modules/audio_device/win/core_audio_input_win.cc

/*
 *  Copyright (c) 2018 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.
 */

#include "modules/audio_device/win/core_audio_input_win.h"

#include <memory>

#include "modules/audio_device/audio_device_buffer.h"
#include "modules/audio_device/fine_audio_buffer.h"
#include "rtc_base/checks.h"
#include "rtc_base/logging.h"
#include "rtc_base/numerics/safe_conversions.h"

using Microsoft::WRL::ComPtr;

namespace webrtc {
namespace webrtc_win {

enum AudioDeviceMessageType : uint32_t {
  kMessageInputStreamDisconnected,
};

CoreAudioInput::CoreAudioInput(bool automatic_restart)
    : CoreAudioBase(
          CoreAudioBase::Direction::kInput,
          automatic_restart,
          [this](uint64_t freq) { return OnDataCallback(freq); },
          [this](ErrorType err) { return OnErrorCallback(err); }) {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  thread_checker_audio_.Detach();
}

CoreAudioInput::~CoreAudioInput() {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
}

int CoreAudioInput::Init() {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return 0;
}

int CoreAudioInput::Terminate() {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  StopRecording();
  return 0;
}

int CoreAudioInput::NumDevices() const {
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return core_audio_utility::NumberOfActiveDevices(eCapture);
}

int CoreAudioInput::SetDevice(int index) {
  RTC_DLOG(INFO) << __FUNCTION__ << ": " << index;
  RTC_DCHECK_GE(index, 0);
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return CoreAudioBase::SetDevice(index);
}

int CoreAudioInput::SetDevice(AudioDeviceModule::WindowsDeviceType device) {
  RTC_DLOG(INFO) << __FUNCTION__ << ": "
                 << ((device == AudioDeviceModule::kDefaultDevice)
                         ? "Default"
                         : "DefaultCommunication");
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return SetDevice((device == AudioDeviceModule::kDefaultDevice) ? 0 : 1);
}

int CoreAudioInput::DeviceName(int index,
                               std::string* name,
                               std::string* guid) {
  RTC_DLOG(INFO) << __FUNCTION__ << ": " << index;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  RTC_DCHECK(name);
  return CoreAudioBase::DeviceName(index, name, guid);
}

void CoreAudioInput::AttachAudioBuffer(AudioDeviceBuffer* audio_buffer) {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  audio_device_buffer_ = audio_buffer;
}

bool CoreAudioInput::RecordingIsInitialized() const {
  RTC_DLOG(INFO) << __FUNCTION__ << ": " << initialized_;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return initialized_;
}

int CoreAudioInput::InitRecording() {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK(!initialized_);
  RTC_DCHECK(!Recording());
  RTC_DCHECK(!audio_capture_client_);

  // Creates an IAudioClient instance and stores the valid interface pointer in
  // `audio_client3_`, `audio_client2_`, or `audio_client_` depending on
  // platform support. The base class will use optimal input parameters and do
  // an event driven shared mode initialization. The utilized format will be
  // stored in `format_` and can be used for configuration and allocation of
  // audio buffers.
  if (!CoreAudioBase::Init()) {
    return -1;
  }
  RTC_DCHECK(audio_client_);

  // Configure the recording side of the audio device buffer using `format_`
  // after a trivial sanity check of the format structure.
  RTC_DCHECK(audio_device_buffer_);
  WAVEFORMATEX* format = &format_.Format;
  RTC_DCHECK_EQ(format->wFormatTag, WAVE_FORMAT_EXTENSIBLE);
  audio_device_buffer_->SetRecordingSampleRate(format->nSamplesPerSec);
  audio_device_buffer_->SetRecordingChannels(format->nChannels);

  // Create a modified audio buffer class which allows us to supply any number
  // of samples (and not only multiple of 10ms) to match the optimal buffer
  // size per callback used by Core Audio.
  // TODO(henrika): can we share one FineAudioBuffer with the output side?
  fine_audio_buffer_ = std::make_unique<FineAudioBuffer>(audio_device_buffer_);

  // Create an IAudioCaptureClient for an initialized IAudioClient.
  // The IAudioCaptureClient interface enables a client to read input data from
  // a capture endpoint buffer.
  ComPtr<IAudioCaptureClient> audio_capture_client =
      core_audio_utility::CreateCaptureClient(audio_client_.Get());
  if (!audio_capture_client) {
    return -1;
  }

  // Query performance frequency.
  LARGE_INTEGER ticks_per_sec = {};
  qpc_to_100ns_.reset();
  if (::QueryPerformanceFrequency(&ticks_per_sec)) {
    double qpc_ticks_per_second =
        rtc::dchecked_cast<double>(ticks_per_sec.QuadPart);
    qpc_to_100ns_ = 10000000.0 / qpc_ticks_per_second;
  }

  // Store valid COM interfaces.
  audio_capture_client_ = audio_capture_client;

  initialized_ = true;
  return 0;
}

int CoreAudioInput::StartRecording() {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK(!Recording());
  RTC_DCHECK(fine_audio_buffer_);
  RTC_DCHECK(audio_device_buffer_);
  if (!initialized_) {
    RTC_DLOG(LS_WARNING)
        << "Recording can not start since InitRecording must succeed first";
    return 0;
  }

  fine_audio_buffer_->ResetRecord();
  if (!IsRestarting()) {
    audio_device_buffer_->StartRecording();
  }

  if (!Start()) {
    return -1;
  }

  is_active_ = true;
  return 0;
}

int CoreAudioInput::StopRecording() {
  RTC_DLOG(INFO) << __FUNCTION__;
  if (!initialized_) {
    return 0;
  }

  // Release resources allocated in InitRecording() and then return if this
  // method is called without any active input audio.
  if (!Recording()) {
    RTC_DLOG(WARNING) << "No input stream is active";
    ReleaseCOMObjects();
    initialized_ = false;
    return 0;
  }

  if (!Stop()) {
    RTC_LOG(LS_ERROR) << "StopRecording failed";
    return -1;
  }

  if (!IsRestarting()) {
    RTC_DCHECK(audio_device_buffer_);
    audio_device_buffer_->StopRecording();
  }

  // Release all allocated resources to allow for a restart without
  // intermediate destruction.
  ReleaseCOMObjects();
  qpc_to_100ns_.reset();

  initialized_ = false;
  is_active_ = false;
  return 0;
}

bool CoreAudioInput::Recording() {
  RTC_DLOG(INFO) << __FUNCTION__ << ": " << is_active_;
  return is_active_;
}

// TODO(henrika): finalize support of audio session volume control. As is, we
// are not compatible with the old ADM implementation since it allows accessing
// the volume control with any active audio output stream.
int CoreAudioInput::VolumeIsAvailable(bool* available) {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return IsVolumeControlAvailable(available) ? 0 : -1;
}

// Triggers the restart sequence. Only used for testing purposes to emulate
// a real event where e.g. an active input device is removed.
int CoreAudioInput::RestartRecording() {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  if (!Recording()) {
    return 0;
  }

  if (!Restart()) {
    RTC_LOG(LS_ERROR) << "RestartRecording failed";
    return -1;
  }
  return 0;
}

bool CoreAudioInput::Restarting() const {
  RTC_DCHECK_RUN_ON(&thread_checker_);
  return IsRestarting();
}

int CoreAudioInput::SetSampleRate(uint32_t sample_rate) {
  RTC_DLOG(INFO) << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_);
  sample_rate_ = sample_rate;
  return 0;
}

void CoreAudioInput::ReleaseCOMObjects() {
  RTC_DLOG(INFO) << __FUNCTION__;
  CoreAudioBase::ReleaseCOMObjects();
  if (audio_capture_client_.Get()) {
    audio_capture_client_.Reset();
  }
}

bool CoreAudioInput::OnDataCallback(uint64_t device_frequency) {
  RTC_DCHECK_RUN_ON(&thread_checker_audio_);

  if (!initialized_ || !is_active_) {
    // This is concurrent examination of state across multiple threads so will
    // be somewhat error prone, but we should still be defensive and not use
    // audio_capture_client_ if we know it's not there.
    return false;
  }
  if (num_data_callbacks_ == 0) {
    RTC_LOG(INFO) << "--- Input audio stream is alive ---";
  }
  UINT32 num_frames_in_next_packet = 0;
  _com_error error =
      audio_capture_client_->GetNextPacketSize(&num_frames_in_next_packet);
  if (error.Error() == AUDCLNT_E_DEVICE_INVALIDATED) {
    // Avoid breaking the thread loop implicitly by returning false and return
    // true instead for AUDCLNT_E_DEVICE_INVALIDATED even it is a valid error
    // message. We will use notifications about device changes instead to stop
    // data callbacks and attempt to restart streaming .
    RTC_DLOG(LS_ERROR) << "AUDCLNT_E_DEVICE_INVALIDATED";
    return true;
  }
  if (FAILED(error.Error())) {
    RTC_LOG(LS_ERROR) << "IAudioCaptureClient::GetNextPacketSize failed: "
                      << core_audio_utility::ErrorToString(error);
    return false;
  }

  // Drain the WASAPI capture buffer fully if audio has been recorded.
  while (num_frames_in_next_packet > 0) {
    uint8_t* audio_data;
    UINT32 num_frames_to_read = 0;
    DWORD flags = 0;
    UINT64 device_position_frames = 0;
    UINT64 capture_time_100ns = 0;
    error = audio_capture_client_->GetBuffer(&audio_data, &num_frames_to_read,
                                             &flags, &device_position_frames,
                                             &capture_time_100ns);
    if (error.Error() == AUDCLNT_S_BUFFER_EMPTY) {
      // The call succeeded but no capture data is available to be read.
      // Return and start waiting for new capture event
      RTC_DCHECK_EQ(num_frames_to_read, 0u);
      return true;
    }
    if (FAILED(error.Error())) {
      RTC_LOG(LS_ERROR) << "IAudioCaptureClient::GetBuffer failed: "
                        << core_audio_utility::ErrorToString(error);
      return false;
    }

    // Update input delay estimate but only about once per second to save
    // resources. The estimate is usually stable.
    if (num_data_callbacks_ % 100 == 0) {
      absl::optional<int> opt_record_delay_ms;
      // TODO(henrika): note that FineAudioBuffer adds latency as well.
      opt_record_delay_ms = EstimateLatencyMillis(capture_time_100ns);
      if (opt_record_delay_ms) {
        latency_ms_ = *opt_record_delay_ms;
      } else {
        RTC_DLOG(LS_WARNING) << "Input latency is set to fixed value";
        latency_ms_ = 20;
      }
    }
    if (num_data_callbacks_ % 500 == 0) {
      RTC_DLOG(INFO) << "latency: " << latency_ms_;
    }

    // The data in the packet is not correlated with the previous packet's
    // device position; possibly due to a stream state transition or timing
    // glitch. The behavior of the AUDCLNT_BUFFERFLAGS_DATA_DISCONTINUITY flag
    // is undefined on the application's first call to GetBuffer after Start.
    if (device_position_frames != 0 &&
        flags & AUDCLNT_BUFFERFLAGS_DATA_DISCONTINUITY) {
      RTC_DLOG(LS_WARNING) << "AUDCLNT_BUFFERFLAGS_DATA_DISCONTINUITY";
    }
    // The time at which the device's stream position was recorded is uncertain.
    // Thus, the client might be unable to accurately set a time stamp for the
    // current data packet.
    if (flags & AUDCLNT_BUFFERFLAGS_TIMESTAMP_ERROR) {
      RTC_DLOG(LS_WARNING) << "AUDCLNT_BUFFERFLAGS_TIMESTAMP_ERROR";
    }

    // Treat all of the data in the packet as silence and ignore the actual
    // data values when AUDCLNT_BUFFERFLAGS_SILENT is set.
    if (flags & AUDCLNT_BUFFERFLAGS_SILENT) {
      rtc::ExplicitZeroMemory(audio_data,
                              format_.Format.nBlockAlign * num_frames_to_read);
      RTC_DLOG(LS_WARNING) << "Captured audio is replaced by silence";
    } else {
      // Copy recorded audio in `audio_data` to the WebRTC sink using the
      // FineAudioBuffer object.
      fine_audio_buffer_->DeliverRecordedData(
          rtc::MakeArrayView(reinterpret_cast<const int16_t*>(audio_data),
                             format_.Format.nChannels * num_frames_to_read),

          latency_ms_);
    }

    error = audio_capture_client_->ReleaseBuffer(num_frames_to_read);
    if (FAILED(error.Error())) {
      RTC_LOG(LS_ERROR) << "IAudioCaptureClient::ReleaseBuffer failed: "
                        << core_audio_utility::ErrorToString(error);
      return false;
    }

    error =
        audio_capture_client_->GetNextPacketSize(&num_frames_in_next_packet);
    if (FAILED(error.Error())) {
      RTC_LOG(LS_ERROR) << "IAudioCaptureClient::GetNextPacketSize failed: "
                        << core_audio_utility::ErrorToString(error);
      return false;
    }
  }
  ++num_data_callbacks_;
  return true;
}

bool CoreAudioInput::OnErrorCallback(ErrorType error) {
  RTC_DLOG(INFO) << __FUNCTION__ << ": " << as_integer(error);
  RTC_DCHECK_RUN_ON(&thread_checker_audio_);
  if (error == CoreAudioBase::ErrorType::kStreamDisconnected) {
    HandleStreamDisconnected();
  } else {
    RTC_DLOG(WARNING) << "Unsupported error type";
  }
  return true;
}

absl::optional<int> CoreAudioInput::EstimateLatencyMillis(
    uint64_t capture_time_100ns) {
  if (!qpc_to_100ns_) {
    return absl::nullopt;
  }
  // Input parameter `capture_time_100ns` contains the performance counter at
  // the time that the audio endpoint device recorded the device position of
  // the first audio frame in the data packet converted into 100ns units.
  // We derive a delay estimate by:
  // - sampling the current performance counter (qpc_now_raw),
  // - converting it into 100ns time units (now_time_100ns), and
  // - subtracting `capture_time_100ns` from now_time_100ns.
  LARGE_INTEGER perf_counter_now = {};
  if (!::QueryPerformanceCounter(&perf_counter_now)) {
    return absl::nullopt;
  }
  uint64_t qpc_now_raw = perf_counter_now.QuadPart;
  uint64_t now_time_100ns = qpc_now_raw * (*qpc_to_100ns_);
  webrtc::TimeDelta delay_us = webrtc::TimeDelta::Micros(
      0.1 * (now_time_100ns - capture_time_100ns) + 0.5);
  return delay_us.ms();
}

// Called from OnErrorCallback() when error type is kStreamDisconnected.
// Note that this method is called on the audio thread and the internal restart
// sequence is also executed on that same thread. The audio thread is therefore
// not stopped during restart. Such a scheme also makes the restart process less
// complex.
// Note that, none of the called methods are thread checked since they can also
// be called on the main thread. Thread checkers are instead added on one layer
// above (in audio_device_module.cc) which ensures that the public API is thread
// safe.
// TODO(henrika): add more details.
bool CoreAudioInput::HandleStreamDisconnected() {
  RTC_DLOG(INFO) << "<<<--- " << __FUNCTION__;
  RTC_DCHECK_RUN_ON(&thread_checker_audio_);
  RTC_DCHECK(automatic_restart());

  if (StopRecording() != 0) {
    return false;
  }

  if (!SwitchDeviceIfNeeded()) {
    return false;
  }

  if (InitRecording() != 0) {
    return false;
  }
  if (StartRecording() != 0) {
    return false;
  }

  RTC_DLOG(INFO) << __FUNCTION__ << " --->>>";
  return true;
}

}  // namespace webrtc_win
}  // namespace webrtc