Moves channel-dependent audio input processing to separate encoder task queue

webrtc/voice_engine/channel.h

Index: webrtc/voice_engine/channel.h
diff --git a/webrtc/voice_engine/channel.h b/webrtc/voice_engine/channel.h
index 68d022d4d1fec4725acfda3b791cd5d6eab8ecff..c0ac046c003a5b12ab1ed19bb404d6affbdb7a60 100644
--- a/webrtc/voice_engine/channel.h
+++ b/webrtc/voice_engine/channel.h
@@ -28,6 +28,7 @@
 #include "webrtc/modules/rtp_rtcp/include/remote_ntp_time_estimator.h"
 #include "webrtc/modules/rtp_rtcp/include/rtp_header_parser.h"
 #include "webrtc/modules/rtp_rtcp/include/rtp_rtcp.h"
+#include "webrtc/voice_engine/audio_frame_pool.h"
 #include "webrtc/voice_engine/audio_level.h"
 #include "webrtc/voice_engine/file_player.h"
 #include "webrtc/voice_engine/file_recorder.h"
@@ -157,8 +158,8 @@ class Channel
                                ProcessThread& moduleProcessThread,
                                AudioDeviceModule& audioDeviceModule,
                                VoiceEngineObserver* voiceEngineObserver,
-                               rtc::CriticalSection* callbackCritSect);
-  int32_t UpdateLocalTimeStamp();
+                               rtc::CriticalSection* callbackCritSect,
+                               rtc::TaskQueue* encoder_queue);
 
   void SetSink(std::unique_ptr<AudioSinkInterface> sink);
 
@@ -352,16 +353,36 @@ class Channel
   }
   RtpRtcp* RtpRtcpModulePtr() const { return _rtpRtcpModule.get(); }
   int8_t OutputEnergyLevel() const { return _outputAudioLevel.Level(); }
-  uint32_t Demultiplex(const AudioFrame& audioFrame);
-  // Demultiplex the data to the channel's |_audioFrame|. The difference
-  // between this method and the overloaded method above is that |audio_data|
-  // does not go through transmit_mixer and APM.
-  void Demultiplex(const int16_t* audio_data,
-                   int sample_rate,
-                   size_t number_of_frames,
-                   size_t number_of_channels);
-  uint32_t PrepareEncodeAndSend(int mixingFrequency);
-  uint32_t EncodeAndSend();
+
+  // ProcessAndEncodeAudio() creates an audio frame copy and posts a task
+  // on the shared encoder task queue, wich in turn calls (on the queue)
+  // ProcessAndEncodeAudioOnTaskQueue() where the actual processing of the
+  // audio takes place. The processing mainly consists of encoding and preparing
+  // the result for sending by adding it to a send queue.
+  // The main reason for using a task queue here is to release the native,
+  // OS-specific, audio capture thread as soon as possible to ensure that it
+  // can go back to sleep and be prepared to deliver an new captured audio
+  // packet.
+  void ProcessAndEncodeAudio(const AudioFrame& audio_input);
+
+  // This version of ProcessAndEncodeAudio() is used by PushCaptureData() in
+  // VoEBase and the audio in |audio_data| has not been subject to any APM
+  // processing. Some extra steps are therfore needed when building up the
+  // audio frame copy before using the same task as in the default call to
+  // ProcessAndEncodeAudio(const AudioFrame& audio_input).
+  void ProcessAndEncodeAudio(const int16_t* audio_data,
+                             int sample_rate,
+                             size_t number_of_frames,
+                             size_t number_of_channels);
+
+  // Called on the encoder task queue when a new input audio frame is ready
+  // for encoding.
+  void ProcessAndEncodeAudioOnTaskQueue(AudioFrame* audio_input);
+
+  // Internal helper methods used by ProcessAndEncodeAudioOnTaskQueue().
+  // Both are called on the encoder task queue.
+  uint32_t PrepareEncodeAndSend(AudioFrame* audio_input);
+  uint32_t EncodeAndSend(AudioFrame* audio_input);
 
   // Associate to a send channel.
   // Used for obtaining RTT for a receive-only channel.
@@ -382,6 +403,8 @@ class Channel
   void OnIncomingFractionLoss(int fraction_lost);
 
  private:
+  class ProcessAndEncodeAudioTask;
+
   bool InputMute() const;
   bool OnRtpPacketWithHeader(const uint8_t* received_packet,
                              size_t length,
@@ -396,7 +419,7 @@ class Channel
   bool IsPacketInOrder(const RTPHeader& header) const;
   bool IsPacketRetransmitted(const RTPHeader& header, bool in_order) const;
   int ResendPackets(const uint16_t* sequence_numbers, int length);
-  int32_t MixOrReplaceAudioWithFile(int mixingFrequency);
+  int32_t MixOrReplaceAudioWithFile(AudioFrame* audio_frame);
   int32_t MixAudioWithFile(AudioFrame& audioFrame, int mixingFrequency);
   void UpdatePlayoutTimestamp(bool rtcp);
   void RegisterReceiveCodecsToRTPModule();
@@ -434,7 +457,6 @@ class Channel
   std::unique_ptr<AudioSinkInterface> audio_sink_;
   AudioLevel _outputAudioLevel;
   bool _externalTransport;
-  AudioFrame _audioFrame;
   // Downsamples to the codec rate if necessary.
   PushResampler<int16_t> input_resampler_;
   std::unique_ptr<FilePlayer> input_file_player_;
@@ -473,6 +495,7 @@ class Channel
   AudioDeviceModule* _audioDeviceModulePtr;
   VoiceEngineObserver* _voiceEngineObserverPtr;  // owned by base
   rtc::CriticalSection* _callbackCritSectPtr;    // owned by base
+  rtc::TaskQueue* encoder_queue_;
   Transport* _transportPtr;  // WebRtc socket or external transport
   RmsLevel rms_level_;
   bool input_mute_ GUARDED_BY(volume_settings_critsect_);
@@ -504,6 +527,10 @@ class Channel
   std::unique_ptr<RtpPacketSenderProxy> rtp_packet_sender_proxy_;
   std::unique_ptr<RateLimiter> retransmission_rate_limiter_;
 
+  // Pool of preallocated audio frames for temporary storage of recorded audio
+  // to be encoded on the encoder queue.
+  AudioFramePool audio_frame_pool_;
+
   // TODO(ossu): Remove once GetAudioDecoderFactory() is no longer needed.
   rtc::scoped_refptr<AudioDecoderFactory> decoder_factory_;