qffmpegaudioencoder.cpp

Go to the documentation of this file.

// Copyright (C) 2024 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only
#include "qffmpegaudioencoder_p.h"
#include "qffmpegrecordingengineutils_p.h"
#include "qffmpegaudioencoderutils_p.h"
#include "qffmpegaudioinput_p.h"
#include "qffmpegencoderoptions_p.h"
#include "qffmpegmuxer_p.h"
#include "qffmpegrecordingengine_p.h"
#include "qffmpegmediaformatinfo_p.h"
#include "qffmpegcodecstorage_p.h"
#include <QtCore/qloggingcategory.h>
 
QT_BEGIN_NAMESPACE
 
namespace QFFmpeg {
 
Q_STATIC_LOGGING_CATEGORY(qLcFFmpegAudioEncoder, "qt.multimedia.ffmpeg.audioencoder");
static constexpr bool audioEncoderExtendedTracing = false;
 
namespace {
void setupStreamParameters(AVStream *stream, const Codec &codec,
                           const AVAudioFormat &requestedAudioFormat)
{
    const auto channelLayouts = codec.channelLayouts();
#if QT_FFMPEG_HAS_AV_CHANNEL_LAYOUT
    stream->codecpar->ch_layout =
            adjustChannelLayout(channelLayouts, requestedAudioFormat.channelLayout);
#else
    stream->codecpar->channel_layout =
            adjustChannelLayout(channelLayouts, requestedAudioFormat.channelLayoutMask);
    stream->codecpar->channels = qPopulationCount(stream->codecpar->channel_layout);
#endif
    const auto sampleRates = codec.sampleRates();
    const auto sampleRate = adjustSampleRate(sampleRates, requestedAudioFormat.sampleRate);
 
    stream->codecpar->sample_rate = sampleRate;
    // Codecs with a variable frame size (e.g. PCM) must keep frame_size at 0;
    // FFmpeg 9 rejects any frame whose nb_samples exceeds a non-zero frame_size,
    // even for such codecs. Codecs with a fixed frame size determine and set
    // their own frame_size during avcodec_open2, overriding this default.
    if (!(codec.capabilities() & AV_CODEC_CAP_VARIABLE_FRAME_SIZE))
        stream->codecpar->frame_size = 1024;
    const auto sampleFormats = codec.sampleFormats();
    stream->codecpar->format = adjustSampleFormat(sampleFormats, requestedAudioFormat.sampleFormat);
 
    stream->time_base = AVRational{ 1, sampleRate };
 
    qCDebug(qLcFFmpegAudioEncoder)
            << "set stream time_base" << stream->time_base.num << "/" << stream->time_base.den;
}
 
bool openCodecContext(AVCodecContext *codecContext, AVStream *stream,
                      const QMediaEncoderSettings &settings)
{
    Q_ASSERT(codecContext);
    codecContext->time_base = stream->time_base;
 
    int status = avcodec_parameters_to_context(codecContext, stream->codecpar);
    if (status < 0) {
        qCWarning(qLcFFmpegAudioEncoder)
                << "Cannot set codec parameters; result:" << AVError(status);
        return false;
    }
 
    // if avcodec_open2 fails, it may clean codecContext->codec
    Codec codec{ codecContext->codec };
 
    AVDictionaryHolder opts;
    applyAudioEncoderOptions(settings, codec.name(), codecContext, opts);
    applyExperimentalCodecOptions(codec, opts);
 
    status = avcodec_open2(codecContext, codec.get(), opts);
 
    if (status != 0) {
        qCWarning(qLcFFmpegAudioEncoder)
                << "Cannot open audio codec" << codec.name() << "; result:" << AVError(status);
        return false;
    }
 
    qCDebug(qLcFFmpegAudioEncoder) << "audio codec params: fmt=" << codecContext->sample_fmt
                                   << "rate=" << codecContext->sample_rate;
 
    status = avcodec_parameters_from_context(stream->codecpar, codecContext);
    if (status < 0) {
        qCWarning(qLcFFmpegAudioEncoder) << "Cannot get parameters from audio codec context"
                                         << codec.name() << "; result:" << AVError(status);
        return false;
    }
 
    return true;
}
 
} // namespace
 
AudioEncoder::AudioEncoder(RecordingEngine &recordingEngine, const QAudioFormat &sourceFormat,
                           const QMediaEncoderSettings &settings)
    : EncoderThread(recordingEngine), m_sourceFormat(sourceFormat), m_settings(settings)
{
    setObjectName(QLatin1String("AudioEncoder"));
    qCDebug(qLcFFmpegAudioEncoder) << "AudioEncoder" << settings.audioCodec();
 
    const AVCodecID codecID = QFFmpegMediaFormatInfo::codecIdForAudioCodec(settings.audioCodec());
    Q_ASSERT(avformat_query_codec(recordingEngine.avFormatContext()->oformat, codecID,
                                  FF_COMPLIANCE_NORMAL));
 
    Q_ASSERT(QFFmpeg::findAVEncoder(codecID));
 
    m_stream = avformat_new_stream(recordingEngine.avFormatContext(), nullptr);
    m_stream->id = recordingEngine.avFormatContext()->nb_streams - 1;
    m_stream->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
    m_stream->codecpar->codec_id = codecID;
}
 
void AudioEncoder::addBuffer(const QAudioBuffer &buffer)
{
    if (!buffer.isValid()) {
        setEndOfSourceStream();
        return;
    }
 
    {
        const std::chrono::microseconds bufferDuration(buffer.duration());
        auto guard = lockLoopData();
 
        resetEndOfSourceStream();
 
        if (m_paused)
            return;
 
        // TODO: apply logic with canPushFrame
 
        m_audioBufferQueue.push(buffer);
        m_queueDuration += bufferDuration;
    }
 
    dataReady();
}
 
QAudioBuffer AudioEncoder::takeBuffer()
{
    auto locker = lockLoopData();
    QAudioBuffer result = dequeueIfPossible(m_audioBufferQueue);
    m_queueDuration -= std::chrono::microseconds(result.duration());
    return result;
}
 
bool AudioEncoder::init()
{
    const AVAudioFormat requestedAudioFormat(m_sourceFormat);
    const AVCodecID codecId = m_stream->codecpar->codec_id;
 
    namespace ranges = QtMultimediaPrivate::ranges;
 
    const auto encoders = findAndScoreEncoders(codecId, [&](const Codec &codec) {
        AVScore result = DefaultAVScore;
 
        // Attempt to find no-conversion format
        if (auto fmts = codec.sampleFormats(); !fmts.empty())
            result += ranges::contains(fmts, requestedAudioFormat.sampleFormat) ? 1 : -1;
 
        if (auto rates = codec.sampleRates(); !rates.empty())
            result += ranges::contains(rates, requestedAudioFormat.sampleRate) ? 1 : -1;
 
#if QT_FFMPEG_HAS_AV_CHANNEL_LAYOUT
        if (auto layouts = codec.channelLayouts(); !layouts.empty())
            result += ranges::contains(layouts, requestedAudioFormat.channelLayout) ? 1 : -1;
#else
        if (auto layouts = codec.channelLayouts(); !layouts.empty())
            result +=
                    ranges::contains(layouts, requestedAudioFormat.channelLayoutMask) ? 1 : -1;
#endif
 
        return result;
    });
 
    for (const auto &[codec, score] : encoders) {
        AVCodecContextUPtr codecContext(avcodec_alloc_context3(codec.get()));
        if (!codecContext)
            continue;
 
        setupStreamParameters(m_stream, Codec{ codecContext->codec }, requestedAudioFormat);
        if (!openCodecContext(codecContext.get(), m_stream, m_settings))
            continue;
 
        m_codecContext = std::move(codecContext);
        break;
    }
 
    if (!m_codecContext) {
        qCWarning(qLcFFmpegAudioEncoder) << "Unable to open any audio codec";
        emit m_recordingEngine.sessionError(QMediaRecorder::FormatError,
                                            QStringLiteral("Cannot open any audio codec"));
        return false;
    }
 
    qCDebug(qLcFFmpegAudioEncoder) << "found audio codec" << m_codecContext->codec->name;
 
    updateResampler(m_sourceFormat);
 
    // TODO: try to address this dependency here.
    if (auto input = qobject_cast<QFFmpegAudioInput *>(source()))
        input->setBufferSize(m_codecContext->frame_size);
 
    return EncoderThread::init();
}
 
void AudioEncoder::cleanup()
{
    while (!m_audioBufferQueue.empty())
        processOne();
 
    if (m_avFrameSamplesOffset) {
        // the size of the last frame can be less than m_codecContext->frame_size
 
        retrievePackets();
        sendPendingFrameToAVCodec();
    }
 
    while (avcodec_send_frame(m_codecContext.get(), nullptr) == AVERROR(EAGAIN))
        retrievePackets();
    retrievePackets();
}
 
bool AudioEncoder::hasData() const
{
    return !m_audioBufferQueue.empty();
}
 
void AudioEncoder::retrievePackets()
{
    while (true) {
        AVPacketUPtr packet(av_packet_alloc());
        int ret = avcodec_receive_packet(m_codecContext.get(), packet.get());
        switch (ret) {
        case 0:
            break;
        case AVERROR(EAGAIN):
        case AVERROR(EOF):
            return;
        default:
            qCDebug(qLcFFmpegAudioEncoder) << "receive packet" << ret << QFFmpeg::AVError{ ret };
            return;
        }
 
        if constexpr (audioEncoderExtendedTracing)
            qCDebug(qLcFFmpegAudioEncoder)
                    << "writing audio packet" << packet->size << packet->pts << packet->dts;
        packet->stream_index = m_stream->id;
        m_recordingEngine.getMuxer()->addPacket(std::move(packet));
    }
}
 
void AudioEncoder::processOne()
{
    QAudioBuffer buffer = takeBuffer();
    Q_ASSERT(buffer.isValid());
 
    if constexpr (audioEncoderExtendedTracing)
        qCDebug(qLcFFmpegAudioEncoder)
                << "new audio buffer" << buffer.byteCount() << buffer.format()
                << buffer.frameCount() << m_codecContext->frame_size;
 
    if (buffer.format() != m_sourceFormat && !updateResampler(buffer.format()))
        return;
 
    int samplesOffset = 0;
    const int bufferSamplesCount = static_cast<int>(buffer.frameCount());
 
    while (samplesOffset < bufferSamplesCount)
        handleAudioData(buffer.constData<uint8_t>(), samplesOffset, bufferSamplesCount);
 
    Q_ASSERT(samplesOffset == bufferSamplesCount);
}
 
bool AudioEncoder::checkIfCanPushFrame() const
{
    if (m_encodingStarted)
        return m_audioBufferQueue.size() <= 1 || m_queueDuration < m_maxQueueDuration;
    if (!isFinished())
        return m_audioBufferQueue.empty();
 
    return false;
}
 
bool AudioEncoder::updateResampler(const QAudioFormat &sourceFormat)
{
    m_resampler.reset();
 
    const AVAudioFormat requestedAudioFormat(sourceFormat);
    const AVAudioFormat codecAudioFormat(m_codecContext.get());
 
    if (requestedAudioFormat != codecAudioFormat) {
        m_resampler = createResampleContext(requestedAudioFormat, codecAudioFormat);
        if (!swr_is_initialized(m_resampler.get())) {
            m_sourceFormat = {};
            qCWarning(qLcFFmpegAudioEncoder) << "Cannot initialize resampler for audio encoder";
            emit m_recordingEngine.sessionError(
                    QMediaRecorder::FormatError,
                    QStringLiteral("Cannot initialize resampler for audio encoder"));
            return false;
        }
        qCDebug(qLcFFmpegAudioEncoder) << "Created resampler with audio formats conversion\n"
                                       << requestedAudioFormat << "->" << codecAudioFormat;
    } else {
        qCDebug(qLcFFmpegAudioEncoder) << "Resampler is not needed due to no-conversion format\n"
                                       << requestedAudioFormat;
    }
 
    m_sourceFormat = sourceFormat;
 
    return true;
}
 
void AudioEncoder::ensurePendingFrame(int availableSamplesCount)
{
    Q_ASSERT(availableSamplesCount >= 0);
 
    if (m_avFrame)
        return;
 
    m_avFrame = makeAVFrame();
 
    m_avFrame->format = m_codecContext->sample_fmt;
#if QT_FFMPEG_HAS_AV_CHANNEL_LAYOUT
    m_avFrame->ch_layout = m_codecContext->ch_layout;
#else
    m_avFrame->channel_layout = m_codecContext->channel_layout;
    m_avFrame->channels = m_codecContext->channels;
#endif
    m_avFrame->sample_rate = m_codecContext->sample_rate;
 
    const bool isFixedFrameSize =
            !(m_codecContext->codec->capabilities & AV_CODEC_CAP_VARIABLE_FRAME_SIZE)
            && m_codecContext->frame_size;
    m_avFrame->nb_samples = isFixedFrameSize ? m_codecContext->frame_size : availableSamplesCount;
    if (m_avFrame->nb_samples) {
        const int status = av_frame_get_buffer(m_avFrame.get(), 0);
        if (status < 0)
            qCWarning(qLcFFmpegAudioEncoder)
                    << "Failed to allocate audio frame buffer:" << AVError(status);
    }
 
    const auto &timeBase = m_stream->time_base;
    const auto pts = timeBase.den && timeBase.num
            ? timeBase.den * m_samplesWritten / (m_codecContext->sample_rate * timeBase.num)
            : m_samplesWritten;
    setAVFrameTime(*m_avFrame, pts, timeBase);
}
 
void AudioEncoder::writeDataToPendingFrame(const uchar *data, int &samplesOffset, int samplesCount)
{
    Q_ASSERT(m_avFrame);
    Q_ASSERT(m_avFrameSamplesOffset <= m_avFrame->nb_samples);
 
    const int bytesPerSample = av_get_bytes_per_sample(m_codecContext->sample_fmt);
    const bool isPlanar = av_sample_fmt_is_planar(m_codecContext->sample_fmt);
 
#if QT_FFMPEG_HAS_AV_CHANNEL_LAYOUT
    const int channelsCount = m_codecContext->ch_layout.nb_channels;
#else
    const int channelsCount = m_codecContext->channels;
#endif
 
    const int audioDataOffset = isPlanar ? bytesPerSample * m_avFrameSamplesOffset
                                         : bytesPerSample * m_avFrameSamplesOffset * channelsCount;
 
    const int planesCount = isPlanar ? channelsCount : 1;
    m_avFramePlanesData.resize(planesCount);
    for (int plane = 0; plane < planesCount; ++plane)
        m_avFramePlanesData[plane] = m_avFrame->extended_data[plane] + audioDataOffset;
 
    const int samplesToWrite = m_avFrame->nb_samples - m_avFrameSamplesOffset;
    int samplesToRead =
            (samplesToWrite * m_sourceFormat.sampleRate() + m_codecContext->sample_rate / 2)
            / m_codecContext->sample_rate;
    // the lower bound is need to get round infinite loops in corner cases
    samplesToRead = qBound(1, samplesToRead, samplesCount - samplesOffset);
 
    data += m_sourceFormat.bytesForFrames(samplesOffset);
 
    if (m_resampler) {
        m_avFrameSamplesOffset += swr_convert(m_resampler.get(), m_avFramePlanesData.data(),
                                              samplesToWrite, &data, samplesToRead);
    } else {
        Q_ASSERT(planesCount == 1);
        m_avFrameSamplesOffset += samplesToRead;
        memcpy(m_avFramePlanesData[0], data, m_sourceFormat.bytesForFrames(samplesToRead));
    }
 
    samplesOffset += samplesToRead;
}
 
void AudioEncoder::sendPendingFrameToAVCodec()
{
    Q_ASSERT(m_avFrame);
    Q_ASSERT(m_avFrameSamplesOffset <= m_avFrame->nb_samples);
 
    m_avFrame->nb_samples = m_avFrameSamplesOffset;
 
    m_samplesWritten += m_avFrameSamplesOffset;
 
    const qint64 time = m_sourceFormat.durationForFrames(
            m_samplesWritten * m_sourceFormat.sampleRate() / m_codecContext->sample_rate);
    m_recordingEngine.newTimeStamp(time / 1000);
 
    if constexpr (audioEncoderExtendedTracing)
        qCDebug(qLcFFmpegAudioEncoder) << "sendPendingFrameToAVCodec" << m_avFrame->nb_samples
                                       << m_codecContext->frame_size << m_avFrame->pts;
 
    int ret = avcodec_send_frame(m_codecContext.get(), m_avFrame.get());
    if (ret < 0)
        qCDebug(qLcFFmpegAudioEncoder) << "error sending frame" << ret << QFFmpeg::AVError(ret);
 
    m_avFrame = nullptr;
    m_avFrameSamplesOffset = 0;
    std::fill(m_avFramePlanesData.begin(), m_avFramePlanesData.end(), nullptr);
}
 
void AudioEncoder::handleAudioData(const uchar *data, int &samplesOffset, int samplesCount)
{
    ensurePendingFrame(samplesCount - samplesOffset);
 
    writeDataToPendingFrame(data, samplesOffset, samplesCount);
 
    // The frame is not ready yet
    if (m_avFrameSamplesOffset < m_avFrame->nb_samples)
        return;
 
    retrievePackets();
 
    sendPendingFrameToAVCodec();
}
 
} // namespace QFFmpeg
 
QT_END_NAMESPACE