Android-Choreographer工作原理

5,059

概述

注:本文基于Android 10源码,为了文章的简洁性,引用源码的地方可能有所删减。文中内容如有错误欢迎指出,共同进步!觉得不错的留个赞再走哈~

在开始解析 Choreographer 源码流程之前,先看一些基础概念和背景,直接参考大佬的总结 -- “终于懂了” 系列:Android屏幕刷新机制—VSync、Choreographer 全面理解!。显示系统一般包括 CPU, GPU 以及 Display 三个部分,CPU负责计算帧数据,把计算好的数据交给GPU,GPU会对图形数据进行渲染,渲染好后放到buffer(图像缓冲区)里存起来,然后Display(屏幕或显示器)负责把buffer里的数据呈现到屏幕上。当然实际上的流程要复杂许多,比如说 APP 绘制过程中会涉及到软件绘制硬件加速,以及通过 Surface 向图形缓存区写入渲染数据,另外在 SurfaceFlinger 进程会将这些不同 Surface(Layer) 的数据合成交给显示器的缓存区等等。

最近在整理 Android 图形系统相关的流程,为了对这些知识点有一个系统的理解,一边阅读别人的博客总结,一边阅读系统源码,之前总结过 SurfaceFlinger 启动与工作流程,这篇文章再看一下 Choreographer 与 Vsync 信号之间的相关工作,后续会更新一下 Surface 及 Layer 在 View 绘制过程中的数据走向,最后将这些内容结合起来,把整个图形系统的工作逻辑串联起来,然后再分开看这些知识点,应该就会有一个比较好的效果。源码解析过程中难免有所错误,如果发现了欢迎指正哈!

Choreographer: 编舞者,用来控制当收到 VSync 信号后才开始绘制任务,保证绘制拥有完整的16.6ms。

入口: scheduleTraversals

Android-Window机制原理 可知,在调用 context.startActivity 后,经过 AMS 的一些处理,后面又通过 Binder 调用目标进程的 ActivityThread.handleResumeActivity 方法,在这个方法里会回调目标 Activity 的 onResume 和 makeVisible 方法,在 makeVisible 方法里完成 WindowManager.addView 的过程,这个过程调用了 ViewRootImpl.setView方 法,内部又调用其 scheduleTraversals 方法,最后会走到 performTraversals 方法,接着就到了熟悉的 measure,layout,draw 三大流程了。

另外,查看 View.invalidate 方法的源码,也可以发现最后会调用到 ViewRootImpl.scheduleTraversals 方法。

// ViewRootImpl
final ViewRootHandler mHandler = new ViewRootHandler();

void scheduleTraversals() {
    if (!mTraversalScheduled) {
        // 保证同时间多次更改只会刷新一次,例如TextView连续两次setText()也只会走一次绘制流程
        mTraversalScheduled = true;
        // 添加同步屏障,保证 VSync 到来立即执行绘制
        mTraversalBarrier = mHandler.getLooper().getQueue().postSyncBarrier();
        mChoreographer.postCallback(Choreographer.CALLBACK_TRAVERSAL, mTraversalRunnable, null);
        // ...
    }
}

// mTraversalRunnable 是一个 Runnable 实例
final class TraversalRunnable implements Runnable {
    @Override
    public void run() {
        doTraversal();
    }
}

void doTraversal() {
    if (mTraversalScheduled) {
        mTraversalScheduled = false;
        // 移除同步屏障
        mHandler.getLooper().getQueue().removeSyncBarrier(mTraversalBarrier);
        // 真正执行View的measure,layout,draw流程
        performTraversals();
    }
}

首先使用 mTraversalScheduled 字段保证同时间多次更改只会刷新一次,然后为当前线程的 MessageQueue 添加同步屏障来屏蔽同步消息,保证 VSync 信号到来后立即执行绘制,而不是要等前面的同步消息。调用 mChoreographer.postCallback() 方法发送了一个会在下一帧执行的回调,即在下一个 VSync 信号到来时会执行TraversalRunnable-->doTraversal()-->performTraversals()-->绘制流程。同步屏障可以参考 Android消息机制

Choreographer实例化

首先看一下 Choreographer 的实例化过程,它在 ViewRootImpl 构造方法中实例化,ViewRootImpl 的实例化时机可以参考 Android-Window机制原理

// ViewRootImpl在WindowManager.addView时创建
public ViewRootImpl(Context context, Display display) {
    // ...
    mChoreographer = Choreographer.getInstance();
    // ...
}

public final class Choreographer {
    private static volatile Choreographer mMainInstance;

    private static final ThreadLocal<Choreographer> sThreadInstance = new ThreadLocal<Choreographer>() {
        @Override
        protected Choreographer initialValue() {
            Looper looper = Looper.myLooper();
            // VSYNC_SOURCE_APP = 0; -- APP
            // VSYNC_SOURCE_SURFACE_FLINGER = 1; -- SurfaceFlinger
            Choreographer choreographer = new Choreographer(looper, VSYNC_SOURCE_APP);
            // ...
            return choreographer;
        }
    };

    public static Choreographer getInstance() {
        return sThreadInstance.get();
    }
}

可知 Choreographer 和 Looper 一样都是线程单例的,由 ThreadLocal 实现,参考ThreadLocal原理

public final class Choreographer {
    // 4.1以上默认是true
    // Enable/disable vsync for animations and drawing.
    private static final boolean USE_VSYNC = SystemProperties.getBoolean("debug.choreographer.vsync", true);

    // VSync事件接收器
    private final FrameDisplayEventReceiver mDisplayEventReceiver;

    private Choreographer(Looper looper, int vsyncSource) {
        mLooper = looper;
        mDisplayEventReceiver = USE_VSYNC ? new FrameDisplayEventReceiver(looper, vsyncSource) : null;
        // ...
    }
}

在 Choreographer 实例化时创建了一个 FrameDisplayEventReceiver 对象,它用来注册 Vsync 信号。

Vsync信号注册

DisplayEventReceiver

mDisplayEventReceiver 是 FrameDisplayEventReceiver 类型的实例,在Choreographer构造方法中实例化,其父类为 DisplayEventReceiver。

public abstract class DisplayEventReceiver {
    public static final int VSYNC_SOURCE_APP = 0;
    private long mReceiverPtr;

    public DisplayEventReceiver(Looper looper, int vsyncSource) {
        mMessageQueue = looper.getQueue();
        // 注册VSYNC信号监听者
        mReceiverPtr = nativeInit(new WeakReference<DisplayEventReceiver>(this), mMessageQueue, vsyncSource);
    }

    private static native long nativeInit(WeakReference<DisplayEventReceiver> receiver, MessageQueue messageQueue, int vsyncSource);
}

nativeInit

nativeInit是一个native方法,其实现在frameworks/base/core/jni/android_view_DisplayEventReceiver.cpp中:

static jlong nativeInit(JNIEnv* env, jclass clazz, jobject receiverWeak,
        jobject messageQueueObj, jint vsyncSource) {
    sp<MessageQueue> messageQueue = android_os_MessageQueue_getMessageQueue(env, messageQueueObj);
    sp<NativeDisplayEventReceiver> receiver = new NativeDisplayEventReceiver(env, receiverWeak, messageQueue, vsyncSource);
    status_t status = receiver->initialize();
    receiver->incStrong(gDisplayEventReceiverClassInfo.clazz); // retain a reference for the object
    return reinterpret_cast<jlong>(receiver.get());
}

NativeDisplayEventReceiver 继承自 DisplayEventDispatcher:

DisplayEventReceiver::DisplayEventReceiver(ISurfaceComposer::VsyncSource vsyncSource) {
    sp<ISurfaceComposer> sf(ComposerService::getComposerService());
    if (sf != NULL) {
        mEventConnection = sf->createDisplayEventConnection(vsyncSource);
        if (mEventConnection != NULL) {
            mDataChannel = std::make_unique<gui::BitTube>();
            mEventConnection->stealReceiveChannel(mDataChannel.get());
        }
    }
}

sp<IDisplayEventConnection> SurfaceFlinger::createDisplayEventConnection(
        ISurfaceComposer::VsyncSource vsyncSource) {
    if (vsyncSource == eVsyncSourceSurfaceFlinger) {
        return mSFEventThread->createEventConnection();
    } else {
        // vsyncSource 是 APP
        return mEventThread->createEventConnection();
    }
}

SurfaceFlinger 启动与工作流程 可以知道 EventThread.createEventConnection 创建了一个对 Vsync 信号感兴趣的连接,具体逻辑可以阅读这篇文章。initialize 方法如下:

// frameworks/base/libs/androidfw/DisplayEventDispatcher.cpp
status_t DisplayEventDispatcher::initialize() {
    // DisplayEventReceiver mReceiver;
    status_t result = mReceiver.initCheck();
    int rc = mLooper->addFd(mReceiver.getFd(), 0, Looper::EVENT_INPUT, this, NULL);
    if (rc < 0) {
        return UNKNOWN_ERROR;
    }
    return OK;
}

mReceiver 是 DisplayEventReceiver 类型实例,位于frameworks/native/libs/gui/DisplayEventReceiver.cppmLooper->addFd(mReceiver.getFd(), 0, Looper::EVENT_INPUT, this, NULL) 用来监听 mReceiver 所获取的文件句柄,当有存在对 Vsync 信号感兴趣的连接接收到了 Vsync 信号时,会发送数据到 mReceiver, 然后回调到 DisplayEventDispatcher 中的 handleEvent 方法,具体源码参考 SurfaceFlinger 启动与工作流程 中 addFd 的解析。

请求Vsync信号

上面已经注册了一个对 Vsync 信号感兴趣的连接,在 Vsync 信号到来后,会回调到 DisplayEventDispatcher.handleEvent 方法。于是接下来我们需要请求 Vsync 信号。看一下上面调用的代码:mChoreographer.postCallback(Choreographer.CALLBACK_TRAVERSAL, mTraversalRunnable, null),它的调用链是: Choreographer.postCallback -> Choreographer.postCallbackDelayedInternal -> Choreographer.scheduleFrameLocked -> Choreographer.scheduleVsyncLocked 方法,节省篇幅,具体代码不贴出了:

private void postCallbackDelayedInternal(int callbackType, Object action, Object token, long delayMillis) {
    synchronized (mLock) {
        final long now = SystemClock.uptimeMillis();
        final long dueTime = now + delayMillis;
        // 对应类型的 CallbackQueue 添加 Callback
        mCallbackQueues[callbackType].addCallbackLocked(dueTime, action, token);
        // ...
    }
}

private void scheduleVsyncLocked() {
    mDisplayEventReceiver.scheduleVsync();
}

// DisplayEventReceiver
public void scheduleVsync() {
    if (mReceiverPtr == 0) {
        // ...
    } else {
        nativeScheduleVsync(mReceiverPtr);
    }
}

接着就到了 native 层代码:

// frameworks/base/core/jni/android_view_DisplayEventReceiver.cpp
static void nativeScheduleVsync(JNIEnv* env, jclass clazz, jlong receiverPtr) {
    sp<NativeDisplayEventReceiver> receiver = reinterpret_cast<NativeDisplayEventReceiver*>(receiverPtr);
    status_t status = receiver->scheduleVsync();
    // ...
}

// frameworks/base/libs/androidfw/DisplayEventDispatcher.cpp
status_t DisplayEventDispatcher::scheduleVsync() {
    if (!mWaitingForVsync) {
        // ...
        // mReceiver 是 DisplayEventReceiver 实例
        status_t status = mReceiver.requestNextVsync();
        mWaitingForVsync = true;
    }
    return OK;
}

// frameworks/native/libs/gui/DisplayEventReceiver.cpp
status_t DisplayEventReceiver::requestNextVsync() {
    if (mEventConnection != NULL) {
        // 请求接收下一次Vsync信号的回调
        mEventConnection->requestNextVsync();
        return NO_ERROR;
    }
    return NO_INIT;
}

可以看到最终调用了 requestNextVsync 函数,关于 requestNextVsync 的逻辑已经在 SurfaceFlinger 启动与工作流程 中解析过了,它用来请求接收下一次 Vsync 信号,可以唤醒 EventThread 线程,等到 Vsync 信号到来后回调给 APP。

Vsync回调流程

在 Vsync 信号到来后,便来到了这里。

DisplayEventDispatcher::handleEvent

int DisplayEventDispatcher::handleEvent(int, int events, void*) {
    // ...
    // 分发Vsync,实现方法在子类中
    dispatchVsync(vsyncTimestamp, vsyncDisplayId, vsyncCount);
    // 这里返回 1 是为了能一直保持 addFd 中添加的监听 callback
    // 具体逻辑可以参考之前 SurfaceFlinger 的解析
    return 1; // keep the callback
}

void NativeDisplayEventReceiver::dispatchVsync(nsecs_t timestamp, int32_t id, uint32_t count) {
    JNIEnv* env = AndroidRuntime::getJNIEnv();
    ScopedLocalRef<jobject> receiverObj(env, jniGetReferent(env, mReceiverWeakGlobal));
    if (receiverObj.get()) {
        env->CallVoidMethod(receiverObj.get(), gDisplayEventReceiverClassInfo.dispatchVsync, timestamp, id, count);
    }
    mMessageQueue->raiseAndClearException(env, "dispatchVsync");
}

int register_android_view_DisplayEventReceiver(JNIEnv* env) {
    // ...
    jclass clazz = FindClassOrDie(env, "android/view/DisplayEventReceiver");
    gDisplayEventReceiverClassInfo.clazz = MakeGlobalRefOrDie(env, clazz);
    gDisplayEventReceiverClassInfo.dispatchVsync = GetMethodIDOrDie(env,
            gDisplayEventReceiverClassInfo.clazz, "dispatchVsync", "(JII)V");
    return res;
}

由上可知,会调用到Java层 android/view/DisplayEventReceiver 的 dispatchVsync 方法:

public abstract class DisplayEventReceiver {
    private void dispatchVsync(long timestampNanos, int builtInDisplayId, int frame) {
        onVsync(timestampNanos, builtInDisplayId, frame);
    }

    public void onVsync(long timestampNanos, int builtInDisplayId, int frame) {
    }
}

onVsync 实现在子类 FrameDisplayEventReceiver 中。

FrameDisplayEventReceiver.onVsync

private final class FrameDisplayEventReceiver extends DisplayEventReceiver implements Runnable {
    private boolean mHavePendingVsync;
    private long mTimestampNanos;
    private int mFrame;

    @Override
    public void onVsync(long timestampNanos, int builtInDisplayId, int frame) {
        if (builtInDisplayId != SurfaceControl.BUILT_IN_DISPLAY_ID_MAIN) {
            // 非主display
            Log.d(TAG, "Received vsync from secondary display, but we don't support "
                    + "this case yet.  Choreographer needs a way to explicitly request "
                    + "vsync for a specific display to ensure it doesn't lose track "
                    + "of its scheduled vsync.");
            scheduleVsync();
            return;
        }

        long now = System.nanoTime();
        if (timestampNanos > now) {
            timestampNanos = now;
        }

        if (mHavePendingVsync) {
            Log.w(TAG, "Already have a pending vsync event.  There should only be one at a time.");
        } else {
            mHavePendingVsync = true;
        }

        mTimestampNanos = timestampNanos;
        mFrame = frame;
        // 会调用run方法
        Message msg = Message.obtain(mHandler, this);
        msg.setAsynchronous(true);
        mHandler.sendMessageAtTime(msg, timestampNanos / TimeUtils.NANOS_PER_MS);
    }

    @Override
    public void run() {
        mHavePendingVsync = false;
        doFrame(mTimestampNanos, mFrame);
    }
}

Choreographer.doFrame

于是接收到 Vsync 信号后,又执行回到了 Choreographer:

private static final int SKIPPED_FRAME_WARNING_LIMIT = SystemProperties.getInt("debug.choreographer.skipwarning", 30);

void doFrame(long frameTimeNanos, int frame) {
    final long startNanos;
    synchronized (mLock) {
        if (!mFrameScheduled) {
            return; // no work to do
        }
        // 计划执行时间
        long intendedFrameTimeNanos = frameTimeNanos;
        startNanos = System.nanoTime();
        final long jitterNanos = startNanos - frameTimeNanos;
        if (jitterNanos >= mFrameIntervalNanos) {
            // 是否超过一帧的时间,因为虽然添加了同步屏障,但是如果有正在执行的同步任务,会导致doFrame延迟执行
            // 计算掉帧数
            final long skippedFrames = jitterNanos / mFrameIntervalNanos;
            if (skippedFrames >= SKIPPED_FRAME_WARNING_LIMIT) {
                // 默认掉帧超过30帧打印日志
                Log.i(TAG, "Skipped " + skippedFrames + " frames!  "
                    + "The application may be doing too much work on its main thread.");
            }
            final long lastFrameOffset = jitterNanos % mFrameIntervalNanos;
            frameTimeNanos = startNanos - lastFrameOffset;
        }

        if (frameTimeNanos < mLastFrameTimeNanos) {
            scheduleVsyncLocked(); // 请求下一次 Vsync 信号
            return;
        }

        if (mFPSDivisor > 1) {
            long timeSinceVsync = frameTimeNanos - mLastFrameTimeNanos;
            if (timeSinceVsync < (mFrameIntervalNanos * mFPSDivisor) && timeSinceVsync > 0) {
                scheduleVsyncLocked(); // 请求下一次 Vsync 信号
                return;
            }
        }

        mFrameInfo.setVsync(intendedFrameTimeNanos, frameTimeNanos);
        mFrameScheduled = false;
        mLastFrameTimeNanos = frameTimeNanos;
    }

    try {
        // 按类型执行,Choreographer中有四种类型
        AnimationUtils.lockAnimationClock(frameTimeNanos / TimeUtils.NANOS_PER_MS);

        mFrameInfo.markInputHandlingStart();
        doCallbacks(Choreographer.CALLBACK_INPUT, frameTimeNanos);

        mFrameInfo.markAnimationsStart();
        doCallbacks(Choreographer.CALLBACK_ANIMATION, frameTimeNanos);

        mFrameInfo.markPerformTraversalsStart();
        doCallbacks(Choreographer.CALLBACK_TRAVERSAL, frameTimeNanos);

        doCallbacks(Choreographer.CALLBACK_COMMIT, frameTimeNanos);
    } finally {
        AnimationUtils.unlockAnimationClock();
    }
}

Choreographer.doCallbacks

void doCallbacks(int callbackType, long frameTimeNanos) {
    CallbackRecord callbacks;
    synchronized (mLock) {
        final long now = System.nanoTime();
        // 根据指定的类型CallbackkQueue中查找到达执行时间的CallbackRecord
        callbacks = mCallbackQueues[callbackType].extractDueCallbacksLocked(now / TimeUtils.NANOS_PER_MS);
        // ...
    }
    // 迭代执行队列所有任务
    for (CallbackRecord c = callbacks; c != null; c = c.next) {
        c.run(frameTimeNanos);
    }
}

private static final class CallbackRecord {
    public CallbackRecord next;
    public long dueTime;
    public Object action; // Runnable or FrameCallback
    public Object token;

    public void run(long frameTimeNanos) {
        if (token == FRAME_CALLBACK_TOKEN) {
            ((FrameCallback)action).doFrame(frameTimeNanos);
        } else { // 直接调用 Runnable 的 run 方法
            ((Runnable)action).run();
        }
    }
}

于是到这里就可以开始执行真正的绘制操作了。

总结

  • Choreographer: 使 CPU/GPU 的绘制是在 VSYNC 到来时开始。Choreographer 初始化时会创建一个表示对 Vsync 信号感兴趣的连接,当有绘制请求时通过 postCallback 方法请求下一次 Vsync 信号,当信号到来后才开始执行绘制任务。
  • 只有当 App 注册监听下一个 Vsync 信号后才能接收到 Vsync 到来的回调。如果界面一直保持不变,那么 App 不会去接收每隔 16.6ms 一次的 Vsync 事件,但底层依旧会以这个频率来切换每一帧的画面(也是通过监听 Vsync 信号实现)。即当界面不变时屏幕也会固定每 16.6ms 刷新,但 CPU/GPU 不走绘制流程。
  • 当 View 请求刷新时,这个任务并不会马上开始,而是需要等到下一个 Vsync 信号到来时才开始;measure/layout/draw 流程运行完后,界面也不会立刻刷新,而会等到下一个 VSync 信号到来时才进行缓存交换和显示。
  • 造成丢帧主要有两个原因:一是遍历绘制 View 树以及计算屏幕数据超过了16.6ms;二是主线程一直在处理其他耗时消息,导致绘制任务迟迟不能开始(同步屏障不能完全解决这个问题)。
  • 可通过Choreographer.getInstance().postFrameCallback()来监听帧率情况,其用法和原理参考 postFrameCallback用法

阅读这篇文章建议先阅读 SurfaceFlinger 启动与工作流程 这篇文章,然后结合 Choreographer 的工作流程,可以对 Vsync 信号是怎么协调 App 端的绘制任务以及 SurfaceFlinger 的合成任务有一个比较清晰的认识。

用一张图总结一下 Choreographer 的工作流程:

Choreographer工作流程