425 lines
16 KiB
C++
425 lines
16 KiB
C++
/*
|
||
* Copyright (c) 2016-present The ZLMediaKit project authors. All Rights Reserved.
|
||
*
|
||
* This file is part of ZLMediaKit(https://github.com/ZLMediaKit/ZLMediaKit).
|
||
*
|
||
* Use of this source code is governed by MIT-like license that can be found in the
|
||
* LICENSE file in the root of the source tree. All contributing project authors
|
||
* may be found in the AUTHORS file in the root of the source tree.
|
||
*/
|
||
|
||
#include "MediaSink.h"
|
||
#include "Common/config.h"
|
||
#include "Extension/Factory.h"
|
||
|
||
#define MUTE_AUDIO_INDEX 0xFFFF
|
||
|
||
using namespace std;
|
||
|
||
namespace mediakit{
|
||
|
||
bool MediaSink::addTrack(const Track::Ptr &track_in) {
|
||
if (_only_audio && track_in->getTrackType() != TrackAudio) {
|
||
InfoL << "Only audio enabled, track ignored: " << track_in->getCodecName();
|
||
return false;
|
||
}
|
||
if (!_enable_audio) {
|
||
// 关闭音频时,加快单视频流注册速度 [AUTO-TRANSLATED:4d5a361d]
|
||
// Speed up single video stream registration when audio is off
|
||
if (track_in->getTrackType() == TrackAudio) {
|
||
// 音频被全局忽略 [AUTO-TRANSLATED:a8134a0b]
|
||
// Audio is globally ignored
|
||
InfoL << "Audio disabled, audio track ignored";
|
||
return false;
|
||
}
|
||
}
|
||
if (_all_track_ready) {
|
||
WarnL << "All track is ready, add track too late: " << track_in->getCodecName();
|
||
return false;
|
||
}
|
||
// 克隆Track,只拷贝其数据,不拷贝其数据转发关系 [AUTO-TRANSLATED:09edaa31]
|
||
// Clone Track, only copy its data, not its data forwarding relationship
|
||
auto track = track_in->clone();
|
||
CHECK(track, "Clone track failed: ", track_in->getCodecName());
|
||
auto index = track->getIndex();
|
||
if (!_track_map.emplace(index, std::make_pair(track, false)).second) {
|
||
WarnL << "Already add a same track: " << track->getIndex() << ", codec: " << track->getCodecName();
|
||
return false;
|
||
}
|
||
_ticker.resetTime();
|
||
_audio_add = track->getTrackType() == TrackAudio ? true : _audio_add;
|
||
_track_ready_callback[index] = [this, track]() { onTrackReady(track); };
|
||
|
||
track->addDelegate([this](const Frame::Ptr &frame) {
|
||
if (_all_track_ready) {
|
||
return onTrackFrame(frame);
|
||
}
|
||
auto &frame_unread = _frame_unread[frame->getIndex()];
|
||
|
||
GET_CONFIG(uint32_t, kMaxUnreadyFrame, General::kUnreadyFrameCache);
|
||
if (frame_unread.size() > kMaxUnreadyFrame) {
|
||
// 未就绪的的track,不能缓存太多的帧,否则可能内存溢出 [AUTO-TRANSLATED:23958376]
|
||
// Unready tracks cannot cache too many frames, otherwise memory may overflow
|
||
frame_unread.clear();
|
||
WarnL << "Cached frame of unready track(" << frame->getCodecName() << ") is too much, now cleared";
|
||
}
|
||
// 还有Track未就绪,先缓存之 [AUTO-TRANSLATED:f96eadfa]
|
||
// There are still unready tracks, cache them first
|
||
frame_unread.emplace_back(Frame::getCacheAbleFrame(frame));
|
||
return true;
|
||
});
|
||
return true;
|
||
}
|
||
|
||
void MediaSink::resetTracks() {
|
||
_audio_add = false;
|
||
_have_video = false;
|
||
_all_track_ready = false;
|
||
_mute_audio_maker = nullptr;
|
||
_ticker.resetTime();
|
||
_track_map.clear();
|
||
_frame_unread.clear();
|
||
_track_ready_callback.clear();
|
||
}
|
||
|
||
bool MediaSink::inputFrame(const Frame::Ptr &frame) {
|
||
auto it = _track_map.find(frame->getIndex());
|
||
if (it == _track_map.end()) {
|
||
return false;
|
||
}
|
||
// got frame
|
||
it->second.second = true;
|
||
auto ret = it->second.first->inputFrame(frame);
|
||
if (_mute_audio_maker && frame->getTrackType() == TrackVideo) {
|
||
// 视频驱动产生静音音频 [AUTO-TRANSLATED:2a8c789c]
|
||
// Video driver generates silent audio
|
||
_mute_audio_maker->inputFrame(frame);
|
||
}
|
||
checkTrackIfReady();
|
||
return ret;
|
||
}
|
||
|
||
void MediaSink::checkTrackIfReady() {
|
||
if (!_all_track_ready && !_track_ready_callback.empty()) {
|
||
for (auto &pr : _track_map) {
|
||
if (pr.second.second && pr.second.first->ready()) {
|
||
// Track由未就绪状态转换成就绪状态,我们就触发onTrackReady回调 [AUTO-TRANSLATED:f8975e53]
|
||
// When a Track transitions from an unready state to a ready state, we trigger the onTrackReady callback
|
||
auto it = _track_ready_callback.find(pr.first);
|
||
if (it != _track_ready_callback.end()) {
|
||
it->second();
|
||
_track_ready_callback.erase(it);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
// 等待音频超时时间
|
||
GET_CONFIG(uint32_t, kWaitAudioTrackDataMS, General::kWaitAudioTrackDataMS);
|
||
if (_max_track_size > 1) {
|
||
for (auto it = _track_map.begin(); it != _track_map.end();) {
|
||
if (it->second.first->getTrackType() == TrackAudio && _ticker.elapsedTime() > kWaitAudioTrackDataMS && !it->second.second) {
|
||
// 音频超时且完全没收到音频数据,忽略音频
|
||
auto index = it->second.first->getIndex();
|
||
WarnL << "Audio track index " << index << " codec " << it->second.first->getCodecName() << " receive no data for long "
|
||
<< _ticker.elapsedTime() << "ms. Ignore it!";
|
||
it = _track_map.erase(it);
|
||
_max_track_size -= 1;
|
||
_track_ready_callback.erase(index);
|
||
} else {
|
||
++it;
|
||
}
|
||
}
|
||
}
|
||
|
||
if (!_all_track_ready) {
|
||
GET_CONFIG(uint32_t, kMaxWaitReadyMS, General::kWaitTrackReadyMS);
|
||
if (_ticker.elapsedTime() > kMaxWaitReadyMS) {
|
||
// 如果超过规定时间,那么不再等待并忽略未准备好的Track [AUTO-TRANSLATED:fd089806]
|
||
// If it exceeds the specified time, then stop waiting and ignore unprepared Tracks
|
||
emitAllTrackReady();
|
||
return;
|
||
}
|
||
|
||
if (!_track_ready_callback.empty()) {
|
||
// 在超时时间内,如果存在未准备好的Track,那么继续等待 [AUTO-TRANSLATED:cfaf3b49]
|
||
// Within the timeout period, if there are unprepared Tracks, then continue waiting
|
||
return;
|
||
}
|
||
|
||
if (_only_audio && _audio_add) {
|
||
// 只开启音频 [AUTO-TRANSLATED:bac07e47]
|
||
// Only enable audio
|
||
emitAllTrackReady();
|
||
return;
|
||
}
|
||
|
||
if (_track_map.size() == _max_track_size) {
|
||
// 如果已经添加了音视频Track,并且不存在未准备好的Track,那么说明所有Track都准备好了 [AUTO-TRANSLATED:6fce8779]
|
||
// If audio and video Tracks have been added, and there are no unprepared Tracks, then all Tracks are ready
|
||
emitAllTrackReady();
|
||
return;
|
||
}
|
||
|
||
GET_CONFIG(uint32_t, kMaxAddTrackMS, General::kWaitAddTrackMS);
|
||
if (_track_map.size() == 1 && (_ticker.elapsedTime() > kMaxAddTrackMS || !_enable_audio)) {
|
||
// 如果只有一个Track,那么在该Track添加后,我们最多还等待若干时间(可能后面还会添加Track) [AUTO-TRANSLATED:5b4bd438]
|
||
// If there is only one Track, then after the Track is added, we wait for a certain amount of time at most (more Tracks may be added later)
|
||
emitAllTrackReady();
|
||
return;
|
||
}
|
||
}
|
||
}
|
||
|
||
void MediaSink::addTrackCompleted() {
|
||
setMaxTrackCount(_track_map.size());
|
||
}
|
||
|
||
void MediaSink::setMaxTrackCount(size_t i) {
|
||
if (_all_track_ready) {
|
||
WarnL << "All track is ready, set max track count ignored";
|
||
return;
|
||
}
|
||
_max_track_size = MAX(i, 1);
|
||
checkTrackIfReady();
|
||
}
|
||
|
||
void MediaSink::emitAllTrackReady() {
|
||
if (_all_track_ready) {
|
||
return;
|
||
}
|
||
|
||
DebugL << "All track ready use " << _ticker.elapsedTime() << "ms";
|
||
if (!_track_ready_callback.empty()) {
|
||
// 这是超时强制忽略未准备好的Track [AUTO-TRANSLATED:d4f57e00]
|
||
// This is a timeout forced ignore of unprepared Tracks
|
||
_track_ready_callback.clear();
|
||
// 移除未准备好的Track [AUTO-TRANSLATED:69965c62]
|
||
// Remove unprepared Tracks
|
||
for (auto it = _track_map.begin(); it != _track_map.end();) {
|
||
if (!it->second.second || !it->second.first->ready()) {
|
||
WarnL << "Track not ready for a long time, ignored: " << it->second.first->getCodecName();
|
||
it = _track_map.erase(it);
|
||
continue;
|
||
}
|
||
++it;
|
||
}
|
||
}
|
||
|
||
if (!_track_map.empty()) {
|
||
// 最少有一个有效的Track [AUTO-TRANSLATED:099adc94]
|
||
// There is at least one valid Track
|
||
onAllTrackReady_l();
|
||
|
||
// 全部Track就绪,我们一次性把之前的帧输出 [AUTO-TRANSLATED:2431422b]
|
||
// All Tracks are ready, we output all the previous frames at once
|
||
for (auto &pr : _frame_unread) {
|
||
if (_track_map.find(pr.first) == _track_map.end()) {
|
||
// 该Track已经被移除 [AUTO-TRANSLATED:d44bf74e]
|
||
// The Track has been removed
|
||
continue;
|
||
}
|
||
pr.second.for_each([&](const Frame::Ptr &frame) { MediaSink::inputFrame(frame); });
|
||
}
|
||
_frame_unread.clear();
|
||
} else {
|
||
throw toolkit::SockException(toolkit::Err_shutdown, "no vaild track data");
|
||
}
|
||
}
|
||
|
||
void MediaSink::onAllTrackReady_l() {
|
||
// 是否添加静音音频 [AUTO-TRANSLATED:bbfbfe73]
|
||
// Whether to add silent audio
|
||
if (_add_mute_audio) {
|
||
addMuteAudioTrack();
|
||
}
|
||
onAllTrackReady();
|
||
_all_track_ready = true;
|
||
_have_video = (bool)getTrack(TrackVideo);
|
||
}
|
||
|
||
vector<Track::Ptr> MediaSink::getTracks(bool ready) const {
|
||
vector<Track::Ptr> ret;
|
||
for (auto &pr : _track_map) {
|
||
if (ready && !pr.second.first->ready()) {
|
||
continue;
|
||
}
|
||
ret.emplace_back(pr.second.first);
|
||
}
|
||
return ret;
|
||
}
|
||
|
||
static uint8_t s_mute_adts[] = {0xff, 0xf1, 0x6c, 0x40, 0x2d, 0x3f, 0xfc, 0x00, 0xe0, 0x34, 0x20, 0xad, 0xf2, 0x3f, 0xb5, 0xdd,
|
||
0x73, 0xac, 0xbd, 0xca, 0xd7, 0x7d, 0x4a, 0x13, 0x2d, 0x2e, 0xa2, 0x62, 0x02, 0x70, 0x3c, 0x1c,
|
||
0xc5, 0x63, 0x55, 0x69, 0x94, 0xb5, 0x8d, 0x70, 0xd7, 0x24, 0x6a, 0x9e, 0x2e, 0x86, 0x24, 0xea,
|
||
0x4f, 0xd4, 0xf8, 0x10, 0x53, 0xa5, 0x4a, 0xb2, 0x9a, 0xf0, 0xa1, 0x4f, 0x2f, 0x66, 0xf9, 0xd3,
|
||
0x8c, 0xa6, 0x97, 0xd5, 0x84, 0xac, 0x09, 0x25, 0x98, 0x0b, 0x1d, 0x77, 0x04, 0xb8, 0x55, 0x49,
|
||
0x85, 0x27, 0x06, 0x23, 0x58, 0xcb, 0x22, 0xc3, 0x20, 0x3a, 0x12, 0x09, 0x48, 0x24, 0x86, 0x76,
|
||
0x95, 0xe3, 0x45, 0x61, 0x43, 0x06, 0x6b, 0x4a, 0x61, 0x14, 0x24, 0xa9, 0x16, 0xe0, 0x97, 0x34,
|
||
0xb6, 0x58, 0xa4, 0x38, 0x34, 0x90, 0x19, 0x5d, 0x00, 0x19, 0x4a, 0xc2, 0x80, 0x4b, 0xdc, 0xb7,
|
||
0x00, 0x18, 0x12, 0x3d, 0xd9, 0x93, 0xee, 0x74, 0x13, 0x95, 0xad, 0x0b, 0x59, 0x51, 0x0e, 0x99,
|
||
0xdf, 0x49, 0x98, 0xde, 0xa9, 0x48, 0x4b, 0xa5, 0xfb, 0xe8, 0x79, 0xc9, 0xe2, 0xd9, 0x60, 0xa5,
|
||
0xbe, 0x74, 0xa6, 0x6b, 0x72, 0x0e, 0xe3, 0x7b, 0x28, 0xb3, 0x0e, 0x52, 0xcc, 0xf6, 0x3d, 0x39,
|
||
0xb7, 0x7e, 0xbb, 0xf0, 0xc8, 0xce, 0x5c, 0x72, 0xb2, 0x89, 0x60, 0x33, 0x7b, 0xc5, 0xda, 0x49,
|
||
0x1a, 0xda, 0x33, 0xba, 0x97, 0x9e, 0xa8, 0x1b, 0x6d, 0x5a, 0x77, 0xb6, 0xf1, 0x69, 0x5a, 0xd1,
|
||
0xbd, 0x84, 0xd5, 0x4e, 0x58, 0xa8, 0x5e, 0x8a, 0xa0, 0xc2, 0xc9, 0x22, 0xd9, 0xa5, 0x53, 0x11,
|
||
0x18, 0xc8, 0x3a, 0x39, 0xcf, 0x3f, 0x57, 0xb6, 0x45, 0x19, 0x1e, 0x8a, 0x71, 0xa4, 0x46, 0x27,
|
||
0x9e, 0xe9, 0xa4, 0x86, 0xdd, 0x14, 0xd9, 0x4d, 0xe3, 0x71, 0xe3, 0x26, 0xda, 0xaa, 0x17, 0xb4,
|
||
0xac, 0xe1, 0x09, 0xc1, 0x0d, 0x75, 0xba, 0x53, 0x0a, 0x37, 0x8b, 0xac, 0x37, 0x39, 0x41, 0x27,
|
||
0x6a, 0xf0, 0xe9, 0xb4, 0xc2, 0xac, 0xb0, 0x39, 0x73, 0x17, 0x64, 0x95, 0xf4, 0xdc, 0x33, 0xbb,
|
||
0x84, 0x94, 0x3e, 0xf8, 0x65, 0x71, 0x60, 0x7b, 0xd4, 0x5f, 0x27, 0x79, 0x95, 0x6a, 0xba, 0x76,
|
||
0xa6, 0xa5, 0x9a, 0xec, 0xae, 0x55, 0x3a, 0x27, 0x48, 0x23, 0xcf, 0x5c, 0x4d, 0xbc, 0x0b, 0x35,
|
||
0x5c, 0xa7, 0x17, 0xcf, 0x34, 0x57, 0xc9, 0x58, 0xc5, 0x20, 0x09, 0xee, 0xa5, 0xf2, 0x9c, 0x6c,
|
||
0x39, 0x1a, 0x77, 0x92, 0x9b, 0xff, 0xc6, 0xae, 0xf8, 0x36, 0xba, 0xa8, 0xaa, 0x6b, 0x1e, 0x8c,
|
||
0xc5, 0x97, 0x39, 0x6a, 0xb8, 0xa2, 0x55, 0xa8, 0xf8};
|
||
|
||
#define MUTE_ADTS_DATA s_mute_adts
|
||
#define MUTE_ADTS_DATA_MS 128
|
||
static uint8_t ADTS_CONFIG[2] = { 0x15, 0x88 };
|
||
|
||
bool MuteAudioMaker::inputFrame(const Frame::Ptr &frame) {
|
||
if (_track_index == -1) {
|
||
// 锁定track [AUTO-TRANSLATED:41aff35e]
|
||
// Lock track
|
||
_track_index = frame->getIndex();
|
||
}
|
||
if (frame->getIndex() != _track_index) {
|
||
// 不是锁定的track [AUTO-TRANSLATED:496bd08b]
|
||
// Not a locked track
|
||
return false;
|
||
}
|
||
auto audio_idx = frame->dts() / MUTE_ADTS_DATA_MS;
|
||
if (_audio_idx != audio_idx) {
|
||
_audio_idx = audio_idx;
|
||
auto aacFrame = std::make_shared<FrameToCache<FrameFromPtr>>(CodecAAC, (char *)MUTE_ADTS_DATA, sizeof(s_mute_adts), _audio_idx * MUTE_ADTS_DATA_MS, 0, 7);
|
||
aacFrame->setIndex(MUTE_AUDIO_INDEX);
|
||
return FrameDispatcher::inputFrame(aacFrame);
|
||
}
|
||
return false;
|
||
}
|
||
|
||
bool MediaSink::addMuteAudioTrack() {
|
||
if (!_enable_audio) {
|
||
return false;
|
||
}
|
||
for (auto &pr : _track_map) {
|
||
if (pr.second.first->getTrackType() == TrackAudio) {
|
||
return false;
|
||
}
|
||
}
|
||
auto audio = Factory::getTrackByCodecId(CodecAAC);
|
||
audio->setIndex(MUTE_AUDIO_INDEX);
|
||
audio->setExtraData(ADTS_CONFIG, 2);
|
||
_track_map[MUTE_AUDIO_INDEX] = std::make_pair(audio, true);
|
||
audio->addDelegate([this](const Frame::Ptr &frame) { return onTrackFrame(frame); });
|
||
_mute_audio_maker = std::make_shared<MuteAudioMaker>();
|
||
_mute_audio_maker->addDelegate([audio](const Frame::Ptr &frame) { return audio->inputFrame(frame); });
|
||
onTrackReady(audio);
|
||
TraceL << "Mute aac track added";
|
||
return true;
|
||
}
|
||
|
||
bool MediaSink::isAllTrackReady() const {
|
||
return _all_track_ready;
|
||
}
|
||
|
||
void MediaSink::enableAudio(bool flag) {
|
||
_enable_audio = flag;
|
||
}
|
||
|
||
void MediaSink::setOnlyAudio() {
|
||
_only_audio = true;
|
||
_enable_audio = true;
|
||
_add_mute_audio = false;
|
||
}
|
||
|
||
void MediaSink::enableMuteAudio(bool flag) {
|
||
_add_mute_audio = flag;
|
||
}
|
||
|
||
bool MediaSink::haveVideo() const {
|
||
return _have_video;
|
||
}
|
||
|
||
///////////////////////////DemuxerSink//////////////////////////////
|
||
|
||
void MediaSinkDelegate::setTrackListener(TrackListener *listener) {
|
||
_listener = listener;
|
||
}
|
||
|
||
bool MediaSinkDelegate::onTrackReady(const Track::Ptr &track) {
|
||
if (_listener) {
|
||
_listener->addTrack(track);
|
||
}
|
||
return true;
|
||
}
|
||
|
||
void MediaSinkDelegate::onAllTrackReady() {
|
||
if (_listener) {
|
||
_listener->addTrackCompleted();
|
||
}
|
||
}
|
||
|
||
void MediaSinkDelegate::resetTracks() {
|
||
MediaSink::resetTracks();
|
||
if (_listener) {
|
||
_listener->resetTracks();
|
||
}
|
||
}
|
||
|
||
///////////////////////////Demuxer//////////////////////////////
|
||
|
||
void Demuxer::setTrackListener(TrackListener *listener, bool wait_track_ready) {
|
||
if (wait_track_ready) {
|
||
auto sink = std::make_shared<MediaSinkDelegate>();
|
||
sink->setTrackListener(listener);
|
||
_sink = std::move(sink);
|
||
}
|
||
_listener = listener;
|
||
}
|
||
|
||
bool Demuxer::addTrack(const Track::Ptr &track) {
|
||
if (!_sink) {
|
||
_origin_track.emplace_back(track);
|
||
return _listener ? _listener->addTrack(track) : false;
|
||
}
|
||
|
||
if (_sink->addTrack(track)) {
|
||
track->addDelegate([this](const Frame::Ptr &frame) { return _sink->inputFrame(frame); });
|
||
return true;
|
||
}
|
||
return false;
|
||
}
|
||
|
||
void Demuxer::addTrackCompleted() {
|
||
if (_sink) {
|
||
_sink->addTrackCompleted();
|
||
} else if (_listener) {
|
||
_listener->addTrackCompleted();
|
||
}
|
||
}
|
||
|
||
void Demuxer::resetTracks() {
|
||
if (_sink) {
|
||
_sink->resetTracks();
|
||
} else if (_listener) {
|
||
_listener->resetTracks();
|
||
}
|
||
}
|
||
|
||
vector<Track::Ptr> Demuxer::getTracks(bool ready) const {
|
||
if (_sink) {
|
||
return _sink->getTracks(ready);
|
||
}
|
||
|
||
vector<Track::Ptr> ret;
|
||
for (auto &track : _origin_track) {
|
||
if (ready && !track->ready()) {
|
||
continue;
|
||
}
|
||
ret.emplace_back(track);
|
||
}
|
||
return ret;
|
||
}
|
||
} // namespace mediakit
|