tdesktop/Telegram/SourceFiles/audio.cpp

709 lines
20 KiB
C++
Raw Normal View History

2014-09-04 07:33:44 +00:00
/*
This file is part of Telegram Desktop,
the official desktop version of Telegram messaging app, see https://telegram.org
2014-09-04 07:33:44 +00:00
Telegram Desktop is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
It is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
Full license: https://github.com/telegramdesktop/tdesktop/blob/master/LICENSE
Copyright (c) 2014 John Preston, https://desktop.telegram.org
2014-09-04 07:33:44 +00:00
*/
#include "stdafx.h"
#include "audio.h"
#include <AL/al.h>
#include <AL/alc.h>
#include <opusfile.h>
#include <ogg/ogg.h>
2014-09-04 07:33:44 +00:00
namespace {
ALCdevice *audioDevice = 0;
ALCcontext *audioContext = 0;
ALuint notifySource = 0;
ALuint notifyBuffer = 0;
QMutex voicemsgsMutex;
VoiceMessages *voicemsgs = 0;
}
bool _checkALCError() {
ALenum errCode;
if ((errCode = alcGetError(audioDevice)) != ALC_NO_ERROR) {
LOG(("Audio Error: (alc) %1").arg((const char *)alcGetString(audioDevice, errCode)));
return false;
}
return true;
}
bool _checkALError() {
ALenum errCode;
if ((errCode = alGetError()) != AL_NO_ERROR) {
LOG(("Audio Error: (al) %1").arg((const char *)alGetString(errCode)));
return false;
}
return true;
}
void audioInit() {
if (audioDevice) return;
audioDevice = alcOpenDevice(NULL);
if (!audioDevice) {
LOG(("Audio Error: default sound device not present."));
return;
}
ALCint attributes[] = { ALC_STEREO_SOURCES, 8, 0 };
audioContext = alcCreateContext(audioDevice, attributes);
alcMakeContextCurrent(audioContext);
if (!_checkALCError()) return audioFinish();
ALfloat v[] = { 0.f, 0.f, -1.f, 0.f, 1.f, 0.f };
alListener3f(AL_POSITION, 0.f, 0.f, 0.f);
alListener3f(AL_VELOCITY, 0.f, 0.f, 0.f);
alListenerfv(AL_ORIENTATION, v);
alDistanceModel(AL_NONE);
alGenSources(1, &notifySource);
alSourcef(notifySource, AL_PITCH, 1.f);
alSourcef(notifySource, AL_GAIN, 1.f);
alSource3f(notifySource, AL_POSITION, 0, 0, 0);
alSource3f(notifySource, AL_VELOCITY, 0, 0, 0);
alSourcei(notifySource, AL_LOOPING, 0);
alGenBuffers(1, &notifyBuffer);
if (!_checkALError()) return audioFinish();
QFile notify(st::newMsgSound);
if (!notify.open(QIODevice::ReadOnly)) return audioFinish();
QByteArray blob = notify.readAll();
const char *data = blob.constData();
if (blob.size() < 44) return audioFinish();
if (*((const uint32*)(data + 0)) != 0x46464952) return audioFinish(); // ChunkID - "RIFF"
if (*((const uint32*)(data + 4)) != uint32(blob.size() - 8)) return audioFinish(); // ChunkSize
2014-09-04 07:33:44 +00:00
if (*((const uint32*)(data + 8)) != 0x45564157) return audioFinish(); // Format - "WAVE"
if (*((const uint32*)(data + 12)) != 0x20746d66) return audioFinish(); // Subchunk1ID - "fmt "
uint32 subchunk1Size = *((const uint32*)(data + 16)), extra = subchunk1Size - 16;
if (subchunk1Size < 16 || (extra && extra < 2)) return audioFinish();
2014-09-04 07:33:44 +00:00
if (*((const uint16*)(data + 20)) != 1) return audioFinish(); // AudioFormat - PCM (1)
uint16 numChannels = *((const uint16*)(data + 22));
if (numChannels != 1 && numChannels != 2) return audioFinish();
uint32 sampleRate = *((const uint32*)(data + 24));
uint32 byteRate = *((const uint32*)(data + 28));
uint16 blockAlign = *((const uint16*)(data + 32));
uint16 bitsPerSample = *((const uint16*)(data + 34));
if (bitsPerSample % 8) return audioFinish();
uint16 bytesPerSample = bitsPerSample / 8;
if (bytesPerSample != 1 && bytesPerSample != 2) return audioFinish();
if (blockAlign != numChannels * bytesPerSample) return audioFinish();
if (byteRate != sampleRate * blockAlign) return audioFinish();
if (extra) {
uint16 extraSize = *((const uint16*)(data + 36));
if (uint32(extraSize + 2) != extra) return audioFinish();
if (uint32(blob.size()) < 44 + extra) return audioFinish();
2014-09-04 07:33:44 +00:00
}
if (*((const uint32*)(data + extra + 36)) != 0x61746164) return audioFinish(); // Subchunk2ID - "data"
uint32 subchunk2Size = *((const uint32*)(data + extra + 40));
if (subchunk2Size % (numChannels * bytesPerSample)) return audioFinish();
uint32 numSamples = subchunk2Size / (numChannels * bytesPerSample);
if (uint32(blob.size()) < 44 + extra + subchunk2Size) return audioFinish();
2014-09-04 07:33:44 +00:00
data += 44 + extra;
ALenum format = 0;
switch (bytesPerSample) {
case 1:
switch (numChannels) {
case 1: format = AL_FORMAT_MONO8; break;
case 2: format = AL_FORMAT_STEREO8; break;
}
break;
case 2:
switch (numChannels) {
case 1: format = AL_FORMAT_MONO16; break;
case 2: format = AL_FORMAT_STEREO16; break;
}
break;
}
if (!format) return audioFinish();
alBufferData(notifyBuffer, format, data, subchunk2Size, sampleRate);
alSourcei(notifySource, AL_BUFFER, notifyBuffer);
if (!_checkALError()) return audioFinish();
voicemsgs = new VoiceMessages();
}
bool audioWorks() {
return !!voicemsgs;
}
void audioPlayNotify() {
if (!audioWorks()) return;
alSourcePlay(notifySource);
}
void audioFinish() {
if (voicemsgs) {
delete voicemsgs;
}
alSourceStop(notifySource);
if (alIsBuffer(notifyBuffer)) {
alDeleteBuffers(1, &notifyBuffer);
notifyBuffer = 0;
}
if (alIsSource(notifySource)) {
alDeleteSources(1, &notifySource);
notifySource = 0;
}
if (audioContext) {
alcMakeContextCurrent(NULL);
alcDestroyContext(audioContext);
audioContext = 0;
}
if (audioDevice) {
alcCloseDevice(audioDevice);
audioDevice = 0;
}
}
VoiceMessages::VoiceMessages() : _current(0),
_fader(new VoiceMessagesFader(&_faderThread)), _loader(new VoiceMessagesLoader(&_loaderThread)) {
connect(this, SIGNAL(faderOnTimer()), _fader, SLOT(onTimer()));
connect(this, SIGNAL(loaderOnStart(AudioData*)), _loader, SLOT(onStart(AudioData*)));
connect(this, SIGNAL(loaderOnCancel(AudioData*)), _loader, SLOT(onCancel(AudioData*)));
connect(&_faderThread, SIGNAL(started()), _fader, SLOT(onInit()));
connect(&_loaderThread, SIGNAL(started()), _loader, SLOT(onInit()));
connect(&_faderThread, SIGNAL(finished()), _fader, SLOT(deleteLater()));
connect(&_loaderThread, SIGNAL(finished()), _loader, SLOT(deleteLater()));
connect(_loader, SIGNAL(needToCheck()), _fader, SLOT(onTimer()));
connect(_loader, SIGNAL(error(AudioData*)), this, SLOT(onError(AudioData*)));
connect(_fader, SIGNAL(needToPreload(AudioData*)), _loader, SLOT(onLoad(AudioData*)));
connect(_fader, SIGNAL(playPositionUpdated(AudioData*)), this, SIGNAL(updated(AudioData*)));
connect(_fader, SIGNAL(audioStopped(AudioData*)), this, SIGNAL(stopped(AudioData*)));
connect(_fader, SIGNAL(error(AudioData*)), this, SLOT(onError(AudioData*)));
_loaderThread.start();
_faderThread.start();
}
VoiceMessages::~VoiceMessages() {
{
QMutexLocker lock(&voicemsgsMutex);
voicemsgs = 0;
}
for (int32 i = 0; i < AudioVoiceMsgSimultaneously; ++i) {
alSourceStop(_data[i].source);
if (alIsBuffer(_data[i].buffers[0])) {
alDeleteBuffers(3, _data[i].buffers);
for (int32 j = 0; j < 3; ++j) {
_data[i].buffers[j] = _data[i].samplesCount[j] = 0;
}
}
if (alIsSource(_data[i].source)) {
alDeleteSources(1, &_data[i].source);
_data[i].source = 0;
}
}
_faderThread.quit();
_loaderThread.quit();
_faderThread.wait();
_loaderThread.wait();
}
void VoiceMessages::onError(AudioData *audio) {
emit stopped(audio);
}
bool VoiceMessages::updateCurrentStarted(int32 pos) {
if (pos < 0) {
if (alIsSource(_data[_current].source)) {
alGetSourcei(_data[_current].source, AL_SAMPLE_OFFSET, &pos);
} else {
pos = 0;
}
}
if (!_checkALError()) {
_data[_current].state = VoiceMessageStopped;
onError(_data[_current].audio);
return false;
}
_data[_current].started = _data[_current].position = pos + _data[_current].skipStart;
return true;
}
void VoiceMessages::play(AudioData *audio) {
QMutexLocker lock(&voicemsgsMutex);
bool startNow = true;
if (_data[_current].audio != audio) {
switch (_data[_current].state) {
case VoiceMessageStarting:
case VoiceMessageResuming:
case VoiceMessagePlaying:
_data[_current].state = VoiceMessageFinishing;
updateCurrentStarted();
startNow = false;
break;
case VoiceMessagePausing: _data[_current].state = VoiceMessageFinishing; startNow = false; break;
case VoiceMessagePaused: _data[_current].state = VoiceMessageStopped; break;
}
if (_data[_current].audio) {
emit loaderOnCancel(_data[_current].audio);
emit faderOnTimer();
}
}
int32 index = 0;
for (; index < AudioVoiceMsgSimultaneously; ++index) {
if (_data[index].audio == audio) {
_current = index;
break;
}
}
if (index == AudioVoiceMsgSimultaneously && ++_current >= AudioVoiceMsgSimultaneously) {
_current -= AudioVoiceMsgSimultaneously;
}
_data[_current].audio = audio;
_data[_current].fname = audio->already(true);
_data[_current].data = audio->data;
if (_data[_current].fname.isEmpty() && _data[_current].data.isEmpty()) {
_data[_current].state = VoiceMessageStopped;
onError(audio);
} else if (updateCurrentStarted(0)) {
_data[_current].state = startNow ? VoiceMessagePlaying : VoiceMessageStarting;
_data[_current].loading = true;
emit loaderOnStart(audio);
}
}
void VoiceMessages::pauseresume() {
QMutexLocker lock(&voicemsgsMutex);
switch (_data[_current].state) {
case VoiceMessagePausing:
case VoiceMessagePaused:
if (_data[_current].state == VoiceMessagePaused) {
updateCurrentStarted();
}
_data[_current].state = VoiceMessageResuming;
alSourcePlay(_data[_current].source);
break;
case VoiceMessageStarting:
case VoiceMessageResuming:
case VoiceMessagePlaying:
_data[_current].state = VoiceMessagePausing;
updateCurrentStarted();
break;
case VoiceMessageFinishing: _data[_current].state = VoiceMessagePausing; break;
}
emit faderOnTimer();
}
void VoiceMessages::currentState(AudioData **audio, VoiceMessageState *state, int64 *position, int64 *duration) {
QMutexLocker lock(&voicemsgsMutex);
if (audio) *audio = _data[_current].audio;
if (state) *state = _data[_current].state;
if (position) *position = _data[_current].position;
if (duration) *duration = _data[_current].duration;
}
VoiceMessages *audioVoice() {
return voicemsgs;
}
VoiceMessagesFader::VoiceMessagesFader(QThread *thread) : _timer(this) {
moveToThread(thread);
_timer.moveToThread(thread);
_timer.setSingleShot(true);
connect(&_timer, SIGNAL(timeout()), this, SLOT(onTimer()));
}
void VoiceMessagesFader::onInit() {
}
void VoiceMessagesFader::onTimer() {
bool hasFading = false, hasPlaying = false;
QMutexLocker lock(&voicemsgsMutex);
VoiceMessages *voice = audioVoice();
if (!voice) return;
for (int32 i = 0; i < AudioVoiceMsgSimultaneously; ++i) {
VoiceMessages::Msg &m(voice->_data[i]);
if (m.state == VoiceMessageStopped || m.state == VoiceMessagePaused || !m.source) continue;
bool playing = false, fading = false;
ALint pos = 0;
ALint state = AL_INITIAL;
alGetSourcei(m.source, AL_SAMPLE_OFFSET, &pos);
alGetSourcei(m.source, AL_SOURCE_STATE, &state);
if (!_checkALError()) {
m.state = VoiceMessageStopped;
emit error(m.audio);
} else {
switch (m.state) {
case VoiceMessageFinishing:
case VoiceMessagePausing:
case VoiceMessageStarting:
case VoiceMessageResuming:
fading = true;
break;
case VoiceMessagePlaying:
playing = true;
break;
}
if (fading && (state == AL_PLAYING || !m.loading)) {
if (state != AL_PLAYING) {
fading = false;
if (m.source) {
alSourcef(m.source, AL_GAIN, 1);
alSourceStop(m.source);
}
m.state = VoiceMessageStopped;
emit audioStopped(m.audio);
} else if (1000 * (pos + m.skipStart - m.started) >= AudioFadeDuration * AudioVoiceMsgFrequency) {
fading = false;
alSourcef(m.source, AL_GAIN, 1);
switch (m.state) {
case VoiceMessageFinishing: alSourceStop(m.source); m.state = VoiceMessageStopped; break;
case VoiceMessagePausing: alSourcePause(m.source); m.state = VoiceMessagePaused; break;
case VoiceMessageStarting:
case VoiceMessageResuming:
m.state = VoiceMessagePlaying;
playing = true;
break;
}
} else {
float64 newGain = 1000. * (pos + m.skipStart - m.started) / (AudioFadeDuration * AudioVoiceMsgFrequency);
if (m.state == VoiceMessagePausing || m.state == VoiceMessageFinishing) {
newGain = 1. - newGain;
}
if (newGain < 0) {
int a = 0, b;
b = a;
}
alSourcef(m.source, AL_GAIN, newGain);
LOG(("Now volume is: %1").arg(newGain));
}
} else if (playing && (state == AL_PLAYING || !m.loading)) {
if (state != AL_PLAYING) {
playing = false;
if (m.source) {
alSourceStop(m.source);
alSourcef(m.source, AL_GAIN, 1);
}
m.state = VoiceMessageStopped;
emit audioStopped(m.audio);
}
}
if (pos + m.skipStart - m.position >= AudioCheckPositionDelta) {
m.position = pos + m.skipStart;
emit playPositionUpdated(m.audio);
}
if (!m.loading && m.skipEnd > 0 && m.position + AudioPreloadSamples + m.skipEnd > m.duration) {
m.loading = true;
emit needToPreload(m.audio);
}
if (playing) hasPlaying = true;
if (fading) hasFading = true;
}
}
if (hasFading) {
_timer.start(AudioFadeTimeout);
} else if (hasPlaying) {
_timer.start(AudioCheckPositionTimeout);
}
}
struct VoiceMessagesLoader::Loader {
QString fname;
QByteArray data;
OggOpusFile *file;
ogg_int64_t pcm_offset;
ogg_int64_t pcm_print_offset;
int prev_li;
Loader() : file(0), pcm_offset(0), pcm_print_offset(0), prev_li(-1) {
}
};
VoiceMessagesLoader::VoiceMessagesLoader(QThread *thread) {
moveToThread(thread);
}
VoiceMessagesLoader::~VoiceMessagesLoader() {
for (Loaders::iterator i = _loaders.begin(), e = _loaders.end(); i != e; ++i) {
delete i.value();
}
_loaders.clear();
}
void VoiceMessagesLoader::onInit() {
}
void VoiceMessagesLoader::onStart(AudioData *audio) {
Loaders::iterator i = _loaders.find(audio);
if (i != _loaders.end()) {
delete (*i);
_loaders.erase(i);
}
onLoad(audio);
}
void VoiceMessagesLoader::loadError(Loaders::iterator i) {
emit error(i.key());
delete (*i);
_loaders.erase(i);
}
void VoiceMessagesLoader::onLoad(AudioData *audio) {
bool started = false;
int32 audioindex = -1;
Loader *l = 0;
Loaders::iterator j = _loaders.end();
{
QMutexLocker lock(&voicemsgsMutex);
VoiceMessages *voice = audioVoice();
if (!voice) return;
for (int32 i = 0; i < AudioVoiceMsgSimultaneously; ++i) {
VoiceMessages::Msg &m(voice->_data[i]);
if (m.audio != audio || !m.loading) continue;
audioindex = i;
j = _loaders.find(audio);
if (j != _loaders.end() && (j.value()->fname != m.fname || j.value()->data.size() != m.data.size())) {
delete j.value();
_loaders.erase(j);
j = _loaders.end();
}
if (j == _loaders.end()) {
l = (j = _loaders.insert(audio, new Loader())).value();
l->fname = m.fname;
l->data = m.data;
int ret;
if (m.data.isEmpty()) {
l->file = op_open_file(m.fname.toUtf8().constData(), &ret);
} else {
l->file = op_open_memory((const unsigned char*)m.data.constData(), m.data.size(), &ret);
}
if (!l->file) {
LOG(("Audio Error: op_open_file failed for '%1', data size '%2', error code %3").arg(m.fname).arg(m.data.size()).arg(ret));
m.state = VoiceMessageStopped;
return loadError(j);
}
ogg_int64_t duration = op_pcm_total(l->file, -1);
if (duration < 0) {
LOG(("Audio Error: op_pcm_total failed to get full duration for '%1', data size '%2', error code %3").arg(m.fname).arg(m.data.size()).arg(duration));
m.state = VoiceMessageStopped;
return loadError(j);
}
m.duration = duration;
m.skipStart = 0;
m.skipEnd = duration;
m.position = 0;
m.started = 0;
started = true;
} else {
if (!m.skipEnd) continue;
l = j.value();
}
break;
}
}
if (j == _loaders.end()) {
LOG(("Audio Error: trying to load part of audio, that is not playing at the moment"));
emit error(audio);
return;
}
if (started) {
l->pcm_offset = op_pcm_tell(l->file);
l->pcm_print_offset = l->pcm_offset - AudioVoiceMsgFrequency;
}
bool finished = false;
DEBUG_LOG(("Audio Info: reading buffer for file '%1', data size '%2', current pcm_offset %3").arg(l->fname).arg(l->data.size()).arg(l->pcm_offset));
2014-09-04 07:33:44 +00:00
QByteArray result;
int64 samplesAdded = 0;
while (result.size() < AudioVoiceMsgBufferSize) {
opus_int16 pcm[AudioVoiceMsgFrequency * AudioVoiceMsgChannels];
int ret = op_read_stereo(l->file, pcm, sizeof(pcm) / sizeof(*pcm));
if (ret < 0) {
{
QMutexLocker lock(&voicemsgsMutex);
VoiceMessages *voice = audioVoice();
if (voice) {
VoiceMessages::Msg &m(voice->_data[audioindex]);
if (m.audio == audio) {
m.state = VoiceMessageStopped;
}
}
}
LOG(("Audio Error: op_read_stereo failed, error code %1").arg(ret));
return loadError(j);
}
int li = op_current_link(l->file);
if (li != l->prev_li) {
const OpusHead *head = op_head(l->file, li);
const OpusTags *tags = op_tags(l->file, li);
for (int32 ci = 0; ci < tags->comments; ++ci) {
const char *comment = tags->user_comments[ci];
if (opus_tagncompare("METADATA_BLOCK_PICTURE", 22, comment) == 0) {
OpusPictureTag pic;
int err = opus_picture_tag_parse(&pic, comment);
if (err >= 0) {
opus_picture_tag_clear(&pic);
}
}
}
if (!op_seekable(l->file)) {
l->pcm_offset = op_pcm_tell(l->file) - ret;
}
}
if (li != l->prev_li || l->pcm_offset >= l->pcm_print_offset + AudioVoiceMsgFrequency) {
l->pcm_print_offset = l->pcm_offset;
}
l->pcm_offset = op_pcm_tell(l->file);
if (!ret) {
DEBUG_LOG(("Audio Info: read completed"));
finished = true;
break;
}
result.append((const char*)pcm, sizeof(*pcm) * ret * AudioVoiceMsgChannels);
l->prev_li = li;
samplesAdded += ret;
{
QMutexLocker lock(&voicemsgsMutex);
VoiceMessages *voice = audioVoice();
if (!voice) return;
VoiceMessages::Msg &m(voice->_data[audioindex]);
if (m.audio != audio || !m.loading || m.fname != l->fname || m.data.size() != l->data.size()) {
LOG(("Audio Error: playing changed while loading"));
m.state = VoiceMessageStopped;
return loadError(j);
}
}
}
QMutexLocker lock(&voicemsgsMutex);
VoiceMessages *voice = audioVoice();
if (!voice) return;
VoiceMessages::Msg &m(voice->_data[audioindex]);
if (m.audio != audio || !m.loading || m.fname != l->fname || m.data.size() != l->data.size()) {
LOG(("Audio Error: playing changed while loading"));
m.state = VoiceMessageStopped;
return loadError(j);
}
if (started) {
if (m.source) {
alSourceStop(m.source);
for (int32 i = 0; i < 3; ++i) {
if (m.samplesCount[i]) {
alSourceUnqueueBuffers(m.source, 1, m.buffers + i);
m.samplesCount[i] = 0;
}
}
m.nextBuffer = 0;
}
}
if (samplesAdded) {
if (!m.source) {
alGenSources(1, &m.source);
alSourcef(m.source, AL_PITCH, 1.f);
alSourcef(m.source, AL_GAIN, 1.f);
alSource3f(m.source, AL_POSITION, 0, 0, 0);
alSource3f(m.source, AL_VELOCITY, 0, 0, 0);
alSourcei(m.source, AL_LOOPING, 0);
}
if (!m.buffers[m.nextBuffer]) alGenBuffers(3, m.buffers);
if (!_checkALError()) {
m.state = VoiceMessageStopped;
return loadError(j);
}
if (m.samplesCount[m.nextBuffer]) {
alSourceUnqueueBuffers(m.source, 1, m.buffers + m.nextBuffer);
m.skipStart += m.samplesCount[m.nextBuffer];
}
m.samplesCount[m.nextBuffer] = samplesAdded;
alBufferData(m.buffers[m.nextBuffer], AL_FORMAT_STEREO16, result.constData(), result.size(), AudioVoiceMsgFrequency);
alSourceQueueBuffers(m.source, 1, m.buffers + m.nextBuffer);
m.skipEnd -= samplesAdded;
m.nextBuffer = (m.nextBuffer + 1) % 3;
2014-09-04 07:33:44 +00:00
if (!_checkALError()) {
m.state = VoiceMessageStopped;
return loadError(j);
}
} else {
finished = true;
}
if (finished) {
m.skipEnd = 0;
m.duration = m.skipStart + m.samplesCount[0] + m.samplesCount[1] + m.samplesCount[2];
}
m.loading = false;
if (m.state == VoiceMessageResuming || m.state == VoiceMessagePlaying || m.state == VoiceMessageStarting) {
ALint state = AL_INITIAL;
alGetSourcei(m.source, AL_SOURCE_STATE, &state);
if (_checkALError()) {
if (state != AL_PLAYING) {
alSourcePlay(m.source);
emit needToCheck();
}
}
}
}
void VoiceMessagesLoader::onCancel(AudioData *audio) {
Loaders::iterator i = _loaders.find(audio);
if (i != _loaders.end()) {
delete (*i);
_loaders.erase(i);
}
QMutexLocker lock(&voicemsgsMutex);
VoiceMessages *voice = audioVoice();
if (!voice) return;
for (int32 i = 0; i < AudioVoiceMsgSimultaneously; ++i) {
VoiceMessages::Msg &m(voice->_data[i]);
if (m.audio == audio) {
m.loading = false;
}
}
}