Avoids irregular OnMoreData callbacks on Windows using Core Audio

media/audio/win/audio_low_latency_output_win.h

 // Copyright (c) 2012 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.
 
 // Implementation of AudioOutputStream for Windows using Windows Core Audio
 // WASAPI for low latency rendering.
 //
 // Overview of operation and performance:
 //
 // - An object of WASAPIAudioOutputStream is created by the AudioManager
 //   factory.
 // - Next some thread will call Open(), at that point the underlying
 //   Core Audio APIs are utilized to create two WASAPI interfaces called
 //   IAudioClient and IAudioRenderClient.
 // - Then some thread will call Start(source).
 //   A thread called "wasapi_render_thread" is started and this thread listens
 //   on an event signal which is set periodically by the audio engine to signal
 //   render events. As a result, OnMoreData() will be called and the registered
 //   client is then expected to provide data samples to be played out.
 // - At some point, a thread will call Stop(), which stops and joins the
 //   render thread and at the same time stops audio streaming.
 // - The same thread that called stop will call Close() where we cleanup
 //   and notify the audio manager, which likely will destroy this object.
-// - Initial tests on Windows 7 shows that this implementation results in a
-//   latency of approximately 35 ms if the selected packet size is less than
-//   or equal to 20 ms. Using a packet size of 10 ms does not result in a
-//   lower latency but only affects the size of the data buffer in each
-//   OnMoreData() callback.
 // - A total typical delay of 35 ms contains three parts:
 //    o Audio endpoint device period (~10 ms).
 //    o Stream latency between the buffer and endpoint device (~5 ms).
 //    o Endpoint buffer (~20 ms to ensure glitch-free rendering).
-// - Note that, if the user selects a packet size of e.g. 100 ms, the total
-//   delay will be approximately 115 ms (10 + 5 + 100).
 //
 // Implementation notes:
 //
 // - The minimum supported client is Windows Vista.
 // - This implementation is single-threaded, hence:
 //    o Construction and destruction must take place from the same thread.
 //    o All APIs must be called from the creating thread as well.
-// - It is recommended to first acquire the native sample rate of the default
-//   input device and then use the same rate when creating this object. Use
-//   WASAPIAudioOutputStream::HardwareSampleRate() to retrieve the sample rate.
+// - It is required to first acquire the native audio parameters of the default
+//   output device and then use the same rate when creating this object. Use
+//   e.g. WASAPIAudioOutputStream::HardwareSampleRate() to retrieve the sample
+//   rate. Open() will fail unless "perfect" audio parameters are utilized.
 // - Calling Close() also leads to self destruction.
-// - Stream switching is not supported if the user shifts the audio device
-//   after Open() is called but before Start() has been called.
-// - Stream switching can fail if streaming starts on one device with a
-//   supported format (X) and the new default device - to which we would like
-//   to switch - uses another format (Y), which is not supported given the
-//   configured audio parameters.
-// - The audio device must be opened with the same number of channels as it
-//   supports natively (see HardwareChannelCount()) otherwise Open() will fail.
 // - Support for 8-bit audio has not yet been verified and tested.
 //
 // Core Audio API details:
 //
 // - The public API methods (Open(), Start(), Stop() and Close()) must be
 //   called on constructing thread. The reason is that we want to ensure that
 //   the COM environment is the same for all API implementations.
 // - Utilized MMDevice interfaces:
 //     o IMMDeviceEnumerator
 //     o IMMDevice
@@ skipping 112 matching lines @@
   // processing/mixing of shared-mode streams for the default endpoint device.
   int GetEndpointChannelCountForTesting() { return format_.Format.nChannels; }
 
  private:
   // DelegateSimpleThread::Delegate implementation.
   virtual void Run() OVERRIDE;
 
   // Issues the OnError() callback to the |sink_|.
   void HandleError(HRESULT err);
 
-  // The Open() method is divided into these sub methods.
-  HRESULT SetRenderDevice();
-  HRESULT ActivateRenderDevice();
-  bool DesiredFormatIsSupported();
-  HRESULT InitializeAudioEngine();
-
-  // Called when the device will be opened in shared mode and use the
-  // internal audio engine's mix format.
-  HRESULT SharedModeInitialization();
-
   // Called when the device will be opened in exclusive mode and use the
   // application specified format.
-  HRESULT ExclusiveModeInitialization();
+  // TODO(henrika): rewrite and move to CoreAudioUtil when removing flag
+  // for exclusive audio mode.
+  HRESULT ExclusiveModeInitialization(IAudioClient* client,
+                                      HANDLE event_handle,
+                                      size_t* endpoint_buffer_size);
 
-  // Converts unique endpoint ID to user-friendly device name.
-  std::string GetDeviceName(LPCWSTR device_id) const;
+  // Fills up the endpoint rendering buffer with silence.
+  bool FillEndpointBufferWithSilence(UINT32* num_written_frames);
 
   // Contains the thread ID of the creating thread.
   base::PlatformThreadId creating_thread_id_;
 
   // Our creator, the audio manager needs to be notified when we close.
   AudioManagerWin* manager_;
 
   // Rendering is driven by this thread (which has no message loop).
   // All OnMoreData() callbacks will be called from this thread.
   scoped_ptr<base::DelegateSimpleThread> render_thread_;
 
   // Contains the desired audio format which is set up at construction.
   // Extended PCM waveform format structure based on WAVEFORMATEXTENSIBLE.
   // Use this for multiple channel and hi-resolution PCM data.
   WAVEFORMATPCMEX format_;
 
-  // Copy of the audio format which we know the audio engine supports.
-  // It is recommended to ensure that the sample rate in |format_| is identical
-  // to the sample rate in |audio_engine_mix_format_|.
-  base::win::ScopedCoMem<WAVEFORMATPCMEX> audio_engine_mix_format_;
-
+  // Set to true when stream is successfully opened.
   bool opened_;
 
-  // Set to true as soon as a new default device is detected, and cleared when
-  // the streaming has switched from using the old device to the new device.
-  // All additional device detections during an active state are ignored to
-  // ensure that the ongoing switch can finalize without disruptions.
-  bool restart_rendering_mode_;
+  // We check if the input audio parameters are identical (bit depth is
+  // excluded) to the preferred (native) audio parameters during construction.
+  // Open() will fail if |audio_parmeters_are_valid_| is false.
+  bool audio_parmeters_are_valid_;
 
   // Volume level from 0 to 1.
   float volume_;
 
-  // Size in bytes of each audio frame (4 bytes for 16-bit stereo PCM).
-  size_t frame_size_;
-
   // Size in audio frames of each audio packet where an audio packet
   // is defined as the block of data which the source is expected to deliver
   // in each OnMoreData() callback.
   size_t packet_size_frames_;
 
   // Size in bytes of each audio packet.
   size_t packet_size_bytes_;
 
   // Size in milliseconds of each audio packet.
   float packet_size_ms_;
 
   // Length of the audio endpoint buffer.
   size_t endpoint_buffer_size_frames_;
 
   // Defines the role that the system has assigned to an audio endpoint device.
   ERole device_role_;
 
   // The sharing mode for the connection.
   // Valid values are AUDCLNT_SHAREMODE_SHARED and AUDCLNT_SHAREMODE_EXCLUSIVE
   // where AUDCLNT_SHAREMODE_SHARED is the default.
   AUDCLNT_SHAREMODE share_mode_;
 
-  // The channel count set by the client in |params| which is provided to the
-  // constructor. The client must feed the AudioSourceCallback::OnMoreData()
-  // callback with PCM-data that contains this number of channels.
-  int client_channel_count_;
-
   // Counts the number of audio frames written to the endpoint buffer.
   UINT64 num_written_frames_;
 
   // Pointer to the client that will deliver audio samples to be played out.
   AudioSourceCallback* source_;
 
   // An IMMDeviceEnumerator interface which represents a device enumerator.
   base::win::ScopedComPtr<IMMDeviceEnumerator> device_enumerator_;
 
-  // An IMMDevice interface which represents an audio endpoint device.
-  base::win::ScopedComPtr<IMMDevice> endpoint_device_;
-
   // An IAudioClient interface which enables a client to create and initialize
   // an audio stream between an audio application and the audio engine.
   base::win::ScopedComPtr<IAudioClient> audio_client_;
 
   // The IAudioRenderClient interface enables a client to write output
   // data to a rendering endpoint buffer.
   base::win::ScopedComPtr<IAudioRenderClient> audio_render_client_;
 
   // The audio engine will signal this event each time a buffer becomes
   // ready to be filled by the client.
   base::win::ScopedHandle audio_samples_render_event_;
 
   // This event will be signaled when rendering shall stop.
   base::win::ScopedHandle stop_render_event_;
 
   // Container for retrieving data from AudioSourceCallback::OnMoreData().
   scoped_ptr<AudioBus> audio_bus_;
 
   DISALLOW_COPY_AND_ASSIGN(WASAPIAudioOutputStream);
 };
 
 }  // namespace media
 
 #endif  // MEDIA_AUDIO_WIN_AUDIO_LOW_LATENCY_OUTPUT_WIN_H_