Mpeg2 TS stream parser for media source.

media/mpeg2/mpeg2ts_stream_parser.cc

Index: media/mpeg2/mpeg2ts_stream_parser.cc
diff --git a/media/mpeg2/mpeg2ts_stream_parser.cc b/media/mpeg2/mpeg2ts_stream_parser.cc
new file mode 100644
index 0000000000000000000000000000000000000000..bb54948274d89739eb5bd75271b5f558f88eccd3
--- /dev/null
+++ b/media/mpeg2/mpeg2ts_stream_parser.cc
@@ -0,0 +1,471 @@
+// Copyright (c) 2013 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 "media/mpeg2/mpeg2ts_stream_parser.h"
+
+#include "base/bind.h"
+#include "base/memory/scoped_ptr.h"
+#include "media/base/audio_decoder_config.h"
+#include "media/base/stream_parser_buffer.h"
+#include "media/base/video_decoder_config.h"
+#include "media/mpeg2/es_parser.h"
+#include "media/mpeg2/es_parser_adts.h"
+#include "media/mpeg2/es_parser_h264.h"
+#include "media/mpeg2/mpeg2ts_common.h"
+#include "media/mpeg2/mpeg2ts_pat.h"
+#include "media/mpeg2/mpeg2ts_pes.h"
+#include "media/mpeg2/mpeg2ts_pmt.h"
+#include "media/mpeg2/mpeg2ts_section_parser.h"
+#include "media/mpeg2/ts_packet.h"
+
+namespace {
+
+enum StreamType {
+  kStreamTypeMpeg1Audio = 0x3,
+  kStreamTypeAAC = 0xf,
+  kStreamTypeAVC = 0x1b,
+};
+
+}
+
+namespace media {
+namespace mpeg2ts {
+
+class PidState {
+ public:
+  int pid;
+  int continuity_counter;
+  scoped_ptr<Mpeg2TsSectionParser> section_parser;
+};
+
+Mpeg2TsStreamParser::Mpeg2TsStreamParser()
+  : pids_deleter_(&pids_),
+    selected_pmt_pid_(-1),
+    selected_audio_pid_(-1),
+    selected_video_pid_(-1),
+    is_initialized_(false),
+    segment_started_(false) {
+}
+
+Mpeg2TsStreamParser::~Mpeg2TsStreamParser() {
+}
+
+void Mpeg2TsStreamParser::Init(
+    const InitCB& init_cb,
+    const NewConfigCB& config_cb,
+    const NewBuffersCB& new_buffers_cb,
+    const NewTextBuffersCB& text_cb,
+    const NeedKeyCB& need_key_cb,
+    const AddTextTrackCB& add_text_track_cb,
+    const NewMediaSegmentCB& new_segment_cb,
+    const base::Closure& end_of_segment_cb,
+    const LogCB& log_cb) {
+  DCHECK(init_cb_.is_null());
+  DCHECK(!init_cb.is_null());
+  DCHECK(!config_cb.is_null());
+  DCHECK(!new_buffers_cb.is_null());
+  DCHECK(!need_key_cb.is_null());
+  DCHECK(!end_of_segment_cb.is_null());
+
+  init_cb_ = init_cb;
+  config_cb_ = config_cb;
+  new_buffers_cb_ = new_buffers_cb;
+  need_key_cb_ = need_key_cb;
+  new_segment_cb_ = new_segment_cb;
+  end_of_segment_cb_ = end_of_segment_cb;
+  log_cb_ = log_cb;
+}
+
+void Mpeg2TsStreamParser::Flush() {
+  // Flush the buffers and reset the pids.
+  for (std::map<int, PidState*>::iterator it = pids_.begin();
+       it != pids_.end(); ++it) {
+    LOG(INFO) << "Flushing PID: " << it->first;
+    PidState* pid_state = it->second;
+    pid_state->section_parser->Flush();
+    delete pid_state;
+  }
+  pids_.clear();
+  EmitRemainingBuffers();
+
+  // End of the segment.
+  end_of_segment_cb_.Run();
+  segment_started_ = false;
+
+  // Remove any bytes left in the TS buffer.
+  // (i.e. any partial TS packet => less than 188 bytes).
+  ts_buffer_.clear();
+
+  // Remove the programs and the corresponding PIDs.
+  programs_.clear();
+
+  // Reset the selected PIDs.
+  selected_pmt_pid_ = -1;
+  selected_audio_pid_ = -1;
+  selected_video_pid_ = -1;
+
+  // Reset the audio and video configs.
+  audio_config_.reset();
+  video_config_.reset();
+
+  // TODO(damienv): What to do with the remaining audio/video buffer queues ?
+  // Note: there should not be any.
+  LOG_IF(WARNING, !audio_buffer_queue_.empty())
+      << "Flush: audio buffer queue not empty";
+  LOG_IF(WARNING, !video_buffer_queue_.empty())
+      << "Flush: video buffer queue not empty";
+  audio_buffer_queue_.clear();
+  video_buffer_queue_.clear();
+}
+
+bool Mpeg2TsStreamParser::Parse(const uint8* buf, int size) {
+  LOG(INFO) << "Mpeg2TsStreamParser::Parse size=" << size;
+
+  // Add the data to the parser state.
+  int old_size = ts_buffer_.size();
+  ts_buffer_.resize(old_size + size);
+  memcpy(&ts_buffer_[old_size], buf, size);
+
+  int remaining_size = ts_buffer_.size();
+  int pos = 0;
+  while (remaining_size >= TsPacket::kPacketSize) {
+    // Synchronization.
+    int skipped_bytes = TsPacket::Sync(&ts_buffer_[pos], remaining_size);
+    if (skipped_bytes > 0) {
+      LOG(WARNING) << "Packet not aligned on a TS syncword:"
+                   << " skipped_bytes=" << skipped_bytes;
+      pos += skipped_bytes;
+      remaining_size -= skipped_bytes;
+      continue;
+    }
+
+    // Parse the TS header.
+    scoped_ptr<TsPacket> ts_packet(
+        TsPacket::Parse(&ts_buffer_[pos], remaining_size));
+    if (!ts_packet.get()) {
+      LOG(WARNING) << "Error: invalid TS packet";
+      pos += 1;
+      remaining_size -= 1;
+      continue;
+    }
+
+    VLOG(LOG_LEVEL_TS)
+        << "Processing PID=" << ts_packet->pid()
+        << " start_unit=" << ts_packet->payload_unit_start_indicator();
+
+    // Parse the section.
+    std::map<int, PidState*>::iterator it =
+        pids_.find(ts_packet->pid());
+    if (it == pids_.end() && ts_packet->pid() == 0) {

[acolwell GONE FROM CHROMIUM, 2013/08/29 20:44:24]

nit: Use Mpeg2TsSectionParser::kPidPat here. You might consider making
SpecialPid values top level constants in this namepace to avoid such a long
name. :)

[damienv1, 2013/09/04 01:37:14]

Done.

+      // Create the PAT state here if needed.
+      LOG(INFO) << "Create a new PAT parser";
+      PidState* pat_state = new PidState;
+      pat_state->section_parser.reset(
+          new Mpeg2TsPatParser(
+              base::Bind(&Mpeg2TsStreamParser::RegisterPmt,
+                         base::Unretained(this))));
+      it = pids_.insert(
+          std::pair<int, PidState*>(ts_packet->pid(), pat_state)).first;
+    }
+
+    LOG_IF(WARNING, it == pids_.end()) << "Ignoring TS packet for pid: "
+                                       << ts_packet->pid();
+    if (it != pids_.end()) {
+      it->second->section_parser->Parse(
+          ts_packet->payload_unit_start_indicator(),
+          &ts_buffer_[pos+ts_packet->GetPayloadOffset()],
+          ts_packet->GetPayloadSize());
+    }
+
+    // Go to the next packet.
+    pos += TsPacket::kPacketSize;
+    remaining_size -= TsPacket::kPacketSize;
+  }
+
+  // Check whether we are on a TS packet boundary.
+  if (remaining_size > 0) {
+    memmove(&ts_buffer_[0], &ts_buffer_[pos], remaining_size);
+    ts_buffer_.resize(remaining_size);
+  } else {
+    ts_buffer_.resize(0);
+  }
+
+  EmitRemainingBuffers();
+
+  return true;
+}
+
+void Mpeg2TsStreamParser::RegisterPmt(int program_number, int pmt_pid) {
+  LOG(INFO) << "RegisterPmt:"
+            << " program_number=" << program_number
+            << " pmt_pid=" << pmt_pid;
+  std::map<int, PidState*>::iterator it = pids_.find(pmt_pid);
+  if (it != pids_.end()) {
+    return;
+  }
+
+  if (selected_pmt_pid_ < 0) {
+    selected_pmt_pid_ = pmt_pid;
+  }
+  LOG_IF(WARNING, selected_pmt_pid_ != pmt_pid)
+      << "More than one program is defined";
+
+  // Create the PMT state here if needed.
+  LOG(INFO) << "Create a new PMT parser";
+  PidState* pmt_state = new PidState;
+  pmt_state->section_parser.reset(
+      new Mpeg2TsPmtParser(
+          base::Bind(&Mpeg2TsStreamParser::RegisterPes,
+                     base::Unretained(this), pmt_pid)));
+  pids_.insert(std::pair<int, PidState*>(pmt_pid, pmt_state));
+}
+
+void Mpeg2TsStreamParser::RegisterPes(int pmt_pid,
+                                      int pes_pid,
+                                      int stream_type) {
+  // TODO(damienv): check there is no mismatch if the entry already exists.
+  LOG(INFO) << "RegisterPes:"
+            << " pes_pid=" << pes_pid
+            << " stream_type=" << std::hex << stream_type << std::dec;
+  std::map<int, PidState*>::iterator it = pids_.find(pes_pid);
+  if (it != pids_.end()) {
+    return;
+  }
+
+  // TODO(damienv): add other formats if needed.
+  bool is_audio = (stream_type == kStreamTypeAAC);
+  bool is_video = (stream_type == kStreamTypeAVC);
+
+  // Update the active tracks.
+  // Select the audio/video tracks with the lowest PID.
+  if (pmt_pid == selected_pmt_pid_) {
+    if (is_audio &&
+        (selected_audio_pid_ < 0 || pes_pid < selected_audio_pid_)) {
+      selected_audio_pid_ = pes_pid;
+    }
+    if (is_video &&
+        (selected_video_pid_ < 0 || pes_pid < selected_video_pid_)) {
+      selected_video_pid_ = pes_pid;
+    }
+  }
+
+  // Create a stream parser corresponding to the stream type.
+  scoped_ptr<EsParser> es_parser;
+  if (stream_type == kStreamTypeAVC) {
+    es_parser.reset(
+        new EsParserH264(
+            base::Bind(&Mpeg2TsStreamParser::OnVideoConfigChanged,
+                       base::Unretained(this),
+                       pes_pid),
+            base::Bind(&Mpeg2TsStreamParser::OnEmitVideoBuffer,
+                       base::Unretained(this),
+                       pes_pid)));
+  } else if (stream_type == kStreamTypeAAC) {
+    es_parser.reset(
+        new EsParserAdts(
+            base::Bind(&Mpeg2TsStreamParser::OnAudioConfigChanged,
+                       base::Unretained(this),
+                       pes_pid),
+            base::Bind(&Mpeg2TsStreamParser::OnEmitAudioBuffer,
+                       base::Unretained(this),
+                       pes_pid)));
+  }
+
+  // Create the PES state here.
+  LOG(INFO) << "Create a new PES state";
+  PidState* pes_state = new PidState;
+  pes_state->section_parser.reset(
+      new Mpeg2TsPesParser(es_parser.release()));
+  pids_.insert(std::pair<int, PidState*>(pes_pid, pes_state));
+}
+
+void Mpeg2TsStreamParser::OnVideoConfigChanged(
+    int pes_pid,
+    const VideoDecoderConfig& video_decoder_config) {
+  DCHECK_GT(selected_video_pid_, 0);
+  LOG(INFO) << "OnVideoConfigChanged for pid=" << pes_pid;
+  if (pes_pid != selected_video_pid_) {
+    return;
+  }
+
+  video_config_.reset(new VideoDecoderConfig());
+  *video_config_ = video_decoder_config;
+
+  OnAudioVideoConfigChanged();
+}
+
+void Mpeg2TsStreamParser::OnAudioConfigChanged(
+    int pes_pid,
+    const AudioDecoderConfig& audio_decoder_config) {
+  LOG(INFO) << "OnAudioConfigChanged";
+  if (pes_pid != selected_audio_pid_) {
+    return;
+  }
+
+  audio_config_.reset(new AudioDecoderConfig());
+  *audio_config_ = audio_decoder_config;
+
+  OnAudioVideoConfigChanged();
+}
+
+void Mpeg2TsStreamParser::OnAudioVideoConfigChanged() {
+  if (selected_audio_pid_ > 0 && !audio_config_) {
+    // Need to get the audio config as well before going any further.
+    return;
+  }
+  if (selected_video_pid_ > 0 && !video_config_) {
+    // Need to get the video config as well before going any further.
+    return;
+  }
+
+  // Emit pending buffers only if already initialized.
+  // Buffers need to be emitted since these buffers do not necesseraly
+  // have the same config.
+  if (is_initialized_) {
+    // TODO(damienv): at initialization time, although unlikely,
+    // some situations like:
+    // VConfig0 VBuffer0 VBuffer1 VConfig1 VBuffer0 AConfig0 ABuffer0
+    // cannot be handled properly since video buffers will be assigned the same
+    // video config.
+    EmitRemainingBuffers();
+  }
+
+  if (selected_audio_pid_ > 0 && selected_video_pid_ > 0) {
+    // Streams with both audio and video.
+    config_cb_.Run(*audio_config_, *video_config_);
+  } else if (selected_video_pid_ > 0) {
+    // Video stream only.
+    AudioDecoderConfig audio_config;
+    config_cb_.Run(audio_config, *video_config_);
+  } else if (selected_audio_pid_ > 0) {
+    // Audio stream only.
+    VideoDecoderConfig video_config;
+    config_cb_.Run(*audio_config_, video_config);
+  } else {
+    // At least one audio or video config should be set.
+    NOTREACHED();
+  }
+
+  if (!is_initialized_) {
+    // For Mpeg2 TS we don't know the stream duration.
+    LOG(INFO) << "Mpeg2TS stream parser initialization done";
+    init_cb_.Run(true, base::TimeDelta());
+    is_initialized_ = true;
+  }
+}
+
+void Mpeg2TsStreamParser::OnEmitAudioBuffer(
+    int pes_pid,
+    scoped_refptr<StreamParserBuffer> stream_parser_buffer) {
+  // Since ChunkDemuxer currently handle only one audio and one video,
+  // the PID filter must be done inside the TS parser.
+  // Eventually, the PID filter will be done on the JS application side
+  // (using the HTML5 video element track selection).
+  if (pes_pid != selected_audio_pid_) {
+    return;
+  }
+
+  VLOG(LOG_LEVEL_ES)
+      << "OnEmitAudioBuffer: "
+      << " size="
+      << stream_parser_buffer->data_size()
+      << " dts="
+      << stream_parser_buffer->GetDecodeTimestamp().InMilliseconds()
+      << " pts="
+      << stream_parser_buffer->timestamp().InMilliseconds();
+  stream_parser_buffer->set_timestamp(
+      stream_parser_buffer->timestamp() - time_offset_);
+  stream_parser_buffer->SetDecodeTimestamp(
+      stream_parser_buffer->GetDecodeTimestamp() - time_offset_);
+  audio_buffer_queue_.push_back(stream_parser_buffer);
+}
+
+void Mpeg2TsStreamParser::OnEmitVideoBuffer(
+    int pes_pid,
+    scoped_refptr<StreamParserBuffer> stream_parser_buffer) {
+  // Since ChunkDemuxer currently handle only one audio and one video,
+  // the PID filter must be done inside the TS parser.
+  // Eventually, the PID filter will be done on the JS application side
+  // (using the HTML5 video element track selection).
+  if (pes_pid != selected_video_pid_) {
+    return;
+  }
+
+  VLOG(LOG_LEVEL_ES)
+      << "OnEmitVideoBuffer"
+      << " size="
+      << stream_parser_buffer->data_size()
+      << " dts="
+      << stream_parser_buffer->GetDecodeTimestamp().InMilliseconds()
+      << " pts="
+      << stream_parser_buffer->timestamp().InMilliseconds()
+      << " IsKeyframe="
+      << stream_parser_buffer->IsKeyframe();
+  stream_parser_buffer->set_timestamp(
+      stream_parser_buffer->timestamp() - time_offset_);
+  stream_parser_buffer->SetDecodeTimestamp(
+      stream_parser_buffer->GetDecodeTimestamp() - time_offset_);
+  video_buffer_queue_.push_back(stream_parser_buffer);
+}
+
+void Mpeg2TsStreamParser::EmitRemainingBuffers() {
+  VLOG(LOG_LEVEL_ES) << "Mpeg2TsStreamParser::EmitRemainingBuffers";
+
+  // Segment cannot start with non key frames
+  // so replace non key frames with the 1st IDR.
+  // TODO(damienv): Non key frames could be replaced with a stuffing NAL
+  // as well, the most important is to avoid any gap between frames
+  // as MSE does not allow sparse video/audio buffers.
+  if (!segment_started_) {
+    StreamParser::BufferQueue::iterator it = video_buffer_queue_.begin();
+    for ( ; it != video_buffer_queue_.end(); ++it) {
+      if ((*it)->IsKeyframe()) {
+        break;
+      }
+    }
+    if (it != video_buffer_queue_.end()) {
+      StreamParser::BufferQueue::iterator it2 = video_buffer_queue_.begin();
+      while (!(*it2)->IsKeyframe()) {
+        base::TimeDelta dts = (*it2)->GetDecodeTimestamp();
+        base::TimeDelta pts = (*it2)->timestamp();
+        scoped_refptr<StreamParserBuffer> stream_parser_buffer =
+            StreamParserBuffer::CopyFrom(
+                (*it)->data(), (*it)->data_size(), true);
+        stream_parser_buffer->set_timestamp(pts);
+        stream_parser_buffer->SetDecodeTimestamp(dts);
+        LOG(WARNING) << "Replacing frame with an IDR @ pts="
+                     << pts.InMilliseconds();
+        *it2 = stream_parser_buffer;
+        ++it2;
+      }
+    } else {
+      // TODO(damienv): is there a better way to handle this situation ?
+      LOG(WARNING) << "Only non key frames in the buffer queue";
+      video_buffer_queue_.clear();
+    }
+  }
+
+  // Possibly start a segment if not done yet.
+  StartSegmentIfNeeded();
+
+  // Finally, add the video and audio buffers.
+  if (!video_buffer_queue_.empty() ||
+      !audio_buffer_queue_.empty()) {
+    new_buffers_cb_.Run(audio_buffer_queue_, video_buffer_queue_);
+    audio_buffer_queue_.clear();
+    video_buffer_queue_.clear();
+  }
+}
+
+void Mpeg2TsStreamParser::StartSegmentIfNeeded() {
+  if (segment_started_) {
+    return;
+  }
+  LOG(INFO) << "Starting a new segment";
+  segment_started_ = true;
+  new_segment_cb_.Run();
+}
+
+}  // namespace mpeg2ts
+}  // namespace media