Implement FrameMarking header extension support

webrtc/modules/rtp_rtcp/source/rtp_header_extensions.cc

Index: webrtc/modules/rtp_rtcp/source/rtp_header_extensions.cc
diff --git a/webrtc/modules/rtp_rtcp/source/rtp_header_extensions.cc b/webrtc/modules/rtp_rtcp/source/rtp_header_extensions.cc
index 2f3feb34cd3851f53da1bbda45d5f90a9f575bd8..71bbd3254cc45bbc2ccc5fbb3090c9c7e49049aa 100644
--- a/webrtc/modules/rtp_rtcp/source/rtp_header_extensions.cc
+++ b/webrtc/modules/rtp_rtcp/source/rtp_header_extensions.cc
@@ -313,4 +313,112 @@ bool RepairedRtpStreamId::Write(uint8_t* data, const std::string& rsid) {
   return RtpStreamId::Write(data, rsid);
 }
 
+// For Frame Marking RTP Header Extension:
+//
+// https://tools.ietf.org/html/draft-ietf-avtext-framemarking-04#page-4
+// This extensions provides meta-information about the RTP streams outside the
+// encrypted media payload, an RTP switch can do codec-agnostic
+// selective forwarding without decrypting the payload.
+//
+// for Non-Scalable Streams:
+//
+//     0                   1
+//     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5
+//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+//    |  ID=? |  L=0  |S|E|I|D|0 0 0 0|
+//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+//
+// for Scalable Streams:
+//
+//     0                   1                   2                   3
+//     0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+//    |  ID=? |  L=2  |S|E|I|D|B| TID |   LID         |    TL0PICIDX  |
+//    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+//
+constexpr RTPExtensionType FrameMarking::kId;
+constexpr const char* FrameMarking::kUri;
+
+bool FrameMarking::Parse(rtc::ArrayView<const uint8_t> data,
+                         FrameMarks* frame_marks) {
+  RTC_DCHECK(frame_marks);
+
+  if (data.empty())
+    return false;
+
+  // Set frame marking data
+  frame_marks->start_of_frame = (data[0] & 0x80) != 0;
+  frame_marks->end_of_frame = (data[0] & 0x40) != 0;
+  frame_marks->independent = (data[0] & 0x20) != 0;
+  frame_marks->discardable = (data[0] & 0x10) != 0;
+
+  // Check variable length.
+  if (data.size() == 1) {
+    // We are non-scalable.
+    frame_marks->base_layer_sync = false;
+    frame_marks->temporal_layer_id = 0;
+    frame_marks->layer_id = 0;
+    frame_marks->tl0_pic_idx = 0;
+  } else if (data.size() == 3) {
+    // Set scalable parts
+    frame_marks->base_layer_sync = (data[0] & 0x08) != 0;
+    frame_marks->temporal_layer_id = data[0] & 0x07;
+    frame_marks->layer_id = data[1];
+    frame_marks->tl0_pic_idx = data[2];
+  } else {
+    // Incorrect length.
+    return false;
+  }
+  return true;
+}
+
+size_t FrameMarking::ValueSize(const FrameMarks& frame_marks) {
+  // Check if it is scalable.
+  if (frame_marks.base_layer_sync ||
+      (frame_marks.temporal_layer_id &&
+       frame_marks.temporal_layer_id != kNoTemporalIdx) ||
+      (frame_marks.layer_id && frame_marks.layer_id != kNoSpatialIdx) ||
+      (frame_marks.tl0_pic_idx &&
+       frame_marks.tl0_pic_idx != static_cast<uint8_t>(kNoTl0PicIdx)))

[danilchap, 2017/06/30 12:53:25]

does it make 0xff picture id invalid?
may be instead convert tl0_pic_idx to int

[sergio.garcia.murillo, 2017/06/30 13:42:00]

Agree, changing it to int so it matches the other code using tl0_pic_idx

+    return 3;
+  else
+    return 1;
+}
+
+bool FrameMarking::Write(uint8_t* data, const FrameMarks& frame_marks) {
+  data[0] = frame_marks.start_of_frame ? 0x80 : 0x00;
+  data[0] |= frame_marks.end_of_frame ? 0x40 : 0x00;
+  data[0] |= frame_marks.independent ? 0x20 : 0x00;
+  data[0] |= frame_marks.discardable ? 0x10 : 0x00;
+
+  // Check if it is scalable.
+  if (frame_marks.base_layer_sync ||
+      (frame_marks.temporal_layer_id &&
+       frame_marks.temporal_layer_id != kNoTemporalIdx) ||
+      (frame_marks.layer_id && frame_marks.layer_id != kNoSpatialIdx) ||
+      (frame_marks.tl0_pic_idx &&
+       frame_marks.tl0_pic_idx != static_cast<uint8_t>(kNoTl0PicIdx))) {

[danilchap, 2017/06/30 12:53:25]

same here (since this check is large and non-trivial may be reuse previous
function:
if (ValueSize(frame_marks) == 3) {
  data[0] |= ...
}

[sergio.garcia.murillo, 2017/06/30 13:42:00]

Done.

+    data[0] |= frame_marks.base_layer_sync ? 0x08 : 0x00;
+    data[0] |= (frame_marks.temporal_layer_id & 0x07);
+    data[1] = frame_marks.layer_id;
+    data[2] = frame_marks.tl0_pic_idx;
+  }
+  return true;
+}
+
+uint8_t FrameMarking::CreateLayerId(RTPVideoHeaderVP9 vp9) {
+  // The following  shows VP9 Layer encoding information (3 bits for
+  // spatial and temporal layer) mapped to the generic LID and TID fields.
+  // The P and U bits MUST match the corresponding bits in the VP9 Payload
+  // Description.
+  //    0                   1                   2                   3
+  //    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+  //   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+  //   |  ID=2 |  L=2  |S|E|I|D|B| TID |0|0|0|P|U| SID |    TL0PICIDX  |

[danilchap, 2017/06/30 12:53:25]

linked spec has different format for VP9 LID Mapping (looking at 3.2.1.2)
Is there more recent spec?

[sergio.garcia.murillo, 2017/06/30 13:42:00]

Current draft has invalid format for LID. Spatial id has only two bits and it
includes a quality layer id, which is not used in vp9, it also misses important
information required by an SFU in order to do the layer selection (the P and U
bits).

I have submitted a patch to the IETF group and it will be included in the next
draft (I think it will be pre-IETF 99)
https://www.ietf.org/mail-archive/web/avtext/current/msg01359.html

+  //   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+  return (vp9.spatial_idx != kNoSpatialIdx ? vp9.spatial_idx & 0x07 : 0x00) |
+         (vp9.temporal_up_switch ? 0x08 : 0x00) |
+         (vp9.inter_pic_predicted ? 0x10 : 0x00);
+}
+
 }  // namespace webrtc