.bits.field_pic_flag == 0' failed.
> 
> SVCHS-2-L1.264:
> lt-gst-launch-1.0: gen75_mfd.c:556: gen75_mfd_avc_img_state: Assertion
> `pic_param->pic_fields.bits.field_pic_flag == 0' failed.
> 

This is the driver assertion.
Because somehow middleware set the frame_mbs_only_flag in pic_params (which
means not an interlaced content) but set the filed_pic_flag (which means frame
has interlaced fields) in slice_header.
Check the parsed values, could be a wrongly encoded stream.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.