【Graphics & SF】【硬件加速】3、DisplayList渲染过程分析【Android 13】

1,824 阅读26分钟

上一篇文章分析了DisplayList的构建流程,即ThreadedRenderer.draw方法中的ThreadedRenderer.updateRootDisplayList方法的内容。

// frameworks\base\core\java\android\view\ThreadedRenderer.java

	void draw(View view, AttachInfo attachInfo, DrawCallbacks callbacks) {
        // ......

        updateRootDisplayList(view, callbacks);

        // ......

        int syncResult = syncAndDrawFrame(frameInfo);
        // ......
    }

这一篇文章继续分析DisplayList的渲染流程,即ThreadedRenderer.syncAndDrawFrame方法的内容。

大概的流程如下:

流程.png

1 ThreadedRenderer.syncAndDrawFrame

    /**
     * Syncs the RenderNode tree to the render thread and requests a frame to be drawn.
     *
     * @hide
     */
    @SyncAndDrawResult
    public int syncAndDrawFrame(@NonNull FrameInfo frameInfo) {
        return nSyncAndDrawFrame(mNativeProxy, frameInfo.frameInfo, frameInfo.frameInfo.length);
    }

将之前在主线程通过ThreadedRenderer.updateRootDisplayList方法构建出的RenderNode树同步给RenderThread,并且请求一帧的绘制。

nSyncAndDrawFrame是JNI函数,对应android_view_ThreadedRenderer_syncAndDrawFrame。

1.1 android_view_ThreadedRenderer_syncAndDrawFrame

// frameworks\base\libs\hwui\jni\android_graphics_HardwareRenderer.cpp

static int android_view_ThreadedRenderer_syncAndDrawFrame(JNIEnv* env, jobject clazz,
                                                          jlong proxyPtr, jlongArray frameInfo,
                                                          jint frameInfoSize) {
    LOG_ALWAYS_FATAL_IF(frameInfoSize != UI_THREAD_FRAME_INFO_SIZE,
                        "Mismatched size expectations, given %d expected %zu", frameInfoSize,
                        UI_THREAD_FRAME_INFO_SIZE);
    RenderProxy* proxy = reinterpret_cast<RenderProxy*>(proxyPtr);
    env->GetLongArrayRegion(frameInfo, 0, frameInfoSize, proxy->frameInfo());
    return proxy->syncAndDrawFrame();
}

这里从上层传入的地址拿到Native层对应的RenderProxy对象,然后调用RenderProxy.syncAndDrawFrame。

1.2 RenderProxy.syncAndDrawFrame

int RenderProxy::syncAndDrawFrame() {
    return mDrawFrameTask.drawFrame();
}

RenderProxy的成员变量mDrawFrameTask是一个DrawFrameTask对象,继续调用DrawFrameTask.drawFrame。

1.3 DrawFrameTask.drawFrame

int DrawFrameTask::drawFrame() {
    LOG_ALWAYS_FATAL_IF(!mContext, "Cannot drawFrame with no CanvasContext!");

    mSyncResult = SyncResult::OK;
    mSyncQueued = systemTime(SYSTEM_TIME_MONOTONIC);
    postAndWait();

    return mSyncResult;
}

继续调用了自身的postAndWait函数。

1.4 DrawFrameTask.postAndWait

void DrawFrameTask::postAndWait() {
    ATRACE_CALL();
    AutoMutex _lock(mLock);
    mRenderThread->queue().post([this]() { run(); });
    mSignal.wait(mLock);
}

1)、向RenderThread的工作队列WorkQueue发布一个任务,根据之前的分析,知道了后续的run函数将会运行在RenderThread自己的线程中,即渲染线程,而非当前的主线程。

2)、DrawFrameTask的成员变量mSignal是一个Condition类型的条件变量,这里调用了其wait函数让主线程等待在此处。

1.5 DrawFrameTask.run

void DrawFrameTask::run() {
   // ......

    bool canUnblockUiThread;
    bool canDrawThisFrame;
    {
        TreeInfo info(TreeInfo::MODE_FULL, *mContext);
        info.forceDrawFrame = mForceDrawFrame;
        mForceDrawFrame = false;
        canUnblockUiThread = syncFrameState(info);
        canDrawThisFrame = info.out.canDrawThisFrame;

        // ......
    }

    // ......

    // From this point on anything in "this" is *UNSAFE TO ACCESS*
    if (canUnblockUiThread) {
        unblockUiThread();
    }

    // ......

    nsecs_t dequeueBufferDuration = 0;
    if (CC_LIKELY(canDrawThisFrame)) {
        dequeueBufferDuration = context->draw();
    } else {
        // ......
    }

    // ......

    if (!canUnblockUiThread) {
        unblockUiThread();
    }
    
    // ......
}

要理解这个函数首先要理解应用程序进程的Main Thread和Render Thread是如何协作的。从前面的分析可以知道,Main Thread请求Render Thread执行Draw Frame Task的时候,不能马上返回,而是进入等待状态。等到Render Thread从Main Thread同步完绘制所需要的信息之后,Main Thread才会被唤醒。

那么,Render Thread要从Main Thread同步什么信息呢?原来,Main Thread和Render Thread都各自维护了一份应用程序窗口视图信息。各自维护了一份应用程序窗口视图信息的目的,就是为了可以互不干扰,进而实现最大程度的并行。其中,Render Thread维护的应用程序窗口视图信息是来自于Main Thread的。因此,当Main Thread维护的应用程序窗口信息发生了变化时,就需要同步到Render Thread去。

应用程序窗口的视图信息包括各个Render Node的DisplayList、RenderProperties等属性,其中DisplayList保存的是绘制命令,RenderProperties保存的则是RenderNode的几何信息。

class RenderNode : public VirtualLightRefBase {
    friend class TestUtils;  // allow TestUtils to access syncDisplayList / syncProperties

public:
    enum DirtyPropertyMask {
        GENERIC = 1 << 1,
        TRANSLATION_X = 1 << 2,
        TRANSLATION_Y = 1 << 3,
        TRANSLATION_Z = 1 << 4,
        SCALE_X = 1 << 5,
        SCALE_Y = 1 << 6,
        ROTATION = 1 << 7,
        ROTATION_X = 1 << 8,
        ROTATION_Y = 1 << 9,
        X = 1 << 10,
        Y = 1 << 11,
        Z = 1 << 12,
        ALPHA = 1 << 13,
        DISPLAY_LIST = 1 << 14,
    };

    // ......

private:
    // ......

    uint32_t mDirtyPropertyFields;
    RenderProperties mProperties;
    RenderProperties mStagingProperties;

    // Owned by UI. Set when DL is set, cleared when DL cleared or when node detached
    // (likely by parent re-record/removal)
    bool mValid = false;

    bool mNeedsDisplayListSync;
    // WARNING: Do not delete this directly, you must go through deleteDisplayList()!
    DisplayList mDisplayList;
    DisplayList mStagingDisplayList;

    // ......

};  // class RenderNode

其中,成员变量mStagingProperties描述的Render Properties和成员变量mStagingDisplayListData描述的Display List Data由Main Thread维护,而成员变量mProperties描述的Render Properties和成员变量mDisplayListData描述的Display List Data由Render Thread维护。

回忆在记录绘制命令的时候,会调用Java层的RenderNode.endRecording方法,将绘制命令记录在RenderNode中,这最终是通过调用C/C++层RenderNode的setStagingDisplayList函数实现的:

void RenderNode::setStagingDisplayList(DisplayList&& newData) {
    mValid = newData.isValid();
    mNeedsDisplayListSync = true;
    mStagingDisplayList = std::move(newData);
}
  • mValid置为true,表示当前RenderNode持有一个DisplayList。

  • mNeedsDisplayListSync置为true,表示需要向RenderThread同步DisplayList的信息。

  • 具体的DisplayList信息则保存在mStagingDisplayList中。

再回到DrawFrameTask类的成员函数run中,它的执行逻辑如下所示:

1)、创建一个TreeInfo对象,模式为MODE_FULL,暂且记下。

2)、调用成员函数syncFrameState将应用程序窗口的Display List、Render Property以及Display List引用的Bitmap等信息从Main Thread同步到Render Thread中。注意,在这个同步过程中,Main Thread是处于等待状态的。

3)、如果成员函数syncFrameState能顺利地完成信息同步,那么它的返回值canUnblockUiThread就会等于true,表示在Render Thread渲染应用程序窗口的下一帧之前,就可以唤醒Main Thread了。否则的话,就要等到Render Thread渲染应用程序窗口的下一帧之后,才能唤醒Main Thread。唤醒Render Thread是通过调用成员函数unblockUiThread来完成的,如下所示:

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

前面Main Thread就刚好是等待在DrawFrameTask类的成员变量mSignal描述的一个条件变量之上的,所以现在Render Thread就通过这个条件变量来唤醒它。

4)、调用成员变量mContext描述的一个CanvasContext对象的成员函数draw渲染应用程序窗口的Display List。注意,这一步执行的时候,Main Thread有可能是处于等待状态,也有可能不是处于等待状态。这取决于前面的信息同步结果。信息同步结果是通过一个TreeInfo来描述的。当Main Thread不是处于等待状态时,它就可以马上处理其它事情了,例如构建应用程序窗口下一帧时使用的Display List。这样就可以做到Render Thread在绘制应用程序窗口的当前帧的同时,Main Thread可以并行地去构建应用程序窗口的下一帧的Display List。这一点也是Android 5.0引进Render Thread的好处所在。

接下来,我们就先分析应用程序窗口绘制信息的同步过程,即DrawFrameTask类的成员函数syncFrameState的实现,接着再分析应用程序窗口的Display List的渲染过程,即CanvasContext类的成员函数draw的实现。

2 DrawFrameTask.syncFrameState

bool DrawFrameTask::syncFrameState(TreeInfo& info) {
    // ......
    
    bool canDraw = mContext->makeCurrent();
    
    // ......
    
    mContext->prepareTree(info, mFrameInfo, mSyncQueued, mTargetNode);

    // ......
    
    // If prepareTextures is false, we ran out of texture cache space
    return info.prepareTextures;
}

1)、调用CanvasContext.makeCurrent。之前在分析硬件加速渲染环境初始化的时候,当时只是创建了一个EglSurface,但是并没有调用eglMakeCurrent将这个EglSurface和EglDisplay以及EglContext进行绑定,而这里则是调用了CanvasContext.makeCurrent,最终将会调用到EglManager.makeCurrent完成绑定。

2)、调用CanvasContext.prepareTree函数,之前分析硬件渲染环境初始化流程的时候,知道DrawFrameTask初始化的时候,窗口的RootRenderNode就保存在了其成员变量mTargetNode中。

3)、TreeInfo的成员变量prepareTextures描述Texture缓存空间是否已经被用尽,因为我们的分析不涉及Texture,因此这里默认为true,表示主线程不需要等待RenderThread绘制完一帧后再被唤醒。

主要看下CanvasContext.prepareTree函数的内容。

2.1 CanvasContext.prepareTree

void CanvasContext::prepareTree(TreeInfo& info, int64_t* uiFrameInfo, int64_t syncQueued,
                                RenderNode* target) {
    // ......
    
    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);
    }

    // ......

    bool postedFrameCallback = false;
    if (info.out.hasAnimations || !info.out.canDrawThisFrame) {
        if (CC_UNLIKELY(!Properties::enableRTAnimations)) {
            info.out.requiresUiRedraw = true;
        }
        if (!info.out.requiresUiRedraw) {
            // If animationsNeedsRedraw is set don't bother posting for an RT anim
            // as we will just end up fighting the UI thread.
            mRenderThread.postFrameCallback(this);
            postedFrameCallback = true;
        }
    }

    // ......
}

1)、之前分析硬件渲染环境初始化流程的时候,知道CanvasContext初始化的时候,窗口的RootRenderNode就保存在了其成员变量mRenderNodes中,另外注意这里的mRenderNodes是一个向量,那就是说mRenderNodes中除了RootRenderNode外可能还有其它的RenderNode,事实上也的确如此,Java层也可以通过addRenderNode的方式向mRenderNodes添加别的RenderNode,主要是BackdropFrameRenderer类,不过这些RenderNode是在窗口resize的时候用来添加背景的,暂不考虑这种情况,那么在此处,从mRenderNodes拿到的就是RootRenderNode,而传参target是DrawFrameTask的成员变量mTargetNode,也是RootRenderNode,那么这里的TreeInfo的mode就是MODE_FULL。

2)、TreeInfo中定义了两种模式:

        // The full monty - sync, push, run animators, etc... Used by DrawFrameTask
        // May only be used if both the UI thread and RT thread are blocked on the
        // prepare
        MODE_FULL,
        // Run only what can be done safely on RT thread. Currently this only means
        // animators, but potentially things like SurfaceTexture updates
        // could be handled by this as well if there are no listeners
        MODE_RT_ONLY,
  • MODE_FULL,表示本次绘制是由主线程驱动的。
  • MODE_RT_ONLY,表示本次绘制是由渲染线程驱动的,主要是动画。

3)、之前分析硬件渲染环境初始化流程的时候,知道CanvasContext初始化的时候,窗口的RootRenderNode就保存在了其成员变量mRenderNodes中,这里继续调用了RenderNode.prepareTree函数。

4)、看一下Out的几个成员变量的定义:

    struct Out {
        // ......
        
        // This is only updated if evaluateAnimations is true
        bool hasAnimations = false;
        // This is set to true if there is an animation that RenderThread cannot
        // animate itself, such as if hasFunctors is true
        // This is only set if hasAnimations is true
        bool requiresUiRedraw = false;
        // This is set to true if draw() can be called this frame
        // false means that we must delay until the next vsync pulse as frame
        // production is outrunning consumption
        // NOTE that if this is false CanvasContext will set either requiresUiRedraw
        // *OR* will post itself for the next vsync automatically, use this
        // only to avoid calling draw()
        bool canDrawThisFrame = true;
        
        // ......
    } out;
  • hasAnimations,动画是否正在运行,即未执行完成。
  • requiresUiRedraw,为true表示动画无法由RenderThread进行驱使,应该由主线程发起渲染请求。
  • canDrawThisFrame,为true表示CanvasContext.draw函数可以在这一帧的时候调用,为false表示推迟本次绘制到下一个Vsync信号到来的时候。

在这个同步的过程中,如果某些View设置了动画,并且这些动还未执行完成,那么参数info指向的TreeInfo对象的成员变量hasAnimations的值就会等于true。这时候如果应用程序窗口的下一帧不可以渲染,即上述TreeInfo对象的成员变量canDrawThisFrame的值等于false,并且所有View设置的动画都是非异步的,即上述TreeInfo对象的成员变量requiresUiRedraw的值等于false,那么就需要解决一个问题,那些未执行完成的动画如何继续执行下去?因为等到当应用程序窗口的下一帧可以渲染时,这些未完成的动画还是需要继续执行的。

我们知道,当TreeInfo对象的成员变量requiresUiRedraw的值等于true时,Main Thread会自动发起渲染应用程序窗口的Display List的命令。在这个命令的执行过程中,未完成的动画是可以继续执行的。但是当TreeInfo对象的成员变量requiresUiRedraw的值等于false时,Main Thread不会自动发起渲染应用程序窗口的Display List的命令,这时候就需要向Render Thread注册一个IFrameCallback接口,这是通过调用CanvasContext类的成员变量mRenderThread指向的一个RenderThread对象的成员函数postFrameCallback实现的。

根据之前的分析,我们知道注册到Render Thread的IFrameCallback接口在下一个Vsync信号到来时,它的成员函数doFrame会被调用,这时候就可以执行渲染应用程序窗口的下一帧了。在渲染的过程中,就可以继续执行那些未完成的动画了。

2.2 RenderNode.prepareTreeImpl

void RenderNode::prepareTree(TreeInfo& info) {
    // ......   
    prepareTreeImpl(observer, info, false);
    // ......
}

/**
 * Traverse down the the draw tree to prepare for a frame.
 *
 * MODE_FULL = UI Thread-driven (thus properties must be synced), otherwise RT driven
 *
 * While traversing down the tree, functorsNeedLayer flag is set to true if anything that uses the
 * stencil buffer may be needed. Views that use a functor to draw will be forced onto a layer.
 */
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);
    }

    if (mDisplayList) {
        // ......
        bool isDirty = mDisplayList.prepareListAndChildren(
                observer, info, childFunctorsNeedLayer,
                [this](RenderNode* child, TreeObserver& observer, TreeInfo& info,
                       bool functorsNeedLayer) {
                    child->prepareTreeImpl(observer, info, functorsNeedLayer);
                    mHasHolePunches |= child->hasHolePunches();
                });
        if (isDirty) {
            damageSelf(info);
        }
    } else {
        // ......
    }

    // ......
}

RenderNode的prepareTree函数继续调用了其prepareTreeImpl函数:

1)、根据之前的分析,知道本次的TreeInfo的模式为MODE_FULL,那么会调用RenderNode.pushStagingPropertiesChanges,该函数主要是调用RenderNode.syncProperties函数将成员变量mStagingProperties的信息同步到成员变量mProperties中。

2)、RenderNode.pushStagingDisplayListChanges函数和RenderNode.pushStagingPropertiesChanges类似,最后会调用RenderNode.syncDisplayList函数,将成员变量mStagingDisplayList的信息同步到成员变量mDisplayList中。

3)、DisplayList.prepareListAndChildren函数将会继续调用封装在其内的SkiaDisplayList.prepareListAndChildren函数,SkiaDisplayList.prepareListAndChildren函数仅由 RenderNode.prepareTree 调用,以便在 UI 线程被阻塞时进行从 UI 线程到渲染线程的信息同步工作。这里看到将对子RenderNode的RenderNode::prepareTreeImpl函数的调用封装为参数传给了DisplayList.prepareListAndChildren。如果返回isDirty为true,那么表示当前RenderNode无效需要重绘,那么就将通过RenderNode.damageSelf将当前RenderNode的脏区域添加到TreeInfo.damageAccumulator,一般来说RenderNode的脏区域就是这个RenderNode对应的RenderProperties的宽高所代表的区域。

2.3 SkiaDisplayList.prepareListAndChildren

// frameworks\base\libs\hwui\DisplayList.h
	bool prepareListAndChildren(
            TreeObserver& observer, TreeInfo& info, bool functorsNeedLayer,
            std::function<void(RenderNode*, TreeObserver&, TreeInfo&, bool)> childFn) {
        return mImpl && mImpl->prepareListAndChildren(
                observer, info, functorsNeedLayer, std::move(childFn));
    }

// frameworks\base\libs\hwui\pipeline\skia\SkiaDisplayList.cpp
bool SkiaDisplayList::prepareListAndChildren(
        TreeObserver& observer, TreeInfo& info, bool functorsNeedLayer,
        std::function<void(RenderNode*, TreeObserver&, TreeInfo&, bool)> childFn) {
    // If the prepare tree is triggered by the UI thread and no previous call to
    // pinImages has failed then we must pin all mutable images in the GPU cache
    // until the next UI thread draw.
#ifdef __ANDROID__ // Layoutlib does not support CanvasContext
    if (info.prepareTextures && !info.canvasContext.pinImages(mMutableImages)) {
        // In the event that pinning failed we prevent future pinImage calls for the
        // remainder of this tree traversal and also unpin any currently pinned images
        // to free up GPU resources.
        info.prepareTextures = false;
        info.canvasContext.unpinImages();
    }
#endif

    // ......

    for (auto& child : mChildNodes) {
        RenderNode* childNode = child.getRenderNode();
        // ......
        childFn(childNode, observer, info, functorsNeedLayer);
        // ......
    }

    // ......
}

之前分析过,DisplayList是对SkiaDisplayList的封装,那么真正得到调用的是SkiaDisplayList.prepareListAndChildren函数。

这里的childFn就是上面封装的:

                [this](RenderNode* child, TreeObserver& observer, TreeInfo& info,
                       bool functorsNeedLayer) {
                    child->prepareTreeImpl(observer, info, functorsNeedLayer);
                    mHasHolePunches |= child->hasHolePunches();
                }

那么会从RootRenderNode开始,递归调用RenderNode树中的每一个RenderNode的RenderNode.prepareTreeImpl函数,完成Properties、DisplayList等信息从主线程到渲染线程的同步,以及脏区域的计算等。

3 CanvasContext.draw

nsecs_t CanvasContext::draw() {
    // .....
    
    SkRect dirty;
    mDamageAccumulator.finish(&dirty);
    
    // ......

    Frame frame = mRenderPipeline->getFrame();
    SkRect windowDirty = computeDirtyRect(frame, &dirty);

    ATRACE_FORMAT("Drawing " RECT_STRING, SK_RECT_ARGS(dirty));

    const auto drawResult = mRenderPipeline->draw(frame, windowDirty, dirty, mLightGeometry,
                                                  &mLayerUpdateQueue, mContentDrawBounds, mOpaque,
                                                  mLightInfo, mRenderNodes, &(profiler()));

    // .....

    bool didSwap = mRenderPipeline->swapBuffers(frame, drawResult.success, windowDirty,
                                                mCurrentFrameInfo, &requireSwap);

    // .....
}

1)、调用SkiaOpenGLPipeline.getFrame,向BufferQueue请求一个图形缓冲区用来进行绘制。

2)、调用CanvasContext.computeDirtyRect,根据上面构建的Frame对象重新计算窗口的脏区域。

3)、调用SkiaOpenGLPipeline.draw,向图形缓冲区绘制内容。

4)、调用SkiaOpenGLPipeline.swapBuffers,向BufferQueue提交已经有内容的图形缓冲区。

下面分别分析这几部分。

4 请求图形缓冲区

请求图形缓冲区的操作是通过调用SkiaOpenGLPipeline.getFrame函数来完成的。

Frame SkiaOpenGLPipeline::getFrame() {
    LOG_ALWAYS_FATAL_IF(mEglSurface == EGL_NO_SURFACE,
                        "drawRenderNode called on a context with no surface!");
    return mEglManager.beginFrame(mEglSurface);
}

之前分析硬件加速渲染环境初始化的时候,知道SkiaOpenGLPipeline的成员变量mEglSurface已经被初始化过了,为EglManager.createSurface函数的返回值。

Frame EglManager::beginFrame(EGLSurface surface) {
    LOG_ALWAYS_FATAL_IF(surface == EGL_NO_SURFACE, "Tried to beginFrame on EGL_NO_SURFACE!");
    makeCurrent(surface);
    Frame frame;
    frame.mSurface = surface;
    eglQuerySurface(mEglDisplay, surface, EGL_WIDTH, &frame.mWidth);
    eglQuerySurface(mEglDisplay, surface, EGL_HEIGHT, &frame.mHeight);
    if (frame.mWidth > 65536 || frame.mHeight > 65536) {
        ALOGE("beginFrame's surface %p dimension exceeds 65536: frame.mWidth=%d, frame.mHeight=%d",
              (void*)surface, frame.mWidth, frame.mHeight);
    }
    frame.mBufferAge = queryBufferAge(surface);
    eglBeginFrame(mEglDisplay, surface);
    return frame;
}

1)、调用EglManager.makeCurrent,最终调用eglMakeCurrent函数重新绑定一下EglSurface和上下文。

2)、创建一个Frame对象并且初始化,Frame类的定义为:

class Frame {
public:
    // ......

private:
    Frame() {}
    friend class EglManager;

    int32_t mWidth;
    int32_t mHeight;
    int32_t mBufferAge;

    EGLSurface mSurface;

    // Maps from 0,0 in top-left to 0,0 in bottom-left
    // If out is not an int32_t[4] you're going to have a bad time
    void map(const SkRect& in, int32_t* out) const;
};

Frame用来描述一帧的信息,宽高之类的不用多说,这里的BufferAge作为缓冲区的年龄,代表了从缓冲区入队以后经过的帧数,App可以通过该信息可重新使用旧帧的内容并在下一帧时最大限度地减少必须重绘的内容。EGLSurface创建的时候以Surface作为参数,渲染该EGLSurface的时候会向BufferQueue请求一个缓冲区保存在Surface中。

这里重要的是对eglQuerySurface函数的调用。该函数的具体内容不再分析,只需要知道这一步最终将会调用BufferQueueProducer.dequeueBuffer向BufferQueue请求一个缓冲区即可,作为我们后续分析dequeueBuffer流程的起点。

#00 pc 0000000000090884  /system/lib64/libgui.so (android::BufferQueueProducer::dequeueBuffer(int*, android::sp<android::Fence>*, unsigned int, unsigned int, int, unsigned long, unsigned long*, android::FrameEventHistoryDelta*)+308)
#01 pc 00000000000f1a28  /system/lib64/libgui.so (android::Surface::dequeueBuffer(ANativeWindowBuffer**, int*)+488)
#02 pc 000000000028d778  /system/lib64/libhwui.so (android::uirenderer::renderthread::ReliableSurface::hook_dequeueBuffer(ANativeWindow*, int (*)(ANativeWindow*, ANativeWindowBuffer**, int*), void*, ANativeWindowBuffer**, int*)+200)
#03 pc 00000000000efb6c  /system/lib64/libgui.so (android::Surface::hook_dequeueBuffer(ANativeWindow*, ANativeWindowBuffer**, int*)+92)
#04 pc 0000000000050298  /vendor/lib64/libIMGegl.so
#05 pc 00000000000232e4  /vendor/lib64/libIMGegl.so (KEGLGetDrawableParameters+316)
#06 pc 000000000002bf7c  /vendor/lib64/libIMGegl.so (IMGeglQuerySurface+1296)
#07 pc 000000000001f89c  /system/lib64/libEGL.so (android::eglQuerySurfaceImpl(void*, void*, int, int*)+204)
#08 pc 000000000001c850  /system/lib64/libEGL.so (eglQuerySurface+64)
#09 pc 000000000028c770  /system/lib64/libhwui.so (android::uirenderer::renderthread::EglManager::beginFrame(void*)+128)
#10 pc 0000000000287be4  /system/lib64/libhwui.so (android::uirenderer::renderthread::CanvasContext::draw()+260)
#11 pc 000000000028a9b0  /system/lib64/libhwui.so (std::__1::__function::__func<android::uirenderer::renderthread::DrawFrameTask::postAndWait()::$_0, std::__1::allocator<android::uirenderer::renderthread::DrawFrameTask::postAndWait()::$_0>, void ()>::operator()() (.c1671e787f244890c877724752face20)+1008)
#12 pc 0000000000279dc4  /system/lib64/libhwui.so (android::uirenderer::WorkQueue::process()+644)
#13 pc 0000000000299970  /system/lib64/libhwui.so (android::uirenderer::renderthread::RenderThread::threadLoop()+560)
#14 pc 0000000000013550  /system/lib64/libutils.so (android::Thread::_threadLoop(void*)+416)
#15 pc 00000000000fc350  /apex/com.android.runtime/lib64/bionic/libc.so (__pthread_start(void*)+208)
#16 pc 000000000008e310  /apex/com.android.runtime/lib64/bionic/libc.so (__start_thread+64)

5 向图形缓冲区绘制内容

绘制内容的操作是通过调用SkiaOpenGLPipeline.draw函数来完成的。

IRenderPipeline::DrawResult SkiaOpenGLPipeline::draw(
        const Frame& frame, const SkRect& screenDirty, const SkRect& dirty,
        const LightGeometry& lightGeometry, LayerUpdateQueue* layerUpdateQueue,
        const Rect& contentDrawBounds, bool opaque, const LightInfo& lightInfo,
        const std::vector<sp<RenderNode>>& renderNodes, FrameInfoVisualizer* profiler) {
    // ......
    
    sk_sp<SkSurface> surface(SkSurface::MakeFromBackendRenderTarget(
            mRenderThread.getGrContext(), backendRT, this->getSurfaceOrigin(), colorType,
            mSurfaceColorSpace, &props));
    
    // ......

    renderFrame(*layerUpdateQueue, dirty, renderNodes, opaque, contentDrawBounds, surface,
                SkMatrix::I());
	// ......
    
    {
        ATRACE_NAME("flush commands");
        surface->flushAndSubmit();
    }
    
    // ......
}

继续调用其父类SkiaPipeline的renderFrame函数。

5.1 SkiaPipeline.renderFrame

  void SkiaPipeline::renderFrame(const LayerUpdateQueue& layers, const SkRect& clip,
                                 const std::vector<sp<RenderNode>>& nodes, bool opaque,
                                 const Rect& contentDrawBounds, sk_sp<SkSurface> surface,
                                 const SkMatrix& preTransform) {
      // ......
  
      renderFrameImpl(clip, nodes, opaque, contentDrawBounds, canvas, preTransform);
  
      // ......
  }

继续调用其SkiaPipeline.renderFrameImpl函数。

5.2 SkiaPipeline.renderFrameImpl

  void SkiaPipeline::renderFrameImpl(const SkRect& clip,
                                     const std::vector<sp<RenderNode>>& nodes, bool opaque,
                                     const Rect& contentDrawBounds, SkCanvas* canvas,
                                     const SkMatrix& preTransform) {
      // ......
  
      if (1 == nodes.size()) {
          if (!nodes[0]->nothingToDraw()) {
              RenderNodeDrawable root(nodes[0].get(), canvas);
              root.draw(canvas);
          }
      } else if (0 == nodes.size()) {
          // nothing to draw
      } else {
          // It there are multiple render nodes, they are laid out as follows:
          // #0 - backdrop (content + caption)
          // #1 - content (local bounds are at (0,0), will be translated and clipped to backdrop)
          // #2 - additional overlay nodes
          // Usually the backdrop cannot be seen since it will be entirely covered by the content.
          // While
          // resizing however it might become partially visible. The following render loop will crop
          // the
          // backdrop against the content and draw the remaining part of it. It will then draw the
          // content
          // cropped to the backdrop (since that indicates a shrinking of the window).
          //
          // Additional nodes will be drawn on top with no particular clipping semantics.
  
          // Usually the contents bounds should be mContentDrawBounds - however - we will
          // move it towards the fixed edge to give it a more stable appearance (for the moment).
          // If there is no content bounds we ignore the layering as stated above and start with 2.
  
          // ......
      }
  }

这里根据传入的nodes,即CanvasContext.mRenderNodes的数量分为多种情况:

1)、数量为1,那么该RenderNode即为RootRenderNode,View层级结构中的根View对应的RenderNode,那么创建一个RenderNodeDrawable对象,并调用其draw函数。

2)、数量为0,没有东西可以绘制。

3)、数量大于1,除了RootRenderNode之外还有其它的RenderNode,需要根据它们的Z轴高度按顺序绘制。

之前我们已经讲过mRenderNodes的数量大于1的情况了,这里我们分析最普通的情况,即只有RootRenderNode的情况。

5.3 SkDrawable.draw

RenderNodeDrawable类的draw是继承其父类SKDrawable的:

// external\skia\src\core\SkDrawable.cpp
void SkDrawable::draw(SkCanvas* canvas, const SkMatrix* matrix) {
    // ......
    
    this->onDraw(canvas);

    // ......
}

又回到了RenderNodeDrawable。

5.4 RenderNodeDrawable.ondraw

void RenderNodeDrawable::onDraw(SkCanvas* canvas) {
    // negative and positive Z order are drawn out of order, if this render node drawable is in
    // a reordering section
    if ((!mInReorderingSection) || MathUtils::isZero(mRenderNode->properties().getZ())) {
        this->forceDraw(canvas);
    }
}

5.5 RenderNodeDrawable.forceDraw

void RenderNodeDrawable::forceDraw(SkCanvas* canvas) const {
    // ......

    if (!properties.getProjectBackwards()) {
        drawContent(canvas);
        // ......
    }
    // ......
}

5.6 RenderNodeDrawable.drawContent

void RenderNodeDrawable::drawContent(SkCanvas* canvas) const {
    // ......

    // TODO should we let the bound of the drawable do this for us?
    const SkRect bounds = SkRect::MakeWH(properties.getWidth(), properties.getHeight());
    bool quickRejected = properties.getClipToBounds() && canvas->quickReject(bounds);
    // ......
    if (!quickRejected) {
        SkiaDisplayList* displayList = renderNode->getDisplayList().asSkiaDl();
        const LayerProperties& layerProperties = properties.layerProperties();
        // composing a hardware layer
        if (renderNode->getLayerSurface() && mComposeLayer) {
			// ......
        } else {
            if (alphaMultiplier < 1.0f) {
                // Non-layer draw for a view with getHasOverlappingRendering=false, will apply
                // the alpha to the paint of each nested draw.
                AlphaFilterCanvas alphaCanvas(canvas, alphaMultiplier);
                displayList->draw(&alphaCanvas);
            } else {
                displayList->draw(canvas);
            }
        }
    }
}

1)、RenderNode本身有一个位置属性,是一个矩阵区域,通过RenderProperties.setTop等函数进行设置,同时也可以再设置一个裁剪区域,通过RenderProperties.setClipBounds函数设置。

如果RenderProperties.getClipToBounds为true,那么这个RenderNode就会被裁剪到这两个区域的相交部分,那么如果getClipToBounds为true且两个区域不相交,则quickRejected为true,表示在裁剪区域内该RenderNode没有可以绘制的部分,当前RenderNode无效无需绘制。

2)、RenderNode.getLayerSurface返回一个屏幕外的渲染Surface,用于渲染到层图中并将图层合成到其父图层中。一般流程下这里返回的是null,那么后续就走到了else流程。

最后调用了SkiaDisplayList.draw函数。

5.7 SkiaDisplayList.draw

// frameworks\base\libs\hwui\pipeline\skia\SkiaDisplayList.h

	void draw(SkCanvas* canvas) { mDisplayList.draw(canvas); }

    DisplayListData mDisplayList;

继续调用其成员变量mDisplayList的DisplayListData.draw函数。

5.8 DisplayListData.draw

// frameworks\base\libs\hwui\RecordingCanvas.cpp

	void DisplayListData::draw(SkCanvas* canvas) const {
      SkAutoCanvasRestore acr(canvas, false);
      this->map(draw_fns, canvas, canvas->getTotalMatrix());
  }

看下这里的draw_fns的定义:

// All ops implement draw().
#define X(T)                                                    \
    [](const void* op, SkCanvas* c, const SkMatrix& original) { \
        ((const T*)op)->draw(c, original);                      \
    },
static const draw_fn draw_fns[] = {
#include "DisplayListOps.in"
};
#undef X

在DisplayListOps.in文件定义了所有的绘制命令:

// frameworks\base\libs\hwui\DisplayListOps.in

X(Flush)
X(Save)
X(Restore)
X(SaveLayer)
X(SaveBehind)
X(Concat)
X(SetMatrix)
X(Scale)
X(Translate)
X(ClipPath)
X(ClipRect)
X(ClipRRect)
X(ClipRegion)
X(ResetClip)
X(DrawPaint)
X(DrawBehind)
X(DrawPath)
X(DrawRect)
X(DrawRegion)
X(DrawOval)
X(DrawArc)
X(DrawRRect)
X(DrawDRRect)
X(DrawAnnotation)
X(DrawDrawable)
X(DrawPicture)
X(DrawImage)
X(DrawImageRect)
X(DrawImageLattice)
X(DrawTextBlob)
X(DrawPatch)
X(DrawPoints)
X(DrawVertices)
X(DrawAtlas)
X(DrawShadowRec)
X(DrawVectorDrawable)
X(DrawRippleDrawable)
X(DrawWebView)

继续看DisplayListData.map函数。

5.9 DisplayListData.map

template <typename Fn, typename... Args>
inline void DisplayListData::map(const Fn fns[], Args... args) const {
    auto end = fBytes.get() + fUsed;
    for (const uint8_t* ptr = fBytes.get(); ptr < end;) {
        auto op = (const Op*)ptr;
        auto type = op->type;
        auto skip = op->skip;
        if (auto fn = fns[type]) {  // We replace no-op functions with nullptrs
            fn(op, args...);        // to avoid the overhead of a pointless call.
        }
        ptr += skip;
    }
}

根据之前的分析,知道之前在Java层View.onDraw方法中调用的Canvas的各种drawXXX的API,最终都是将绘制命令保存在了这里的以fBytes为起点的内存空间中了,那么这里就是从将这些绘制命令从内存空间中取出,如果绘制类型符合,那么就调用相应的绘制命令。

比如之前的Java层调用Canvas.drawRect方法绘制一个矩形,最终会调用到DisplayListData.drawRect:

void DisplayListData::drawRect(const SkRect& rect, const SkPaint& paint) {
    this->push<DrawRect>(0, rect, paint);
}

当时向内存空间中缓存了一条DrawRect的命令,那么此时就可以将这条DrawRect命令从内存空间取出并进行调用。

struct DrawRect final : Op {
    static const auto kType = Type::DrawRect;
    DrawRect(const SkRect& rect, const SkPaint& paint) : rect(rect), paint(paint) {}
    SkRect rect;
    SkPaint paint;
    void draw(SkCanvas* c, const SkMatrix&) const { c->drawRect(rect, paint); }
};

具体调用的则是SKCanvas.drawRect函数,SKCanvas的部分因为能力不够暂不分析。

5.10 小结

回到我们分析的起点,SkiaOpenGLPipeline.draw函数。

IRenderPipeline::DrawResult SkiaOpenGLPipeline::draw(
        const Frame& frame, const SkRect& screenDirty, const SkRect& dirty,
        const LightGeometry& lightGeometry, LayerUpdateQueue* layerUpdateQueue,
        const Rect& contentDrawBounds, bool opaque, const LightInfo& lightInfo,
        const std::vector<sp<RenderNode>>& renderNodes, FrameInfoVisualizer* profiler) {
    // ......
    
    sk_sp<SkSurface> surface(SkSurface::MakeFromBackendRenderTarget(
            mRenderThread.getGrContext(), backendRT, this->getSurfaceOrigin(), colorType,
            mSurfaceColorSpace, &props));
    
    // ......

    renderFrame(*layerUpdateQueue, dirty, renderNodes, opaque, contentDrawBounds, surface,
                SkMatrix::I());
	// ......
    
    {
        ATRACE_NAME("flush commands");
        surface->flushAndSubmit();
    }
    
    // ......
}

1)、之前在Java层View.onDraw方法中调用的Canvas的各种绘制命令保存在了DisplayListData中,这里通过renderFrame函数将这些绘制命令转化为对SKCanvas绘制函数的调用。

2)、继续调用SKSurface.flushAndSubmit,相当于根据上一步的SkCanvas的绘制命令生成GL指令,绘制相关内容到图形缓冲区。

6 提交图形缓冲区

提交图形缓冲区的操作是通过调用SkiaOpenGLPipeline.swapBuffers函数完成的。

bool SkiaOpenGLPipeline::swapBuffers(const Frame& frame, bool drew, const SkRect& screenDirty,
                                     FrameInfo* currentFrameInfo, bool* requireSwap) {
    // ......

    if (*requireSwap && (CC_UNLIKELY(!mEglManager.swapBuffers(frame, screenDirty)))) {
        return false;
    }

    return *requireSwap;
}

这里继续调用了EglManager.swapBuffers函数:

bool EglManager::swapBuffers(const Frame& frame, const SkRect& screenDirty) {
    // ......
    
    eglSwapBuffersWithDamageKHR(mEglDisplay, frame.mSurface, rects, screenDirty.isEmpty() ? 0 : 1);

    // ......
}

这里调用了eglApi中的eglSwapBuffersWithDamageKHR函数。

// frameworks/native/opengl/libs/EGL/eglApi.cpp

  EGLBoolean eglSwapBuffersWithDamageKHR(EGLDisplay dpy, EGLSurface draw, EGLint* rects,
                                         EGLint n_rects) {
      ATRACE_CALL();
      clearError();
  
      egl_connection_t* const cnx = &gEGLImpl;
      return cnx->platform.eglSwapBuffersWithDamageKHR(dpy, draw, rects, n_rects);
  }

其实现为egl_platform_entries.cpp的eglSwapBuffersWithDamageKHRImpl函数,后续为各个平台的具体实现代码,不再叙述。

我们分析这一部分是要为了引起出后续的BufferQueue的queueBuffer和acquireBuffer流程,作为分析后续这两个流程的起点。

queueBuffer:

#00 pc 0000000000092790  /system/lib64/libgui.so (android::BufferQueueProducer::queueBuffer(int, android::IGraphicBufferProducer::QueueBufferInput const&, android::IGraphicBufferProducer::QueueBufferOutput*)+848)
#01 pc 00000000000f521c  /system/lib64/libgui.so (android::Surface::queueBuffer(ANativeWindowBuffer*, int)+1228)
#02 pc 00000000000efcec  /system/lib64/libgui.so (android::Surface::hook_queueBuffer(ANativeWindow*, ANativeWindowBuffer*, int)+92)
#03 pc 0000000000051b88  /vendor/lib64/libIMGegl.so
#04 pc 0000000000030438  /vendor/lib64/libIMGegl.so (IMGeglSwapBuffersWithDamageKHR+1372)
#05 pc 00000000000205fc  /system/lib64/libEGL.so (android::eglSwapBuffersWithDamageKHRImpl(void*, void*, int*, int)+524)
#06 pc 000000000001cc48  /system/lib64/libEGL.so (eglSwapBuffersWithDamageKHR+72)
#07 pc 000000000028c9dc  /system/lib64/libhwui.so (android::uirenderer::renderthread::EglManager::swapBuffers(android::uirenderer::renderthread::Frame const&, SkRect const&)+108)
#08 pc 0000000000280418  /system/lib64/libhwui.so (android::uirenderer::skiapipeline::SkiaOpenGLPipeline::swapBuffers(android::uirenderer::renderthread::Frame const&, bool, SkRect const&, android::uirenderer::FrameInfo*, bool*)+120)
#09 pc 0000000000287ff8  /system/lib64/libhwui.so (android::uirenderer::renderthread::CanvasContext::draw()+1304)
#10 pc 000000000028a9b0  /system/lib64/libhwui.so (std::__1::__function::__func<android::uirenderer::renderthread::DrawFrameTask::postAndWait()::$_0, std::__1::allocator<android::uirenderer::renderthread::DrawFrameTask::postAndWait()::$_0>, void ()>::operator()() (.c1671e787f244890c877724752face20)+1008)
#11 pc 0000000000279dc4  /system/lib64/libhwui.so (android::uirenderer::WorkQueue::process()+644)
#12 pc 0000000000299970  /system/lib64/libhwui.so (android::uirenderer::renderthread::RenderThread::threadLoop()+560)
#13 pc 0000000000013550  /system/lib64/libutils.so (android::Thread::_threadLoop(void*)+416)
#14 pc 00000000000fc350  /apex/com.android.runtime/lib64/bionic/libc.so (__pthread_start(void*)+208)
#15 pc 000000000008e310  /apex/com.android.runtime/lib64/bionic/libc.so (__start_thread+64)

acquireBuffer:

#00 pc 0000000000088204  /system/lib64/libgui.so (android::BufferQueueConsumer::acquireBuffer(android::BufferItem*, long, unsigned long)+388)
#01 pc 00000000000bb134  /system/lib64/libgui.so (android::ConsumerBase::acquireBufferLocked(android::BufferItem*, long, unsigned long)+84)
#02 pc 00000000000b920c  /system/lib64/libgui.so (android::BufferItemConsumer::acquireBuffer(android::BufferItem*, long, bool)+76)
#03 pc 00000000000a93c8  /system/lib64/libgui.so (android::BLASTBufferQueue::acquireNextBufferLocked(std::__1::optional<android::SurfaceComposerClient::Transaction*>)+232)
#04 pc 00000000000ac964  /system/lib64/libgui.so (android::BLASTBufferQueue::onFrameAvailable(android::BufferItem const&)+1780)
#05 pc 00000000000ba48c  /system/lib64/libgui.so (android::ConsumerBase::onFrameAvailable(android::BufferItem const&)+172)
#06 pc 0000000000087068  /system/lib64/libgui.so (android::BufferQueue::ProxyConsumerListener::onFrameAvailable(android::BufferItem const&)+104)
#07 pc 0000000000092cf0  /system/lib64/libgui.so (android::BufferQueueProducer::queueBuffer(int, android::IGraphicBufferProducer::QueueBufferInput const&, android::IGraphicBufferProducer::QueueBufferOutput*)+2224)
#08 pc 00000000000f521c  /system/lib64/libgui.so (android::Surface::queueBuffer(ANativeWindowBuffer*, int)+1228)
#09 pc 00000000000efcec  /system/lib64/libgui.so (android::Surface::hook_queueBuffer(ANativeWindow*, ANativeWindowBuffer*, int)+92)
#10 pc 0000000000051b88  /vendor/lib64/libIMGegl.so
#11 pc 0000000000030438  /vendor/lib64/libIMGegl.so (IMGeglSwapBuffersWithDamageKHR+1372)
#12 pc 00000000000205fc  /system/lib64/libEGL.so (android::eglSwapBuffersWithDamageKHRImpl(void*, void*, int*, int)+524)
#13 pc 000000000001cc48  /system/lib64/libEGL.so (eglSwapBuffersWithDamageKHR+72)
#14 pc 000000000028c9dc  /system/lib64/libhwui.so (android::uirenderer::renderthread::EglManager::swapBuffers(android::uirenderer::renderthread::Frame const&, SkRect const&)+108)
#15 pc 0000000000280418  /system/lib64/libhwui.so (android::uirenderer::skiapipeline::SkiaOpenGLPipeline::swapBuffers(android::uirenderer::renderthread::Frame const&, bool, SkRect const&, android::uirenderer::FrameInfo*, bool*)+120)
#16 pc 0000000000287ff8  /system/lib64/libhwui.so (android::uirenderer::renderthread::CanvasContext::draw()+1304)
#17 pc 000000000028a9b0  /system/lib64/libhwui.so (std::__1::__function::__func<android::uirenderer::renderthread::DrawFrameTask::postAndWait()::$_0, std::__1::allocator<android::uirenderer::renderthread::DrawFrameTask::postAndWait()::$_0>, void ()>::operator()() (.c1671e787f244890c877724752face20)+1008)
#18 pc 0000000000279dc4  /system/lib64/libhwui.so (android::uirenderer::WorkQueue::process()+644)
#19 pc 0000000000299970  /system/lib64/libhwui.so (android::uirenderer::renderthread::RenderThread::threadLoop()+560)
#20 pc 0000000000013550  /system/lib64/libutils.so (android::Thread::_threadLoop(void*)+416)
#21 pc 00000000000fc350  /apex/com.android.runtime/lib64/bionic/libc.so (__pthread_start(void*)+208)
#22 pc 000000000008e310  /apex/com.android.runtime/lib64/bionic/libc.so (__start_thread+64)

7 总结

这次主要分析了DrawFrameTask的run函数的主要内容,即绘制流程从主线程切换到渲染线程后,渲染线程做了哪些工作:

1)、之前我们构建RenderNode树的工作是在主线程上进行的,但是后续的绘制工作是在渲染线程上进行的,那么就需要将主线程收集到的RenderNode的信息同步到渲染线程,包括Properties和DisplayList等信息。

2)、绘制之前,还需要向BufferQeueu请求一个图形缓冲区,对应BufferQueue的dequeueBuffer流程。

3)、开始绘制,向图形缓冲区填充内容。

4)、绘制结束,得到了一个有内容的图形缓冲区,那么下一步就是把这个缓冲区推入BufferQueue,并提交给SurfaceFlinger进行合成,对应BufferQueue的queueBuffer流程和acquireBuffer流程。

总结完毕后我们下一步要分析的内容也很明朗了,BufferQueue机制。