view libmpdemux/demux_rtp_codec.cpp @ 16429:84174804804b

Updates to NUT spec: 1. remove average_bitrate 2. add other_stream_header, for subtitles and metadata 3. add max_pts to index 4. index_ptr - a 64 bit integer to say the total length of all index packets 5. specify how to write "multiple" indexes 6. change forward_ptr behavior, starts right after forward_ptr, ends after checksum 7. remove stream_id <-> stream_class limitation. 8. time_base_nom must also be non zero. 9. rename time_base_nom and time_base_denom, now timebase means the length of a tick, not amounts of ticks 10. remove (old?) sample_rate_mul stuff. 11. specify what exactly the checksum covers. 12. specify that stream classes which have multiple streams must have an info packet.. (in new Semantic requirements section) 13. Rename 'timestamp' to pts. 14. Change date of draft... 15. Add myself to authors...

////////// Codec-specific routines used to interface between "MPlayer"
////////// and the "LIVE.COM Streaming Media" libraries:

#include "demux_rtp_internal.h"
extern "C" {
#include "stheader.h"
}

static void
needVideoFrameRate(demuxer_t* demuxer, MediaSubsession* subsession); // forward
static Boolean
parseQTState_video(QuickTimeGenericRTPSource::QTState const& qtState,
		   unsigned& fourcc); // forward
static Boolean
parseQTState_audio(QuickTimeGenericRTPSource::QTState const& qtState,
		   unsigned& fourcc, unsigned& numChannels); // forward
		       
void rtpCodecInitialize_video(demuxer_t* demuxer,
			      MediaSubsession* subsession,
			      unsigned& flags) {
  flags = 0;
  // Create a dummy video stream header
  // to make the main MPlayer code happy:
  sh_video_t* sh_video = new_sh_video(demuxer,0);
  BITMAPINFOHEADER* bih
    = (BITMAPINFOHEADER*)calloc(1,sizeof(BITMAPINFOHEADER));
  bih->biSize = sizeof(BITMAPINFOHEADER);
  sh_video->bih = bih;
  demux_stream_t* d_video = demuxer->video;
  d_video->sh = sh_video; sh_video->ds = d_video;
  
  // Map known video MIME types to the BITMAPINFOHEADER parameters
  // that this program uses.  (Note that not all types need all
  // of the parameters to be set.)
  if (strcmp(subsession->codecName(), "MPV") == 0) {
    flags |= RTPSTATE_IS_MPEG12_VIDEO;
  } else if (strcmp(subsession->codecName(), "MP1S") == 0 ||
	     strcmp(subsession->codecName(), "MP2T") == 0) {
    flags |= RTPSTATE_IS_MPEG12_VIDEO|RTPSTATE_IS_MULTIPLEXED;
  } else if (strcmp(subsession->codecName(), "H263") == 0 ||
	     strcmp(subsession->codecName(), "H263-1998") == 0) {
    bih->biCompression = sh_video->format
      = mmioFOURCC('H','2','6','3');
    needVideoFrameRate(demuxer, subsession);
  } else if (strcmp(subsession->codecName(), "H261") == 0) {
    bih->biCompression = sh_video->format
      = mmioFOURCC('H','2','6','1');
    needVideoFrameRate(demuxer, subsession);
  } else if (strcmp(subsession->codecName(), "JPEG") == 0) {
    bih->biCompression = sh_video->format
      = mmioFOURCC('M','J','P','G');
    needVideoFrameRate(demuxer, subsession);
  } else if (strcmp(subsession->codecName(), "MP4V-ES") == 0) {
    bih->biCompression = sh_video->format
      = mmioFOURCC('m','p','4','v');
    // For the codec to work correctly, it may need a 'VOL Header' to be
    // inserted at the front of the data stream.  Construct this from the
    // "config" MIME parameter, which was present (hopefully) in the
    // session's SDP description:
    unsigned configLen;
    unsigned char* configData
      = parseGeneralConfigStr(subsession->fmtp_config(), configLen);
    insertRTPData(demuxer, demuxer->video, configData, configLen);
    needVideoFrameRate(demuxer, subsession);
  } else if (strcmp(subsession->codecName(), "X-QT") == 0 ||
	     strcmp(subsession->codecName(), "X-QUICKTIME") == 0) {
    // QuickTime generic RTP format, as described in
    // http://developer.apple.com/quicktime/icefloe/dispatch026.html

    // We can't initialize this stream until we've received the first packet
    // that has QuickTime "sdAtom" information in the header.  So, keep
    // reading packets until we get one:
    unsigned char* packetData; unsigned packetDataLen; float pts;
    QuickTimeGenericRTPSource* qtRTPSource
      = (QuickTimeGenericRTPSource*)(subsession->rtpSource());
    unsigned fourcc;
    do {
      if (!awaitRTPPacket(demuxer, demuxer->video,
			  packetData, packetDataLen, pts)) {
	return;
      }
    } while (!parseQTState_video(qtRTPSource->qtState, fourcc));

    bih->biCompression = sh_video->format = fourcc;
  } else {
    fprintf(stderr,
	    "Unknown MPlayer format code for MIME type \"video/%s\"\n",
	    subsession->codecName());
  }
}

void rtpCodecInitialize_audio(demuxer_t* demuxer,
			      MediaSubsession* subsession,
			      unsigned& flags) {
  flags = 0;
  // Create a dummy audio stream header
  // to make the main MPlayer code happy:
  sh_audio_t* sh_audio = new_sh_audio(demuxer,0);
  WAVEFORMATEX* wf = (WAVEFORMATEX*)calloc(1,sizeof(WAVEFORMATEX));
  sh_audio->wf = wf;
  demux_stream_t* d_audio = demuxer->audio;
  d_audio->sh = sh_audio; sh_audio->ds = d_audio;
  
  wf->nChannels = subsession->numChannels();

  // Map known audio MIME types to the WAVEFORMATEX parameters
  // that this program uses.  (Note that not all types need all
  // of the parameters to be set.)
  wf->nSamplesPerSec
    = subsession->rtpSource()->timestampFrequency(); // by default
  if (strcmp(subsession->codecName(), "MPA") == 0 ||
      strcmp(subsession->codecName(), "MPA-ROBUST") == 0 ||
      strcmp(subsession->codecName(), "X-MP3-DRAFT-00") == 0) {
    wf->wFormatTag = sh_audio->format = 0x55;
    // Note: 0x55 is for layer III, but should work for I,II also
    wf->nSamplesPerSec = 0; // sample rate is deduced from the data
  } else if (strcmp(subsession->codecName(), "AC3") == 0) {
    wf->wFormatTag = sh_audio->format = 0x2000;
    wf->nSamplesPerSec = 0; // sample rate is deduced from the data
  } else if (strcmp(subsession->codecName(), "L16") == 0) {
    wf->wFormatTag = sh_audio->format = 0x736f7774; // "twos"
    wf->nBlockAlign = 1;
    wf->wBitsPerSample = 16;
    wf->cbSize = 0;
  } else if (strcmp(subsession->codecName(), "L8") == 0) {
    wf->wFormatTag = sh_audio->format = 0x20776172; // "raw "
    wf->nBlockAlign = 1;
    wf->wBitsPerSample = 8;
    wf->cbSize = 0;
  } else if (strcmp(subsession->codecName(), "PCMU") == 0) {
    wf->wFormatTag = sh_audio->format = 0x7;
    wf->nAvgBytesPerSec = 8000;
    wf->nBlockAlign = 1;
    wf->wBitsPerSample = 8;
    wf->cbSize = 0;
  } else if (strcmp(subsession->codecName(), "PCMA") == 0) {
    wf->wFormatTag = sh_audio->format = 0x6;
    wf->nAvgBytesPerSec = 8000;
    wf->nBlockAlign = 1;
    wf->wBitsPerSample = 8;
    wf->cbSize = 0;
  } else if (strcmp(subsession->codecName(), "GSM") == 0) {
    wf->wFormatTag = sh_audio->format = mmioFOURCC('a','g','s','m');
    wf->nAvgBytesPerSec = 1650;
    wf->nBlockAlign = 33;
    wf->wBitsPerSample = 16;
    wf->cbSize = 0;
  } else if (strcmp(subsession->codecName(), "QCELP") == 0) {
    wf->wFormatTag = sh_audio->format = mmioFOURCC('Q','c','l','p');
    wf->nAvgBytesPerSec = 1750;
    wf->nBlockAlign = 35;
    wf->wBitsPerSample = 16;
    wf->cbSize = 0;
  } else if (strcmp(subsession->codecName(), "MP4A-LATM") == 0) {
    wf->wFormatTag = sh_audio->format = mmioFOURCC('m','p','4','a');
    // For the codec to work correctly, it needs "AudioSpecificConfig"
    // data, which is parsed from the "StreamMuxConfig" string that
    // was present (hopefully) in the SDP description:
    unsigned codecdata_len;
    sh_audio->codecdata
      = parseStreamMuxConfigStr(subsession->fmtp_config(),
				codecdata_len);
    sh_audio->codecdata_len = codecdata_len;
  } else if (strcmp(subsession->codecName(), "MPEG4-GENERIC") == 0) {
    wf->wFormatTag = sh_audio->format = mmioFOURCC('m','p','4','a');
    // For the codec to work correctly, it needs "AudioSpecificConfig"
    // data, which was present (hopefully) in the SDP description:
    unsigned codecdata_len;
    sh_audio->codecdata
      = parseGeneralConfigStr(subsession->fmtp_config(),
			      codecdata_len);
    sh_audio->codecdata_len = codecdata_len;
  } else if (strcmp(subsession->codecName(), "X-QT") == 0 ||
	     strcmp(subsession->codecName(), "X-QUICKTIME") == 0) {
    // QuickTime generic RTP format, as described in
    // http://developer.apple.com/quicktime/icefloe/dispatch026.html

    // We can't initialize this stream until we've received the first packet
    // that has QuickTime "sdAtom" information in the header.  So, keep
    // reading packets until we get one:
    unsigned char* packetData; unsigned packetDataLen; float pts;
    QuickTimeGenericRTPSource* qtRTPSource
      = (QuickTimeGenericRTPSource*)(subsession->rtpSource());
    unsigned fourcc, numChannels;
    do {
      if (!awaitRTPPacket(demuxer, demuxer->audio,
			  packetData, packetDataLen, pts)) {
	return;
      }
    } while (!parseQTState_audio(qtRTPSource->qtState, fourcc, numChannels));

    wf->wFormatTag = sh_audio->format = fourcc;
    wf->nChannels = numChannels;
  } else {
    fprintf(stderr,
	    "Unknown MPlayer format code for MIME type \"audio/%s\"\n",
	    subsession->codecName());
  }
}

static void needVideoFrameRate(demuxer_t* demuxer,
			       MediaSubsession* subsession) {
  // For some codecs, MPlayer's decoding software can't (or refuses to :-)
  // figure out the frame rate by itself, so (unless the user specifies
  // it manually, using "-fps") we figure it out ourselves here, using the
  // presentation timestamps in successive packets,
  extern float force_fps; if (force_fps != 0.0) return; // user used "-fps"

  demux_stream_t* d_video = demuxer->video;
  sh_video_t* sh_video = (sh_video_t*)(d_video->sh);

  // If we already know the subsession's video frame rate, use it:
  int fps = (int)(subsession->videoFPS());
  if (fps != 0) {
    sh_video->fps = fps;
    return;
  }
  
  // Keep looking at incoming frames until we see two with different,
  // non-zero "pts" timestamps:
  unsigned char* packetData; unsigned packetDataLen;
  float lastPTS = 0.0, curPTS;
  unsigned const maxNumFramesToWaitFor = 300;
  for (unsigned i = 0; i < maxNumFramesToWaitFor; ++i) {
    if (!awaitRTPPacket(demuxer, d_video, packetData, packetDataLen, curPTS)) {
      break;
    }

    if (curPTS > lastPTS && lastPTS != 0.0) {
      // Use the difference between these two "pts"s to guess the frame rate.
      // (should really check that there were no missing frames inbetween)#####
      // Guess the frame rate as an integer.  If it's not, use "-fps" instead.
      fps = (int)(1/(curPTS-lastPTS) + 0.5); // rounding
      fprintf(stderr, "demux_rtp: Guessed the video frame rate as %d frames-per-second.\n\t(If this is wrong, use the \"-fps <frame-rate>\" option instead.)\n", fps);
      sh_video->fps = fps;
      return;
    }
    lastPTS = curPTS;
  }
  fprintf(stderr, "demux_rtp: Failed to guess the video frame rate\n");
}

static Boolean
parseQTState_video(QuickTimeGenericRTPSource::QTState const& qtState,
		   unsigned& fourcc) {
  // qtState's "sdAtom" field is supposed to contain a QuickTime video
  // 'sample description' atom.  This atom's name is the 'fourcc' that we want:
  char const* sdAtom = qtState.sdAtom;
  if (sdAtom == NULL || qtState.sdAtomSize < 2*4) return False;

  fourcc = *(unsigned*)(&sdAtom[4]); // put in host order
  return True;
}

static Boolean
parseQTState_audio(QuickTimeGenericRTPSource::QTState const& qtState,
		   unsigned& fourcc, unsigned& numChannels) {
  // qtState's "sdAtom" field is supposed to contain a QuickTime audio
  // 'sample description' atom.  This atom's name is the 'fourcc' that we want.
  // Also, the top half of the 5th word following the atom name should
  // contain the number of channels ("numChannels") that we want:
  char const* sdAtom = qtState.sdAtom;
  if (sdAtom == NULL || qtState.sdAtomSize < 7*4) return False;

  fourcc = *(unsigned*)(&sdAtom[4]); // put in host order

  char const* word7Ptr = &sdAtom[6*4];
  numChannels = (word7Ptr[0]<<8)|(word7Ptr[1]);
  return True;
}