1 | // Copyright (C) 2022 The Qt Company Ltd. |
2 | // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only |
3 | #include "qffmpegresampler_p.h" |
4 | #include "playbackengine/qffmpegcodeccontext_p.h" |
5 | #include "qffmpegmediaformatinfo_p.h" |
6 | #include <qloggingcategory.h> |
7 | |
8 | static Q_LOGGING_CATEGORY(qLcResampler, "qt.multimedia.ffmpeg.resampler" ) |
9 | static Q_LOGGING_CATEGORY(qLcResamplerTrace, "qt.multimedia.ffmpeg.resampler.trace" ) |
10 | |
11 | QT_BEGIN_NAMESPACE |
12 | |
13 | using namespace QFFmpeg; |
14 | |
15 | QFFmpegResampler::QFFmpegResampler(const QAudioFormat &inputFormat, const QAudioFormat &outputFormat) : |
16 | m_inputFormat(inputFormat), m_outputFormat(outputFormat) |
17 | { |
18 | Q_ASSERT(inputFormat.isValid()); |
19 | Q_ASSERT(outputFormat.isValid()); |
20 | |
21 | const auto inputAvFormat = AVAudioFormat(m_inputFormat); |
22 | const auto outputAvFormat = AVAudioFormat(m_outputFormat); |
23 | |
24 | m_resampler = createResampleContext(inputFormat: inputAvFormat, outputFormat: outputAvFormat); |
25 | |
26 | qCDebug(qLcResampler) << "Created QFFmpegResampler with offset" << m_startTime |
27 | << "us. Converting from" << inputAvFormat << "to" << outputAvFormat; |
28 | } |
29 | |
30 | QFFmpegResampler::QFFmpegResampler(const CodecContext *codecContext, |
31 | const QAudioFormat &outputFormat, |
32 | qint64 startTime) |
33 | : m_outputFormat(outputFormat), m_startTime(startTime) |
34 | { |
35 | Q_ASSERT(codecContext); |
36 | |
37 | const AVStream *audioStream = codecContext->stream(); |
38 | |
39 | if (!m_outputFormat.isValid()) |
40 | // want the native format |
41 | m_outputFormat = QFFmpegMediaFormatInfo::audioFormatFromCodecParameters(codecPar: *audioStream->codecpar); |
42 | |
43 | const auto inputAvFormat = AVAudioFormat(audioStream->codecpar); |
44 | const auto outputAvFormat = AVAudioFormat(m_outputFormat); |
45 | |
46 | m_resampler = createResampleContext(inputFormat: inputAvFormat, outputFormat: outputAvFormat); |
47 | |
48 | qCDebug(qLcResampler).nospace() << "Created QFFmpegResampler. Offset: " << m_startTime |
49 | << "us. From: " << inputAvFormat << " to: " << outputAvFormat; |
50 | } |
51 | |
52 | QFFmpegResampler::~QFFmpegResampler() = default; |
53 | |
54 | QAudioBuffer QFFmpegResampler::resample(const char* data, size_t size) |
55 | { |
56 | if (!m_inputFormat.isValid()) |
57 | return {}; |
58 | |
59 | return resample(inputData: reinterpret_cast<const uint8_t **>(&data), |
60 | inputSamplesCount: m_inputFormat.framesForBytes(byteCount: static_cast<qint32>(size))); |
61 | } |
62 | |
63 | QAudioBuffer QFFmpegResampler::resample(const AVFrame *frame) |
64 | { |
65 | return resample(inputData: const_cast<const uint8_t **>(frame->extended_data), inputSamplesCount: frame->nb_samples); |
66 | } |
67 | |
68 | QAudioBuffer QFFmpegResampler::resample(const uint8_t **inputData, int inputSamplesCount) |
69 | { |
70 | const int maxOutSamples = adjustMaxOutSamples(inputSamplesCount); |
71 | |
72 | QByteArray samples(m_outputFormat.bytesForFrames(frameCount: maxOutSamples), Qt::Uninitialized); |
73 | auto *out = reinterpret_cast<uint8_t *>(samples.data()); |
74 | const int outSamples = |
75 | swr_convert(s: m_resampler.get(), out: &out, out_count: maxOutSamples, in: inputData, in_count: inputSamplesCount); |
76 | |
77 | samples.resize(size: m_outputFormat.bytesForFrames(frameCount: outSamples)); |
78 | |
79 | const qint64 startTime = m_outputFormat.durationForFrames(frameCount: m_samplesProcessed) + m_startTime; |
80 | m_samplesProcessed += outSamples; |
81 | |
82 | qCDebug(qLcResamplerTrace).nospace() |
83 | << "Created output buffer. Time stamp: " << startTime |
84 | << "us. Samples in: " << inputSamplesCount |
85 | << ", Samples out: " << outSamples << ", Max samples: " << maxOutSamples; |
86 | return QAudioBuffer(samples, m_outputFormat, startTime); |
87 | } |
88 | |
89 | int QFFmpegResampler::adjustMaxOutSamples(int inputSamplesCount) |
90 | { |
91 | int maxOutSamples = swr_get_out_samples(s: m_resampler.get(), in_samples: inputSamplesCount); |
92 | |
93 | const auto remainingCompensationDistance = m_endCompensationSample - m_samplesProcessed; |
94 | |
95 | if (remainingCompensationDistance > 0 && maxOutSamples > remainingCompensationDistance) { |
96 | // If the remaining compensation distance less than output frame, |
97 | // the ffmpeg resampler bufferises the rest of frames that makes |
98 | // unexpected delays on large frames. |
99 | // The hack might cause some compensation bias on large frames, |
100 | // however it's not significant for our logic, in fact. |
101 | // TODO: probably, it will need some improvements |
102 | setSampleCompensation(delta: 0, distance: 0); |
103 | maxOutSamples = swr_get_out_samples(s: m_resampler.get(), in_samples: inputSamplesCount); |
104 | } |
105 | |
106 | return maxOutSamples; |
107 | } |
108 | |
109 | void QFFmpegResampler::setSampleCompensation(qint32 delta, quint32 distance) |
110 | { |
111 | const int res = swr_set_compensation(s: m_resampler.get(), sample_delta: delta, compensation_distance: static_cast<int>(distance)); |
112 | if (res < 0) |
113 | qCWarning(qLcResampler) << "swr_set_compensation fail:" << res; |
114 | else { |
115 | m_sampleCompensationDelta = delta; |
116 | m_endCompensationSample = m_samplesProcessed + distance; |
117 | } |
118 | } |
119 | |
120 | qint32 QFFmpegResampler::activeSampleCompensationDelta() const |
121 | { |
122 | return m_samplesProcessed < m_endCompensationSample ? m_sampleCompensationDelta : 0; |
123 | } |
124 | |
125 | QT_END_NAMESPACE |
126 | |