一、概述:
声明:下面我们看的是Android 4.0版本下的Media播放器的框架,我们首先从一个简单的media播放器apk源码入手,从Java->JNI->C/C++一步步研究Android是如何通过Java一个MediaPlayer实现解码到屏幕的输出。
通常在Android中播放视频用到的是MediaPlayer类,展示视频使用的是SurfaceView控件。
二、apk实现:
我们首先在main.xml布局文件中添加用于视频画面绘制的SurfaceView控件:
<SurfaceView android:layout_width="fill_parent"android:layout_height="240dip"android:id="@+id/surfaceView"/>
例如我们编写一个简单的视频播放器调用的常用方法如下。
SurfaceView surfaceView = (SurfaceView)this.findViewById(R.id.surfaceView); surfaceView.getHolder().setFixedSize(720, 576); //设置分辨率 /*下面设置Surface不维护自己的缓冲区,而是等待屏幕的渲染引擎将内容推送到用户面前*/ surfaceView.getHolder().setType(SurfaceHolder.SURFACE_TYPE_PUSH_BUFFERS); /* new 一个播放器 mediaPlayer */ MediaPlayer mediaPlayer = new MediaPlayer(); mediaPlayer.reset(); //重置为初始状态 mediaPlayer.setAudioStreamType(AudioManager.STREAM_MUSIC); /* 设置Video影片以SurfaceHolder播放 */ mediaPlayer.setDisplay(surfaceView.getHolder()); mediaPlayer.setDataSource("/mnt/sdcard/test.ts"); mediaPlayer.prepare(); //缓冲 mediaPlayer.start(); //播放 mediaPlayer.pause(); //暂停播放 mediaPlayer.start(); //恢复播放 mediaPlayer.stop(); //停止播放 mediaPlayer.release(); //释放资源
三、源码分析
我们首先来看MediaPlayer类,Java层的MediaPlayer.java位于frameworks/base/media/java/android/media/目录下:
public class MediaPlayer { ... static { System.loadLibrary("media_jni"); native_init(); } private int mNativeContext; private int mNativeSurfaceTexture; private int mListenerContext; private SurfaceHolder mSurfaceHolder; private EventHandler mEventHandler; public MediaPlayer() { Looper loop; if((loop = Looper.myLooper()) != null) { mEventHandler = new EventHandler(this, looper); } else if((looper = Looper.getMainLooper()) != null) { mEventHandler = new EventHandler(this. looper); } else { mEventHandler = null; } native_setup(new WeakReference<MediaPlayer>(this)); } .... }
(一) static代码块
首先会加载libmedia_jni.so库,调用native_init()方法,对应JNI接口为
android_media_MediaPlayer_native_init(JNIEnv* env) { jclass clazz; class = env->FindClass("android/media/MediaPlayer"); fields.context = env->GetFieldID(clazz, "mNativeContext"); // Java类中保存JNI层的mediaplayer对象 /* JNI 事件通知Java,static 函数 */ fields.post_event = env->GetStaticMethodID(clazz, "postEventFromNative", "(Ljava/lang/Object;IIILjava/lang/Object;)V"); fields.surface_texture = env->GetFieldID(clazz, "mNativeSurfaceTexture", "I"); jclass surface = env->FindClass("android/view/Surface"); fields.bitmapClazz = env->FindClass("android/graphics/Bitmap"); fields.bitmapContstructor = env->GetMethodID(fields.bitmapClazz, "<init>", "(I[BZ[BI)V"); // 找到Bitmap的构造函数 }
(二) 构造方法 MediaPlayer
这里在MediaPlayer构造函数中会new 一个 EventHandler,其中EventHandler是MediaPlayer的一个内部类,继承于Handler。用于处理各种消息:MEDIA_PREPARED、MEDIA_PLAYBACK_COMPLETE、MEDIA_BUFFERING_UPDATE、MEDIA_SEEK_COMPLETE、MEDIA_SET_VIDEO_SIZE、MEDIA_ERROR、MEDIA_INFO、MEDIA_TIMED_TEXT、MEDIA_NOP等消息,对此分别调用接口OnPreparedListener的onPrepared()、OnCompletionListener的onCompletion()、OnBufferingUpdateListener的onBufferingUpdate()。。。等方法来处理,而这些方法我们都可以通过实现相应的接口来处理。
重点还是native_setup(new WeakReference<MediaPlayer>(this))这句话,调用的JNI方法:
android_media_MediaPlayer_setup(JNIEnv* env, JObject thiz, jobject weak_this) { /* 这里参数中:thiz代码Java层的MediaPlayer对象,weak_this表示对Java层MediaPlayer对象的弱引用*/ /* 这里首先在JNI层 new 一个 MediaPlayer对象 */ sp<MediaPlayer> mp = new MediaPlayer(); // Create new listener and give it to MediaPlayer sp<JNIMediaPlayerListener> listener = new JNIMediaPlayerListener(env, thiz, weak_this); mp->setListener(listener); setMediaPlayer(env, thiz, mp); // Stow out new C++ MediaPlayer }
这里我们首先看看JNIMediaPlayerListener类:
class JNIMediaPlayerListener : public MediaPlayerListener { public: JNIMediaPlayerListener(JNIEnv* env, jobject thiz, jobject weak_thiz); ~JNIMediaPlayerListener(); virtual void notify(int msg, int ext1, int ext2, const Parcel *obj = NULL); private: JNIMediaPlayerListener(); jclass mClass; // 对 MediaPlayer类的引用 jobject mObject; // 对Java层MediaPlayer对象的弱引用 }
我们只需要重点关注下notify这个函数,用于JNI层向Java层通知事件,以后在分析底层播放的时候我们会用到:
void JNIMediaPlayerListener::notify(int msg, int ext1, int ext2, const Parcel *obj = NULL) { JNIEnv *env = AndroidRuntime::getJNIEnvt(); if(obj && obj->dataSize() > 0) { jbyteArray jArray = env->NewByteArray(obj->dataSize()); jbyte *nArray = env->GetByteArrayElements(jArray, NULL); memcpy(nArray, obj->data(), obj->dataSize()); env->ReleaseByteArrayElements(jArray, nArray, 0); env->CallStaticVoidMethod(mClass, fields.post_event, mObject, msg, ext1, ext2, jArray); env->DeleteLocalRef(jArray); } else env->CallStaticVoidMethod(mClass, fields.post_event, mObject, msg, ext1, ext2, NULL); }
我们在前面native_init的JNI实现中初始化了fields.post_event对应的是Java层MediaPlayer对象中的postEventFromNative,
private static void postEventFromNative(Object mediaplayer_ref, int what, int arg1, int arg2, Object obj) { MediaPlayer mp = (MediaPlayer)((WeakReference)mediaplayer_ref).get(); Message m = mp.mEventHandler.obtainMessage(what, arg1, arg2, obj); mp.mEventHandler.sendMessage(m); } 最后是通过EventHandler来处理。
现在我们看看构造JNI层的MediaPlayer类:定义在frameworks/base/include/media/mediaplayer.h中
class MediaPlayer : public BnMediaPlayerClient, public virtual IMediaDeathNotifier { public: MediaPlayer(); ~MediaPlayer(); void died(); void disconnect(); status_t setDataSource(const char* url, const KeyedVector<String8, String8> *headers); ... status_t setVideoSurfaceTexture(const sp<ISurfaceTexture>& surfaceTexture); status_t setListener(const sp<MediaPlayerListener>& listener); status_t prepare(); status_t start(); status_t stop(); status_t pause(); ... void notify(int msg, int ext1, int ext2, const Parcel &obj = NULL); static sp<IMemory> decode(const char* url, uint32_t *pSampleRate, int *pNumChannels, int *pFormat); status_t setAudioSessionID(int sessionId); .... private: sp<IMediaPlayer> mPlayer; // 对应着MediaPlayerService内部类Client在客户端的代理,相当于BpMediaPlayer thread_id_t mLockThreadId; sp<MediaPlayerListener> mListener; .... };
new 完MediaPlayer对象之后,设置其监听变量mListener = listener,然后将其保存到Java层对象中。
static sp<MediaPlayer> setMediaPlayer(JNIEnv* env, jobject thiz, const sp<MediaPlayer>& player) { Mutex::Autolock l(sLock); sp<MediaPlayer> old = (MediaPlayer*)env>GetIntField(thiz, fields.context); if(player.get()) player->incStrong(thiz); // 增加player对象强引用计数 if(old != 0) old->decStrong(thiz); // 对原来的Java层保存的JNI层MediaPlayer对象减少强引用计数 // 将新的player对象保存到Java层的thiz对象的fields.context对应的变量mNativeContext中 env->SetIntField(thiz, fields.context, (int)player.get()); return old; }
(三)设置播放器参数
好了,前面这么多就是我们在Java代码中调用MediaPlayer mp = new MediaPlayer(),Android整个Media所需要执行的过程,下面我们继续执行mp.reset() 和 mp.setAudioStreamType(AudioManager.STREAM_MUSIC);这两个函数比较简单都是直接调用到JNI层函数,我们直接跳到JNI函数:
static void android_media_MediaPlayer_reset(JNIEnv *env, jobject thiz) { sp<MediaPlayer> mp = getMediaPlayer(env, thiz); process_media_player_call(env, thiz, mp->reset(), NULL, NULL); }调用的C++层MediaPlayer的reset()函数,其reset()就是设置相应的播放器状态等,如mCurrentState = MEDIA_PLAYER_IDLE将MediaPlayer类的mPlayer对象设置为0.(这里的sp<IMediaPlayer> mPlayer对应的就是IMediaPlayer的Bp客户端的代理,实际类型为MediaPlayerService的内部类Client。)
status MediaPlayer::setAudioStreamType(int type)就是根据当前播放器的状态类配置mStreamType变量,如果当前MediaPlayer对象已经调用过了prepare(),播放器进入了MEDIA_PLAYER_PREPARED状态则无法设置streamType。
下面我们看看mp.setDisplay()设置播放器显示输出。
public vodi setDisplay(SurfaceHolder sh) { mSurfaceHolder = sh; Surface surface; surface = sh.getSurface(); _setVideoSurface(surface); // 调用JNI函数 updateSurfaceScreenOn(); }
这里的SurfaceHolder实际是一个接口,需要继承类去实现,定义frameworks/base/core/java/android/view/SurfaceHolder.java
我们看到注释介绍:Abstract interface to someone holding a display surface. Allows you to control the surface size and format, edit the pixels in the surface, and monitor changes to the surface.
原来是一个抽象接口类,里面封装了Surface,主要用于对Surface的控制操作如:改变大小、格式、像素等。
_setVideoSurface()在JNI层实现为:
static void setVideoSurface(JNIEnv* env, jobject thiz, jobject jsurface, jboolean mediaPlayerMustBeAlive = true) { sp<MediaPlayer> mp = getMediaPlayer(env, thiz); decVideoSurfaceRef(env, thiz); // 减少Java类中mNativeSurfaceTexture保存的JNI层对之前ISurfaceTexture对象的弱引用 // Surface 是Android中比较复杂的一个模块,我们以后再分析,现在只需要知道是一块显示区域就行了。 sp<ISurfaceTexture> new_st; sp<Surface> surface(Surface_getSurface(env, jsurface)); new_st = surface->getSurfaceTexture(); new_st->incStrong(thiz); env->SetIntField(thiz, fields.surface_texture, (int)new_st.get()); // 重新设置到Java类中mNativeSurfaceTexture保存的JNI层对象的引用 mp->setVideoSurfaceTexture(new_st); }
(四) 设置播放源setDataSource
setDataSource()设置播放源,对应的JNI函数为: static void android_media_MediaPlayer_setDataSource(JNIEnv *env, jobject thiz, jstring path) { android_media_MediaPlayer_setDataSourceAndHeaders(env, thiz, path, NULL, NULL); }
调用的是mp.setDataSource(path, NULL);
status_t MediaPlayer::setDataSource(const char* url, const KeyedVector<String8, String8>* headers) { const sp<IMediaPlayerService>& service(getMediaPlayerService()); sp<IMediaPlayer> player(service->create(getpid(), this, mAudioSessionId)); player->setDataSource(url, headers); attachNewPlayer(player); }
这里首先 MediaPlayer的基类IMediaDeathNotifier的static函数getMediaPlayerService()函数获取MediaPlayerService的代理BpMediaPlayerService,然后通过Binder通信调用IMediaPlayerService的create()函数,传入的参数分别为当前线程的PID,当前MediaPlayer对象和前面获取的mAudioSessionId,返回一个IMediaPlayer对象实际类型为MediaPlayerService内部类型Client。
sp<IMediaPlayerService> IMediaDeathNotifier::sMediaPlayerService; const sp<IMediaPlayerService>& IMediaDeathNotifier::getMediaPlayerService() { if(sMediaPlayerService.get() == 0) { sp<IServiceManager> sm = defaultServiceManager(); sp<IBinder> binder; do { binder = sm->getService(String16("media.player")); if(binder != 0) break; usleep(500000); // 0.5 s } while(true); if(sDeathNotifier == NULL) sDeathNotifier = new DeathNotifier(); } binder->linkToDeath(sDeathNotifier); sMediaPlayerService = interface_cast<IMediaPlayerService>(binder); return sMediaPlayerService; }
这里调用的service->create(getpid(), this, mAudioSessionId); 是通过Binder通信调用到MediaPlayerService::create()。
@frameworks/base/media/libmediaplayerservice/MediaPlayerService.cpp
class MediaPlayerService : public BnMediaPlayerService { class Client; class AudioOutput : public MediaPlayerBase::AudioSink { public: AudioOutput(int sessionId); virtual ~AutioOutput(); virtual status_t open(uint32_t sampleRate, int channelCount, int format, int bufferCount, AudioCallback cb, void *cookie); virtual void start(); virtual ssize_t write(const void* buffer, size_t size); ... private: AudioTrack* mTrack; AudioCallback mCallback; ... }; class AudioCache : public MediaPlayserBase::AudioSink { ... }; public: static void instantiate(); // start the MediaPlayerService virtual sp<IMediaPlayer> create(pid_t pid, cosnt sp<IMediaPlayerClient>& client, int audioSessionId); virtual sp<IOMX> getOMX(); ... private: virtual MediaPlayerService(); virtual ~MediaPlayerService(); SortedVector< wp<Client> > mClient; SortedVector< wp<MediaRecorderClient> > mMediaRecorderClients; sp<IOMX> mOMX; // 有个内部类 Client class Client : public BnMediaPlayer { public: virtual status_t setVideoSurfaceTexture(const sp<ISurfaceTexture>& surfaceTexture); virtual status_t prepareAsync(); virtual status_t start(); ... sp<MediaPlayerBase> createPlayer(player_type playerType); virtual status_t setDataSource(const char* url, ...); static void notify(void* cookie, int msg, int ext1, int ext2, const Parcel *obj); private: friend class MediaPlayerService; Client(const sp<MediaPlayerService>& service, pid_t pid, int32_t connId, const sp<IMediaPlayerClient>& client, int audioSessionId, uid_t uid); sp<MediaPlayerBase> mPlayer; sp<MediaPlayerService> mService; sp<IMediaPlayerClient> mClient; }; };
这里看下MediaPlayerService的create函数:
sp<IMediaPlayer> create(pid_t pid, cosnt sp<IMediaPlayerClient>& client, int audioSessionId) { int32_t connId = android_atomic_inc(&mNextConnId); // 穿入参数分别为:当前MediaPlayerService对象、客户端进程pid、客户端MediaPlayer对象的引用、audioSessionId等 sp<Client> c = new Client(this, pid, client, audioSessionId, IPCThreadState::self()->getCallingUid()); wp<Client> w = c; // Client的构造函数就是一些简单的给变量赋值操作了 mClients.add(w); return c; }
分析到这里我们在Java中调用setDataSource()的时候底层MediaPlayer类首先通过Binder获取MediaPlayerService的代理BpMediaPlayerService,然后调用其create函数,由MediaPlayerService来处理,返回一个匿名Binder对象Client,返回类型为IMediaPlayer,保存在MediaPlayer类的sp<IMediaPlayer> mPlayer中。后续我们就可以直接使用这个mPlayer对象了,调用它的setDataSource、prepare、start、stop等。
下面我们可以直接跳到MediaPlayerService内部类Client里面去看setDataSource():
MediaPlayerService::Client::setDataSource(const char* url, const KeyedVector<String8, String8> *headers) { if(strncmp(url, "http://", 7) == 0 || strncmp(url, "https://", 8)==0 || strncmp(url, "rtsp://", 7) == 0) { checkPermission("android.permission.INTERNET"); // 检查是否具有网络权限 } if(strncmp(url, "content://", 10) == 0) { String16 url16(url); int fd = android::openContentProviderFile(url16); setDataSource(fd, 0, 0x7ffffffffLL); close(fd); return mStatus; } else { player_type playerType = getPlayerType(url); // 根据URL获取播放器类型 sp<MediaPlayerBase> p = createPlayer(playerType); // 根据播放器类型创建播放器 // 这里我们分析Android的StagefrightPlayer播放器 if(!p->hardwareOutput()) { mAudioOutput = new AudioOutput(mAudioSessionId); static_cast<MediaPlayerInterface*>(p.get())->setAudioSink(mAudioOutput); } } mStatus = p->setDataSource(url, headers); mPlayer = p; return mStatus; }
分析代码我们发现StagefrightPlayer其实就是AwesomePlayer的封装,基本上StagefrightPlayer所有的方法都是调用AwesomePlayer来实现的。AwesomePlayer类比较复杂,涉及到获取音视频流的格式、找到并打开相应的解码器、缓冲音视频数据送到解码器解码都是在这个类里面完成,主要通过mVideoEvent、mStreamDoneEvent、mBufferingEvent、mAsyncPrepareEvent等几个事件队列来进行驱动和调用完成视频的播放。