上一篇我们讲了mediaplayer播放的第一步骤setdataSource,下面我们来讲解preparesync的流程,在prepare前我们还有setDisplay这一步,即获取surfacetexture来进行画面的展示
setVideoSurface(JNIEnv *env, jobject thiz, jobject jsurface, jboolean mediaPlayerMustBeAlive) { sp<MediaPlayer> mp = getMediaPlayer(env, thiz); ……… sp<ISurfaceTexture> new_st; if (jsurface) { sp<Surface> surface(Surface_getSurface(env, jsurface)); if (surface != NULL) { new_st = surface->getSurfaceTexture(); ---通过surface获取surfaceTexture new_st->incStrong(thiz); ………. }…………. mp->setVideoSurfaceTexture(new_st); }
为什么用surfaceTexture不用surface来展示呢?ICS之前都用的是surfaceview来展示video或者openGL的内容,surfacaview render在surface上,textureview render在surfaceTexture,textureview和surfaceview 这两者有什么区别呢?surfaceview跟应用的视窗不是同一个视窗,它自己new了一个window来展示openGL或者video的内容,这样做有一个好处就是不用重绘应用的视窗,本身就可以不停的更新,但这也带来一些局限性,surfaceview不是依附在应用视窗中,也就不能移动、缩放、旋转,应用ListView或者 ScrollView就比较费劲。Textureview就很好的解决了这些问题。它拥有surfaceview的一切特性外,它也拥有view的一切行为,可以当个view使用。
获取完surfaceTexture,我们就可以prepare/prepareAsync了,先给大伙看个大体时序图吧:
JNI的部分我们跳过,直接进入libmedia下的mediaplayer.cpp的 prepareAsync_l方法,prepare是个同步的过程,所以要加锁,prepareAsync_l后缀加_l就是表面是同步的过程。
status_t MediaPlayer::prepareAsync_l() { if ( (mPlayer != 0) && ( mCurrentState & ( MEDIA_PLAYER_INITIALIZED | MEDIA_PLAYER_STOPPED) ) ) { mPlayer->setAudioStreamType(mStreamType); mCurrentState = MEDIA_PLAYER_PREPARING; return mPlayer->prepareAsync(); } ALOGE("prepareAsync called in state %d", mCurrentState); return INVALID_OPERATION; }
在上面的代码中,我们看到有个mPlayer,看过前一章的朋友都会记得,就是我们从Mediaplayerservice获得的BpMediaplayer.通过BpMediaplayer我们就可以长驱直入,直捣Awesomeplayer这条干实事的黄龙,前方的mediaplayerservice:client和stagefrightplayer都是些通风报信的料,不值得我们去深入研究,无非是些接口而已。进入了prepareAsync_l方法,我们的播放器所处的状态就是MEDIA_PLAYER_PREPARING了。好了,我们就来看看Awesomeplayer到底做了啥吧.
代码定位于:frameworks/av/media/libstagefright/Awesomeplayer.cpp
先看下prepareAsync_l吧:
status_t AwesomePlayer::prepareAsync_l() { if (mFlags & PREPARING) { return UNKNOWN_ERROR; // async prepare already pending } if (!mQueueStarted) { mQueue.start(); mQueueStarted = true; } modifyFlags(PREPARING, SET); mAsyncPrepareEvent = new AwesomeEvent( this, &AwesomePlayer::onPrepareAsyncEvent); mQueue.postEvent(mAsyncPrepareEvent); return OK; }
这里我们涉及到了TimeEventQueue,即时间事件队列模型,Awesomeplayer里面类似Handler的东西,它的实现方式是把事件响应时间和事件本身封装成一个queueItem,通过postEvent 插入队列,时间到了就会根据事件id进行相应的处理。
首先我们来看下TimeEventQueue的start(mQueue.start();)方法都干了什么:
frameworks/av/media/libstagefright/TimedEventQueue.cpp void TimedEventQueue::start() { if (mRunning) { return; } …….. pthread_create(&mThread, &attr, ThreadWrapper, this); ……… }
目的很明显就是在主线程创建一个子线程,可能很多没有写过C/C++的人对ptread_create这个创建线程的方法有点陌生,我们就来分析下:
int pthread_create(pthread_t *thread, pthread_addr_t *arr, void* (*start_routine)(void *), void *arg); thread :用于返回创建的线程的ID arr : 用于指定的被创建的线程的属性 start_routine : 这是一个函数指针,指向线程被创建后要调用的函数 arg : 用于给线程传递参数 分析完了,我们就看下创建线程后调用的函数ThreadWrapper吧: // static void *TimedEventQueue::ThreadWrapper(void *me) { …… static_cast<TimedEventQueue *>(me)->threadEntry(); return NULL; }
跟踪到threadEntry:
frameworks/av/media/libstagefright/TimedEventQueue.cpp void TimedEventQueue::threadEntry() { prctl(PR_SET_NAME, (unsigned long)"TimedEventQueue", 0, 0, 0); for (;;) { int64_t now_us = 0; sp<Event> event; { Mutex::Autolock autoLock(mLock); if (mStopped) { break; } while (mQueue.empty()) { mQueueNotEmptyCondition.wait(mLock); } event_id eventID = 0; for (;;) { if (mQueue.empty()) { // The only event in the queue could have been cancelled // while we were waiting for its scheduled time. break; } List<QueueItem>::iterator it = mQueue.begin(); eventID = (*it).event->eventID(); …………………………… static int64_t kMaxTimeoutUs = 10000000ll; // 10 secs …………….. status_t err = mQueueHeadChangedCondition.waitRelative( mLock, delay_us * 1000ll); if (!timeoutCapped && err == -ETIMEDOUT) { // We finally hit the time this event is supposed to // trigger. now_us = getRealTimeUs(); break; } } ………………………. event = removeEventFromQueue_l(eventID); } if (event != NULL) { // Fire event with the lock NOT held. event->fire(this, now_us); } } }
从代码我们可以了解到,主要目的是检查queue是否为空,刚开始肯定是为空了,等待队列不为空时的条件成立,即有queueIten进入进入队列中。这个事件应该就是
mQueue.postEvent(mAsyncPrepareEvent);
在讲postEvent前,我们先来看看mAsyncPrepareEvent这个封装成AwesomeEvent的Event。
frameworks/av/media/libstagefright/TimedEventQueue.cpp void TimedEventQueue::threadEntry() { prctl(PR_SET_NAME, (unsigned long)"TimedEventQueue", 0, 0, 0); for (;;) { int64_t now_us = 0; sp<Event> event; { Mutex::Autolock autoLock(mLock); if (mStopped) { break; } while (mQueue.empty()) { mQueueNotEmptyCondition.wait(mLock); } event_id eventID = 0; for (;;) { if (mQueue.empty()) { // The only event in the queue could have been cancelled // while we were waiting for its scheduled time. break; } List<QueueItem>::iterator it = mQueue.begin(); eventID = (*it).event->eventID(); …………………………… static int64_t kMaxTimeoutUs = 10000000ll; // 10 secs …………….. status_t err = mQueueHeadChangedCondition.waitRelative( mLock, delay_us * 1000ll); if (!timeoutCapped && err == -ETIMEDOUT) { // We finally hit the time this event is supposed to // trigger. now_us = getRealTimeUs(); break; } } ………………………. event = removeEventFromQueue_l(eventID); } if (event != NULL) { // Fire event with the lock NOT held. event->fire(this, now_us); } } }
从这个结构体我们可以知道当这个event被触发时将会执行Awesomeplayer的某个方法,我们看下mAsyncPrepareEvent:
mAsyncPrepareEvent = new AwesomeEvent(
this, &AwesomePlayer::onPrepareAsyncEvent);
mAsyncPrepareEvent被触发时也就触发了onPrepareAsyncEvent方法。
好了,回到我们的postEvent事件,我们开始说的TimeEventQueue,即时间事件队列模型,刚刚我们说了Event, 但是没有看到delay time啊?会不会在postEvent中加入呢?跟下去看看:
TimedEventQueue::event_id TimedEventQueue::postEvent(const sp<Event> &event) { // Reserve an earlier timeslot an INT64_MIN to be able to post // the StopEvent to the absolute head of the queue. return postTimedEvent(event, INT64_MIN + 1); }
终于看到delay时间了INT64_MIN + 1。重点在postTimedEvent,它把post过来的event和时间封装成queueItem加入队列中,并通知Queue为空的条件不成立,线程解锁,允许thread继续进行,经过delay time后pull event_id所对应的event。
frameworks/av/media/libstagefright/TimedEventQueue.cpp
TimedEventQueue::event_id TimedEventQueue::postTimedEvent( const sp<Event> &event, int64_t realtime_us) { Mutex::Autolock autoLock(mLock); event->setEventID(mNextEventID++); …………………. QueueItem item; item.event = event; item.realtime_us = realtime_us; if (it == mQueue.begin()) { mQueueHeadChangedCondition.signal(); } mQueue.insert(it, item); mQueueNotEmptyCondition.signal(); return event->eventID(); }
到此,我们的TimeEventQueue,即时间事件队列模型讲完了。实现机制跟handle的C/C++部分类似。
在我们setdataSource实例化Awesomeplayer的时候,我们还顺带创建了如下几个event
sp<TimedEventQueue::Event> mVideoEvent;
sp<TimedEventQueue::Event> mStreamDoneEvent;
sp<TimedEventQueue::Event> mBufferingEvent;
sp<TimedEventQueue::Event> mCheckAudioStatusEvent;
sp<TimedEventQueue::Event> mVideoLagEvent;
具体都是实现了什么功能呢?我们在具体调用的时候再深入讲解。
接下来我们就来讲讲onPrepareAsyncEvent方法了。
frameworks/av/media/libstagefight/AwesomePlayer.cpp
void AwesomePlayer::onPrepareAsyncEvent() { Mutex::Autolock autoLock(mLock); ………………………… if (mUri.size() > 0) { status_t err = finishSetDataSource_l();----这个不会走了,如果是本地文件的话 ………………………… if (mVideoTrack != NULL && mVideoSource == NULL) { status_t err = initVideoDecoder();-----------如果有videotrack初始化video的解码器 ………………………… if (mAudioTrack != NULL && mAudioSource == NULL) { status_t err = initAudioDecoder();---------------如果有audiotrack初始化audio解码器 …………………….. modifyFlags(PREPARING_CONNECTED, SET); if (isStreamingHTTP()) { postBufferingEvent_l(); ------一般不会走了 } else { finishAsyncPrepare_l();----------对外宣布prepare完成,并从timeeventqueue中移除该queueitem,mAsyncPrepareEvent=null } }
我们终于知道prepare主要目的了,根据类型找到解码器并初始化对应的解码器。那我们首先就来看看有videotrack的媒体文件是如何找到并初始化解码器吧。
先看图吧,了解大概步骤:
看完图就开讲了:
iniVideoDecoder目的是初始化解码器,取得已解码器的联系,解码数据输出格式等等。
frameworks/av/media/libstagefright/Awesomeplayer.cpp
status_t AwesomePlayer::initVideoDecoder(uint32_t flags) { ………… mVideoSource = OMXCodec::Create( mClient.interface(), mVideoTrack->getFormat(), false, // createEncoder mVideoTrack, NULL, flags, USE_SURFACE_ALLOC ? mNativeWindow : NULL); ………….. status_t err = mVideoSource->start(); }
我们先来看create函数到底干了啥吧:
frameworks/av/media/libstagefright/OMXCodec.cpp
sp<MediaSource> OMXCodec::Create( const sp<IOMX> &omx, const sp<MetaData> &meta, bool createEncoder, const sp<MediaSource> &source, const char *matchComponentName, uint32_t flags, const sp<ANativeWindow> &nativeWindow) { ………….. bool success = meta->findCString(kKeyMIMEType, &mime); …………… (1) findMatchingCodecs( mime, createEncoder, matchComponentName, flags, &matchingCodecs, &matchingCodecQuirks); ………. (2) sp<OMXCodecObserver> observer = new OMXCodecObserver; (3) status_t err = omx->allocateNode(componentName, observer, &node); ………. (4) sp<OMXCodec> codec = new OMXCodec( omx, node, quirks, flags, createEncoder, mime, componentName, source, nativeWindow); (5) observer->setCodec(codec); (6)err = codec->configureCodec(meta); ………… }
首先看下findMatchingCodecs,原来是根据mimetype找到匹配的解码组件,android4.1的寻找组件有了很大的变化,以前都是把codecinfo都写在代码上了,现在把他们都放到media_codec.xml文件中,full build 后会保存在“/etc/media_codecs.xml”,这个xml由各个芯片厂商来提供,这样以后添加起来就很方便,不用改代码了。一般是原生态的代码都是软解码。解码器的匹配方式是排名制,因为一般厂商的配置文件都有很多的同类型的编码器,谁排前面就用谁的。
frameworks/av/media/libstagefright/OMXCodec.cpp
void OMXCodec::findMatchingCodecs( const char *mime, bool createEncoder, const char *matchComponentName, uint32_t flags, Vector<String8> *matchingCodecs, Vector<uint32_t> *matchingCodecQuirks) { ………… const MediaCodecList *list = MediaCodecList::getInstance(); ……… for (;;) { ssize_t matchIndex = list->findCodecByType(mime, createEncoder, index); ……………….. matchingCodecs->push(String8(componentName)); ……………. } frameworks/av/media/libstagefright/MediaCodecList.cpp onst MediaCodecList *MediaCodecList::getInstance() { .. if (sCodecList == NULL) { sCodecList = new MediaCodecList; } return sCodecList->initCheck() == OK ? sCodecList : NULL; } MediaCodecList::MediaCodecList() : mInitCheck(NO_INIT) { FILE *file = fopen("/etc/media_codecs.xml", "r"); if (file == NULL) { ALOGW("unable to open media codecs configuration xml file."); return; } parseXMLFile(file); }
有了匹配的componentName,我们就可以创建ComponentInstance,这由allocateNode方法来实现。
frameworks/av/media/libstagefright/omx/OMX.cpp
status_t OMX::allocateNode( const char *name, const sp<IOMXObserver> &observer, node_id *node) { …………………… OMXNodeInstance *instance = new OMXNodeInstance(this, observer); OMX_COMPONENTTYPE *handle; OMX_ERRORTYPE err = mMaster->makeComponentInstance( name, &OMXNodeInstance::kCallbacks, instance, &handle); …………………………… *node = makeNodeID(instance); mDispatchers.add(*node, new CallbackDispatcher(instance)); instance->setHandle(*node, handle); mLiveNodes.add(observer->asBinder(), instance); observer->asBinder()->linkToDeath(this); return OK; }
在allocateNode,我们要用到mMaster来创建component,但是这个mMaster什么时候初始化了呢?我们看下OMX的构造函数:
OMX::OMX()
: mMaster(new OMXMaster),-----------原来在这呢!
mNodeCounter(0) {
}
但是我们前面没有讲到OMX什么时候构造的啊?我们只能往回找了,原来我们在初始化Awesomeplayer的时候忽略掉了,罪过啊:
AwesomePlayer::AwesomePlayer() : mQueueStarted(false), mUIDValid(false), mTimeSource(NULL), mVideoRendererIsPreview(false), mAudioPlayer(NULL), mDisplayWidth(0), mDisplayHeight(0), mVideoScalingMode(NATIVE_WINDOW_SCALING_MODE_SCALE_TO_WINDOW), mFlags(0), mExtractorFlags(0), mVideoBuffer(NULL), mDecryptHandle(NULL), mLastVideoTimeUs(-1), mTextDriver(NULL) { CHECK_EQ(mClient.connect(), (status_t)OK) 这个就是创建的地方 mClient是OMXClient, status_t OMXClient::connect() { sp<IServiceManager> sm = defaultServiceManager(); sp<IBinder> binder = sm->getService(String16("media.player")); sp<IMediaPlayerService> service = interface_cast<IMediaPlayerService>(binder);---很熟悉吧,获得BpMediaplayerservice CHECK(service.get() != NULL); mOMX = service->getOMX(); CHECK(mOMX.get() != NULL); if (!mOMX->livesLocally(NULL /* node */, getpid())) { ALOGI("Using client-side OMX mux."); mOMX = new MuxOMX(mOMX); } return OK; }
好了,我们直接进入mediaplayerservice.cpp看个究竟吧:
sp<IOMX> MediaPlayerService::getOMX() { Mutex::Autolock autoLock(mLock); if (mOMX.get() == NULL) { mOMX = new OMX; } return mOMX; }
终于看到了OMX的创建了,哎以后得注意看代码才行!!!
我们搞了那么多探究OMXMaster由来有什么用呢?
OMXMaster::OMXMaster() : mVendorLibHandle(NULL) { addVendorPlugin(); addPlugin(new SoftOMXPlugin); } void OMXMaster::addVendorPlugin() { addPlugin("libstagefrighthw.so"); }
原来是用来加载各个厂商的解码器(libstagefrighthw.so),还有就是把google本身的软解码器(SoftOMXPlugin)也加载了进来。那么这个libstagefrighthw.so在哪?我找了半天终于找到了,每个芯片厂商对应自己的libstagefrighthw
hardware/XX/media/libstagefrighthw/xxOMXPlugin
如何实例化自己解码器的component?我们以高通为例:
void OMXMaster::addPlugin(const char *libname) { mVendorLibHandle = dlopen(libname, RTLD_NOW); ……………………………. if (createOMXPlugin) { addPlugin((*createOMXPlugin)());-----创建OMXPlugin,并添加进我们的列表里 } }
hardware/qcom/media/libstagefrighthw/ QComOMXPlugin.cpp
OMXPluginBase *createOMXPlugin() { return new QComOMXPlugin; } QComOMXPlugin::QComOMXPlugin() : mLibHandle(dlopen("libOmxCore.so", RTLD_NOW)),----载入自己的omx API mInit(NULL), mDeinit(NULL), mComponentNameEnum(NULL), mGetHandle(NULL), mFreeHandle(NULL), mGetRolesOfComponentHandle(NULL) { if (mLibHandle != NULL) { mInit = (InitFunc)dlsym(mLibHandle, "OMX_Init"); mDeinit = (DeinitFunc)dlsym(mLibHandle, "OMX_DeInit"); mComponentNameEnum = (ComponentNameEnumFunc)dlsym(mLibHandle, "OMX_ComponentNameEnum"); mGetHandle = (GetHandleFunc)dlsym(mLibHandle, "OMX_GetHandle"); mFreeHandle = (FreeHandleFunc)dlsym(mLibHandle, "OMX_FreeHandle"); mGetRolesOfComponentHandle = (GetRolesOfComponentFunc)dlsym( mLibHandle, "OMX_GetRolesOfComponent"); (*mInit)(); } }
以上我们就可以用高通的解码器了。我们在创建component的时候就可以创建高通相应的component实例了:
OMX_ERRORTYPE OMXMaster::makeComponentInstance( const char *name, const OMX_CALLBACKTYPE *callbacks, OMX_PTR appData, OMX_COMPONENTTYPE **component) { Mutex::Autolock autoLock(mLock); *component = NULL; ssize_t index = mPluginByComponentName.indexOfKey(String8(name)); ----根据我们在media_codec.xml的解码器名字,在插件列表找到其索引 OMXPluginBase *plugin = mPluginByComponentName.valueAt(index); --根据索引找到XXOMXPlugin OMX_ERRORTYPE err = plugin->makeComponentInstance(name, callbacks, appData, component); -----创建组件 mPluginByInstance.add(*component, plugin); return err; }
hardware/qcom/media/libstagefrighthw/ QComOMXPlugin.cpp
OMX_ERRORTYPE QComOMXPlugin::makeComponentInstance( const char *name, const OMX_CALLBACKTYPE *callbacks, OMX_PTR appData, OMX_COMPONENTTYPE **component) { if (mLibHandle == NULL) { return OMX_ErrorUndefined; } String8 tmp; RemovePrefix(name, &tmp); name = tmp.string(); return (*mGetHandle)( reinterpret_cast<OMX_HANDLETYPE *>(component), const_cast<char *>(name), appData, const_cast<OMX_CALLBACKTYPE *>(callbacks)); }
哈哈,我们终于完成了app到寻找到正确解码器的工程了!!!
ComponentInstance, OMXCodecObserver,omxcodec,omx的关系和联系,我写了篇文章,可以到链接进去看看:
http://blog.csdn.net/tjy1985/article/details/7397752
OMXcodec通过binder(IOMX)跟omx建立了联系,解码器则通过注册的几个回调事件OMX_CALLBACKTYPE OMXNodeInstance::kCallbacks = {
&OnEvent, &OnEmptyBufferDone, &OnFillBufferDone
}往OMXNodeInstance这个接口上报消息,OMX通过消息分发机制往OMXCodecObserver发消息,它再给注册进observer的omxcodec(observer->setCodec(codec);)进行最后的处理!
stagefright 通过OpenOMX联通解码器的过程至此完毕。
create最后一步就剩下configureCodec(meta),主要是设置下输出的宽高和initNativeWindow。
忘了个事,就是OMXCOdec的状态:
enum State {
DEAD,
LOADED,
LOADED_TO_IDLE,
IDLE_TO_EXECUTING,
EXECUTING,
EXECUTING_TO_IDLE,
IDLE_TO_LOADED,
RECONFIGURING,
ERROR
};
在我们实例化omxcodec的时候该状态处于LOADED状态。
LOADER后应该就是LOADER_TO_IDLE,那什么时候进入该状态呢,就是我们下面讲的start方法:
status_t err = mVideoSource->start();
mVideoSource就是omxcodec,我们进入omxcodec.cpp探个究竟:
status_t OMXCodec::start(MetaData *meta) { …. return init(); } status_t OMXCodec::init() { …….. err = allocateBuffers(); err = mOMX->sendCommand(mNode, OMX_CommandStateSet, OMX_StateIdle); setState(LOADED_TO_IDLE); …………………… } start原来做了三件事啊, 1:allocateBuffers给输入端放入缓存的数据,给输出端准备匹配的native window status_t OMXCodec::allocateBuffers() { status_t err = allocateBuffersOnPort(kPortIndexInput); if (err != OK) { return err; } return allocateBuffersOnPort(kPortIndexOutput); }
2:分配完后通知解码器器端进入idle状态,sendCommand的流程可以参考http://blog.csdn.net/tjy1985/article/details/7397752
emptyBuffer过程
3:本身也处于IDLE。
到此我们的initVideoDecoder就完成了,initAudioDecoder流程也差不多一致,这里就不介绍了,有兴趣的可以自己跟进去看看。
prepare的最后一步finishAsyncPrepare_l(),对外宣布prepare完成,并从timeeventqueue中移除该queueitem,mAsyncPrepareEvent=null。
费了很多的口舌和时间,我们终于完成了prepare的过程,各路信息通道都打开了,往下就是播放的过程了。
转载请注明出处:太妃糖出品