V4L2VDA: Add more NALU types that indicate end of current frame data.

content/common/gpu/media/v4l2_video_decode_accelerator.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include <dlfcn.h>
 #include <errno.h>
 #include <fcntl.h>
 #include <linux/videodev2.h>
 #include <poll.h>
 #include <sys/eventfd.h>
@@ skipping 633 matching lines @@
           // the eighth data bit of the NAL; a zero value is encoded with a
           // leading '1' bit in the byte, which we can detect as the byte being
           // (unsigned) greater than or equal to 0x80.
           if (nalu.data[1] >= 0x80) {
             end_of_frame = true;
             break;
           }
           break;
         case media::H264NALU::kSPS:
         case media::H264NALU::kPPS:
+        case media::H264NALU::kAUD:
         case media::H264NALU::kEOSeq:

[acolwell GONE FROM CHROMIUM, 2014/05/08 17:56:34]

I think you need to also include kSEIMessage & kReserved14-kReserved18 to this
list. See Section 7.4.1.2.3 of the H.264 spec.

[Pawel Osciak, 2014/05/09 00:37:14]

Oh, I didn't notice you added new defines to the enum yet. Should have noticed
that CL. We could add them here as well then.

By the way, this change doesn't actually fix 370758, but I thought I'd throw
this in since we have it now.

         case media::H264NALU::kEOStream:
           // These unconditionally signal a frame boundary.
           end_of_frame = true;
           break;
         default:
           // For all others, keep going.
           break;
       }
       if (end_of_frame) {
         if (!decoder_partial_frame_pending_ && *endpos == 0) {

[acolwell GONE FROM CHROMIUM, 2014/05/08 17:56:34]

I'm a little fuzzy on how this code is supposed to work. Does this properly
identify the sequence {AUD, SPS, PPS, IDRSlice} as a single frame and {IDRSlice,
SPS, PPS, IDRSlice}  as 2 frames?

[Pawel Osciak, 2014/05/09 00:37:14]

It actually splits everything but slices of the same frame. It's too much and I
have a rewrite of this in the works, but it will touch quite a few places in the
class and I didn't want to submit that kind of a change just before branch
point.

           // The frame was previously restarted, and we haven't filled the
           // current frame with any contents yet.  Start the new frame here and
           // continue parsing NALs.
         } else {
           // The frame wasn't previously restarted and/or we have contents for
           // the current frame; signal the start of a new frame here: we don't
           // have a partial frame anymore.
           decoder_partial_frame_pending_ = false;
           return true;
         }
@@ skipping 1246 matching lines @@
   gfx::Size new_size(base::checked_cast<int>(format.fmt.pix_mp.width),
                      base::checked_cast<int>(format.fmt.pix_mp.height));
   if (frame_buffer_size_ != new_size) {
     DVLOG(3) << "IsResolutionChangeNecessary(): Resolution change detected";
     return true;
   }
   return false;
 }
 
 }  // namespace content