android9.0 硬件加速HWUI经验分享

作者: 新进取者 | 来源:发表于2022-05-17 19:57 被阅读0次

一.硬件加速初始化

Canvas API用来绘制应用程序的UI元素，在硬件加速渲染环境中，这些Canvas API调用最终会转化为Open GL API调用（转化过程对应用程序来说是透明的）。因此，新的Activity启动的时候初始化好Open GL环境（又称Open GL渲染上下文）尤为重要。
下面展示下hwui 过程图：

hwui过程.png

一个Activity在OpenGL环境中对应一个ANativeWindow，ANativeWindow从SF中dequeueBuffer得到GraphicBuffer之后通过OpenGL绘制完成后queueBuffer到SF进行合成显示。

1） Open GL渲染上下文只能与一个线程关联，避免多线程冲突，与只能在UI线程中更新UI一个意思，因此，初始化过程任务之一就是要创建一个Render Thread；
2）一个Android应用程序可能存在多个Activity组件，当Main Thread向Render Thread发出渲染命令时，Render Thread要知道当前要渲染的窗口是哪个，因此，初始化任务之二就是要告诉Render Thread当前要渲染的窗口是哪个。

下面就从这2个方面介绍hwui 初始化过程：

1.RenderThread初始化

1.1 java层分析

从ViewRootImpl的setView函数开始说起。在该函数内部会判断有些不会走hwui，比如：canvas api不支持转换成opengl函数的；还有些不需要hwui绘制的（因为hwui会增加内存开销）。

frameworks/base/core/java/android/view/ViewRootImpl.java
public void setView(View view, WindowManager.LayoutParams attrs, View panelParentView) {
...
// If the application owns the surface, don't enable hardware acceleration
                if (mSurfaceHolder == null) {
                    // While this is supposed to enable only, it can effectively disable
                    // the acceleration too.
                    enableHardwareAcceleration(attrs);

SurfaceView是完全由应用程序自己来控制自己的渲染,因此不需要开启硬件加速。

frameworks/base/core/java/android/view/ViewRootImpl.java
private void enableHardwareAcceleration(WindowManager.LayoutParams attrs) {
...
// Don't enable hardware acceleration when the application is in compatibility mode
        if (mTranslator != null) return;
// Try to enable hardware acceleration if requested
        final boolean hardwareAccelerated =
                (attrs.flags & WindowManager.LayoutParams.FLAG_HARDWARE_ACCELERATED) != 0;

        if (hardwareAccelerated) {
           if (!ThreadedRenderer.isAvailable()) {
                return;
            }

            // Persistent processes (including the system) should not do
            // accelerated rendering on low-end devices.  In that case,
            // sRendererDisabled will be set.  In addition, the system process
            // itself should never do accelerated rendering.  In that case, both
            // sRendererDisabled and sSystemRendererDisabled are set.  When
            // sSystemRendererDisabled is set, PRIVATE_FLAG_FORCE_HARDWARE_ACCELERATED
           // can be used by code on the system process to escape that and enable
            // HW accelerated drawing.  (This is basically for the lock screen.)

            final boolean fakeHwAccelerated = (attrs.privateFlags &
                    WindowManager.LayoutParams.PRIVATE_FLAG_FAKE_HARDWARE_ACCELERATED) != 0;
            final boolean forceHwAccelerated = (attrs.privateFlags &
                    WindowManager.LayoutParams.PRIVATE_FLAG_FORCE_HARDWARE_ACCELERATED) != 0;

            if (fakeHwAccelerated) {
                // This is exclusively for the preview windows the window manager
                // shows for launching applications, so they will look more like
                // the app being launched.
                mAttachInfo.mHardwareAccelerationRequested = true;
            } else if (!ThreadedRenderer.sRendererDisabled
                    || (ThreadedRenderer.sSystemRendererDisabled && forceHwAccelerated)) {
...
mAttachInfo.mThreadedRenderer = ThreadedRenderer.create(mContext, translucent,
                        attrs.getTitle().toString());

兼容模式下不走hwui；
硬件需要支持hwui，通过isAvailable体现；
fakeHwAccelerated（true）代表的是“Starting Window xxx” layer；
sRendererDisabled（true）代表Persistent进程（一般系统级别的应用可以在Manifest中配置persistent属性），sSystemRendererDisabled && forceHwAccelerated代表锁屏场景，即当既不是Persistent进程，或者是system进程（其中有很多线程是需要显示UI的，但是这些UI一般都是比较简单的）但是是锁屏场景的话会走hwui。

    ThreadedRenderer(Context context, boolean translucent, String name) {
      ...
        long rootNodePtr = nCreateRootRenderNode();
        mRootNode = RenderNode.adopt(rootNodePtr);
        mRootNode.setClipToBounds(false);
        mNativeProxy = nCreateProxy(translucent, rootNodePtr);
        nSetName(mNativeProxy, name);
        ProcessInitializer.sInstance.init(context, mNativeProxy);
        loadSystemProperties();
    }

java层的ThreadedRenderer初始化主要完成native层renderthread线程的创建以及RenderProxy的创建，后者主要用于向renderthread中post消息，具体过程如下：

ThreadedRenderer.jpg

1.2 native层分析

根据上图，可以发现：此时renderthread线程已经起来了，继续往下分析，在分析之前看下androirdP上新增的WorkQueue，在RenderProxy、RenderThread和CanvasContext之间增加了WorkQueue机制，具体工作原理如下：

WorkQueue.jpg

RenderThread线程起来后首先会设置线程优先级，然后会初始化一些对象，下面看下initThreadLocals做了哪些初始化动作：

299bool RenderThread::threadLoop() {
300    setpriority(PRIO_PROCESS, 0, PRIORITY_DISPLAY);
301    if (gOnStartHook) {
302        gOnStartHook();
303    }
304    initThreadLocals();
305
306    while (true) {
307        waitForWork();
308        processQueue();
309
310        if (mPendingRegistrationFrameCallbacks.size() && !mFrameCallbackTaskPending) {
311            drainDisplayEventQueue();
312            mFrameCallbacks.insert(mPendingRegistrationFrameCallbacks.begin(),
313                                   mPendingRegistrationFrameCallbacks.end());
314            mPendingRegistrationFrameCallbacks.clear();
315            requestVsync();
316        }
317
318        if (!mFrameCallbackTaskPending && !mVsyncRequested && mFrameCallbacks.size()) {
319            // TODO: Clean this up. This is working around an issue where a combination
320            // of bad timing and slow drawing can result in dropping a stale vsync
321            // on the floor (correct!) but fails to schedule to listen for the
322            // next vsync (oops), so none of the callbacks are run.
323            requestVsync();
324        }
325    }
326
327    return false;
328}

在initThreadLocals做了一些animation的动作、初始化EglManager、RenderState、VulkanManager以及CacheManager。
下面主要分析下initThreadLocals过程，具体如下：

andorid_9.0_hwui_initThreadLocals.jpg

1）创建的DisplayEventReceiver用于请求和接收vsync，与Choreographer中提及到的java层DisplayEventReceiver应该是一个用处 ->std::make_unique<DisplayEventReceiver>；
2）创建的DisplayEventReceiver对象所关联的文件描述符被注册到了Render Thread的消息循环中 ->addFd；
优点：surfaceflinger分发vsync的时候会借助fd去唤醒renderthread线程，接着调用displayEventReceiverCallback；
3）紧接着Renderthread::drainDisplayEventQueue去处理vsync：通过DisplayEventReceiverWrapper对象获取最近一次的vsync时间，>0的话表示有效的vsync，然后将mVsyncRequested置为false，表示上次上传的vsync已经接收到了。接下来看DispatchFrameCallbacks 的task(9.0之前有很多task，如：drawFrameTask等，9.0就没有了，可以简单的理解为WorkQueue替换了TaskQueue)是否已经添加了，如果添加了mFrameCallbackTaskPending就等于true，就不会执行RenderThread::dispatchFrameCallbacks。
而dispatchFrameCallbacks主要用来干嘛的呢？答：用来显示动画的～

下面研究下dispatchFrameCallbacks函数

定义一个指向IFrameCallback的set mPendingRegistrationFrameCallbacks，
在post的时候向其中插入数据，在pushBack中也向其中插入数据，唯一的不同的会删除“前者”mFrameCallbacks中的callBack，可以把mPendingRegistrationFrameCallbacks理解为“Back Buffer”，mFrameCallbacks理解为“Front Buffer”。在remove的时候删除“前后对象”中的数据，后面会有交换过程分析。

std::set<IFrameCallback*> mPendingRegistrationFrameCallbacks;
frameworks/base/libs/hwui/renderthread/RenderThread.cpp
330void RenderThread::postFrameCallback(IFrameCallback* callback) {
331    mPendingRegistrationFrameCallbacks.insert(callback);
332}
333
334bool RenderThread::removeFrameCallback(IFrameCallback* callback) {
335    size_t erased;
336    erased = mFrameCallbacks.erase(callback);
337    erased |= mPendingRegistrationFrameCallbacks.erase(callback);
338    return erased;
339}
340
341void RenderThread::pushBackFrameCallback(IFrameCallback* callback) {
342    if (mFrameCallbacks.erase(callback)) {
343        mPendingRegistrationFrameCallbacks.insert(callback);
344    }
345}

在RenderThread等到有处理的task的时候会处理callback，将mPendingRegistrationFrameCallbacks中的数据全部copy到mFrameCallbacks中去，同时会清空mPendingRegistrationFrameCallbacks中数据。

310        if (mPendingRegistrationFrameCallbacks.size() && !mFrameCallbackTaskPending) {
311            drainDisplayEventQueue();
312            mFrameCallbacks.insert(mPendingRegistrationFrameCallbacks.begin(),
313                                   mPendingRegistrationFrameCallbacks.end());
314            mPendingRegistrationFrameCallbacks.clear();
315            requestVsync();
316        }

那么mFrameCallbacks作用是什么呢？
将mFrameCallbacks数据交换到临时变量callbacks中去，如果有数据的话就会取出来调用doFrame，那么mFrameCallbacks中保存的是什么呢？
搜索发现只有CanvasContext会继承IFrameCallback，那么回到前面看下什么时候post和pushBack的～

274void RenderThread::dispatchFrameCallbacks() {
275    ATRACE_CALL();
276    mFrameCallbackTaskPending = false;
277
278    std::set<IFrameCallback*> callbacks;
279    mFrameCallbacks.swap(callbacks);
280
281    if (callbacks.size()) {
282        // Assume one of them will probably animate again so preemptively
283        // request the next vsync in case it occurs mid-frame
284        requestVsync();
285        for (std::set<IFrameCallback*>::iterator it = callbacks.begin(); it != callbacks.end();
286             it++) {
287            (*it)->doFrame();
288        }
289    }
290}

简单说post进的数据都是借助于CanvasContext：
在prepareTree的时候会postFrameCallback，上层即Java层注册一个动画类型的Render Node到Render Thread时，一个类型为IFrameCallback的回调接口就会通过RenderThread类的成员函数postFrameCallback；
在notifyFramePending会调用pushBackFrameCallback，上层的触发处在scheduleTraversals中。

总结下:
1）displayEventReceiverCallback主要用于处理动画，将动画的每一帧同步到Vsync信号来显示；
2）renderthread此处渲染的是下一帧数据，即还未显示的；
3）在接收到本地的vsync后会做doFrame，然后请求下一个vsync。

继续往下走：看下RenderProxy还干了什么？
创建完renderthread后，开始创建CanvasContext，即窗口的画布，后期会分析怎么关联到窗口上，其中主要是确定pipline方式～
有没有思考过new RenderThread后才会new CanvasContext，那么之前分析会用到CanvasContext，怎么回事呢？

2.绑定窗口到RenderThread

一旦Render Thread知道了当前要渲染的窗口，它就将可以将该窗口绑定到Open GL渲染上下文中去，从而使得后面的渲染操作都是针对被绑定的窗口的。

2.1 java层分析

上面分析基于ViewRootImpl的setView的基础上，现在开始到了真正的绘制阶段了，即ViewRootImpl的performTraversals函数中，执行measure、layout、draw动作。在绘制之前要获取一个surface，获取成功后再绑定到对应的renderThread线程中去。

frameworks/base/core/java/android/view/ViewRootImpl.java
392    public final Surface mSurface = new Surface();
...
1676    private void performTraversals() {
...
2083                if (!hadSurface) {
2084                    if (mSurface.isValid()) {
...
2092                        newSurface = true;
2093                        mFullRedrawNeeded = true;
2094                        mPreviousTransparentRegion.setEmpty();
2095
2096                        // Only initialize up-front if transparent regions are not
2097                        // requested, otherwise defer to see if the entire window
2098                        // will be transparent
2099                        if (mAttachInfo.mThreadedRenderer != null) {
2100                            try {
2101                                hwInitialized = mAttachInfo.mThreadedRenderer.initialize(
2102                                        mSurface);
...
2301            performMeasure(childWidthMeasureSpec, childHeightMeasureSpec);
...
2320            performLayout(lp, mWidth, mHeight);
...
2477        if (!cancelDraw && !newSurface) {
...
2485            performDraw();

如果这个Surface是新创建的，那么会将该surface(mSurface)通过initialize将它绑定到Render Thread中去，绑定完成后才会做measure、layout、draw的动作。

2.2 native层分析

下面主要从C++层分析下绑定窗口的过程：

frameworks/base/core/jni/android_view_ThreadedRenderer.cpp
689static void android_view_ThreadedRenderer_initialize(JNIEnv* env, jobject clazz,
690        jlong proxyPtr, jobject jsurface) {
691    RenderProxy* proxy = reinterpret_cast<RenderProxy*>(proxyPtr);
692    sp<Surface> surface = android_view_Surface_getSurface(env, jsurface);
693    proxy->initialize(surface);
694}

frameworks/base/libs/hwui/renderthread/RenderProxy.cpp
85void RenderProxy::initialize(const sp<Surface>& surface) {
86    mRenderThread.queue().post(
87            [ this, surf = surface ]() mutable { mContext->setSurface(std::move(surf)); });
88}

直接将上层的surface通过setSurface函数经过workQueue传到CanvasContext中去，成功的话返回true，即拥有了一个新的surface。

frameworks/base/libs/hwui/renderthread/CanvasContext.cpp
186void CanvasContext::setSurface(sp<Surface>&& surface) {
187    ATRACE_CALL();
188
189    mNativeSurface = std::move(surface);
190
191    ColorMode colorMode = mWideColorGamut ? ColorMode::WideColorGamut : ColorMode::Srgb;
192    bool hasSurface = mRenderPipeline->setSurface(mNativeSurface.get(), mSwapBehavior, colorMode);
193
194    mFrameNumber = -1;
195
196    if (hasSurface) {
197        mHaveNewSurface = true;
198        mSwapHistory.clear();

CanvasContext::setSurface有systrace 标签，通过systrace可以看出具体过程来，如下：

setSurface.png

再往下到具体的Pipeline去setSurface，这里以默认的Pipeline介绍：
首先判断EglSurface是否已经存在，如果存在先destroy，然后再去create一个EglSurface，该mEglSurface表示的是一个绘图表面，有了这个mEglSurface之后，当执行Open GL命令的时候，就可以知道这些命令是作用在哪个窗口上的了。

frameworks/base/libs/hwui/renderthread/OpenGLPipeline.cpp
146bool OpenGLPipeline::setSurface(Surface* surface, SwapBehavior swapBehavior, ColorMode colorMode) {
147    if (mEglSurface != EGL_NO_SURFACE) {
148        mEglManager.destroySurface(mEglSurface);
149        mEglSurface = EGL_NO_SURFACE;
150    }
151
152    if (surface) {
153        const bool wideColorGamut = colorMode == ColorMode::WideColorGamut;
154        mEglSurface = mEglManager.createSurface(surface, wideColorGamut);
155    }
156
157    if (mEglSurface != EGL_NO_SURFACE) {
158        const bool preserveBuffer = (swapBehavior != SwapBehavior::kSwap_discardBuffer);
159        mBufferPreserved = mEglManager.setPreserveBuffer(mEglSurface, preserveBuffer);
160        return true;
161    }
162
163    return false;
164}

在EglManager中createSurface中首先会调用EglManager::initialize（关键哦～），其中完成EGL的初始化动作（eglGetDisplay、eglInitialize、eglChooseConfig、eglCreateContext、eglCreatePbufferSurface、eglMakeCurrent、eglSwapInterval），在makeCurrent的时候首先看是不是surface没有改变，如果是的话就不需要重新eglMakeCurrent了，此处makeCurrent的是当前的mPBufferSurface。然后才是真正的创建当前surface的eglCreateWindowSurface；最后调用eglSurfaceAttrib。
在EglManager中setPreserveBuffer如果不是SwapBehavior::Preserved就直接返回false了，目前很多平台应该都是SwapBehavior::BufferAge。
那么问题来了，前面创建的surface什么时候设置到上下文环境中的呢？
查看代码发现：
DrawFrameTask::syncFrameState -> CanvasContext::makeCurrent ->OpenGLPipeline::makeCurrent
bool haveNewSurface = mEglManager.makeCurrent(mEglSurface, &error); //mEglSurface就是前在setSurface中绑定的surface
这样的话就绑定到了前面创建的surface了。

继续看下EglManager::makeCurrent发现：
1）在OpenGLPipeline::onStop、EglManager::destroySurface的时候设定上下文surface为EGL_NO_SURFACE
2）在EglManager::initialize时候会绑定PBSurface（前面已经介绍了）
3）在EglManager::beginFrame时候会绑定传入的surface，此处值得深入跟踪下，
mEglManager.beginFrame(mEglSurface) //mEglSurface就是前在setSurface中绑定的surface
CanvasContext::draw ->OpenGLPipeline::getFrame ->EglManager::beginFrame->makeCurrent(surface)
因此，将当前surface设置到OpenGL渲染上下文中总共做了2次,在syncFrameState做1次，在CanvasContext::draw时做1次。在makeCurrent的实现的地方会判断if (isCurrent(surface)) return false; 也就是说后面1次的makeCurrent可能直接返回false。
总结下：初始化过程中会通过eglCreateWindowSurface创建上层surface对应的底层surface（EglSurface），makeCurrent当前的PBSurface，然后在syncFrameSate时会makeCurrent之前创建的底层surface（EglSurface），最后在draw的时候同样也会makeCurrent一次，但是直接返回了，没有执行真正的eglMakeCurrent。

到此，RenderThread线程已经run了，OpenGL、EGL环境也已经准备就绪，上层Surface也已经创建完成并成功绑定到HWUI层pipeline中mEglSurface。

二.资源地图集服务

以下内容描述都是7.0平台上的，androidO上开始已经没有Asset Atlas Service。不清楚androidO上叫什么～有知道的同学回复呀～
Android启动的时候会预加载一些资源，方便应用的后期快速访问，同时达到共享目的。hwui中做了进一步优化，将预加载资源合成为一个纹理上传到GPU去，并且能够在所有的应用程序之间进行共享。
资源预加载发生在Zygote进程的，然后Zygote进程fork了应用程序进程，这样就保证了资源的共享，但是在hwui中，如果每个应用都去使用预加载的资源的话，那么每个应用都要将资源作为纹理传入到GPU中，这样太浪费GPU内存了，这块是不是有问题需要优化呢，这就是本节的重点。
Zygote进程将预加载的资源作为texture传到system进程中去，System进程中运行了一个Asset Atlas Service，该service就是上面提到的将预加载资源合成为一个纹理上传到GPU去。这样的话，app的renderthread线程直接请求Asset Atlas Service得到纹理即可，无需单独上传纹理到GPU。

1.Zygote

加载资源
启动system_server
接收AMS的请求创建应用进程

下面主要分析加载资源环节,preloadClasses、preloadResources、nativePreloadAppProcessHALs、 preloadOpenGL、preloadSharedLibraries、preloadTextResources，主要分析preloadResources。
而在preloadResources中会分别调用preloadDrawables（对应R.array.preloaded_drawables）和preloadColorStateLists（对应R.array.preloaded_color_state_lists），主要跟踪preloadDrawables。
mResources.getDrawable加载所有drawable～
Zygote预加载的Drawable将会被运行在System进程里面的Asset Atlas Service合成一个地图集，最后作为纹理上传到GPU去，因此，接下来我们就继续分析Asset Atlas Service的实现。

2. system_server

Zygote进程启动System进程，System进程会加载系统服务，其中就包括Asset Atlas Service。
在system进程启动服务之前会进行一些属性的设置，在startOtherServices函数中会启动Asset Atlas Service，Asset Atlas Service是一个非系统服务，工厂模式下不会启用。
注：7.0、7.1还存在AssetAtlasService，到8.0开始就已经取消了～
AssetAtlasService计算将所有预加载的Drawable资源合成在一张图片中所需要的最小宽度和高度值，有了这两个值之后创建一块Graphic Buffer。然后将Drawable渲染到该Buffer中去。最后上传到GPU中去。

三. android4.4 hwui

这边插一章节介绍下旧版本hwui的实现过程，相对比较简单，主要过程如下：

android4.4_setView.png

3.1 hardware draw

下面主要分析下hardware draw过程，引用一个大神的图，关键的几个步骤做简单分析：

1）beginFrame主要完成EGLDisplay(用于显示) 和一个EGLSurface（OpenGL将在这个Surface上进行绘图），然后eglBeginFrame主要是校验参数的合法性；
2）buildDisplayList主要完成录制过程，构建native层的DisplayList；
3）prepareFrame完成dirtyRegion的构建；

4）onPostDraw会进行OpenGLRenderer的finish过程；
5）swapBuffer完成对buffer向SurfaceFlinger的递交，注意在java层调用的哦～

android4.4_draw.png

四.DisplayList的构建过程

在上面初始化过程中会调用performTraversals，然后调用initialize完成初始化过程，之后在performTraversals中会调用performMeasure、performLayout、performDraw，本节主要介绍performDraw过程：

frameworks/base/core/java/android/view/ViewRootImpl.java
private boolean draw(boolean fullRedrawNeeded) {
    ...
    if (!dirty.isEmpty() || mIsAnimating || accessibilityFocusDirty) {
            if (mAttachInfo.mThreadedRenderer != null && mAttachInfo.mThreadedRenderer.isEnabled()) {
        ...
         mAttachInfo.mThreadedRenderer.draw(mView, mAttachInfo, this, callback);
    } else {
        ...
        if (!drawSoftware(surface, mAttachInfo, xOffset, yOffset,
                        scalingRequired, dirty, surfaceInsets)) {
                    return false;
        }
    ...
}

frameworks/base/core/java/android/view/ThreadedRenderer.java
    void draw(View view, AttachInfo attachInfo, HardwareDrawCallbacks callbacks) {
        ......
 
        updateRootDisplayList(view, callbacks);
        ......
 
        if (attachInfo.mPendingAnimatingRenderNodes != null) {
            final int count = attachInfo.mPendingAnimatingRenderNodes.size();
            for (int i = 0; i < count; i++) {
                registerAnimatingRenderNode(
                        attachInfo.mPendingAnimatingRenderNodes.get(i));
            }
            attachInfo.mPendingAnimatingRenderNodes.clear();
            // We don't need this anymore as subsequent calls to
            // ViewRootImpl#attachRenderNodeAnimator will go directly to us.
            attachInfo.mPendingAnimatingRenderNodes = null;
        }
        ...
        int syncResult = nSyncAndDrawFrame(mNativeProxy, frameInfo, frameInfo.length);
        if ((syncResult & SYNC_LOST_SURFACE_REWARD_IF_FOUND) != 0) {
            ...
            attachInfo.mViewRootImpl.invalidate();
        }
    }
    ...
}

下面主要分析updateRootDisplayList：
可以看到在该函数中存在“Record View#draw()”的trace tag，首先完成updateViewTreeDisplayList，稍后分析；然后判断rootNode是否有更新或者rootNode是否无效。
那么什么时候isValid为true呢？mRootNode.end结束时候该值会被置为true，详细分析下面会有介绍。

frameworks/base/core/java/android/view/ThreadedRenderer.java 
private void updateRootDisplayList(View view, DrawCallbacks callbacks) {
        Trace.traceBegin(Trace.TRACE_TAG_VIEW, "Record View#draw()");
        updateViewTreeDisplayList(view);

        if (mRootNodeNeedsUpdate || !mRootNode.isValid()) {
            DisplayListCanvas canvas = mRootNode.start(mSurfaceWidth, mSurfaceHeight);
            try {
                final int saveCount = canvas.save();
                canvas.translate(mInsetLeft, mInsetTop);
                callbacks.onPreDraw(canvas);

                canvas.insertReorderBarrier();
                canvas.drawRenderNode(view.updateDisplayListIfDirty());
                canvas.insertInorderBarrier();

                callbacks.onPostDraw(canvas);
                canvas.restoreToCount(saveCount);
                mRootNodeNeedsUpdate = false;
            } finally {
                mRootNode.end(canvas);
            }
        }
        Trace.traceEnd(Trace.TRACE_TAG_VIEW);
    }

mRootNode.isValid值来自于native层，当第一次设置的时候RenderNode中displayList不为空，因此isValid值为true；当renderNode被destroy时该值为false。

frameworks/base/core/java/android/view/RenderNode.java
public void end(DisplayListCanvas canvas) {
    long displayList = canvas.finishRecording();
    nSetDisplayList(mNativeRenderNode, displayList);
    canvas.recycle();
}
frameworks/base/core/jni/android_view_RenderNode.cpp
static void android_view_RenderNode_setDisplayList(JNIEnv* env,
        jobject clazz, jlong renderNodePtr, jlong displayListPtr) {
    RenderNode* renderNode = reinterpret_cast<RenderNode*>(renderNodePtr);
    DisplayList* newData = reinterpret_cast<DisplayList*>(displayListPtr);
    renderNode->setStagingDisplayList(newData);
}

frameworks/base/libs/hwui/RenderNode.cpp
void RenderNode::destroyHardwareResources(TreeInfo* info) {
    ...
    setStagingDisplayList(nullptr);

frameworks/base/libs/hwui/RenderNode.cpp
void RenderNode::setStagingDisplayList(DisplayList* displayList) {
    mValid = (displayList != nullptr);
    mNeedsDisplayListSync = true;
    delete mStagingDisplayList;
    mStagingDisplayList = displayList;
}

好了，下面开始分析DisplayList的构建过程中重要的几个环节：

1.updateViewTreeDisplayList

view.updateDisplayListIfDirty();

2.start

start时候会在native层创建RecordingCanvas、DisplayList，而9.0默认使用skiagl，那么native层就对应SkiaRecordingCanvas、SkiaDisplayList。

frameworks/base/core/java/android/view/RenderNode.java
public DisplayListCanvas start(int width, int height) {
    return DisplayListCanvas.obtain(this, width, height);
}

接下来是对应的canvas做save操作，在RecordingCanvas逻辑中会创建对应的Snapshot，在skiagl中走的是SkiaCanvas::save，最终会设置到SkCanvas中。
后面的translate函数和上面的save效果类似。

3.drawRenderNode

在drawRenderNode的前后分别insertBarrier目的是创建一个新的chunk，然后向displayList中加入RenderNodeOp，而skiagl中会走SkiaRecordingCanvas::drawRenderNode。

4.end

得到native层的DisplayList对象地址displayList，最后将displayList设置到mStagingDisplayList。

frameworks/base/core/java/android/view/RenderNode.java
public void end(DisplayListCanvas canvas) {
        long displayList = canvas.finishRecording();
        nSetDisplayList(mNativeRenderNode, displayList);
        canvas.recycle();
    }

5.addOp

将addOp单独拉出来讲，因为上面drawRenderNode以及在updateViewTreeDisplayList中的drawBitmap都会往DisplayList中写入对应的Op。
当上层触发drawColor、drawRect等操作时会调用native层的addOp（具体可参见之前博客：HWUI绘制系列——从java到C++），传入对应的op参数进来，下面就详细分析下addOp函数，因为该函数为后面渲染做铺垫。
可以将op理解为一个人，首先判断这个人的rect是否为空，然后得到当前的DisplayList中的ops的last值的索引，将当前的op加到ops中去。
判断mDeferredBarrierType，研究下该值，初始化的时候该值为DeferredBarrierType::None，当resetRecording时会重置为DeferredBarrierType::InOrder，还有就是在insertReorderBarrier时会进行重新赋值。那么有什么意义呢？

1）刚初始化renderthread的时候会new RecordingCanvas，这样会resetRecording，表示要重新建一个Chunk了，还有就是进程再次obtain也会强制做resetRecording；
2）在updateViewTreeDisplayList结束后，在drawRenderNode前后会分别调用insertReorderBarrier（true）和insertInorderBarrier（false）进行mDeferredBarrierType重置，也就是说drawRenderNode时会新建一个chunk。
3）大小关系：ops > chunk > children > op

frameworks/base/libs/hwui/RecordingCanvas.cpp
int RecordingCanvas::addOp(RecordedOp* op) {
    // skip op with empty clip
    if (op->localClip && op->localClip->rect.isEmpty()) {
        // NOTE: this rejection happens after op construction/content ref-ing, so content ref'd
        // and held by renderthread isn't affected by clip rejection.
        // Could rewind alloc here if desired, but callers would have to not touch op afterwards.
        return -1;
    }

    int insertIndex = mDisplayList->ops.size();
    mDisplayList->ops.push_back(op);
    if (mDeferredBarrierType != DeferredBarrierType::None) {
        // op is first in new chunk
        mDisplayList->chunks.emplace_back();
        DisplayList::Chunk& newChunk = mDisplayList->chunks.back();
        newChunk.beginOpIndex = insertIndex;
        newChunk.endOpIndex = insertIndex + 1;
        newChunk.reorderChildren = (mDeferredBarrierType == DeferredBarrierType::OutOfOrder);
        newChunk.reorderClip = mDeferredBarrierClip;

        int nextChildIndex = mDisplayList->children.size();
        newChunk.beginChildIndex = newChunk.endChildIndex = nextChildIndex;
        mDeferredBarrierType = DeferredBarrierType::None;
    } else {
        // standard case - append to existing chunk
        mDisplayList->chunks.back().endOpIndex = insertIndex + 1;
    }
    return insertIndex;
}

void RecordingCanvas::insertReorderBarrier(bool enableReorder) {
    if (enableReorder) {
        mDeferredBarrierType = DeferredBarrierType::OutOfOrder;
        mDeferredBarrierClip = getRecordedClip();
    } else {
        mDeferredBarrierType = DeferredBarrierType::InOrder;
        mDeferredBarrierClip = nullptr;
    }
}

frameworks/base/core/java/android/view/DisplayListCanvas.java
    static DisplayListCanvas obtain(@NonNull RenderNode node, int width, int height) {
        if (node == null) throw new IllegalArgumentException("node cannot be null");
        DisplayListCanvas canvas = sPool.acquire();
        if (canvas == null) {
            canvas = new DisplayListCanvas(node, width, height);
        } else {
            nResetDisplayListCanvas(canvas.mNativeCanvasWrapper, node.mNativeRenderNode,
                    width, height);
        }

frameworks/base/core/java/android/view/ThreadedRenderer.java
    private void updateRootDisplayList(View view, DrawCallbacks callbacks) {
        Trace.traceBegin(Trace.TRACE_TAG_VIEW, "Record View#draw()");
        updateViewTreeDisplayList(view);

        if (mRootNodeNeedsUpdate || !mRootNode.isValid()) {
            DisplayListCanvas canvas = mRootNode.start(mSurfaceWidth, mSurfaceHeight);
            try {
                final int saveCount = canvas.save();
                canvas.translate(mInsetLeft, mInsetTop);
                callbacks.onPreDraw(canvas);

                canvas.insertReorderBarrier();
                canvas.drawRenderNode(view.updateDisplayListIfDirty());
                canvas.insertInorderBarrier();

                callbacks.onPostDraw(canvas);
                canvas.restoreToCount(saveCount);
                mRootNodeNeedsUpdate = false;
            } finally {
                mRootNode.end(canvas);
            }
        }
        Trace.traceEnd(Trace.TRACE_TAG_VIEW);
    }

五.绘制过程介绍

1.syncFrameState

应用主线程向RT线程的workQueue中post消息并等待消息处理完毕唤醒UI线程，等到RT 执行该消息时会回调run方法，虽然回调了但是不会立刻唤醒UI线程。
DrawFrameTask中只有两处持锁，下面会分析什么时候会调用unblockUiThread，当unblockUiThread时会唤醒UI线程继续往下执行～
先剧透下：当syncFrameState完成后会唤醒UI线程，还有就是本次draw完后会唤醒（但是这种情况是不理想的，不应该存在的）～

frameworks/base/libs/hwui/renderthread/DrawFrameTask.cpp
void DrawFrameTask::postAndWait() {
    AutoMutex _lock(mLock);
    mRenderThread->queue().post([this]() { run(); });
    mSignal.wait(mLock);
}

void DrawFrameTask::unblockUiThread() {
    AutoMutex _lock(mLock);
    mSignal.signal();
}

下面看下syncFrameState过程：
1）首先同步当前vsync到TimeLord中的mFrameTimeNanos，即更新上一次的vsync时间；
2）然后是makeCurrent，当VRI发出setStop的时候会停止makeCurrent，也就停止渲染，否则一直往EglManager去makeCurrent，此处会判断当前的surface是不是已经makeCurrent过了，如果已经makeCurrent了，那么就不去调用eglMakeCurrent，或者没有surface的话会makeCurrent pbSurface。那么看下makeCurrent的位置有哪些？
a.EglManager::initialize时候makeCurrent(mPBufferSurface)；
b.EglManager::beginFrame时候makeCurrent(surface)，而beginFrame是在CanvasContext::draw时候调用的；
c.EglManager::destroySurface时候makeCurrent(EGL_NO_SURFACE)。
因此，正常情况下在初始化RT时候会OpenGLPipeline::setSurface将当前待渲染的surface设置进来，然后在syncFrameState时候将surface设置为opengl上下文中。
3）unpinImages主要是为了提高hwui精度的，对每个object做cache，然后让cache去unpin：caches.textureCache.resetMarkInUse(this);前面一步应该也有这个过程；

frameworks/base/libs/hwui/renderthread/DrawFrameTask.cpp
std::vector<sp<DeferredLayerUpdater> > mLayers;
Rect mContentDrawBounds;
bool DrawFrameTask::syncFrameState(TreeInfo& info) {
    ATRACE_CALL();
    int64_t vsync = mFrameInfo[static_cast<int>(FrameInfoIndex::Vsync)];
    mRenderThread->timeLord().vsyncReceived(vsync);
    bool canDraw = mContext->makeCurrent();
    mContext->unpinImages();

    for (size_t i = 0; i < mLayers.size(); i++) {
        mLayers[i]->apply();
    }
    mLayers.clear();
    mContext->setContentDrawBounds(mContentDrawBounds);
    mContext->prepareTree(info, mFrameInfo, mSyncQueued, mTargetNode);
    ...
    if (info.out.hasAnimations) {
        if (info.out.requiresUiRedraw) {
            mSyncResult |= SyncResult::UIRedrawRequired;
        }
    }
    if (!info.out.canDrawThisFrame) {
        mSyncResult |= SyncResult::FrameDropped;
    }
    // If prepareTextures is false, we ran out of texture cache space
    return info.prepareTextures;                                       //构造TreeInfo的时候赋值的：true
}

4）处理layer，在TextureLayer中会将layer借助于ThreadedRenderer传到native层的DrawFrameTask中并用mLayers保存起来。
是不是很想知道DeferredLayerUpdater类中是否有保存layer的name呢？很可惜没有，只有getWidth()和getHeight() 。

frameworks/base/libs/hwui/renderthread/DrawFrameTask.cpp
void DrawFrameTask::pushLayerUpdate(DeferredLayerUpdater* layer) {
    LOG_ALWAYS_FATAL_IF(!mContext,
                        "Lifecycle violation, there's no context to pushLayerUpdate with!");

    for (size_t i = 0; i < mLayers.size(); i++) {
        if (mLayers[i].get() == layer) {
            return;
        }
    }
    mLayers.push_back(layer);
}

void DrawFrameTask::removeLayerUpdate(DeferredLayerUpdater* layer) {
    for (size_t i = 0; i < mLayers.size(); i++) {
        if (mLayers[i].get() == layer) {
            mLayers.erase(mLayers.begin() + i);
            return;
        }
    }
}

frameworks/base/core/java/android/view/ThreadedRenderer.java
void pushLayerUpdate(TextureLayer layer) {
    nPushLayerUpdate(mNativeProxy, layer.getDeferredLayerUpdater());
}

5）然后再看下layer的apply过程：
a.首先mCreateLayerFn创建一个layer，该为函数指针，在pipeline中传入的，那么就会调用OpenGLPipeline的createLayer方法，根据传入的变量去生成纹理（glActiveTexture、glGenTextures）。
b.在setRenderTarget的时候glBindTexture(target, texture)同时glTexParameteri
到此，layer纹理也绑定结束了～

frameworks/base/libs/hwui/DeferredLayerUpdater.cpp
Layer* mLayer;
CreateLayerFn mCreateLayerFn;
void DeferredLayerUpdater::apply() {
    if (!mLayer) {
        mLayer = mCreateLayerFn(mRenderState, mWidth, mHeight, mColorFilter, mAlpha, mMode, mBlend);
    }

    mLayer->setColorFilter(mColorFilter);
    mLayer->setAlpha(mAlpha, mMode);

    if (mSurfaceTexture.get()) {
        if (mLayer->getApi() == Layer::Api::Vulkan) {
            if (mUpdateTexImage) {
                mUpdateTexImage = false;
                doUpdateVkTexImage();
            }
        } else {
            LOG_ALWAYS_FATAL_IF(mLayer->getApi() != Layer::Api::OpenGL,
                                "apply surfaceTexture with non GL backend %x, GL %x, VK %x",
                                mLayer->getApi(), Layer::Api::OpenGL, Layer::Api::Vulkan);
            if (!mGLContextAttached) {
                mGLContextAttached = true;
                mUpdateTexImage = true;
                mSurfaceTexture->attachToContext(static_cast<GlLayer*>(mLayer)->getTextureId());
            }
            if (mUpdateTexImage) {
                mUpdateTexImage = false;
                doUpdateTexImage();
            }
            GLenum renderTarget = mSurfaceTexture->getCurrentTextureTarget();
            static_cast<GlLayer*>(mLayer)->setRenderTarget(renderTarget);
        }
        if (mTransform) {
            mLayer->getTransform().load(*mTransform);
            setTransform(nullptr);
        }
    }
}

frameworks/base/libs/hwui/renderthread/OpenGLPipeline.cpp
DeferredLayerUpdater* OpenGLPipeline::createTextureLayer() {
    mEglManager.initialize();
    return new DeferredLayerUpdater(mRenderThread.renderState(), createLayer, Layer::Api::OpenGL);
}
static Layer* createLayer(RenderState& renderState, uint32_t layerWidth, uint32_t layerHeight,
                          sk_sp<SkColorFilter> colorFilter, int alpha, SkBlendMode mode,
                          bool blend) {
    GlLayer* layer =
            new GlLayer(renderState, layerWidth, layerHeight, colorFilter, alpha, mode, blend);
    Caches::getInstance().textureState().activateTexture(0);
    layer->generateTexture();
    return layer;
}

frameworks/base/libs/hwui/renderstate/TextureState.cpp
void TextureState::activateTexture(GLuint textureUnit) {
    LOG_ALWAYS_FATAL_IF(textureUnit >= kTextureUnitsCount,
                        "Tried to use texture unit index %d, only %d exist", textureUnit,
                        kTextureUnitsCount);
    if (mTextureUnit != textureUnit) {
        glActiveTexture(kTextureUnits[textureUnit]);
        mTextureUnit = textureUnit;
    }
}

frameworks/base/libs/hwui/GlLayer.cpp
void GlLayer::generateTexture() {
    if (!texture.mId) {
        glGenTextures(1, &texture.mId);
    }
}

6）再往下setContentDrawBounds设置绘制区域大小，初始化时mContentDrawBounds(0, 0, 0, 0)，在VRI中updateContentDrawBounds时会设置bounds。

frameworks/base/libs/hwui/renderthread/CanvasContext.h
void setContentDrawBounds(const Rect& bounds) { mContentDrawBounds = bounds; }

frameworks/base/libs/hwui/renderthread/DrawFrameTask.h
void setContentDrawBounds(int left, int top, int right, int bottom) {
        mContentDrawBounds.set(left, top, right, bottom);
}

frameworks/base/core/java/android/view/ThreadedRenderer.java
public void setContentDrawBounds(int left, int top, int right, int bottom) {
    nSetContentDrawBounds(mNativeProxy, left, top, right, bottom);
}

7）最后一步就是prepareTree过程了,这一步主要完成每个renderNode的prepareTree过程：

frameworks/base/libs/hwui/renderthread/CanvasContext.cpp
void CanvasContext::prepareTree(TreeInfo& info, int64_t* uiFrameInfo, int64_t syncQueued,
                                RenderNode* target) {
    mRenderThread.removeFrameCallback(this);
    for (const sp<RenderNode>& node : mRenderNodes) {
        // Only the primary target node will be drawn full - all other nodes would get drawn in
        // real time mode. In case of a window, the primary node is the window content and the other
        // node(s) are non client / filler nodes.
        info.mode = (node.get() == target ? TreeInfo::MODE_FULL : TreeInfo::MODE_RT_ONLY);
        node->prepareTree(info);
        GL_CHECKPOINT(MODERATE);
    }
    ...
    freePrefetchedLayers();
    ...
    } else {
        info.out.canDrawThisFrame = true;
    }
    ...
}

a.prepareTree过程，首先看下mRenderNodes是如何构建起来的，在CanvasContext初始化时将rootRenderNode加进来，之后通过addRenderNode加入，通过removeRenderNode移除。

//加入rootRenderNode
frameworks/base/libs/hwui/renderthread/CanvasContext.cpp
CanvasContext::CanvasContext(...RenderNode* rootRenderNode,...){
    ...
    mRenderNodes.emplace_back(rootRenderNode);
    ...
}
frameworks/base/core/jni/android_view_ThreadedRenderer.cpp
static jlong android_view_ThreadedRenderer_createRootRenderNode(JNIEnv* env, jobject clazz) {
    RootRenderNode* node = new RootRenderNode(env);
    node->incStrong(0);
    node->setName("RootRenderNode");
    return reinterpret_cast<jlong>(node);
}

void CanvasContext::addRenderNode(RenderNode* node, bool placeFront) {
    int pos = placeFront ? 0 : static_cast<int>(mRenderNodes.size());
    node->makeRoot();
    mRenderNodes.emplace(mRenderNodes.begin() + pos, node);
}
void CanvasContext::removeRenderNode(RenderNode* node) {
    node->clearRoot();
    mRenderNodes.erase(std::remove(mRenderNodes.begin(), mRenderNodes.end(), node),
                       mRenderNodes.end());
}

在pushStagingDisplayListChanges中调用syncDisplayList，在该函数中取走mStagingDisplayList（mDisplayList = mStagingDisplayList;），该displaylist在setDisplayList中赋值的。
这样的话displayList就取到了。

frameworks/base/libs/hwui/RenderNode.cpp
void RenderNode::prepareTreeImpl(TreeObserver& observer, TreeInfo& info, bool functorsNeedLayer) {
    ...
    if (info.mode == TreeInfo::MODE_FULL) {
        pushStagingPropertiesChanges(info);
    }
    ...
    if (info.mode == TreeInfo::MODE_FULL) {
        pushStagingDisplayListChanges(observer, info);
    }
    ...
}

b.清空mPrefetchedLayers中保存的RenderNode，那么什么时候insert呢？答案：CanvasContext::buildLayer(RenderNode* node)，java层触发。

frameworks/base/libs/hwui/renderthread/CanvasContext.cpp
std::set<RenderNode*> mPrefetchedLayers;
void CanvasContext::freePrefetchedLayers() {
    if (mPrefetchedLayers.size()) {
        for (auto& node : mPrefetchedLayers) {
            ALOGW("Incorrectly called buildLayer on View: %s, destroying layer...",
                  node->getName());
            node->destroyLayers();
            node->decStrong(nullptr);
        }
        mPrefetchedLayers.clear();
    }
}

2.现在看下是draw之前的deferLayers，

frameworks/base/libs/hwui/FrameBuilder.cpp
void FrameBuilder::deferLayers(const LayerUpdateQueue& layers) {
    // Render all layers to be updated, in order. Defer in reverse order, so that they'll be
    // updated in the order they're passed in (mLayerBuilders are issued to Renderer in reverse)
    for (int i = layers.entries().size() - 1; i >= 0; i--) {
        RenderNode* layerNode = layers.entries()[i].renderNode.get();
        // only schedule repaint if node still on layer - possible it may have been
        // removed during a dropped frame, but layers may still remain scheduled so
        // as not to lose info on what portion is damaged
        OffscreenBuffer* layer = layerNode->getLayer();
        if (CC_LIKELY(layer)) {
            ATRACE_FORMAT("Optimize HW Layer DisplayList %s %ux%u", layerNode->getName(),
                          layerNode->getWidth(), layerNode->getHeight());

            Rect layerDamage = layers.entries()[i].damage;
            // TODO: ensure layer damage can't be larger than layer
            layerDamage.doIntersect(0, 0, layer->viewportWidth, layer->viewportHeight);
            layerNode->computeOrdering();

            // map current light center into RenderNode's coordinate space
            Vector3 lightCenter = mCanvasState.currentSnapshot()->getRelativeLightCenter();
            layer->inverseTransformInWindow.mapPoint3d(lightCenter);

            saveForLayer(layerNode->getWidth(), layerNode->getHeight(), 0, 0, layerDamage,
                         lightCenter, nullptr, layerNode);

            if (layerNode->getDisplayList()) {
                deferNodeOps(*layerNode);
            }
            restoreForLayer();
        }
    }
}

1）是不是很想知道传入的参数const LayerUpdateQueue& layers从哪里得到的？很诱人吧，来来来，我们来揭晓下：
先看下LayerUpdateQueue类，发现它有成员变量mEntries，表示保存所有的layer信息（RenderNode
和damage）。
现在来分析下参数的由来，当RenderNode::pushLayerUpdate（最开始是prepareTree）时会传入RenderNode对象和待更新区域dirty，这样赋值给LayerUpdateQueue中的mEntries。

//LayerUpdateQueue类定义处：
class LayerUpdateQueue {
public:
    struct Entry {
        Entry(RenderNode* renderNode, const Rect& damage)
                : renderNode(renderNode), damage(damage) {}
        sp<RenderNode> renderNode;
        Rect damage;
    };
    LayerUpdateQueue() {}
    void enqueueLayerWithDamage(RenderNode* renderNode, Rect dirty);
    void clear();
    const std::vector<Entry>& entries() const { return mEntries; }
private:
    std::vector<Entry> mEntries;
};
//参数的由来
void RenderNode::pushLayerUpdate(TreeInfo& info) {
    ...
    info.layerUpdateQueue->enqueueLayerWithDamage(this, dirty);  //有很多RenderNode，但是只有一个info.layerUpdateQueue
    ...
}
frameworks/base/libs/hwui/LayerUpdateQueue.cpp
void LayerUpdateQueue::enqueueLayerWithDamage(RenderNode* renderNode, Rect damage) {
    ...
    if (!damage.isEmpty()) {
        for (Entry& entry : mEntries) {
            if (CC_UNLIKELY(entry.renderNode == renderNode)) {
                entry.damage.unionWith(damage);
                return;
            }
        }
        mEntries.emplace_back(renderNode, damage);
    }
}

2）再看下renderNode中OffscreenBuffer对象的由来，同样在RenderNode::pushLayerUpdate（最开始是prepareTree）且在1）之前会构建OffscreenBuffer对象，在OpenGLPipeline中创建并通过setLayer设置到renderNode中去，然后deferLayers就可以通过getLayer得到得到RenderNode的成员变量mLayer，下面列举了RenderNode的部分成员变量：

class RenderNode : public VirtualLightRefBase {
    String8 mName;
    DisplayList* mDisplayList;
    DisplayList* mStagingDisplayList;
    OffscreenBuffer* mLayer = nullptr;
    RenderProperties mProperties;
    RenderProperties mStagingProperties;
}
frameworks/base/libs/hwui/RenderNode.cpp
void RenderNode::pushLayerUpdate(TreeInfo& info) {
    LayerType layerType = properties().effectiveLayerType();   //softlayer不做处理，直接返回
    ...
    if (info.canvasContext.createOrUpdateLayer(this, *info.damageAccumulator, info.errorHandler)) {
        damageSelf(info);
    }

    if (!hasLayer()) {
        return;
    }

    SkRect dirty;
    info.damageAccumulator->peekAtDirty(&dirty);
    info.layerUpdateQueue->enqueueLayerWithDamage(this, dirty);     //构建LayerUpdateQueue中的成员变量mEntries

    // There might be prefetched layers that need to be accounted for.
    // That might be us, so tell CanvasContext that this layer is in the
    // tree and should not be destroyed.
    info.canvasContext.markLayerInUse(this);
}
frameworks/base/libs/hwui/renderthread/OpenGLPipeline.cpp
bool OpenGLPipeline::createOrUpdateLayer(RenderNode* node,
                                         const DamageAccumulator& damageAccumulator,
                                         bool wideColorGamut,
                                         ErrorHandler* errorHandler) {
    RenderState& renderState = mRenderThread.renderState();
    OffscreenBufferPool& layerPool = renderState.layerPool();
    bool transformUpdateNeeded = false;
    if (node->getLayer() == nullptr) {
        node->setLayer(
                layerPool.get(renderState, node->getWidth(), node->getHeight(), wideColorGamut));
        transformUpdateNeeded = true;
    }

3）继续往下，对layerDamage做交集处理，保证damage的区域在layer范围内。然后计算每个renderNode的ordering。再继续map 当前的light center到renderNode坐标空间中。
接下来调用saveForLayer，看起来应该不错，那就看一下呗。
mCanvasState.save会构建一个SnapShot，然后在writableSnapshot时获取该snapshot，每一个renderNode对应一个snapShot，然后将参数设置到snapshot中去。
将当前的mLayerBuilders个数加入到mLayerStack中（只有2个值，一个0，一个size），那么是不是想知道mLayerBuilders之前怎么构建出来的呢？
那么你就来对地方了，看下FrameBuidler的构造函数就知道了，会构建一个fbo0的LayerBuilder。
继续往下看：
重新构建一个LayerBuilder，再加入到mLayerBuilders中，那么此时mLayerBuilders就有1个fbo0和N个renderNode对应的LayerBuilder。而mLayerStack对应它们的索引，是不是很神奇呢？

frameworks/base/libs/hwui/FrameBuilder.cpp
void FrameBuilder::saveForLayer(uint32_t layerWidth, uint32_t layerHeight, float contentTranslateX,
                                float contentTranslateY, const Rect& repaintRect,
                                const Vector3& lightCenter, const BeginLayerOp* beginLayerOp,
                                RenderNode* renderNode) {
    mCanvasState.save(SaveFlags::MatrixClip);
    mCanvasState.writableSnapshot()->initializeViewport(layerWidth, layerHeight);
    mCanvasState.writableSnapshot()->roundRectClipState = nullptr;
    mCanvasState.writableSnapshot()->setRelativeLightCenter(lightCenter);
    mCanvasState.writableSnapshot()->transform->loadTranslate(contentTranslateX, contentTranslateY,
                                                              0);
    mCanvasState.writableSnapshot()->setClip(repaintRect.left, repaintRect.top, repaintRect.right,
                                             repaintRect.bottom);

    // create a new layer repaint, and push its index on the stack
    mLayerStack.push_back(mLayerBuilders.size());
    auto newFbo = mAllocator.create<LayerBuilder>(layerWidth, layerHeight, repaintRect,
                                                  beginLayerOp, renderNode);
    mLayerBuilders.push_back(newFbo);
}
//1.save的过程就是构建一个Snapshot过程：
frameworks/base/libs/hwui/CanvasState.cpp
int CanvasState::save(int flags) {
    return saveSnapshot(flags);
}
int CanvasState::saveSnapshot(int flags) {
    mSnapshot = allocSnapshot(mSnapshot, flags);
    return mSaveCount++;
}
Snapshot* CanvasState::allocSnapshot(Snapshot* previous, int savecount) {
    void* memory;
    if (mSnapshotPool) {
        memory = mSnapshotPool;
        mSnapshotPool = mSnapshotPool->previous;
        mSnapshotPoolCount--;
    } else {
        memory = malloc(sizeof(Snapshot));
    }
    return new (memory) Snapshot(previous, savecount);
}
frameworks/base/libs/hwui/CanvasState.h
inline Snapshot* writableSnapshot() { return mSnapshot; }

frameworks/base/libs/hwui/FrameBuilder.h
LinearStdAllocator<void*> mStdAllocator;
LinearAllocator mAllocator;
LsaVector<size_t> mLayerStack;
LsaVector<LayerBuilder*> mLayerBuilders;
FrameBuilder::FrameBuilder(const SkRect& clip, uint32_t viewportWidth, uint32_t viewportHeight,
                           const LightGeometry& lightGeometry, Caches& caches)
        : mStdAllocator(mAllocator)
        , mLayerBuilders(mStdAllocator)
        , mLayerStack(mStdAllocator)
        , mCanvasState(*this)
        , mCaches(caches)
        , mLightRadius(lightGeometry.radius)
        , mDrawFbo0(true) {
    // Prepare to defer Fbo0
    auto fbo0 = mAllocator.create<LayerBuilder>(viewportWidth, viewportHeight, Rect(clip));
    mLayerBuilders.push_back(fbo0);
    mLayerStack.push_back(0);
    mCanvasState.initializeSaveStack(viewportWidth, viewportHeight, clip.fLeft, clip.fTop,
                                     clip.fRight, clip.fBottom, lightGeometry.center);
}
//如果FrameBuilder中没有指定viewportWidth、viewportHeight和clip，那么选择1替代：
auto fbo0 = mAllocator.create<LayerBuilder>(1, 1, Rect(1, 1));

4）接下来看下getDisplayList动作，我们知道在end的时候会将之前创建的DisplayList对象设置到mStagingDisplayList，（具体可参见之前博客：HWUI绘制系列——从java到C++），这边得到的就是该displayList对象。

frameworks/base/libs/hwui/RenderNode.h    
    DisplayList* mDisplayList;
    DisplayList* mStagingDisplayList;
    const DisplayList* getDisplayList() const { return mDisplayList; }

frameworks/base/libs/hwui/RenderNode.cpp
void RenderNode::prepareTreeImpl(TreeObserver& observer, TreeInfo& info, bool functorsNeedLayer) {
    ...
    pushStagingDisplayListChanges(observer, info); 
    ...
    pushLayerUpdate(info);
    ...
}
void RenderNode::pushStagingDisplayListChanges(TreeObserver& observer, TreeInfo& info) {
    if (mNeedsDisplayListSync) {
        mNeedsDisplayListSync = false;
        // Damage with the old display list first then the new one to catch any
        // changes in isRenderable or, in the future, bounds
        damageSelf(info);
        syncDisplayList(observer, &info);
        damageSelf(info);
    }
}
void RenderNode::syncDisplayList(TreeObserver& observer, TreeInfo* info) {
    // Make sure we inc first so that we don't fluctuate between 0 and 1,
    // which would thrash the layer cache
    if (mStagingDisplayList) {
        mStagingDisplayList->updateChildren([](RenderNode* child) { child->incParentRefCount(); });
    }
    deleteDisplayList(observer, info);
    mDisplayList = mStagingDisplayList;
    mStagingDisplayList = nullptr;
    if (mDisplayList) {
        mDisplayList->syncContents();
    }
}

void RenderNode::setStagingDisplayList(DisplayList* displayList) {
    mValid = (displayList != nullptr);
    mNeedsDisplayListSync = true;
    delete mStagingDisplayList;
    mStagingDisplayList = displayList;
}

介绍几个大神的博客：
http://blog.csdn.net/guoqifa29/article/details/45131099
http://blog.csdn.net/wind_hzx?viewmode=contents
http://www.tuicool.com/articles/bEjYbqN（android 5.0）(简书地址：http://www.jianshu.com/p/bc1c1d2fadd1)
http://blog.csdn.net/jinzhuojun/article/details/54234354（android 7.0）