Bug 1240201: [vorbis] P1. Properly determine sample duration and time in webm demuxer. r=kinetik

dom/media/platforms/agnostic/VorbisDecoder.cpp

@@ -17,9 +17,63 @@ extern mozilla::LogModule* GetPDMLog();
 
 namespace mozilla {
 
-ogg_packet InitVorbisPacket(const unsigned char* aData, size_t aLength,
-                         bool aBOS, bool aEOS,
-                         int64_t aGranulepos, int64_t aPacketNo)
+VorbisPacketSampleCounter::VorbisPacketSampleCounter()
+  : mError(0)
+  , mVorbisPacketCount(0)
+{
+  vorbis_info_init(&mVorbisInfo);
+  vorbis_comment_init(&mVorbisComment);
+}
+
+bool
+VorbisPacketSampleCounter::Init(const nsTArray<const uint8_t*>& aHeaders,
+                                const nsTArray<size_t>& aHeaderLens)
+{
+  for (size_t i = 0; i < aHeaders.Length(); i++) {
+    ogg_packet pkt =
+      VorbisDataDecoder::InitVorbisPacket(aHeaders[i], aHeaderLens[i],
+                                          i == 0, false, 0, mVorbisPacketCount++);
+    if ((mError = vorbis_synthesis_headerin(&mVorbisInfo, &mVorbisComment, &pkt))) {
+      return false;
+    }
+  }
+  return true;
+}
+
+int
+VorbisPacketSampleCounter::GetNumSamples(const uint8_t* aData, size_t aLength)
+{
+  if (mError) {
+    return -1;
+  }
+  ogg_packet pkt =
+    VorbisDataDecoder::InitVorbisPacket(aData, aLength, false, false, -1, mVorbisPacketCount++);
+  long blockSize = vorbis_packet_blocksize(&mVorbisInfo, &pkt);
+  if (blockSize < 0) {
+    return blockSize;
+  }
+  int nsamples;
+  if (mVorbisLastBlockSize) {
+    nsamples = mVorbisLastBlockSize.ref() / 4 + blockSize / 4;
+  } else {
+    // The first packet will output no audio, so set its count as 0.
+    nsamples = 0;
+  }
+  mVorbisLastBlockSize = Some(blockSize);
+  return nsamples;
+}
+
+VorbisPacketSampleCounter::~VorbisPacketSampleCounter()
+{
+  vorbis_info_clear(&mVorbisInfo);
+  vorbis_comment_clear(&mVorbisComment);
+}
+
+/* static */
+ogg_packet
+VorbisDataDecoder::InitVorbisPacket(const unsigned char* aData, size_t aLength,
+                                    bool aBOS, bool aEOS,
+                                    int64_t aGranulepos, int64_t aPacketNo)
 {
   ogg_packet packet;
   packet.packet = const_cast<unsigned char*>(aData);

dom/media/platforms/agnostic/VorbisDecoder.h

@@ -17,6 +17,25 @@
 
 namespace mozilla {
 
+class VorbisPacketSampleCounter
+{
+public:
+  VorbisPacketSampleCounter();
+  ~VorbisPacketSampleCounter();
+
+  bool Init(const nsTArray<const uint8_t*>& aHeaders,
+            const nsTArray<size_t>& aHeaderLens);
+  int GetNumSamples(const uint8_t* aData, size_t aLength);
+  void Reset() { mVorbisLastBlockSize.reset(); }
+
+private:
+  int mError;
+  vorbis_info mVorbisInfo;
+  vorbis_comment mVorbisComment;
+  int64_t mVorbisPacketCount;
+  Maybe<long> mVorbisLastBlockSize;
+};
+
 class VorbisDataDecoder : public MediaDataDecoder
 {
 public:
@@ -34,6 +53,10 @@ public:
   // Return true if mimetype is Vorbis
   static bool IsVorbis(const nsACString& aMimeType);
 
+  static ogg_packet InitVorbisPacket(const unsigned char* aData, size_t aLength,
+                                     bool aBOS, bool aEOS,
+                                     int64_t aGranulepos, int64_t aPacketNo);
+
 private:
   nsresult DecodeHeader(const unsigned char* aData, size_t aLength);
 

dom/media/webm/WebMDemuxer.cpp

@@ -12,11 +12,13 @@
 #include "WebMBufferedParser.h"
 #include "gfx2DGlue.h"
 #include "mozilla/Endian.h"
+#include "mozilla/PodOperations.h"
 #include "mozilla/Preferences.h"
 #include "mozilla/SharedThreadPool.h"
 #include "MediaDataDemuxer.h"
 #include "nsAutoRef.h"
 #include "NesteggPacketHolder.h"
+#include "VideoUtils.h"
 #include "XiphExtradata.h"
 #include "prprf.h"
 
@@ -417,6 +419,13 @@ WebMDemuxer::ReadMetadata()
         mInfo.mAudio.mCodecSpecificConfig->AppendElements(headers[0],
                                                           headerLens[0]);
       }
+
+      if (mAudioCodec == NESTEGG_CODEC_VORBIS) {
+        if (!mVorbisCounter.Init(headers, headerLens)) {
+          return NS_ERROR_FAILURE;
+        }
+      }
+
       uint64_t duration = 0;
       r = nestegg_duration(mContext, &duration);
       if (!r) {
@@ -542,6 +551,7 @@ WebMDemuxer::GetNextPacket(TrackInfo::TrackType aType, MediaRawDataQueue *aSampl
   (void) nestegg_packet_discard_padding(holder->Packet(), &discardPadding);
 
   for (uint32_t i = 0; i < count; ++i) {
+    int64_t duration = next_tstamp - tstamp;
     unsigned char* data;
     size_t length;
     r = nestegg_packet_data(holder->Packet(), i, &data, &length);
@@ -552,6 +562,26 @@ WebMDemuxer::GetNextPacket(TrackInfo::TrackType aType, MediaRawDataQueue *aSampl
     bool isKeyframe = false;
     if (aType == TrackInfo::kAudioTrack) {
       isKeyframe = true;
+      // Check if we have more than one audio frame in a webm block, if so
+      // determine the duration of each individual frame.
+      // We make the last frame end time match the start time of the next block
+      // unless it would make the duration of that frame negative.
+      if (count > 1 && (i != count - 1 || duration <= 0)) {
+        switch (mAudioCodec) {
+          case NESTEGG_CODEC_VORBIS:
+          {
+            int nsamples = mVorbisCounter.GetNumSamples(data, length);
+            if (nsamples < 0) {
+              return false;
+            }
+            duration =
+              FramesToTimeUnit(nsamples, mInfo.mAudio.mRate).ToMicroseconds();
+            break;
+          }
+          default:
+            break;
+        }
+      }
     } else if (aType == TrackInfo::kVideoTrack) {
       vpx_codec_stream_info_t si;
       PodZero(&si);
@@ -572,7 +602,7 @@ WebMDemuxer::GetNextPacket(TrackInfo::TrackType aType, MediaRawDataQueue *aSampl
     RefPtr<MediaRawData> sample = new MediaRawData(data, length);
     sample->mTimecode = tstamp;
     sample->mTime = tstamp;
-    sample->mDuration = next_tstamp - tstamp;
+    sample->mDuration = duration;
     sample->mOffset = holder->Offset();
     sample->mKeyframe = isKeyframe;
     if (discardPadding) {
@@ -582,6 +612,9 @@ WebMDemuxer::GetNextPacket(TrackInfo::TrackType aType, MediaRawDataQueue *aSampl
       sample->mExtraData->AppendElements(&c[0], 8);
     }
     aSamples->Push(sample);
+    if (mAudioCodec == NESTEGG_CODEC_VORBIS) {
+      tstamp += duration;
+    }
   }
   return true;
 }
@@ -721,6 +754,7 @@ WebMDemuxer::SeekInternal(const media::TimeUnit& aTarget)
 
   mLastAudioFrameTime.reset();
   mLastVideoFrameTime.reset();
+  mVorbisCounter.Reset();
 
   return NS_OK;
 }

dom/media/webm/WebMDemuxer.h

@@ -9,8 +9,10 @@
 #include "nsTArray.h"
 #include "MediaDataDemuxer.h"
 #include "NesteggPacketHolder.h"
+#include "mozilla/Maybe.h"
 #include "mozilla/Move.h"
 
+#include "VorbisDecoder.h"
 typedef struct nestegg nestegg;
 
 namespace mozilla {
@@ -199,6 +201,8 @@ private:
   // as nestegg only performs 1-byte read at a time.
   int64_t mLastWebMBlockOffset;
   const bool mIsMediaSource;
+
+  VorbisPacketSampleCounter mVorbisCounter;
 };
 
 class WebMTrackDemuxer : public MediaTrackDemuxer