qaudioengine_threaded.cpp

Go to the documentation of this file.

// Copyright (C) 2026 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-3.0-only
 
#include "qaudioengine_threaded_p.h"
 
#include <QtMultimedia/qmediadevices.h>
#include <QtMultimedia/qaudiosink.h>
#include <QtMultimedia/private/qaudio_qspan_support_p.h>
#include <QtMultimedia/private/qaudiohelpers_p.h>
#ifdef Q_OS_WIN
#  include <QtMultimedia/private/qwindows_wasapi_warmup_client_p.h>
#endif
#include <QtSpatialAudio/qambientsound.h>
#include <QtSpatialAudio/qaudiolistener.h>
#include <QtSpatialAudio/private/qaudioroom_p.h>
#include <QtSpatialAudio/private/qambisonicdecoder_p.h>
#include <QtSpatialAudio/private/qspatialaudiosound_p.h>
#include <QtCore/qiodevice.h>
#include <QtCore/qdebug.h>
#include <QtCore/qelapsedtimer.h>
 
#include <resonance_audio.h>
 
#include <memory>
#include <q20vector.h>
 
QT_BEGIN_NAMESPACE
 
// We'd like to have short buffer times, so the sound adjusts itself to changes
// quickly, but times below 100ms seem to give stuttering on macOS.
// It might be possible to set this value lower on other OSes.
const int bufferTimeMs = 100;
 
// This class lives in the audioThread, but pulls data from QAudioEnginePrivate
// which lives in the mainThread.
class QAudioOutputStream : public QIODevice
{
public:
    explicit QAudioOutputStream(QAudioEngineThreaded *d) : d(d) { open(QIODevice::ReadOnly); }
    ~QAudioOutputStream() override;
 
    qint64 readData(char *data, qint64 len) override;
 
    qint64 writeData(const char *, qint64) override;
 
    qint64 size() const override { return 0; }
    qint64 bytesAvailable() const override {
        return std::numeric_limits<qint64>::max();
    }
    bool isSequential() const override {
        return true;
    }
    bool atEnd() const override {
        return false;
    }
    qint64 pos() const override {
        return m_pos;
    }
 
    void startOutput()
    {
        d->mutex.lock();
        Q_ASSERT(!sink);
        auto channelConfig = d->outputMode() == QAudioEngine::Surround
                ? d->m_device.channelConfiguration()
                : QAudioFormat::ChannelConfigStereo;
 
        QAudioFormat format;
        if (channelConfig != QAudioFormat::ChannelConfigUnknown)
            format.setChannelConfig(channelConfig);
        else
            format.setChannelCount(d->m_device.preferredFormat().channelCount());
        format.setSampleRate(d->sampleRate());
        format.setSampleFormat(QAudioFormat::Int16);
 
        ambisonicDecoder = std::make_unique<QAmbisonicDecoder>(
                QAmbisonicDecoder::AmbisonicOrder::HighQuality, format.sampleRate(),
                format.channelCount(), format.channelConfig());
 
        sink = std::make_unique<QAudioSink>(d->m_device, format);
        const qsizetype bufferSize = format.bytesForDuration(bufferTimeMs * 1000);
        sink->setBufferSize(bufferSize);
        d->mutex.unlock();
        // It is important to unlock the mutex before starting the sink, as the sink will
        // call readData() in the audio thread, which will try to lock the mutex (again)
        sink->start(this);
 
#ifdef Q_OS_WIN
        QtMultimediaPrivate::refreshWarmupClient();
#endif
    }
 
    void stopOutput()
    {
        if (!sink)
            return;
        sink->stop();
        sink.reset();
        ambisonicDecoder.reset();
    }
 
    void setPaused(bool paused) {
        if (paused)
            sink->suspend();
        else
            sink->resume();
    }
 
private:
    qint64 m_pos = 0;
    QAudioEngineThreaded *d = nullptr;
    std::unique_ptr<QAudioSink> sink;
    std::unique_ptr<QAmbisonicDecoder> ambisonicDecoder;
};
 
QAudioOutputStream::~QAudioOutputStream() = default;
 
qint64 QAudioOutputStream::writeData(const char *, qint64)
{
    return 0;
}
 
 
qint64 QAudioOutputStream::readData(char *data, const qint64 len)
{
    constexpr int framesPerBuffer = qToUnderlying(QAudioEnginePrivate::framesPerBuffer);
    static constexpr std::array<float, 2 * framesPerBuffer> nullBuffer{};
 
    if (d->m_paused.loadRelaxed())
        return 0;
 
    QSpan<short> outputBuffer((short *)data, len / sizeof(short));
 
    QMutexLocker l(&d->mutex);
 
    int nChannels = ambisonicDecoder ? ambisonicDecoder->nOutputChannels() : 2;
    if (outputBuffer.size() < nChannels * framesPerBuffer)
        return 0;
 
    using QtMultimediaPrivate::drop;
    using QtMultimediaPrivate::take;
    using namespace QAudioHelperInternal;
 
    const std::unique_ptr<vraudio::ResonanceAudioApi> &api = d->resonanceAudio->api;
 
    bool ok = true;
    while (outputBuffer.size() >= nChannels * framesPerBuffer) {
 
        // Fill input buffers
        for (auto &&[source, playbackState] : d->playbackStates) {
            Q_ASSERT(source->nchannels <= 2);
            if (playbackState) {
                Q_ASSERT(playbackState->format().channelCount() <= 2);
                std::array<float, 2 * framesPerBuffer> buf;
 
                playbackState->getBuffer(
                        take(QSpan<float>{ buf },
                             playbackState->format().channelCount() * framesPerBuffer));
                api->SetInterleavedBuffer(source->sourceId, buf.data(), source->nchannels,
                                          framesPerBuffer);
            } else {
                api->SetInterleavedBuffer(source->sourceId, nullBuffer.data(), source->nchannels,
                                          framesPerBuffer);
            }
        }
 
        if (ambisonicDecoder && d->outputMode() == QAudioEngine::Surround) {
            std::array<const float *, QAmbisonicDecoder::maxAmbisonicChannels> channels;
            std::array<const float *, 2> reverbBuffers{};
            int nFrames = d->resonanceAudio->getAmbisonicOutput(
                    channels.data(), reverbBuffers.data(), ambisonicDecoder->nInputChannels());
 
            if (nFrames < 0) {
                // If we get here, it means that resonanceAudio did not actually fill the buffer.
                // Sometimes this is expected, for example if resonanceAudio does not have any sources.
                // In this case we just fill the buffer with silence.
                std::fill(outputBuffer.begin(), outputBuffer.end(), 0);
                break;
            }
 
            Q_ASSERT(ambisonicDecoder->nOutputChannels() <= 8);
            int nSamples = ambisonicDecoder->outputSamples(nFrames);
 
            constexpr size_t reverbBufferSize =
                    framesPerBuffer * QAmbisonicDecoder::maxAmbisonicChannels;
            std::array<float, reverbBufferSize> reverbFloatBuffers;
            QSpan<float> reverbOutputSpan = take(QSpan{ reverbFloatBuffers }, nSamples);
            QSpan<short> currentOutput = take(outputBuffer, nSamples);
 
            ambisonicDecoder->processBufferWithReverb(
                    QSpan{ channels.data(), ambisonicDecoder->nInputChannels() }, reverbBuffers,
                    reverbOutputSpan);
 
            convertSampleFormat(as_bytes(reverbOutputSpan), NativeSampleFormat::float32_t,
                                as_writable_bytes(currentOutput), NativeSampleFormat::int16_t);
            outputBuffer = drop(outputBuffer, nSamples);
        } else {
            QSpan<short> currentOutput = take(outputBuffer, nChannels * framesPerBuffer);
            ok = d->resonanceAudio->api->FillInterleavedOutputBuffer(2, framesPerBuffer,
                                                                     currentOutput.data());
            if (!ok) {
                // If we get here, it means that resonanceAudio did not actually fill the buffer.
                // Sometimes this is expected, for example if resonanceAudio does not have any sources.
                // In this case we just fill the buffer with silence.
                if (d->playbackStates.empty()) {
                    std::fill(currentOutput.begin(), currentOutput.end(), 0);
                } else {
                    // If we get here, it means that something unexpected happened, so bail.
                    qWarning() << "    Reading failed!";
                    break;
                }
            }
            outputBuffer = drop(outputBuffer, nChannels * framesPerBuffer);
        }
    }
 
    qint64 bytesProcessed = len - outputBuffer.size_bytes();
    m_pos += bytesProcessed;
    return bytesProcessed;
}
 
QAudioEngineThreaded::QAudioEngineThreaded(int sampleRate) : QAudioEnginePrivate(sampleRate)
{
    audioThread.setObjectName(u"QAudioThread");
    m_device = QMediaDevices::defaultAudioOutput();
}
 
QAudioEngineThreaded::~QAudioEngineThreaded()
{
    stop();
}
 
void QAudioEngineThreaded::start()
{
    if (outputStream)
        return; // already started
 
    outputStream = std::make_unique<QAudioOutputStream>(this);
    outputStream->moveToThread(&audioThread);
    audioThread.start(QThread::TimeCriticalPriority);
 
    QMetaObject::invokeMethod(outputStream.get(), &QAudioOutputStream::startOutput);
}
 
void QAudioEngineThreaded::stop()
{
    if (!outputStream)
        return; // already stopped
 
    QMetaObject::invokeMethod(outputStream.get(), &QAudioOutputStream::stopOutput,
                              Qt::BlockingQueuedConnection);
    outputStream.reset();
    audioThread.exit(0);
    audioThread.wait();
}
 
void QAudioEngineThreaded::setPaused(bool paused)
{
    if (!outputStream)
        return; // can't pause if not started
 
    bool old = m_paused.fetchAndStoreRelaxed(paused);
    if (old != paused) {
        if (outputStream)
            outputStream->setPaused(paused);
        Q_Q(QAudioEngine);
        emit q->pausedChanged();
    }
}
 
bool QAudioEngineThreaded::isPaused() const
{
    return m_paused.loadRelaxed();
}
 
void QAudioEngineThreaded::setOutputDevice(const QAudioDevice &device)
{
    if (m_device == device)
        return;
    if (outputStream) {
        qWarning() << "Changing device on a running engine not implemented";
        return;
    }
    m_device = device;
    Q_Q(QAudioEngine);
    emit q->outputDeviceChanged();
}
 
void QAudioEngineThreaded::addSound(QSpatialAudioSoundPrivate *sound)
{
    std::lock_guard l(mutex);
    playbackStates.emplace(sound, nullptr);
}
 
void QAudioEngineThreaded::removeSound(QSpatialAudioSoundPrivate *sound)
{
    std::lock_guard l(mutex);
    playbackStates.erase(sound);
}
 
void QAudioEngineThreaded::setSoundPlaybackData(QSpatialAudioSoundPrivate *sound,
                                                SharedPlaybackState state)
{
    std::lock_guard l(mutex);
    playbackStates.insert_or_assign(sound, std::move(state));
}
 
void QAudioEngineThreaded::updateRoomEffects()
{
    std::lock_guard l(mutex);
    for (auto [sound, key] : playbackStates)
        sound->updateRoomEffects();
}
 
QT_END_NAMESPACE