quicktimefilesink.cpp

H.264 RTSP 串流(live 555)視窗版本

      }
      case fourChar('h','2','6','3'): {
	fQTTimeUnitsPerSample = fQTTimeScale/fOurSink.fMovieFPS;
	break;
      }
      }
    }
  } else if (fQTMediaDataAtomCreator == &QuickTimeFileSink::addAtom_Qclp) {
    // For QCELP data, make a note of the frame size (even though it's the
    // same as the packet data size), because it varies depending on the
    // 'rate' of the stream, and this size gets used later when setting up
    // the 'Qclp' QuickTime atom:
    fQTBytesPerFrame = packetDataSize;
  }

  useFrame(*fBuffer);
  if (fOurSink.fPacketLossCompensate) {
    // Save this frame, in case we need it for recovery:
    SubsessionBuffer* tmp = fPrevBuffer; // assert: != NULL
    fPrevBuffer = fBuffer;
    fBuffer = tmp;
  }
  fBuffer->reset(); // for the next input

  // Now, try getting more frames:
  fOurSink.continuePlaying();
}

void SubsessionIOState::useFrame(SubsessionBuffer& buffer) {
  unsigned char* const frameSource = buffer.dataStart();
  unsigned const frameSize = buffer.bytesInUse();
  struct timeval const& presentationTime = buffer.presentationTime();
  int64_t const destFileOffset = TellFile64(fOurSink.fOutFid);
  unsigned sampleNumberOfFrameStart = fQTTotNumSamples + 1;
  Boolean avcHack = fQTMediaDataAtomCreator == &QuickTimeFileSink::addAtom_avc1;

  // If we're not syncing streams, or this subsession is not video, then
  // just give this frame a fixed duration:
  if (!fOurSink.fSyncStreams
      || fQTcomponentSubtype != fourChar('v','i','d','e')) {
    unsigned const frameDuration = fQTTimeUnitsPerSample*fQTSamplesPerFrame;
    unsigned frameSizeToUse = frameSize;
    if (avcHack) frameSizeToUse += 4; // H.264/AVC gets the frame size prefix

    fQTTotNumSamples += useFrame1(frameSizeToUse, presentationTime, frameDuration, destFileOffset);
  } else {
    // For synced video streams, we use the difference between successive
    // frames' presentation times as the 'frame duration'.  So, record
    // information about the *previous* frame:
    struct timeval const& ppt = fPrevFrameState.presentationTime; //abbrev
    if (ppt.tv_sec != 0 || ppt.tv_usec != 0) {
      // There has been a previous frame.
      double duration = (presentationTime.tv_sec - ppt.tv_sec)
	+ (presentationTime.tv_usec - ppt.tv_usec)/1000000.0;
      if (duration < 0.0) duration = 0.0;
      unsigned frameDuration
	= (unsigned)((2*duration*fQTTimeScale+1)/2); // round
      unsigned frameSizeToUse = fPrevFrameState.frameSize;
      if (avcHack) frameSizeToUse += 4; // H.264/AVC gets the frame size prefix

      unsigned numSamples
	= useFrame1(frameSizeToUse, ppt, frameDuration, fPrevFrameState.destFileOffset);
      fQTTotNumSamples += numSamples;
      sampleNumberOfFrameStart = fQTTotNumSamples + 1;
    }

    if (avcHack && (*frameSource == H264_IDR_FRAME)) {
      SyncFrame* newSyncFrame = new SyncFrame(fQTTotNumSamples + 1);
      if (fTailSyncFrame == NULL) {
        fHeadSyncFrame = newSyncFrame;
      } else {
        fTailSyncFrame->nextSyncFrame = newSyncFrame;
      }
      fTailSyncFrame = newSyncFrame;
    }

    // Remember the current frame for next time:
    fPrevFrameState.frameSize = frameSize;
    fPrevFrameState.presentationTime = presentationTime;
    fPrevFrameState.destFileOffset = destFileOffset;
  }

  if (avcHack) fOurSink.addWord(frameSize);

  // Write the data into the file:
  fwrite(frameSource, 1, frameSize, fOurSink.fOutFid);

  // If we have a hint track, then write to it also:
  if (hasHintTrack()) {
    // Because presentation times are used for RTP packet timestamps,
    // we don't starting writing to the hint track until we've been synced:
    if (!fHaveBeenSynced) {
      fHaveBeenSynced
	= fOurSubsession.rtpSource()->hasBeenSynchronizedUsingRTCP();
    }
    if (fHaveBeenSynced) {
      fHintTrackForUs->useFrameForHinting(frameSize, presentationTime,
					  sampleNumberOfFrameStart);
    }
  }
}

void SubsessionIOState::useFrameForHinting(unsigned frameSize,
					   struct timeval presentationTime,
					   unsigned startSampleNumber) {
  // At this point, we have a single, combined frame - not individual packets.
  // For the hint track, we need to split the frame back up into separate packets.
  // However, for some RTP sources, then we also need to reuse the special
  // header bytes that were at the start of each of the RTP packets.
  Boolean hack263 = strcmp(fOurSubsession.codecName(), "H263-1998") == 0;
  Boolean hackm4a_generic = strcmp(fOurSubsession.mediumName(), "audio") == 0
    && strcmp(fOurSubsession.codecName(), "MPEG4-GENERIC") == 0;
  Boolean hackm4a_latm = strcmp(fOurSubsession.mediumName(), "audio") == 0
    && strcmp(fOurSubsession.codecName(), "MP4A-LATM") == 0;
  Boolean hackm4a = hackm4a_generic || hackm4a_latm;
  Boolean haveSpecialHeaders = (hack263 || hackm4a_generic);

  // If there has been a previous frame, then output a 'hint sample' for it.
  // (We use the current frame's presentation time to compute the previous
  // hint sample's duration.)
  RTPSource* const rs = fOurSubsession.rtpSource(); // abbrev
  struct timeval const& ppt = fPrevFrameState.presentationTime; //abbrev
  if (ppt.tv_sec != 0 || ppt.tv_usec != 0) {
    double duration = (presentationTime.tv_sec - ppt.tv_sec)
      + (presentationTime.tv_usec - ppt.tv_usec)/1000000.0;
    if (duration < 0.0) duration = 0.0;
    unsigned msDuration = (unsigned)(duration*1000); // milliseconds
    if (msDuration > fHINF.dmax) fHINF.dmax = msDuration;
    unsigned hintSampleDuration
      = (unsigned)((2*duration*fQTTimeScale+1)/2); // round
    if (hackm4a) {
      // Because multiple AAC frames can appear in a RTP packet, the presentation
      // times of the second and subsequent frames will not be accurate.
      // So, use the known "hintSampleDuration" instead:
      hintSampleDuration = fTrackHintedByUs->fQTTimeUnitsPerSample;

      // Also, if the 'time scale' was different from the RTP timestamp frequency,
      // (as can happen with aacPlus), then we need to scale "hintSampleDuration"
      // accordingly:
      if (fTrackHintedByUs->fQTTimeScale != fOurSubsession.rtpTimestampFrequency()) {
	unsigned const scalingFactor
	  = fOurSubsession.rtpTimestampFrequency()/fTrackHintedByUs->fQTTimeScale ;
	hintSampleDuration *= scalingFactor;
      }
    }

    int64_t const hintSampleDestFileOffset = TellFile64(fOurSink.fOutFid);

    unsigned const maxPacketSize = 1450;
    unsigned short numPTEntries
      = (fPrevFrameState.frameSize + (maxPacketSize-1))/maxPacketSize; // normal case
    unsigned char* immediateDataPtr = NULL;
    unsigned immediateDataBytesRemaining = 0;
    if (haveSpecialHeaders) { // special case
      numPTEntries = fPrevFrameState.numSpecialHeaders;
      immediateDataPtr = fPrevFrameState.specialHeaderBytes;
      immediateDataBytesRemaining
	= fPrevFrameState.specialHeaderBytesLength;
    }
    unsigned hintSampleSize
      = fOurSink.addHalfWord(numPTEntries);// Entry count
    hintSampleSize += fOurSink.addHalfWord(0x0000); // Reserved

    unsigned offsetWithinSample = 0;
    for (unsigned i = 0; i < numPTEntries; ++i) {
      // Output a Packet Table entry (representing a single RTP packet):
      unsigned short numDTEntries = 1;
      unsigned short seqNum = fPrevFrameState.seqNum++;
          // Note: This assumes that the input stream had no packets lost #####
      unsigned rtpHeader = fPrevFrameState.rtpHeader;
      if (i+1 < numPTEntries) {
	// This is not the last RTP packet, so clear the marker bit:
	rtpHeader &=~ (1<<23);
      }
      unsigned dataFrameSize = (i+1 < numPTEntries)
	? maxPacketSize : fPrevFrameState.frameSize - i*maxPacketSize; // normal case
      unsigned sampleNumber = fPrevFrameState.startSampleNumber;

      unsigned char immediateDataLen = 0;
      if (haveSpecialHeaders) { // special case
	++numDTEntries; // to include a Data Table entry for the special hdr
	if (immediateDataBytesRemaining > 0) {
	  if (hack263) {
	    immediateDataLen = *immediateDataPtr++;
	    --immediateDataBytesRemaining;
	    if (immediateDataLen > immediateDataBytesRemaining) {
	      // shouldn't happen (length byte was bad)
	      immediateDataLen = immediateDataBytesRemaining;
	    }
	  } else {
	    immediateDataLen = fPrevFrameState.specialHeaderBytesLength;
	  }
	}
	dataFrameSize = fPrevFrameState.packetSizes[i] - immediateDataLen;

	if (hack263) {
	  Boolean PbitSet
	    = immediateDataLen >= 1 && (immediateDataPtr[0]&0x4) != 0;
	  if (PbitSet) {
	    offsetWithinSample += 2; // to omit the two leading 0 bytes
	  }
	}
      }

      // Output the Packet Table:
      hintSampleSize += fOurSink.addWord(0); // Relative transmission time
      hintSampleSize += fOurSink.addWord(rtpHeader|seqNum);
          // RTP header info + RTP sequence number
      hintSampleSize += fOurSink.addHalfWord(0x0000); // Flags
      hintSampleSize += fOurSink.addHalfWord(numDTEntries); // Entry count
      unsigned totalPacketSize = 0;

      // Output the Data Table:
      if (haveSpecialHeaders) {
	//   use the "Immediate Data" format (1):
	hintSampleSize += fOurSink.addByte(1); // Source
	unsigned char len = immediateDataLen > 14 ? 14 : immediateDataLen;
	hintSampleSize += fOurSink.addByte(len); // Length
	totalPacketSize += len; fHINF.dimm += len;
	unsigned char j;
	for (j = 0; j < len; ++j) {
	  hintSampleSize += fOurSink.addByte(immediateDataPtr[j]); // Data
	}
	for (j = len; j < 14; ++j) {
	  hintSampleSize += fOurSink.addByte(0); // Data (padding)
	}

	immediateDataPtr += immediateDataLen;
	immediateDataBytesRemaining -= immediateDataLen;
      }
      //   use the "Sample Data" format (2):
      hintSampleSize += fOurSink.addByte(2); // Source
      hintSampleSize += fOurSink.addByte(0); // Track ref index
      hintSampleSize += fOurSink.addHalfWord(dataFrameSize); // Length
      totalPacketSize += dataFrameSize; fHINF.dmed += dataFrameSize;
      hintSampleSize += fOurSink.addWord(sampleNumber); // Sample number
      hintSampleSize += fOurSink.addWord(offsetWithinSample); // Offset
      // Get "bytes|samples per compression block" from the hinted track:
      unsigned short const bytesPerCompressionBlock
	= fTrackHintedByUs->fQTBytesPerFrame;
      unsigned short const samplesPerCompressionBlock
	= fTrackHintedByUs->fQTSamplesPerFrame;
      hintSampleSize += fOurSink.addHalfWord(bytesPerCompressionBlock);
      hintSampleSize += fOurSink.addHalfWord(samplesPerCompressionBlock);

      offsetWithinSample += dataFrameSize;// for the next iteration (if any)

      // Tally statistics for this packet:
      fHINF.nump += 1;
      fHINF.tpyl += totalPacketSize;
      totalPacketSize += 12; // add in the size of the RTP header
      fHINF.trpy += totalPacketSize;
      if (totalPacketSize > fHINF.pmax) fHINF.pmax = totalPacketSize;
    }

    // Make note of this completed hint sample frame:
    fQTTotNumSamples += useFrame1(hintSampleSize, ppt, hintSampleDuration,
				  hintSampleDestFileOffset);
  }

  // Remember this frame for next time:
  fPrevFrameState.frameSize = frameSize;
  fPrevFrameState.presentationTime = presentationTime;
  fPrevFrameState.startSampleNumber = startSampleNumber;
  fPrevFrameState.rtpHeader
    = rs->curPacketMarkerBit()<<23
    | (rs->rtpPayloadFormat()&0x7F)<<16;
  if (hack263) {
    H263plusVideoRTPSource* rs_263 = (H263plusVideoRTPSource*)rs;
    fPrevFrameState.numSpecialHeaders = rs_263->fNumSpecialHeaders;
    fPrevFrameState.specialHeaderBytesLength = rs_263->fSpecialHeaderBytesLength;
    unsigned i;
    for (i = 0; i < rs_263->fSpecialHeaderBytesLength; ++i) {
      fPrevFrameState.specialHeaderBytes[i] = rs_263->fSpecialHeaderBytes[i];
    }
    for (i = 0; i < rs_263->fNumSpecialHeaders; ++i) {
      fPrevFrameState.packetSizes[i] = rs_263->fPacketSizes[i];
    }
  } else if (hackm4a_generic) {
    // Synthesize a special header, so that this frame can be in its own RTP packet.
    unsigned const sizeLength = fOurSubsession.fmtp_sizelength();
    unsigned const indexLength = fOurSubsession.fmtp_indexlength();
    if (sizeLength + indexLength != 16) {
      envir() << "Warning: unexpected 'sizeLength' " << sizeLength
	      << " and 'indexLength' " << indexLength
	      << "seen when creating hint track\n";
    }
    fPrevFrameState.numSpecialHeaders = 1;
    fPrevFrameState.specialHeaderBytesLength = 4;
    fPrevFrameState.specialHeaderBytes[0] = 0; // AU_headers_length (high byte)
    fPrevFrameState.specialHeaderBytes[1] = 16; // AU_headers_length (low byte)
    fPrevFrameState.specialHeaderBytes[2] = ((frameSize<<indexLength)&0xFF00)>>8;
    fPrevFrameState.specialHeaderBytes[3] = (frameSize<<indexLength);
    fPrevFrameState.packetSizes[0]
      = fPrevFrameState.specialHeaderBytesLength + frameSize;
  }
}

unsigned SubsessionIOState::useFrame1(unsigned sourceDataSize,
				      struct timeval presentationTime,
				      unsigned frameDuration,
				      int64_t destFileOffset) {
  // Figure out the actual frame size for this data:
  unsigned frameSize = fQTBytesPerFrame;
  if (frameSize == 0) {
    // The entire packet data is assumed to be a frame:
    frameSize = sourceDataSize;
  }
  unsigned const numFrames = sourceDataSize/frameSize;
  unsigned const numSamples = numFrames*fQTSamplesPerFrame;

  // Record the information about which 'chunk' this data belongs to:
  ChunkDescriptor* newTailChunk;
  if (fTailChunk == NULL) {
    newTailChunk = fHeadChunk
      = new ChunkDescriptor(destFileOffset, sourceDataSize,
			    frameSize, frameDuration, presentationTime);
  } else {
    newTailChunk = fTailChunk->extendChunk(destFileOffset, sourceDataSize,
					   frameSize, frameDuration,
					   presentationTime);
  }
  if (newTailChunk != fTailChunk) {
   // This data created a new chunk, rather than extending the old one
    ++fNumChunks;
    fTailChunk = newTailChunk;
  }

  return numSamples;
}

void SubsessionIOState::onSourceClosure() {
  fOurSourceIsActive = False;
  fOurSink.onSourceClosure1();
}

Boolean SubsessionIOState::syncOK(struct timeval presentationTime) {
  QuickTimeFileSink& s = fOurSink; // abbreviation
  if (!s.fSyncStreams) return True; // we don't care

  if (s.fNumSyncedSubsessions < s.fNumSubsessions) {
    // Not all subsessions have yet been synced.  Check whether ours was
    // one of the unsynced ones, and, if so, whether it is now synced:
    if (!fHaveBeenSynced) {
      // We weren't synchronized before
      if (fOurSubsession.rtpSource()->hasBeenSynchronizedUsingRTCP()) {
	// H264 ?
	if (fQTMediaDataAtomCreator == &QuickTimeFileSink::addAtom_avc1) {
	  // special case: audio + H264 video: wait until audio is in sync
	  if ((s.fNumSubsessions == 2) && (s.fNumSyncedSubsessions < (s.fNumSubsessions - 1))) return False;

	  // if audio is in sync, wait for the next IDR frame to start
	  unsigned char* const frameSource = fBuffer->dataStart();
	  if (*frameSource != H264_IDR_FRAME) return False;
	}
	// But now we are
	fHaveBeenSynced = True;
	fSyncTime = presentationTime;
	++s.fNumSyncedSubsessions;

	if (timevalGE(fSyncTime, s.fNewestSyncTime)) {
	  s.fNewestSyncTime = fSyncTime;
	}
      }
    }
  }

  // Check again whether all subsessions have been synced:
  if (s.fNumSyncedSubsessions < s.fNumSubsessions) return False;