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

webrtc/voice_engine/channel.h

 /*
  *  Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
 #ifndef WEBRTC_VOICE_ENGINE_CHANNEL_H_
 #define WEBRTC_VOICE_ENGINE_CHANNEL_H_
 
 #include <memory>
 
 #include "webrtc/api/audio/audio_mixer.h"
 #include "webrtc/api/call/audio_sink.h"
 #include "webrtc/base/criticalsection.h"
+#include "webrtc/base/event.h"
 #include "webrtc/base/optional.h"
 #include "webrtc/base/thread_checker.h"
 #include "webrtc/common_audio/resampler/include/push_resampler.h"
 #include "webrtc/common_types.h"
 #include "webrtc/modules/audio_coding/acm2/codec_manager.h"
 #include "webrtc/modules/audio_coding/acm2/rent_a_codec.h"
 #include "webrtc/modules/audio_coding/include/audio_coding_module.h"
 #include "webrtc/modules/audio_conference_mixer/include/audio_conference_mixer_defines.h"
 #include "webrtc/modules/audio_processing/rms_level.h"
 #include "webrtc/modules/rtp_rtcp/include/remote_ntp_time_estimator.h"
@@ skipping 50 matching lines @@
 // Example: a member can be set on thread T1 and read by an internal audio
 // thread T2. Accessing the member via this class ensures that we are
 // safe and also avoid TSan v2 warnings.
 class ChannelState {
  public:
   struct State {
     bool output_file_playing = false;
     bool input_file_playing = false;
     bool playing = false;
     bool sending = false;
+    bool sending_has_been_activated = false;
   };
 
   ChannelState() {}
   virtual ~ChannelState() {}
 
   void Reset() {
     rtc::CritScope lock(&lock_);
     state_ = State();
   }
 
@@ skipping 13 matching lines @@
   }
 
   void SetPlaying(bool enable) {
     rtc::CritScope lock(&lock_);
     state_.playing = enable;
   }
 
   void SetSending(bool enable) {
     rtc::CritScope lock(&lock_);
     state_.sending = enable;
+    state_.sending_has_been_activated = enable;
   }
 
  private:
   rtc::CriticalSection lock_;
   State state_;
 };
 
 class Channel
     : public RtpData,
       public RtpFeedback,
@@ skipping 19 matching lines @@
           uint32_t instanceId,
           const VoEBase::ChannelConfig& config);
   int32_t Init();
   void RegisterLegacyCodecs();
   void Terminate();
   int32_t SetEngineInformation(Statistics& engineStatistics,
                                OutputMixer& outputMixer,
                                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);
 
   // TODO(ossu): Don't use! It's only here to confirm that the decoder factory
   // passed into AudioReceiveStream is the same as the one set when creating the
   // ADM. Once Channel creation is moved into Audio{Send,Receive}Stream this can
   // go.
   const rtc::scoped_refptr<AudioDecoderFactory>& GetAudioDecoderFactory() const;
 
   void SetReceiveCodecs(const std::map<int, SdpAudioFormat>& codecs);
@@ skipping 175 matching lines @@
   uint32_t InstanceId() const { return _instanceId; }
   int32_t ChannelId() const { return _channelId; }
   bool Playing() const { return channel_state_.Get().playing; }
   bool Sending() const { return channel_state_.Get().sending; }
   bool ExternalTransport() const {
     rtc::CritScope cs(&_callbackCritSect);
     return _externalTransport;
   }
   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();
+  PushResampler<int16_t>* input_resampler() { return &input_resampler_; }
+
+  // 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.
   void set_associate_send_channel(const ChannelOwner& channel);
   // Disassociate a send channel if it was associated.
   void DisassociateSendChannel(int channel_id);
 
   // Set a RtcEventLog logging object.
   void SetRtcEventLog(RtcEventLog* event_log);
 
   void SetRtcpRttStats(RtcpRttStats* rtcp_rtt_stats);
   void SetTransportOverhead(size_t transport_overhead_per_packet);
 
   // From OverheadObserver in the RTP/RTCP module
   void OnOverheadChanged(size_t overhead_bytes_per_packet) override;
 
   // The existence of this function alongside OnUplinkPacketLossRate is
   // a compromise. We want the encoder to be agnostic of the PLR source, but
   // we also don't want it to receive conflicting information from TWCC and
   // from RTCP-XR.
   void OnTwccBasedUplinkPacketLossRate(float packet_loss_rate);
 
   void OnRecoverableUplinkPacketLossRate(float recoverable_packet_loss_rate);
 
  private:
+  class ProcessAndEncodeAudioTask;

[aleloi, 2017/03/24 16:59:50]

This declaration is only letting the task DCHECK that it runs on the private
encoder_queue_? I wonder if it's necessary to declare it here.

[henrika_webrtc, 2017/03/24 17:09:36]

Sorry but I don't understand. What do you suggest I do instead?

+
   void OnUplinkPacketLossRate(float packet_loss_rate);
-
   bool InputMute() const;
   bool OnRtpPacketWithHeader(const uint8_t* received_packet,
                              size_t length,
                              RTPHeader *header);
   bool ReceivePacket(const uint8_t* packet,
                      size_t packet_length,
                      const RTPHeader& header,
                      bool in_order);
   bool HandleRtxPacket(const uint8_t* packet,
                        size_t packet_length,
                        const RTPHeader& header);
   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();
 
   int SetSendRtpHeaderExtension(bool enable,
                                 RTPExtensionType type,
                                 unsigned char id);
 
   void UpdateOverheadForEncoder()
       EXCLUSIVE_LOCKS_REQUIRED(overhead_per_packet_lock_);
@@ skipping 17 matching lines @@
   std::unique_ptr<ReceiveStatistics> rtp_receive_statistics_;
   std::unique_ptr<RtpReceiver> rtp_receiver_;
   TelephoneEventHandler* telephone_event_handler_;
   std::unique_ptr<RtpRtcp> _rtpRtcpModule;
   std::unique_ptr<AudioCodingModule> audio_coding_;
   acm2::CodecManager codec_manager_;
   acm2::RentACodec rent_a_codec_;
   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_;
   std::unique_ptr<FilePlayer> output_file_player_;
   std::unique_ptr<FileRecorder> output_file_recorder_;
   int _inputFilePlayerId;
   int _outputFilePlayerId;
   int _outputFileRecorderId;
   bool _outputFileRecording;
   uint32_t _timeStamp;
@@ skipping 18 matching lines @@
   // frame.
   int64_t capture_start_ntp_time_ms_ GUARDED_BY(ts_stats_lock_);
 
   // uses
   Statistics* _engineStatisticsPtr;
   OutputMixer* _outputMixerPtr;
   ProcessThread* _moduleProcessThreadPtr;
   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_);
   bool previous_frame_muted_;  // Only accessed from PrepareEncodeAndSend().
   float _outputGain GUARDED_BY(volume_settings_critsect_);
   // VoEBase
   bool _mixFileWithMicrophone;
   // VoeRTP_RTCP
   uint32_t _lastLocalTimeStamp;
   int8_t _lastPayloadType;
@@ skipping 17 matching lines @@
   std::unique_ptr<TransportSequenceNumberProxy> seq_num_allocator_proxy_;
   std::unique_ptr<RtpPacketSenderProxy> rtp_packet_sender_proxy_;
   std::unique_ptr<RateLimiter> retransmission_rate_limiter_;
 
   // TODO(ossu): Remove once GetAudioDecoderFactory() is no longer needed.
   rtc::scoped_refptr<AudioDecoderFactory> decoder_factory_;
 
   rtc::ThreadChecker construction_thread_;
 
   const bool use_twcc_plr_for_ana_;
+
+  rtc::Event stop_send_event_;
 };
 
 }  // namespace voe
 }  // namespace webrtc
 
 #endif  // WEBRTC_VOICE_ENGINE_CHANNEL_H_