Add audio playback (all platforms) to cast_receiver_app.

media/cast/test/receiver.cc

Index: media/cast/test/receiver.cc
diff --git a/media/cast/test/receiver.cc b/media/cast/test/receiver.cc
index f55656b7e9e25f963a65baaaafb5c639f4a3a2cd..a8eb7dd651cca344a22b9bc7b1128d336025a3f9 100644
--- a/media/cast/test/receiver.cc
+++ b/media/cast/test/receiver.cc
@@ -6,7 +6,9 @@
 #include <climits>
 #include <cstdarg>
 #include <cstdio>
+#include <deque>
 #include <string>
+#include <utility>
 
 #include "base/at_exit.h"
 #include "base/command_line.h"
@@ -14,9 +16,17 @@
 #include "base/memory/ref_counted.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/message_loop/message_loop.h"
+#include "base/synchronization/lock.h"
+#include "base/synchronization/waitable_event.h"
 #include "base/threading/thread.h"
 #include "base/time/default_tick_clock.h"
+#include "base/timer/timer.h"
+#include "media/audio/audio_io.h"
+#include "media/audio/audio_manager.h"
+#include "media/audio/audio_parameters.h"
+#include "media/audio/fake_audio_log_factory.h"
 #include "media/base/audio_bus.h"
+#include "media/base/channel_layout.h"
 #include "media/base/video_frame.h"
 #include "media/cast/cast_config.h"
 #include "media/cast/cast_environment.h"
@@ -127,62 +137,312 @@ VideoReceiverConfig GetVideoReceiverConfig() {
   return video_config;
 }
 
-// An InProcessReceiver that renders video frames to a LinuxOutputWindow.  While
-// it does receive audio frames, it does not play them.
-class ReceiverDisplay : public InProcessReceiver {
+AudioParameters ToAudioParameters(const AudioReceiverConfig& config) {
+  const int samples_in_10ms = config.frequency / 100;
+  return AudioParameters(AudioParameters::AUDIO_PCM_LOW_LATENCY,
+                         GuessChannelLayout(config.channels),
+                         config.frequency, 32, samples_in_10ms);
+}
+
+// An InProcessReceiver that renders video frames to a LinuxOutputWindow and
+// audio frames via Chromium's audio stack.
+//
+// InProcessReceiver pushes audio and video frames to this subclass, and these
+// frames are pushed into a queue.  Then, for audio, the Chromium audio stack
+// will make polling calls on a separate, unknown thread whereby audio frames
+// are pulled out of the audio queue as needed.  For video, however, NaivePlayer
+// is responsible for scheduling updates to the screen itself.  For both, the
+// queues are pruned (i.e., received frames are skipped) when the system is not
+// able to play back as fast as frames are entering the queue.
+//
+// This is NOT a good reference implementation for a Cast receiver player since:
+// 1. It only skips frames to handle slower-than-expected playout, or halts
+//    playback to handle frame underruns.
+// 2. It makes no attempt to synchronize the timing of playout of the video
+//    frames with the audio frames.
+// 3. It does nothing to smooth or hide discontinuities in playback due to
+//    timing issues or missing frames.
+class NaivePlayer : public InProcessReceiver,
+                    public AudioOutputStream::AudioSourceCallback {
  public:
-  ReceiverDisplay(const scoped_refptr<CastEnvironment>& cast_environment,
-                  const net::IPEndPoint& local_end_point,
-                  const net::IPEndPoint& remote_end_point,
-                  const AudioReceiverConfig& audio_config,
-                  const VideoReceiverConfig& video_config)
+  NaivePlayer(const scoped_refptr<CastEnvironment>& cast_environment,
+              const net::IPEndPoint& local_end_point,
+              const net::IPEndPoint& remote_end_point,
+              const AudioReceiverConfig& audio_config,
+              const VideoReceiverConfig& video_config)
       : InProcessReceiver(cast_environment,
                           local_end_point,
                           remote_end_point,
                           audio_config,
                           video_config),
+        max_frame_age_(base::TimeDelta::FromSeconds(1) * 3 /

[imcheng, 2014/04/09 01:34:45]

Could you add a comment here about how you arrived at this number?

[miu, 2014/04/09 04:07:19]

Done.

+                           video_config.max_frame_rate),
 #if defined(OS_LINUX)
         render_(0, 0, kVideoWindowWidth, kVideoWindowHeight, "Cast_receiver"),
 #endif  // OS_LINUX
-        last_playout_time_(),
-        last_render_time_() {
+        num_video_frames_processed_(0),
+        num_audio_frames_processed_(0),
+        currently_playing_audio_frame_start_(-1) {}
+
+  virtual ~NaivePlayer() {}
+
+  virtual void Start() OVERRIDE {
+    AudioManager::Get()->GetTaskRunner()->PostTask(
+        FROM_HERE,
+        base::Bind(&NaivePlayer::StartAudioOutputOnAudioManagerThread,
+                   base::Unretained(this)));
+    InProcessReceiver::Start();

[imcheng, 2014/04/09 01:34:45]

Do we need to wait for StartAudioOutputOnAudioManagerThread to finish before
calling this, similar to Stop() below?

[miu, 2014/04/09 04:07:19]

Added comment to explain.

   }
 
-  virtual ~ReceiverDisplay() {}
+  virtual void Stop() OVERRIDE {
+    // First, stop audio output to the Chromium audio stack.
+    base::WaitableEvent done(false, false);
+    DCHECK(!AudioManager::Get()->GetTaskRunner()->BelongsToCurrentThread());

[imcheng, 2014/04/09 01:34:45]

Do we also need this DCHECK in Start()?

[miu, 2014/04/09 04:07:19]

No, it doesn't matter there since the thread is not explicitly blocked on
anything.

+    AudioManager::Get()->GetTaskRunner()->PostTask(
+        FROM_HERE,
+        base::Bind(&NaivePlayer::StopAudioOutputOnAudioManagerThread,
+                   base::Unretained(this),
+                   &done));
+    done.Wait();
+
+    // Now, stop receiving new frames.
+    InProcessReceiver::Stop();
+
+    // Finally, clear out any frames remaining in the queues.
+    while (!audio_playout_queue_.empty()) {

[imcheng, 2014/04/09 01:34:45]

If linked_ptr is used, then you won't have do this manually.

+      const scoped_ptr<AudioBus> to_be_deleted(
+          audio_playout_queue_.front().second);
+      audio_playout_queue_.pop_front();
+    }
+    video_playout_queue_.clear();
+  }
 
- protected:
-  virtual void OnVideoFrame(const scoped_refptr<media::VideoFrame>& video_frame,
-                            const base::TimeTicks& render_time,
-                            bool is_continuous) OVERRIDE {
-#ifdef OS_LINUX
-    render_.RenderFrame(video_frame);
-#endif  // OS_LINUX
-    // Print out the delta between frames.
-    if (!last_render_time_.is_null()) {
-      base::TimeDelta time_diff = render_time - last_render_time_;
-      VLOG(2) << "Size = " << video_frame->coded_size().ToString()
-              << "; RenderDelay[mS] =  " << time_diff.InMilliseconds();
+ private:
+  void StartAudioOutputOnAudioManagerThread() {
+    DCHECK(AudioManager::Get()->GetTaskRunner()->BelongsToCurrentThread());
+    audio_output_stream_.reset(AudioManager::Get()->MakeAudioOutputStreamProxy(

[imcheng, 2014/04/09 01:34:45]

It looks like you will need to call Stop() and Close() before resetting, just in
case this function is called while the stream is running.

[miu, 2014/04/09 04:07:19]

Instead, I added a DCHECK(!audio_output_stream_) since it should be invalid to
call Start() more than once.

+        ToAudioParameters(audio_config()), ""));
+    if (audio_output_stream_.get() && audio_output_stream_->Open()) {
+      audio_output_stream_->Start(this);
+    } else {
+      LOG(ERROR) << "Failed to open an audio output stream.  "
+                 << "Audio playback disabled.";
+      audio_output_stream_.reset();
     }
-    last_render_time_ = render_time;
+  }
+
+  void StopAudioOutputOnAudioManagerThread(base::WaitableEvent* done) {
+    DCHECK(AudioManager::Get()->GetTaskRunner()->BelongsToCurrentThread());
+    if (audio_output_stream_.get()) {
+      audio_output_stream_->Stop();
+      audio_output_stream_->Close();
+      audio_output_stream_.reset();
+    }
+    done->Signal();
+  }
+
+  ////////////////////////////////////////////////////////////////////
+  // InProcessReceiver overrides.
+
+  virtual void OnVideoFrame(const scoped_refptr<VideoFrame>& video_frame,
+                            const base::TimeTicks& playout_time,
+                            bool is_continuous) OVERRIDE {
+    DCHECK(cast_env()->CurrentlyOn(CastEnvironment::MAIN));
+    LOG_IF(WARNING, !is_continuous)
+        << "Video: Discontinuity in received frames.";
+    video_playout_queue_.push_back(std::make_pair(playout_time, video_frame));
+    ScheduleVideoPlayout();
   }
 
   virtual void OnAudioFrame(scoped_ptr<AudioBus> audio_frame,
                             const base::TimeTicks& playout_time,
                             bool is_continuous) OVERRIDE {
-    // For audio just print the playout delta between audio frames.
-    if (!last_playout_time_.is_null()) {
-      base::TimeDelta time_diff = playout_time - last_playout_time_;
-      VLOG(2) << "SampleRate = " << audio_config().frequency
-              << "; PlayoutDelay[mS] =  " << time_diff.InMilliseconds();
+    DCHECK(cast_env()->CurrentlyOn(CastEnvironment::MAIN));
+    LOG_IF(WARNING, !is_continuous)
+        << "Audio: Discontinuity in received frames.";
+    base::AutoLock auto_lock(audio_lock_);
+    audio_playout_queue_.push_back(
+        std::make_pair(playout_time, audio_frame.release()));
+  }
+
+  // End of InProcessReceiver overrides.
+  ////////////////////////////////////////////////////////////////////
+
+  ////////////////////////////////////////////////////////////////////
+  // AudioSourceCallback implementation.
+
+  virtual int OnMoreData(AudioBus* dest, AudioBuffersState buffers_state)
+      OVERRIDE {
+    // Note: This method is being invoked by a separate thread unknown to us
+    // (i.e., outside of CastEnvironment).
+
+    int samples_remaining = dest->frames();
+
+    while (samples_remaining > 0) {
+      // Get next audio frame ready for playout.
+      if (!currently_playing_audio_frame_.get()) {
+        base::AutoLock auto_lock(audio_lock_);
+
+        // Prune the queue, skipping entries that are too old.
+        // TODO(miu): Use |buffers_state| to account for audio buffering delays
+        // upstream.
+        const base::TimeTicks earliest_time_to_play =
+            cast_env()->Clock()->NowTicks() - max_frame_age_;
+        while (!audio_playout_queue_.empty() &&
+               audio_playout_queue_.front().first < earliest_time_to_play) {
+          PopOneAudioFrame(true);
+        }
+        if (audio_playout_queue_.empty())
+          break;
+
+        currently_playing_audio_frame_.reset(
+            audio_playout_queue_.front().second);
+        currently_playing_audio_frame_start_ = 0;
+        PopOneAudioFrame(false);
+      }
+
+      // Copy some or all of the samples in |currently_playing_audio_frame_| to
+      // |dest|.  Once all samples in |currently_playing_audio_frame_| have been
+      // consumed, release it.
+      const int num_samples_to_copy =
+          std::min(samples_remaining,
+                   currently_playing_audio_frame_->frames() -
+                       currently_playing_audio_frame_start_);
+      currently_playing_audio_frame_->CopyPartialFramesTo(
+          currently_playing_audio_frame_start_,
+          num_samples_to_copy,
+          0,
+          dest);
+      samples_remaining -= num_samples_to_copy;
+      currently_playing_audio_frame_start_ += num_samples_to_copy;
+      if (currently_playing_audio_frame_start_ ==
+              currently_playing_audio_frame_->frames()) {
+        currently_playing_audio_frame_.reset();
+      }
+    }
+
+    // If |dest| has not been fully filled, then an underrun has occurred; and
+    // fill the remainder of |dest| with zeros.
+    if (samples_remaining > 0) {
+      // Note: Only logging underruns after the first frame has been received.
+      LOG_IF(WARNING, currently_playing_audio_frame_start_ != -1)
+          << "Audio: Playback underrun of " << samples_remaining << " samples!";
+      dest->ZeroFramesPartial(dest->frames() - samples_remaining,
+                              samples_remaining);
+    }
+
+    return dest->frames();
+  }
+
+  virtual int OnMoreIOData(AudioBus* source,
+                           AudioBus* dest,
+                           AudioBuffersState buffers_state) OVERRIDE {
+    return OnMoreData(dest, buffers_state);
+  }
+
+  virtual void OnError(AudioOutputStream* stream) OVERRIDE {
+    LOG(ERROR) << "AudioOutputStream reports an error.  "
+               << "Playback is unlikely to continue.";
+  }
+
+  // End of AudioSourceCallback implementation.
+  ////////////////////////////////////////////////////////////////////
+
+  void ScheduleVideoPlayout() {
+    DCHECK(cast_env()->CurrentlyOn(CastEnvironment::MAIN));
+
+    // Prune the queue, skipping entries that are too old.
+    const base::TimeTicks now = cast_env()->Clock()->NowTicks();
+    const base::TimeTicks earliest_time_to_play = now - max_frame_age_;
+    while (!video_playout_queue_.empty() &&
+           video_playout_queue_.front().first < earliest_time_to_play) {
+      PopOneVideoFrame(true);
+    }
+
+    // If the queue is not empty, schedule playout of its first frame.
+    if (video_playout_queue_.empty()) {
+      video_playout_timer_.Stop();
+    } else {
+      video_playout_timer_.Start(
+          FROM_HERE,
+          video_playout_queue_.front().first - now,
+          base::Bind(&NaivePlayer::PlayNextVideoFrame,
+                     base::Unretained(this)));
+    }
+  }
+
+  void PlayNextVideoFrame() {
+    DCHECK(cast_env()->CurrentlyOn(CastEnvironment::MAIN));
+    if (!video_playout_queue_.empty()) {
+      const scoped_refptr<VideoFrame> video_frame =
+          video_playout_queue_.front().second;
+      PopOneVideoFrame(false);
+#ifdef OS_LINUX
+      render_.RenderFrame(video_frame);
+#endif  // OS_LINUX
+    }
+    ScheduleVideoPlayout();
+  }
+
+  void PopOneVideoFrame(bool is_being_skipped) {
+    DCHECK(cast_env()->CurrentlyOn(CastEnvironment::MAIN));
+
+    if (is_being_skipped) {
+      VLOG(1) << "VideoFrame[" << num_video_frames_processed_ << "]: Skipped.";
+    } else {
+      VLOG(1) << "VideoFrame[" << num_video_frames_processed_ << "]: Playing "
+              << (cast_env()->Clock()->NowTicks() -
+                      video_playout_queue_.front().first).InMicroseconds()
+              << " usec later than intended.";
+    }
+
+    video_playout_queue_.pop_front();
+    ++num_video_frames_processed_;
+  }
+
+  void PopOneAudioFrame(bool was_skipped) {
+    audio_lock_.AssertAcquired();
+
+    if (was_skipped) {
+      VLOG(1) << "AudioFrame[" << num_audio_frames_processed_ << "]: Skipped";
+    } else {
+      VLOG(1) << "AudioFrame[" << num_audio_frames_processed_ << "]: Playing "
+              << (cast_env()->Clock()->NowTicks() -
+                      audio_playout_queue_.front().first).InMicroseconds()
+              << " usec later than intended.";
     }
-    last_playout_time_ = playout_time;
+
+    audio_playout_queue_.pop_front();

[imcheng, 2014/04/09 01:34:45]

Do you need to destroy audio_playout_queue_.front().second in cases where you
are just throwing away the frames? Would it work beter if it uses linked_ptrs?
(see below)

[miu, 2014/04/09 04:07:19]

Good catch.  Fixed this memory-leak.

+    ++num_audio_frames_processed_;
   }
 
+  // Frames in the queue older than this (relative to NowTicks()) will be
+  // dropped (i.e., playback is falling behind).
+  const base::TimeDelta max_frame_age_;
+
+  // Outputs created, started, and destroyed by this NaivePlayer.
 #ifdef OS_LINUX
   test::LinuxOutputWindow render_;
 #endif  // OS_LINUX
-  base::TimeTicks last_playout_time_;
-  base::TimeTicks last_render_time_;
+  scoped_ptr<AudioOutputStream> audio_output_stream_;
+
+  // Video playout queue.
+  typedef std::pair<base::TimeTicks, scoped_refptr<VideoFrame> >
+      VideoQueueEntry;
+  std::deque<VideoQueueEntry> video_playout_queue_;
+  int64 num_video_frames_processed_;
+
+  base::OneShotTimer<NaivePlayer> video_playout_timer_;
+
+  // Audio playout queue, synchronized by |audio_lock_|.
+  base::Lock audio_lock_;
+  typedef std::pair<base::TimeTicks, AudioBus*> AudioQueueEntry;

[imcheng, 2014/04/09 01:34:45]

Can you make this into <base::TimeTicks, linked_ptr<AudioBus> > ?

[miu, 2014/04/09 04:07:19]

Unfortunately, no.  AudioBus defines a private destructor that can only be
invoked by scoped_ptr<>.

+  std::deque<AudioQueueEntry> audio_playout_queue_;
+  int64 num_audio_frames_processed_;
+
+  // These must only be used on the audio thread calling OnMoreData().
+  scoped_ptr<AudioBus> currently_playing_audio_frame_;
+  int currently_playing_audio_frame_start_;
 };
 
 }  // namespace cast
@@ -196,43 +456,46 @@ int main(int argc, char** argv) {
   scoped_refptr<media::cast::CastEnvironment> cast_environment(
       new media::cast::StandaloneCastEnvironment);
 
+  // Start up Chromium audio system.
+  media::FakeAudioLogFactory fake_audio_log_factory_;
+  const scoped_ptr<media::AudioManager> audio_manager(
+      media::AudioManager::Create(&fake_audio_log_factory_));
+  CHECK(media::AudioManager::Get());
+
   media::cast::AudioReceiverConfig audio_config =
       media::cast::GetAudioReceiverConfig();
   media::cast::VideoReceiverConfig video_config =
       media::cast::GetVideoReceiverConfig();
 
+  // Determine local and remote endpoints.
   int remote_port, local_port;
   media::cast::GetPorts(&remote_port, &local_port);
   if (!local_port) {
     LOG(ERROR) << "Invalid local port.";
     return 1;
   }
-
   std::string remote_ip_address = media::cast::GetIpAddress("Enter remote IP.");
   std::string local_ip_address = media::cast::GetIpAddress("Enter local IP.");
   net::IPAddressNumber remote_ip_number;
   net::IPAddressNumber local_ip_number;
-
   if (!net::ParseIPLiteralToNumber(remote_ip_address, &remote_ip_number)) {
     LOG(ERROR) << "Invalid remote IP address.";
     return 1;
   }
-
   if (!net::ParseIPLiteralToNumber(local_ip_address, &local_ip_number)) {
     LOG(ERROR) << "Invalid local IP address.";
     return 1;
   }
-
   net::IPEndPoint remote_end_point(remote_ip_number, remote_port);
   net::IPEndPoint local_end_point(local_ip_number, local_port);
 
-  media::cast::ReceiverDisplay* const receiver_display =
-      new media::cast::ReceiverDisplay(cast_environment,
-                                       local_end_point,
-                                       remote_end_point,
-                                       audio_config,
-                                       video_config);
-  receiver_display->Start();
+  // Create and start the player.
+  media::cast::NaivePlayer player(cast_environment,
+                                  local_end_point,
+                                  remote_end_point,
+                                  audio_config,
+                                  video_config);
+  player.Start();
 
   base::MessageLoop().Run();  // Run forever (i.e., until SIGTERM).
   NOTREACHED();