Android P 图形显示系统(八) SurfaceFlinger合成流程(三)

[TOC]

SurfaceFlinger合成流程(三)

配置硬件合成 setUpHWComposer

回到handleMessageRefresh,继续看Refresh消息的处理。此时需要进行合成显示的数据,在rebuildLayerStacks时,已经被更新到每个Display各自的layersSortedByZ中。Layer栈创建完成后,进行HWC 合成的设置。

setUpHWComposer的代码比较长,我们分段看,在setUpHWComposer中,主要做了以下几件事:

1.DisplayDevice beginFrame

void SurfaceFlinger::setUpHWComposer() {
    ... ...

    for (size_t dpy=0 ; dpy<mDisplays.size() ; dpy++) {
        bool dirty = !mDisplays[dpy]->getDirtyRegion(false).isEmpty();
        bool empty = mDisplays[dpy]->getVisibleLayersSortedByZ().size() == 0;
        bool wasEmpty = !mDisplays[dpy]->lastCompositionHadVisibleLayers;

        //  判断是否需要重新合成
        bool mustRecompose = dirty && !(empty && wasEmpty);

        ... ...

        mDisplays[dpy]->beginFrame(mustRecompose);

        if (mustRecompose) {
            mDisplays[dpy]->lastCompositionHadVisibleLayers = !empty;
        }
    }

Android对每一块显示屏的处理都是分开的。这里主要是调Display的beginFrame函数。

status_t DisplayDevice::beginFrame(bool mustRecompose) const {
    return mDisplaySurface->beginFrame(mustRecompose);
}

mDisplaySurface根据屏幕有所不同。

主显和外显用的FramebufferSurface,需显示用的VirtualDisplaySurface,我们这里先不关虚显。

status_t FramebufferSurface::beginFrame(bool /*mustRecompose*/) {
    return NO_ERROR;
}

FramebufferSurface在beginFrame是每一做什么过多的处理。

回到setUpHWComposer函数

2.创建工作列表

void SurfaceFlinger::setUpHWComposer() {
     ... ...

    // build the h/w work list
    if (CC_UNLIKELY(mGeometryInvalid)) {
        mGeometryInvalid = false;
        for (size_t dpy=0 ; dpy<mDisplays.size() ; dpy++) {
            sp<const DisplayDevice> displayDevice(mDisplays[dpy]);
            const auto hwcId = displayDevice->getHwcDisplayId();
            if (hwcId >= 0) {
                const Vector<sp<Layer>>& currentLayers(
                        displayDevice->getVisibleLayersSortedByZ());
                for (size_t i = 0; i < currentLayers.size(); i++) {
                    const auto& layer = currentLayers[i];
                    if (!layer->hasHwcLayer(hwcId)) {
                        if (!layer->createHwcLayer(getBE().mHwc.get(), hwcId)) {
                            layer->forceClientComposition(hwcId);
                            continue;
                        }
                    }

                    layer->setGeometry(displayDevice, i);
                    if (mDebugDisableHWC || mDebugRegion) {
                        layer->forceClientComposition(hwcId);
                    }
                }
            }
        }
    }

对每个Display中的每个Layer创建对应的HWC Layer,注意hwcId,只是对hwcId大于零的Layer才会创建HWC Layer。Layer的createHwcLayer函数如下:

bool Layer::createHwcLayer(HWComposer* hwc, int32_t hwcId) {
    LOG_ALWAYS_FATAL_IF(getBE().mHwcLayers.count(hwcId) != 0,
                        "Already have a layer for hwcId %d", hwcId);
    HWC2::Layer* layer = hwc->createLayer(hwcId);
    if (!layer) {
        return false;
    }
    LayerBE::HWCInfo& hwcInfo = getBE().mHwcLayers[hwcId];
    hwcInfo.hwc = hwc;
    hwcInfo.layer = layer;
    layer->setLayerDestroyedListener(
            [this, hwcId](HWC2::Layer* /*layer*/) { getBE().mHwcLayers.erase(hwcId); });
    return true;
}

根据hwcId来创建,也就是说,HWC上层(SurfaceFlinger)的每个Layer,都会为每个Display创建一个HWC Layer。HWComposer 根据hwcId找对HWC2的Display,再通过具体的HWC2::Dispaly去创建自己的HWC Layer;

Layer通过HWComposer来创建,

HWC2::Layer* HWComposer::createLayer(int32_t displayId) {
    if (!isValidDisplay(displayId)) {
        ALOGE("Failed to create layer on invalid display %d", displayId);
        return nullptr;
    }
    auto display = mDisplayData[displayId].hwcDisplay;
    HWC2::Layer* layer;
    auto error = display->createLayer(&layer);
    if (error != HWC2::Error::None) {
        ALOGE("Failed to create layer on display %d: %s (%d)", displayId,
                to_string(error).c_str(), static_cast<int32_t>(error));
        return nullptr;
    }
    return layer;
}

创建HWC Layer的实现,最终是在Vendor实现的HAL中来完成的。对应的HWC2command为HWC2_FUNCTION_CREATE_LAYER

Error HwcHal::createLayer(Display display, Layer* outLayer)
{
    int32_t err = mDispatch.createLayer(mDevice, display, outLayer);
    return static_cast<Error>(err);
}

创建的HWC Layer在HWC2::Dispaly中也保存了一个引用:

* frameworks/native/services/surfaceflinger/DisplayHardware/HWC2.cpp

Error Display::createLayer(Layer** outLayer)
{
    if (!outLayer) {
        return Error::BadParameter;
    }
    hwc2_layer_t layerId = 0;
    auto intError = mComposer.createLayer(mId, &layerId);
    auto error = static_cast<Error>(intError);
    if (error != Error::None) {
        return error;
    }

    auto layer = std::make_unique<Layer>(
            mComposer, mCapabilities, mId, layerId);
    *outLayer = layer.get();
    mLayers.emplace(layerId, std::move(layer));
    return Error::None;
}

如果hwclayer没有创建成功,那么这一层Layer就强制用Client方式合成forceClientComposition。

void Layer::forceClientComposition(int32_t hwcId) {
    if (getBE().mHwcLayers.count(hwcId) == 0) {
        ALOGE("forceClientComposition: no HWC layer found (%d)", hwcId);
        return;
    }

    getBE().mHwcLayers[hwcId].forceClientComposition = true;
}

另外,如果是Disable掉HWC合成,或者调试Region,也强制用Client方式合成:

mDebugDisableHWC || mDebugRegion

这两个调试方式可以在系统设置,开发者选项中进行设置。

创建完hwcLayer后,设置Layer的几何尺寸:

void Layer::setGeometry(const sp<const DisplayDevice>& displayDevice, uint32_t z)
{
     ... ... //注意,我们这里的数据都是来源于DrawingState
    const State& s(getDrawingState());
    ... ...
    if (!isOpaque(s) || getAlpha() != 1.0f) {
        blendMode =
                mPremultipliedAlpha ? HWC2::BlendMode::Premultiplied : HWC2::BlendMode::Coverage;
    }
    auto error = hwcLayer->setBlendMode(blendMode);

    // 计算displayFrame
    Rect frame{t.transform(computeBounds(activeTransparentRegion))};
    ... ...
    const Transform& tr(displayDevice->getTransform());
    Rect transformedFrame = tr.transform(frame);
    error = hwcLayer->setDisplayFrame(transformedFrame);
    ... ...

    // 计算sourceCrop
    FloatRect sourceCrop = computeCrop(displayDevice);
    error = hwcLayer->setSourceCrop(sourceCrop);
    ... ...

    // 设置Alpha
    float alpha = static_cast<float>(getAlpha());
    error = hwcLayer->setPlaneAlpha(alpha);
    ... ...

    // 设置z-order
    error = hwcLayer->setZOrder(z);
    ... ...

    int type = s.type;
    int appId = s.appId;
    sp<Layer> parent = mDrawingParent.promote();
    if (parent.get()) {
        auto& parentState = parent->getDrawingState();
        type = parentState.type;
        appId = parentState.appId;
    }

    // 设置Layer的信息
    error = hwcLayer->setInfo(type, appId);
    ALOGE_IF(error != HWC2::Error::None, "[%s] Failed to set info (%d)", mName.string(),
             static_cast<int32_t>(error));

    // 设置transform
    const uint32_t orientation = transform.getOrientation();
    if (orientation & Transform::ROT_INVALID) {
        // we can only handle simple transformation
        hwcInfo.forceClientComposition = true;
    } else {
        auto transform = static_cast<HWC2::Transform>(orientation);
        auto error = hwcLayer->setTransform(transform);
        ... ...
    }
}

注意,我们这里的数据都是来源于DrawingState,setGeometry函数中,主要做了以下几件事:

  • 确认HWCLayer的混合模式
    混合模式,是两个Layer直接的混合方式,主要下面的几种:
/* Blend modes, settable per layer */
typedef enum {
    HWC2_BLEND_MODE_INVALID = 0,

    /* colorOut = colorSrc */
    HWC2_BLEND_MODE_NONE = 1,

    /* colorOut = colorSrc + colorDst * (1 - alphaSrc) */
    HWC2_BLEND_MODE_PREMULTIPLIED = 2,

    /* colorOut = colorSrc * alphaSrc + colorDst * (1 - alphaSrc) */
    HWC2_BLEND_MODE_COVERAGE = 3,
} hwc2_blend_mode_t;

HWC2_BLEND_MODE_NONE,不混合,源是什么样,输出就是什么样的。
HWC2_BLEND_MODE_PREMULTIPLIED,预乘,Dst需要做Alpha的处理。
HWC2_BLEND_MODE_COVERAGE,覆盖方式,源和Dst都需要做Alpha的处理。

  • 计算displayFrame并设置给hwcLayer
    displayFrame通过transform转换过的

  • 计算sourceCrop并设置给hwcLayer
    sourceCrop是上层传下来的,再和Dispaly,其他Layer的属性进行计算。

  • 设置Alpha值

  • 设置z-Order

  • 设置Layer的信息
    type和appId是Android Framework层创建SurfaceControl时设置的,可以搜搜"new SurfaceControl"就出来了。type是类型,比如ScreenshotSurfaceBackground等;appId是应用的进程号。

  • 设置变换信息transform

Layer信息的设置,是通过CommandBuffer的读写来完成的,比如,设置混合模式,最终是调的HwcHal的setLayerBlendMode方法。

Error HwcHal::setLayerBlendMode(Display display, Layer layer, int32_t mode)
{
    int32_t err = mDispatch.setLayerBlendMode(mDevice, display, layer, mode);
    return static_cast<Error>(err);
}

回到setUpHWComposer函数

3.设置每层Layer的Frame数据

void SurfaceFlinger::setUpHWComposer() {
     ... ...
    mat4 colorMatrix = mColorMatrix * computeSaturationMatrix() * mDaltonizer();

    // Set the per-frame data
    for (size_t displayId = 0; displayId < mDisplays.size(); ++displayId) {
        auto& displayDevice = mDisplays[displayId];
        const auto hwcId = displayDevice->getHwcDisplayId();

        ... ... // 设置每个Dispaly的颜色矩阵
        if (colorMatrix != mPreviousColorMatrix) {
            status_t result = getBE().mHwc->setColorTransform(hwcId, colorMatrix);
            ... ...
        }
        for (auto& layer : displayDevice->getVisibleLayersSortedByZ()) {
            if (layer->getForceClientComposition(hwcId)) {
                ALOGV("[%s] Requesting Client composition", layer->getName().string());
                layer->setCompositionType(hwcId, HWC2::Composition::Client);
                continue;
            }

            layer->setPerFrameData(displayDevice);
        }

        if (hasWideColorDisplay) {
            android_color_mode newColorMode;
            android_dataspace newDataSpace = HAL_DATASPACE_V0_SRGB;

            for (auto& layer : displayDevice->getVisibleLayersSortedByZ()) {
                newDataSpace = bestTargetDataSpace(layer->getDataSpace(), newDataSpace);
                ALOGV("layer: %s, dataspace: %s (%#x), newDataSpace: %s (%#x)",
                      layer->getName().string(), dataspaceDetails(layer->getDataSpace()).c_str(),
                      layer->getDataSpace(), dataspaceDetails(newDataSpace).c_str(), newDataSpace);
            }
            newColorMode = pickColorMode(newDataSpace);

            setActiveColorModeInternal(displayDevice, newColorMode);
        }
    }

    mPreviousColorMatrix = colorMatrix;

设置Frame数据时,主要做了以下几件事:

  • 设置每个Dispaly的颜色矩阵
    可以在开发这选项中设置,模拟颜色空间。其支持的transform主要有:
typedef enum {
    HAL_COLOR_TRANSFORM_IDENTITY = 0,
    HAL_COLOR_TRANSFORM_ARBITRARY_MATRIX = 1,
    HAL_COLOR_TRANSFORM_VALUE_INVERSE = 2,
    HAL_COLOR_TRANSFORM_GRAYSCALE = 3,
    HAL_COLOR_TRANSFORM_CORRECT_PROTANOPIA = 4,
    HAL_COLOR_TRANSFORM_CORRECT_DEUTERANOPIA = 5,
    HAL_COLOR_TRANSFORM_CORRECT_TRITANOPIA = 6,
} android_color_transform_t;

colorTransform主要是用以做颜色变换,以模拟或帮助色盲患者等。

  • 设置每一层Layer的显示数据
    setPerFrameData BufferLayer和ColorLayer的实现不一样。ColorLayer 的逻辑比较简单:
void ColorLayer::setPerFrameData(const sp<const DisplayDevice>& displayDevice) {
    ... ...
    // 设置可见区域
    auto error = hwcLayer->setVisibleRegion(visible);

    // 制定合成方式
    setCompositionType(hwcId, HWC2::Composition::SolidColor);

    half4 color = getColor();
    // 设置颜色
    error = hwcLayer->setColor({static_cast<uint8_t>(std::round(255.0f * color.r)),
                                static_cast<uint8_t>(std::round(255.0f * color.g)),
                                static_cast<uint8_t>(std::round(255.0f * color.b)), 255});
    ... ...// 去掉变换矩阵
    error = hwcLayer->setTransform(HWC2::Transform::None);
    ... ...
}

ColorLayer,主要有4个操纵:设置可见区域,这个前面已经就算好了,但是这里要确保它在Dispaly的视窗里;指定合成方式,默认采用SolidColor方式合成;设置颜色,指定该Layer的颜色,RGBA的格式,Alpha默认为255,全透;最后,ColorLayer不需要transform,去掉。

BufferLayer的setPerFrameData处理如下:

void BufferLayer::setPerFrameData(const sp<const DisplayDevice>& displayDevice) {
    // 设置可见区域
    auto& hwcLayer = hwcInfo.layer;
    auto error = hwcLayer->setVisibleRegion(visible);

    ... ... //设置Damage区域
    error = hwcLayer->setSurfaceDamage(surfaceDamageRegion);
    ... ...

    // Sideband layers处理,默认合成类型为Sideband
    if (getBE().compositionInfo.hwc.sidebandStream.get()) {
        setCompositionType(hwcId, HWC2::Composition::Sideband);
        // 制定Sideband流
        error = hwcLayer->setSidebandStream(getBE().compositionInfo.hwc.sidebandStream->handle());
        ... ...
        return; // Sideband layers处理完成后直接返回了。
    }

    // 如果是Cursor Layer,合成类似为Cursor,其他为Device
    if (mPotentialCursor) {
        ALOGV("[%s] Requesting Cursor composition", mName.string());
        setCompositionType(hwcId, HWC2::Composition::Cursor);
    } else {
        ALOGV("[%s] Requesting Device composition", mName.string());
        setCompositionType(hwcId, HWC2::Composition::Device);
    }

    // 设置数据空间dataspace
    error = hwcLayer->setDataspace(mCurrentState.dataSpace);
    if (error != HWC2::Error::None) {
        ALOGE("[%s] Failed to set dataspace %d: %s (%d)", mName.string(), mCurrentState.dataSpace,
              to_string(error).c_str(), static_cast<int32_t>(error));
    }

    // 获取GraphicBuffer
    uint32_t hwcSlot = 0;
    sp<GraphicBuffer> hwcBuffer;
    hwcInfo.bufferCache.getHwcBuffer(getBE().compositionInfo.mBufferSlot,
                                     getBE().compositionInfo.mBuffer, &hwcSlot, &hwcBuffer);

    //获取Fence
    auto acquireFence = mConsumer->getCurrentFence();
    // 设置Buffer
    error = hwcLayer->setBuffer(hwcSlot, hwcBuffer, acquireFence);
    if (error != HWC2::Error::None) {
        ALOGE("[%s] Failed to set buffer %p: %s (%d)", mName.string(),
              getBE().compositionInfo.mBuffer->handle, to_string(error).c_str(),
              static_cast<int32_t>(error));
    }
}

BufferLayer的处理比ColorLayer多,Sideband,Cursor和其他的UI图层都属于BufferLayer,每种类型Layer处理不台一样。这里比较难的是Fence的处理。Buffer也只是将Buffer的handle传给底层的HWC,并没有传Buffer里面的内容。

Error Composer::setLayerBuffer(Display display, Layer layer,
        uint32_t slot, const sp<GraphicBuffer>& buffer, int acquireFence)
{
    mWriter.selectDisplay(display);
    mWriter.selectLayer(layer);
    if (mIsUsingVrComposer && buffer.get()) {
        ... ...//VR
    }

    const native_handle_t* handle = nullptr;
    if (buffer.get()) {
        handle = buffer->getNativeBuffer()->handle;
    }

    mWriter.setLayerBuffer(slot, handle, acquireFence);
    return Error::NONE;
}

所以,每层Layer的数据,要么的Buffer,要么是固定的颜色。在处理每一层的数据时,还要处理widecolor。首先通过bestTargetDataSpace找到每层Layer的最佳DataSpace,然后再通过pickColorMode选取颜色模式,最后通过setActiveColorModeInternal函数设置。

void SurfaceFlinger::setActiveColorModeInternal(const sp<DisplayDevice>& hw,
        android_color_mode_t mode) {
    int32_t type = hw->getDisplayType();
    android_color_mode_t currentMode = hw->getActiveColorMode();

    ... ...

    hw->setActiveColorMode(mode);
    getHwComposer().setActiveColorMode(type, mode);
}

回到setUpHWComposer函数

4.prepareFrame准备数据

    for (size_t displayId = 0; displayId < mDisplays.size(); ++displayId) {
        auto& displayDevice = mDisplays[displayId];
        if (!displayDevice->isDisplayOn()) {
            continue;
        }

        status_t result = displayDevice->prepareFrame(*getBE().mHwc);
        ALOGE_IF(result != NO_ERROR, "prepareFrame for display %zd failed:"
                " %d (%s)", displayId, result, strerror(-result));
    }
}

Prepare流程,现在写的有点隐晦,以前都是直接在SurfaceFlinger中调的Prepare。现在通过DisplayDevice来完成,调用每个DisplayDevice的prepareFrame。

prepareFrame函数如下:

status_t DisplayDevice::prepareFrame(HWComposer& hwc) {
    status_t error = hwc.prepare(*this);
    if (error != NO_ERROR) {
        return error;
    }

    DisplaySurface::CompositionType compositionType;
    bool hasClient = hwc.hasClientComposition(mHwcDisplayId);
    bool hasDevice = hwc.hasDeviceComposition(mHwcDisplayId);
    if (hasClient && hasDevice) {
        compositionType = DisplaySurface::COMPOSITION_MIXED;
    } else if (hasClient) {
        compositionType = DisplaySurface::COMPOSITION_GLES;
    } else if (hasDevice) {
        compositionType = DisplaySurface::COMPOSITION_HWC;
    } else {
        // Nothing to do -- when turning the screen off we get a frame like
        // this. Call it a HWC frame since we won't be doing any GLES work but
        // will do a prepare/set cycle.
        compositionType = DisplaySurface::COMPOSITION_HWC;
    }
    return mDisplaySurface->prepareFrame(compositionType);
}

设置Layer数据时,已经指定每一层的合成方式,但是那是SurfaceFlinger的一厢情愿,还得看HWC接受不接受。HWComposer的prepare函数如下:

status_t HWComposer::prepare(DisplayDevice& displayDevice) {
    ATRACE_CALL();

    Mutex::Autolock _l(mDisplayLock);
    auto displayId = displayDevice.getHwcDisplayId();
    ... ...

    auto& displayData = mDisplayData[displayId];
    auto& hwcDisplay = displayData.hwcDisplay;
    if (!hwcDisplay->isConnected()) {
        return NO_ERROR;
    }

    ... ...
    if (!displayData.hasClientComposition) {
        sp<android::Fence> outPresentFence;
        uint32_t state = UINT32_MAX;
        error = hwcDisplay->presentOrValidate(&numTypes, &numRequests, &outPresentFence , &state);
        if (error != HWC2::Error::None && error != HWC2::Error::HasChanges) {
            ALOGV("skipValidate: Failed to Present or Validate");
            return UNKNOWN_ERROR;
        }
        if (state == 1) { //Present Succeeded.
            std::unordered_map<HWC2::Layer*, sp<Fence>> releaseFences;
            error = hwcDisplay->getReleaseFences(&releaseFences);
            displayData.releaseFences = std::move(releaseFences);
            displayData.lastPresentFence = outPresentFence;
            displayData.validateWasSkipped = true;
            displayData.presentError = error;
            return NO_ERROR;
        }
        // Present failed but Validate ran.
    } else {
        error = hwcDisplay->validate(&numTypes, &numRequests);
    }
    ... ...

    std::unordered_map<HWC2::Layer*, HWC2::Composition> changedTypes;
    changedTypes.reserve(numTypes);
    error = hwcDisplay->getChangedCompositionTypes(&changedTypes);
    ... ...

    displayData.displayRequests = static_cast<HWC2::DisplayRequest>(0);
    std::unordered_map<HWC2::Layer*, HWC2::LayerRequest> layerRequests;
    layerRequests.reserve(numRequests);
    error = hwcDisplay->getRequests(&displayData.displayRequests,
            &layerRequests);
    if (error != HWC2::Error::None) {
        ALOGE("prepare: getRequests failed on display %d: %s (%d)", displayId,
                to_string(error).c_str(), static_cast<int32_t>(error));
        return BAD_INDEX;
    }

    displayData.hasClientComposition = false;
    displayData.hasDeviceComposition = false;
    for (auto& layer : displayDevice.getVisibleLayersSortedByZ()) {
        auto hwcLayer = layer->getHwcLayer(displayId);

        if (changedTypes.count(hwcLayer) != 0) {
            // We pass false so we only update our state and don't call back
            // into the HWC device
            validateChange(layer->getCompositionType(displayId),
                    changedTypes[hwcLayer]);
            layer->setCompositionType(displayId, changedTypes[hwcLayer], false);
        }

        switch (layer->getCompositionType(displayId)) {
            ... ...
        }

        if (layerRequests.count(hwcLayer) != 0 &&
                layerRequests[hwcLayer] ==
                        HWC2::LayerRequest::ClearClientTarget) {
            layer->setClearClientTarget(displayId, true);
        } else {
            if (layerRequests.count(hwcLayer) != 0) {
                ALOGE("prepare: Unknown layer request: %s",
                        to_string(layerRequests[hwcLayer]).c_str());
            }
            layer->setClearClientTarget(displayId, false);
        }
    }

    error = hwcDisplay->acceptChanges();
    ... ...

    return NO_ERROR;
}

prepare流程如下:

  • 首先尝试Prepare和Present一次处理完成
    如果SurfaceFlinger没有指定得有Client端合成hasClientComposition为false,首先通过presentOrValidate接口尝试直接present,如果HWC不能直接显示,再执行validate操纵,这时的流程和validate是类似的。如果成功,那么此次数据就显示了,不用再 继续后续的处理。

  • validate刷新

Error Display::validate(uint32_t* outNumTypes, uint32_t* outNumRequests)
{
    uint32_t numTypes = 0;
    uint32_t numRequests = 0;
    auto intError = mComposer.validateDisplay(mId, &numTypes, &numRequests);
    auto error = static_cast<Error>(intError);
    if (error != Error::None && error != Error::HasChanges) {
        return error;
    }

    *outNumTypes = numTypes;
    *outNumRequests = numRequests;
    return error;
}

注意Composer的 validateDisplay 函数,和其他Composer的区别:

Error Composer::validateDisplay(Display display, uint32_t* outNumTypes,
        uint32_t* outNumRequests)
{
    mWriter.selectDisplay(display);
    mWriter.validateDisplay();

    Error error = execute();
    if (error != Error::NONE) {
        return error;
    }

    mReader.hasChanges(display, outNumTypes, outNumRequests);

    return Error::NONE;
}

validateDisplay 也是通过CommandWriter写Buffer的方式调用到HWC中的,但是这里多了一个execute函数。其实,validateDisplay之前的通过,Buffer命令的调用,都还没有真正的调到HWC中,只是将命令写到了Buffer中。这里的execute才真正的调用,这里将触发HWC的服务端去解析Buffer命令,再分别去调HWC中对应的实现函数。

比如设置 z-order的解析如下:

bool ComposerClient::CommandReader::parseSetLayerZOrder(uint16_t length)
{
    if (length != CommandWriterBase::kSetLayerZOrderLength) {
        return false;
    }

    auto err = mHal.setLayerZOrder(mDisplay, mLayer, read());
    if (err != Error::NONE) {
        mWriter.setError(getCommandLoc(), err);
    }

    return true;
}
  • 获取HWC的validate结果
    如果SurfaceFlinger指定的合成方式HWC不能处理,通过getChangedCompositionTypes函数获取到HWC对合成方式的修改,保存在 changedTypes 中。获取LayerRequest,保存在layerRequests中。layerRequests和changedTypes都是以HWC2::Layer作为key的map。

  • 修改合成方式
    如果合成方式HWC不接受,SurfaceFlinger中修改根据HWC的反馈进行修改,也就是changedTypes中的Layer进行修改。修改的函数为setCompositionType,注意这里的callIntoHwc参数为false。

  • 响应layerRequests
    layerRequests主要是决定是否需要清楚Client端的Target,也就是Client的合成结果,留意clearClientTarget,看看后续的流程是怎么处理的。

void Layer::setClearClientTarget(int32_t hwcId, bool clear) {
    if (getBE().mHwcLayers.count(hwcId) == 0) {
        ALOGE("setClearClientTarget called without a valid HWC layer");
        return;
    }
    getBE().mHwcLayers[hwcId].clearClientTarget = clear;
}
  • 最后接受修改
    通过HWC,SurfaceFlinger接受修改。
Error Display::acceptChanges()
{
    auto intError = mComposer.acceptDisplayChanges(mId);
    return static_cast<Error>(intError);
}

到此setUpHWComposer结束,此时,我们需要显示的数据已经送到HWC,且每一层Layer的合成方式已经确定。如果是HWC能支持更新和显示同时完成,那么此时数据已经开始显示。

回到handleMessageRefresh函数,接下来是doDebugFlashRegions。doDebugFlashRegions只是一个debug的功能,其目前就是更新的区域不停的闪烁,收mDebugRegion的控制。

void SurfaceFlinger::doDebugFlashRegions()
{
    // is debugging enabled
    if (CC_LIKELY(!mDebugRegion))
        return;

}

接下来的doTracing也是一个debug的辅助功能。

void SurfaceFlinger::doTracing(const char* where) {
    ATRACE_CALL();
    ATRACE_NAME(where);
    if (CC_UNLIKELY(mTracing.isEnabled())) {
        mTracing.traceLayers(where, dumpProtoInfo(LayerVector::StateSet::Drawing));
    }
}

合成处理 doComposition

如果present和validate没有一起完成,那么此时我们需要显示的数据已经送到HWC,且每一层Layer的合成方式已经确定。接下来的合成处理流程在doComposition中完成。

doComposition函数:

void SurfaceFlinger::doComposition() {
    ATRACE_CALL();
    ALOGV("doComposition");

    const bool repaintEverything = android_atomic_and(0, &mRepaintEverything);
    for (size_t dpy=0 ; dpy<mDisplays.size() ; dpy++) {
        const sp<DisplayDevice>& hw(mDisplays[dpy]);
        if (hw->isDisplayOn()) {
            // transform the dirty region into this screen's coordinate space
            const Region dirtyRegion(hw->getDirtyRegion(repaintEverything));

            // repaint the framebuffer (if needed)
            doDisplayComposition(hw, dirtyRegion);

            hw->dirtyRegion.clear();
            hw->flip();
        }
    }
    postFramebuffer();
}

合成处理也是每个Display各自 进行的,合成处理主要步骤如下:

  1. 获取脏区域
    在前面重构Layer时,Display的脏区域dirtyRegion已经计算出来。如果是重画,mRepaintEverything为true,那么脏区域就是整个屏幕的大小。
Region DisplayDevice::getDirtyRegion(bool repaintEverything) const {
    Region dirty;
    if (repaintEverything) {
        dirty.set(getBounds());
    } else {
        const Transform& planeTransform(mGlobalTransform);
        dirty = planeTransform.transform(this->dirtyRegion);
        dirty.andSelf(getBounds());
    }
    return dirty;
}

2.Display合成处理
doDisplayComposition函数如下:

void SurfaceFlinger::doDisplayComposition(
        const sp<const DisplayDevice>& displayDevice,
        const Region& inDirtyRegion)
{
    // 需要HWC处理,或者脏区域不为空
    bool isHwcDisplay = displayDevice->getHwcDisplayId() >= 0;
    if (!isHwcDisplay && inDirtyRegion.isEmpty()) {
        ALOGV("Skipping display composition");
        return;
    }

    ALOGV("doDisplayComposition");
    if (!doComposeSurfaces(displayDevice)) return;

    // swap buffers (presentation)
    displayDevice->swapBuffers(getHwComposer());
}

合成操纵主要在doComposeSurfaces函数中完成,合成方式,主要就两种,一种Client端用GPU合成;另外一种,Device端合成,用的是HWC硬件。doComposeSurfaces主要是处理Client端合成,Client通过RenderEngine用GPU来进行合成。

doComposeSurfaces 函数如下,我们分段看:

  • RenderEngine 的初始化
bool SurfaceFlinger::doComposeSurfaces(const sp<const DisplayDevice>& displayDevice)
{
    ... ...

    const Region bounds(displayDevice->bounds());
    const DisplayRenderArea renderArea(displayDevice);
    const auto hwcId = displayDevice->getHwcDisplayId();

    mat4 oldColorMatrix;
    const bool applyColorMatrix = !getBE().mHwc->hasDeviceComposition(hwcId) &&
            !getBE().mHwc->hasCapability(HWC2::Capability::SkipClientColorTransform);
    if (applyColorMatrix) {
        mat4 colorMatrix = mColorMatrix * mDaltonizer();
        oldColorMatrix = getRenderEngine().setupColorTransform(colorMatrix);
    }

    bool hasClientComposition = getBE().mHwc->hasClientComposition(hwcId);
    if (hasClientComposition) {
        ALOGV("hasClientComposition");

        getBE().mRenderEngine->setWideColor(
                displayDevice->getWideColorSupport() && !mForceNativeColorMode);
        getBE().mRenderEngine->setColorMode(mForceNativeColorMode ?
                HAL_COLOR_MODE_NATIVE : displayDevice->getActiveColorMode());
        if (!displayDevice->makeCurrent()) {
            ... ...
        }

        const bool hasDeviceComposition = getBE().mHwc->hasDeviceComposition(hwcId);
        if (hasDeviceComposition) {
            getBE().mRenderEngine->clearWithColor(0, 0, 0, 0);
        } else {
            const Region letterbox(bounds.subtract(displayDevice->getScissor()));

            // compute the area to clear
            Region region(displayDevice->undefinedRegion.merge(letterbox));

            // screen is already cleared here
            if (!region.isEmpty()) {
                // can happen with SurfaceView
                drawWormhole(displayDevice, region);
            }
        }

        if (displayDevice->getDisplayType() != DisplayDevice::DISPLAY_PRIMARY) {

            const Rect& bounds(displayDevice->getBounds());
            const Rect& scissor(displayDevice->getScissor());
            if (scissor != bounds) {
                const uint32_t height = displayDevice->getHeight();
                getBE().mRenderEngine->setScissor(scissor.left, height - scissor.bottom,
                        scissor.getWidth(), scissor.getHeight());
            }
        }
    }

RenderEngine的初始化包括:

  • 指定颜色矩阵 setupColorTransform
  • 指定是否用WideColor setWideColor
  • 指定颜色模式 setColorMode
  • 设置FBTarget Surface,视窗,投影矩阵等
    这个过程在DisplayDevice的makeCurrent中完成:
bool DisplayDevice::makeCurrent() const {
    bool success = mFlinger->getRenderEngine().setCurrentSurface(mSurface);
    setViewportAndProjection();
    return success;
}
  • FBTarget 处理背景
    如果是混合模式,也就是hasClientComposition和hasDeviceComposition,先清掉FBTarget背景。一般情况,合成很少采用这种方式。基本都是通过drawWormhole将屏幕填充为RGBA_0000。

  • 设置剪切区 setScissor
    对于非主屏,通过setScissor设置Display的剪切区

到此,初始化完成~

  • 将Client端的Layer渲染到FBTarget
bool SurfaceFlinger::doComposeSurfaces(const sp<const DisplayDevice>& displayDevice)
{
    ... ...

    const Transform& displayTransform = displayDevice->getTransform();
    if (hwcId >= 0) {
        // hwcId >=0 我们使用HWC
        bool firstLayer = true;
        for (auto& layer : displayDevice->getVisibleLayersSortedByZ()) {
            const Region clip(bounds.intersect(
                    displayTransform.transform(layer->visibleRegion)));

            if (!clip.isEmpty()) {
                switch (layer->getCompositionType(hwcId)) {
                    case HWC2::Composition::Cursor:
                    case HWC2::Composition::Device:
                    case HWC2::Composition::Sideband:
                    case HWC2::Composition::SolidColor: {
                        const Layer::State& state(layer->getDrawingState());
                        if (layer->getClearClientTarget(hwcId) && !firstLayer &&
                                layer->isOpaque(state) && (state.color.a == 1.0f)
                                && hasClientComposition) {
                            // never clear the very first layer since we're
                            // guaranteed the FB is already cleared
                            layer->clearWithOpenGL(renderArea);
                        }
                        break;
                    }
                    case HWC2::Composition::Client: {
                        layer->draw(renderArea, clip);
                        break;
                    }
                    default:
                        break;
                }
            } else {
                ALOGV("  Skipping for empty clip");
            }
            firstLayer = false;
        }
    } else {
        // we're not using h/w composer
        for (auto& layer : displayDevice->getVisibleLayersSortedByZ()) {
            const Region clip(bounds.intersect(
                    displayTransform.transform(layer->visibleRegion)));
            if (!clip.isEmpty()) {
                layer->draw(renderArea, clip);
            }
        }
    }

    if (applyColorMatrix) {
        getRenderEngine().setupColorTransform(oldColorMatrix);
    }

    // disable scissor at the end of the frame
    getBE().mRenderEngine->disableScissor();
    return true;
}

hwcId >= 0说明我们用到了HWC,大多数情况都会走到这里。对于很多VirtualDisplay的情况,hwcId为-1。

用到hwc时,首先计算每一层Layer的可见区域在Display中的区域clip,如果Layer是Client合成,那么直接调layer->draw,将Layer中clip区域绘制到FBTarget上。如果不是Client合成,但是有其他Layer是Client合成时,需要将Layer在 FBTarget中对应的区域清理掉clearWithOpenGL,清理掉的区域HWC合成。最终,FBTarget的内容和HWC中的内容再合成为最后的显示数据。

没有用到hwc时,直接调layer->draw,将Layer中clip区域绘制到FBTarget上。

到此,doComposeSurfaces完成,这里主要是处理和Client端合成相关的流程~

3.Display 交换Buffer

void DisplayDevice::swapBuffers(HWComposer& hwc) const {
    if (hwc.hasClientComposition(mHwcDisplayId)) {
        mSurface.swapBuffers();
    }

    status_t result = mDisplaySurface->advanceFrame();
    if (result != NO_ERROR) {
        ALOGE("[%s] failed pushing new frame to HWC: %d",
                mDisplayName.string(), result);
    }
}

如果有Client合成,调eglSwapBuffers交换Buffer

void Surface::swapBuffers() const {
    if (!eglSwapBuffers(mEGLDisplay, mEGLSurface)) {
        ... ...
    }
}

mDisplaySurface的advanceFrame方法,虚显用的VirtualDisplaySurface,非虚显用的FramebufferSurface。advanceFrame获取 FBTarget 的数据,我们看非虚显:

status_t FramebufferSurface::advanceFrame() {
    uint32_t slot = 0;
    sp<GraphicBuffer> buf;
    sp<Fence> acquireFence(Fence::NO_FENCE);
    android_dataspace_t dataspace = HAL_DATASPACE_UNKNOWN;
    status_t result = nextBuffer(slot, buf, acquireFence, dataspace);
    mDataSpace = dataspace;
    if (result != NO_ERROR) {
        ALOGE("error latching next FramebufferSurface buffer: %s (%d)",
                strerror(-result), result);
    }
    return result;
}

主要在nextBuffer函数中完成:

status_t FramebufferSurface::nextBuffer(uint32_t& outSlot,
        sp<GraphicBuffer>& outBuffer, sp<Fence>& outFence,
        android_dataspace_t& outDataspace) {
    Mutex::Autolock lock(mMutex);

    BufferItem item;
    status_t err = acquireBufferLocked(&item, 0);
    ... ...
    if (mCurrentBufferSlot != BufferQueue::INVALID_BUFFER_SLOT &&
        item.mSlot != mCurrentBufferSlot) {
        mHasPendingRelease = true;
        mPreviousBufferSlot = mCurrentBufferSlot;
        mPreviousBuffer = mCurrentBuffer;
    }
    mCurrentBufferSlot = item.mSlot;
    mCurrentBuffer = mSlots[mCurrentBufferSlot].mGraphicBuffer;
    mCurrentFence = item.mFence;

    outFence = item.mFence;
    mHwcBufferCache.getHwcBuffer(mCurrentBufferSlot, mCurrentBuffer,
            &outSlot, &outBuffer);
    outDataspace = item.mDataSpace;
    status_t result =
            mHwc.setClientTarget(mDisplayType, outSlot, outFence, outBuffer, outDataspace);
    ... ...
}

nextBuffer函数中:

  • 获取一个Buffer
    如果是Client合成,swapBuffer时,将调用queueBuffer,queue到FrameBufferSurface的BufferQueue中。这里的acquireBufferLocked 将从BufferQueue中获取一个Buffer。

  • 替换Buffer
    当前Buffer的序号mCurrentBufferSlot,当前Buffer mCurrentBuffer,对应的Fence mCurrentFence;如果新获取到的Buffer不一样,释放旧的。Buffer都被cache到mHwcBufferCache中。

  • 将 FBTarget 设置给HWC
    关键代码mHwc.setClientTarget

status_t HWComposer::setClientTarget(int32_t displayId, uint32_t slot,
        const sp<Fence>& acquireFence, const sp<GraphicBuffer>& target,
        android_dataspace_t dataspace) {
    ... ...

    auto& hwcDisplay = mDisplayData[displayId].hwcDisplay;
    auto error = hwcDisplay->setClientTarget(slot, target, acquireFence, dataspace);
    if (error != HWC2::Error::None) {
        ALOGE("Failed to set client target for display %d: %s (%d)", displayId,
                to_string(error).c_str(), static_cast<int32_t>(error));
        return BAD_VALUE;
    }

    return NO_ERROR;
}

FBTarget 也是通过Command Buffer的方式传到HWC中的。在hwc1.x的版本中,在创建工作列表时也为FBTarget创建了一个Layer,HWC2版本直接传递FBTarget。

回到doComposition函数中,DisplayDevice的flip函数,将记录flip的次数mPageFlipCount。

void DisplayDevice::flip() const
{
    mFlinger->getRenderEngine().checkErrors();
    mPageFlipCount++;
}

4.提交Framebuffer
到此,我们显示的数据成什么样了?需要Client合成的,已经合成完了,合成后的结果FBTarget已传给HWC。需要Device合成的数据之前也提交给HWC了。但是数据还没有最终合成显示出来。postFramebuffer 函数就是告诉HWC开始做最后的合成了。

postFramebuffer函数如下:

void SurfaceFlinger::postFramebuffer()
{
    ... ...

    for (size_t displayId = 0; displayId < mDisplays.size(); ++displayId) {
        auto& displayDevice = mDisplays[displayId];
        if (!displayDevice->isDisplayOn()) {
            continue;
        }
        const auto hwcId = displayDevice->getHwcDisplayId();
        if (hwcId >= 0) {
            getBE().mHwc->presentAndGetReleaseFences(hwcId);
        }
        displayDevice->onSwapBuffersCompleted();
        displayDevice->makeCurrent();
        for (auto& layer : displayDevice->getVisibleLayersSortedByZ()) {
            auto hwcLayer = layer->getHwcLayer(hwcId);
            sp<Fence> releaseFence = getBE().mHwc->getLayerReleaseFence(hwcId, hwcLayer);

            if (layer->getCompositionType(hwcId) == HWC2::Composition::Client) {
                releaseFence = Fence::merge("LayerRelease", releaseFence,
                        displayDevice->getClientTargetAcquireFence());
            }

            layer->onLayerDisplayed(releaseFence);
        }

        if (!displayDevice->getLayersNeedingFences().isEmpty()) {
            sp<Fence> presentFence = getBE().mHwc->getPresentFence(hwcId);
            for (auto& layer : displayDevice->getLayersNeedingFences()) {
                layer->onLayerDisplayed(presentFence);
            }
        }

        if (hwcId >= 0) {
            getBE().mHwc->clearReleaseFences(hwcId);
        }
    }

    mLastSwapBufferTime = systemTime() - now;
    mDebugInSwapBuffers = 0;

    // |mStateLock| not needed as we are on the main thread
    uint32_t flipCount = getDefaultDisplayDeviceLocked()->getPageFlipCount();
    if (flipCount % LOG_FRAME_STATS_PERIOD == 0) {
        logFrameStats();
    }
}

postFramebuffer的流程如下:

  • 通过presentAndGetReleaseFences显示获取releaseFence
status_t HWComposer::presentAndGetReleaseFences(int32_t displayId) {
    ... ...

    auto& displayData = mDisplayData[displayId];
    auto& hwcDisplay = displayData.hwcDisplay;

    if (displayData.validateWasSkipped) {
        // explicitly flush all pending commands
        auto error = mHwcDevice->flushCommands();
        ... ...
    }

    auto error = hwcDisplay->present(&displayData.lastPresentFence);
    if (error != HWC2::Error::None) {
        ... ...
    }

    std::unordered_map<HWC2::Layer*, sp<Fence>> releaseFences;
    error = hwcDisplay->getReleaseFences(&releaseFences);
    if (error != HWC2::Error::None) {
        ... ...
    }

    displayData.releaseFences = std::move(releaseFences);

    return NO_ERROR;
}

如果之前Prepare过程中,presentOrValidate成功,validateWasSkipped为 true,那么直接刷掉command Buffer中的命令,让你执行。就没有后续的处理了,presentOrValidate成功,是没有Client合成的,也就没有所谓的FBTarget。

如果之前presentOrValidate没有成功,很有可能是需要Client端做合成的,也就是present没有完成,那么这里需要走present的流程。

present操纵也是先写到Command Buffer中。最后调的execute。

Error Composer::presentDisplay(Display display, int* outPresentFence)
{
    mWriter.selectDisplay(display);
    mWriter.presentDisplay();

    Error error = execute();
    if (error != Error::NONE) {
        return error;
    }

    mReader.takePresentFence(display, outPresentFence);

    return Error::NONE;
}

设置FBTarget时,只是写到Buffer中,execute时,设置FBTarget的操纵才一起生效,设置到了Server端。Server再进行最后的合成。

present完成后,通过getReleaseFences获取releaseFence,保存在displayData中。注意这里的release是每个Layer的release fence,这是8.0之前的版本没有的流程,之前的releasefence只有一个,所以Layer只有一个。而present时的lastPresentFence就是FBTarget的releasefence。

回到postFramebuffer函数。

  • DisplayDevice处理FBTarget
    释放上一帧的Buffer
void DisplayDevice::onSwapBuffersCompleted() const {
    mDisplaySurface->onFrameCommitted();
}

主要在onFrameCommitted函数中完成:

void FramebufferSurface::onFrameCommitted() {
    if (mHasPendingRelease) {
        sp<Fence> fence = mHwc.getPresentFence(mDisplayType);
        if (fence->isValid()) {
            status_t result = addReleaseFence(mPreviousBufferSlot,
                    mPreviousBuffer, fence);
            ... ...
        }
        status_t result = releaseBufferLocked(mPreviousBufferSlot, mPreviousBuffer);
        ... ...

        mPreviousBuffer.clear();
        mHasPendingRelease = false;
    }
}

makeCurrent为新的合成做准备

bool DisplayDevice::makeCurrent() const {
    bool success = mFlinger->getRenderEngine().setCurrentSurface(mSurface);
    setViewportAndProjection();
    return success;
}
  • 给每一层Layer设置releaseFence
void BufferLayer::onLayerDisplayed(const sp<Fence>& releaseFence) {
    mConsumer->setReleaseFence(releaseFence);
}

如果Layer需要Fence,给它presentFence,也就是FBTarget的Fence。最后清除HWComposer的mDisplayData中的releaseFence,因为他们已经传给Layer去了。

到此合成处理结束~

合成后处理 postComposition

void SurfaceFlinger::postComposition(nsecs_t refreshStartTime)
{
    // Release any buffers which were replaced this frame
    nsecs_t dequeueReadyTime = systemTime();
    for (auto& layer : mLayersWithQueuedFrames) {
        layer->releasePendingBuffer(dequeueReadyTime);
    }

    // 处理Timeline
    ... ...

    // 记录Buffer状态
    mDrawingState.traverseInZOrder([&](Layer* layer) {
        bool frameLatched = layer->onPostComposition(glCompositionDoneFenceTime,
                presentFenceTime, compositorTiming);
        if (frameLatched) {
            recordBufferingStats(layer->getName().string(),
                    layer->getOccupancyHistory(false));
        }
    });

    // Vsync的同步
    if (presentFenceTime->isValid()) {
        if (mPrimaryDispSync.addPresentFence(presentFenceTime)) {
            enableHardwareVsync();
        } else {
            disableHardwareVsync(false);
        }
    }

    if (!hasSyncFramework) {
        if (hw->isDisplayOn()) {
            enableHardwareVsync();
        }
    }

    // 动画合成处理
    if (mAnimCompositionPending) {
        mAnimCompositionPending = false;

        if (presentFenceTime->isValid()) {
            mAnimFrameTracker.setActualPresentFence(
                    std::move(presentFenceTime));
        } else {
            // The HWC doesn't support present fences, so use the refresh
            // timestamp instead.
            nsecs_t presentTime =
                    getBE().mHwc->getRefreshTimestamp(HWC_DISPLAY_PRIMARY);
            mAnimFrameTracker.setActualPresentTime(presentTime);
        }
        mAnimFrameTracker.advanceFrame();
    }

    // 时间记录
}

中主要做了下列几件事:

  • 释放待释放的Buffer
    这一帧合成完成后,将被替代的Buffer释放掉~
void BufferLayer::releasePendingBuffer(nsecs_t dequeueReadyTime) {
    if (!mConsumer->releasePendingBuffer()) {
        return;
    }

    auto releaseFenceTime =
            std::make_shared<FenceTime>(mConsumer->getPrevFinalReleaseFence());
    mReleaseTimeline.updateSignalTimes();
    mReleaseTimeline.push(releaseFenceTime);

    Mutex::Autolock lock(mFrameEventHistoryMutex);
    if (mPreviousFrameNumber != 0) {
        mFrameEventHistory.addRelease(mPreviousFrameNumber, dequeueReadyTime,
                                      std::move(releaseFenceTime));
    }
}
  • 处理Timeline

  • 记录Buffer状态

void SurfaceFlinger::recordBufferingStats(const char* layerName,
        std::vector<OccupancyTracker::Segment>&& history) {
    Mutex::Autolock lock(getBE().mBufferingStatsMutex);
    auto& stats = getBE().mBufferingStats[layerName];
    for (const auto& segment : history) {
        if (!segment.usedThirdBuffer) {
            stats.twoBufferTime += segment.totalTime;
        }
        if (segment.occupancyAverage < 1.0f) {
            stats.doubleBufferedTime += segment.totalTime;
        } else if (segment.occupancyAverage < 2.0f) {
            stats.tripleBufferedTime += segment.totalTime;
        }
        ++stats.numSegments;
        stats.totalTime += segment.totalTime;
    }
}
  • Vsync的同步
    平常我们用的Vsync都是mPrimaryDispSync分发出来的,并不是每一次都是从底层硬件上报的,所以mPrimaryDispSync需要和底层的硬件Vsync保持同步

  • 动画合成处理

  • 处理时间的记录

到此一次合成处理完成,REFRESH处理完成。下一个Vsync到来时,新的一次合成又将开始。

Client合成

硬件HWC合成是Vendor实现的,各个Vendor不一样。而Client合成是Android自带的,我们接下来就来看看Android的Client端的合成。

Client端合成,本质是采用GPU进程合成,SurfaceFlinger中封装了RenderEngine进行具体的实现,相关的代码在如下位置:

frameworks/native/services/surfaceflinger/RenderEngine

我们来看看看相关的类:


RenderEngine
  • RenderEngine 是对GPU渲染的封装,包括了 EGLDisplay,EGLContext, EGLConfig,EGLSurface。注意每个Display的EGLSurface不是同一个,各自有各自的EGLSurface。

  • GLES20RenderEngine 继承RenderEngine,是GELS的2.0版本实现。其Program采用ProgramCache进行cache。状态用Description进描述。

  • 每个BufferLayer 都有专门的Texture进行纹理的描述,GLES20RenderEngine 支持纹理贴图。合成时,将GraphicBuffer转换为纹理,进行混合。

下面我们来看具体的流程,Client端GPU合成相关的流程如下:

1.创建 RenderEngine
RenderEngine 是在SurfaceFlinger初始化时,创建的。

void SurfaceFlinger::init() {
    ... ...
    getBE().mRenderEngine = RenderEngine::create(HAL_PIXEL_FORMAT_RGBA_8888,
            hasWideColorDisplay ? RenderEngine::WIDE_COLOR_SUPPORT : 0);

create函数如下:

std::unique_ptr<RenderEngine> RenderEngine::create(int hwcFormat, uint32_t featureFlags) {
    // 初始化EGLDisplay
    EGLDisplay display = eglGetDisplay(EGL_DEFAULT_DISPLAY);
    if (!eglInitialize(display, NULL, NULL)) {
        LOG_ALWAYS_FATAL("failed to initialize EGL");
    }

    // GLExtensions处理

    EGLint renderableType = 0;
    if (config == EGL_NO_CONFIG) {
        renderableType = EGL_OPENGL_ES2_BIT;
    } else if (!eglGetConfigAttrib(display, config, EGL_RENDERABLE_TYPE, &renderableType)) {
        LOG_ALWAYS_FATAL("can't query EGLConfig RENDERABLE_TYPE");
    }
    EGLint contextClientVersion = 0;
    if (renderableType & EGL_OPENGL_ES2_BIT) {
        contextClientVersion = 2;
    } else if (renderableType & EGL_OPENGL_ES_BIT) {
        contextClientVersion = 1;
    } else {
        LOG_ALWAYS_FATAL("no supported EGL_RENDERABLE_TYPEs");
    }

    // 初始化Attributes
    std::vector<EGLint> contextAttributes;
    ... ...

    // 创建EGLContext
    EGLContext ctxt = eglCreateContext(display, config, NULL, contextAttributes.data());

    ... ...

    // 创建PBuffer
    EGLint attribs[] = {EGL_WIDTH, 1, EGL_HEIGHT, 1, EGL_NONE, EGL_NONE};
    EGLSurface dummy = eglCreatePbufferSurface(display, dummyConfig, attribs);

    // 
    EGLBoolean success = eglMakeCurrent(display, dummy, dummy, ctxt);
    LOG_ALWAYS_FATAL_IF(!success, "can't make dummy pbuffer current");

    ... ...

    std::unique_ptr<RenderEngine> engine;
    switch (version) {
        ... ...
        case GLES_VERSION_3_0:
            engine = std::make_unique<GLES20RenderEngine>(featureFlags);
            break;
    }
    // 设置EGL信息
    engine->setEGLHandles(display, config, ctxt);

    ... ...

    eglMakeCurrent(display, EGL_NO_SURFACE, EGL_NO_SURFACE, EGL_NO_CONTEXT);
    eglDestroySurface(display, dummy);

    return engine;
}

RenderEngine的初始化过程,就是GPU渲染初始化的过程,做过OpenGL编程的同学来说,小case。其大概的流程如下:

  • 创建 EGLDisplay
    eglGetDisplay

  • 初始化 EGLDisplay
    eglInitialize

  • 选择 EGLConfig
    chooseEglConfig

  • 获取renderableType
    eglGetConfigAttrib

  • 初始化Context属性
    contextAttributes

  • 创建EGLContext
    eglCreateContext

  • 创建 PBuffer
    eglCreatePbufferSurface

  • MakeCurrent
    eglMakeCurrent这是为虚拟的PBuffercheck状态。

  • 创建RenderEngine
    这里,目前值支持GELS2.0,对应的Render GLES20RenderEngine

  • 设置设置EGL信息
    将创建的EGL对象设置到我们创建的GLES20RenderEngine中。

void RenderEngine::setEGLHandles(EGLDisplay display, EGLConfig config, EGLContext ctxt) {
    mEGLDisplay = display;
    mEGLConfig = config;
    mEGLContext = ctxt;
}

2.创建Surface FBTarget
在RenderEngine创建时,初始化了EGLDisplaym,EGLConfig,EGLContext。这些都是所有Display共用的,但是Surface每个Display的是自己的。

在DisplayDevice创建时,创建对应的Surface

DisplayDevice::DisplayDevice(
       ... ...
      mSurface{flinger->getRenderEngine()},
      ... ...
{
    // clang-format on
    Surface* surface;
    mNativeWindow = surface = new Surface(producer, false);
    ANativeWindow* const window = mNativeWindow.get();

    ... ...
    mSurface.setCritical(mType == DisplayDevice::DISPLAY_PRIMARY);
    mSurface.setAsync(mType >= DisplayDevice::DISPLAY_VIRTUAL);
    mSurface.setNativeWindow(window);
    mDisplayWidth = mSurface.queryWidth();
    mDisplayHeight = mSurface.queryHeight();

    ... ...

    if (useTripleFramebuffer) {
        surface->allocateBuffers();
    }
}

注意mSurface.setNativeWindow,通过ANativeWindow,Surface就和DisplayDevice的BufferQueue建立了联系。

void Surface::setNativeWindow(ANativeWindow* window) {
    if (mEGLSurface != EGL_NO_SURFACE) {
        eglDestroySurface(mEGLDisplay, mEGLSurface);
        mEGLSurface = EGL_NO_SURFACE;
    }

    mWindow = window;
    if (mWindow) {
        mEGLSurface = eglCreateWindowSurface(mEGLDisplay, mEGLConfig, mWindow, nullptr);
    }
}

创建的EGLSurface mEGLSurface和nativewindow mWindow 关联。这个GPU就可以通过nativewindow,从BufferQueue中dequeue Buffer进行渲染,swapBuffer时,也queue到Bufferqueu中。这里的ANativeWindow,本质就是 FBTarget。

  1. 创建Texture
    BufferLayer创建时,创建Texture:
BufferLayer::BufferLayer(SurfaceFlinger* flinger, const sp<Client>& client, const String8& name,
                         uint32_t w, uint32_t h, uint32_t flags)
      : Layer(flinger, client, name, w, h, flags),
        ... ...

    mFlinger->getRenderEngine().genTextures(1, &mTextureName);
    mTexture.init(Texture::TEXTURE_EXTERNAL, mTextureName);
}

通过glGenTextures函数创建Texture。

void RenderEngine::genTextures(size_t count, uint32_t* names) {
    glGenTextures(count, names);
}

且在创建BufferLayerConsumer时,传到了Consumer中,对应的值为mTexName。

glGenTextures生成的Texture,在BufferLayer中,保存在mTexture中。

4.开始合成 doComposeSurfaces
合成是在SurfaceFlinger的doComposeSurfaces中进的,首先先makeCurrent。每个Display有自己的Surface,所以,每个Display做具体合成时,需要给RenderEngine指定Surface,视窗,投影矩阵等,告诉RenderEngine合成到哪个Surface上。

bool DisplayDevice::makeCurrent() const {
    bool success = mFlinger->getRenderEngine().setCurrentSurface(mSurface);
    setViewportAndProjection();
    return success;
}

setCurrentSurface函数如下:

bool RenderEngine::setCurrentSurface(const RE::Surface& surface) {
    bool success = true;
    EGLSurface eglSurface = surface.getEGLSurface();
    if (eglSurface != eglGetCurrentSurface(EGL_DRAW)) {
        success = eglMakeCurrent(mEGLDisplay, eglSurface, eglSurface, mEGLContext) == EGL_TRUE;
        if (success && surface.getAsync()) {
            eglSwapInterval(mEGLDisplay, 0);
        }
    }

    return success;
}

GPU不支持多线程,所以需要通过eglMakeCurrent切换GPU的工作线程,eglMakeCurrent后,GPU将处理我们当前线程的OpenGL绘图操纵。

5.Layer的合成
合成时,每个Display的每个Layer都合成到Display对应的Surface上。主要是在Layer的draw方法中完成:

void Layer::draw(const RenderArea& renderArea, const Region& clip) const {
    onDraw(renderArea, clip, false);
}

BufferLayer和ColorLayer实现各自的onDraw方法。我们先来看BufferLayer,BufferLayer比较复杂。

BufferLayer的合成onDraw处理流程如下:

  • 绑定Texture
void BufferLayer::onDraw(const RenderArea& renderArea, const Region& clip,
                         bool useIdentityTransform) const {
    ATRACE_CALL();

    if (CC_UNLIKELY(getBE().compositionInfo.mBuffer == 0)) {
        ... ...
        return;
    }

    // 绑定Texture
    status_t err = mConsumer->bindTextureImage();
    ... ...
status_t BufferLayerConsumer::bindTextureImage() {
    Mutex::Autolock lock(mMutex);
    return bindTextureImageLocked();
}

绑定Texture主要在bindTextureImageLocked中完成:

status_t BufferLayerConsumer::bindTextureImageLocked() {
    mRE.checkErrors();

    if (mCurrentTexture == BufferQueue::INVALID_BUFFER_SLOT && mCurrentTextureImage == NULL) {
        ... ...
        return NO_INIT;
    }

    const Rect& imageCrop = canUseImageCrop(mCurrentCrop) ? mCurrentCrop : Rect::EMPTY_RECT;
    status_t err = mCurrentTextureImage->createIfNeeded(imageCrop);
    if (err != NO_ERROR) {
        ... ...
        return UNKNOWN_ERROR;
    }

    mRE.bindExternalTextureImage(mTexName, mCurrentTextureImage->image());

    return doFenceWaitLocked();
}

mCurrentTextureImage是合成开始时,acquireBuffer时更新的。通过createIfNeeded创建Image。

status_t BufferLayerConsumer::Image::createIfNeeded(const Rect& imageCrop) {
    const int32_t cropWidth = imageCrop.width();
    const int32_t cropHeight = imageCrop.height();
    if (mCreated && mCropWidth == cropWidth && mCropHeight == cropHeight) {
        return OK;
    }

    mCreated = mImage.setNativeWindowBuffer(mGraphicBuffer->getNativeBuffer(),
                                            mGraphicBuffer->getUsage() & GRALLOC_USAGE_PROTECTED,
                                            cropWidth, cropHeight);
    if (mCreated) {
        ... ...

    return mCreated ? OK : UNKNOWN_ERROR;
}

image的创建在setNativeWindowBuffer函数中完成:

bool Image::setNativeWindowBuffer(ANativeWindowBuffer* buffer, bool isProtected, int32_t cropWidth,
                                  int32_t cropHeight) {
    if (mEGLImage != EGL_NO_IMAGE_KHR) {
        ... //release pre mEGLImage
    }

    if (buffer) {
        std::vector<EGLint> attrs = buildAttributeList(isProtected, cropWidth, cropHeight);
        mEGLImage = eglCreateImageKHR(mEGLDisplay, EGL_NO_CONTEXT, EGL_NATIVE_BUFFER_ANDROID,
                                      static_cast<EGLClientBuffer>(buffer), attrs.data());
        if (mEGLImage == EGL_NO_IMAGE_KHR) {
            ALOGE("failed to create EGLImage: %#x", eglGetError());
            return false;
        }
    }

    return true;
}

setNativeWindowBuffer时,先释放掉旧的mEGLImage。再创建新的mEGLImage。注意eglCreateImageKHR的参数。这里的buffer就是我们acquireBuffer时,获取到的GraphicBuffer。eglCreateImageKHR函数根据GraphicBuffer创建了一个mEGLImage。

回到bindTextureImageLocked函数,创建的EglImage通过bindExternalTextureImage函数绑定。

void RenderEngine::bindExternalTextureImage(uint32_t texName, const RE::Image& image) {
    const GLenum target = GL_TEXTURE_EXTERNAL_OES;

    glBindTexture(target, texName);
    if (image.getEGLImage() != EGL_NO_IMAGE_KHR) {
        glEGLImageTargetTexture2DOES(target, static_cast<GLeglImageOES>(image.getEGLImage()));
    }
}

最终通过glEGLImageTargetTexture2DOES函数,将创建的EglImage和Texture mTexName进行绑定。这样我们的Layer数据就送到了GPU进行处理。

回到onDraw方法:

  • DRM处理
    如果是受保护的内容,或者是Secure的内容想显示在非安全的Display上,都是不允许的。这个时候,相关的区域显示为黑色
void GLES20RenderEngine::setupLayerBlackedOut() {
    glBindTexture(GL_TEXTURE_2D, mProtectedTexName);
    Texture texture(Texture::TEXTURE_2D, mProtectedTexName);
    texture.setDimensions(1, 1); // FIXME: we should get that from somewhere
    mState.setTexture(texture);
}
  • 获取textureMatrix
void BufferLayer::onDraw(const RenderArea& renderArea, const Region& clip,
                         bool useIdentityTransform) const {
    ... ...
    bool blackOutLayer = isProtected() || (isSecure() && !renderArea.isSecure());

    RenderEngine& engine(mFlinger->getRenderEngine());

    if (!blackOutLayer) {
        const bool useFiltering = getFiltering() || needsFiltering(renderArea) || isFixedSize();

        // Query the texture matrix given our current filtering mode.
        float textureMatrix[16];
        mConsumer->setFilteringEnabled(useFiltering);
        mConsumer->getTransformMatrix(textureMatrix);

        if (getTransformToDisplayInverse()) {
            // 处理Inverse翻转
        }

        // Set things up for texturing.
        mTexture.setDimensions(getBE().compositionInfo.mBuffer->getWidth(),
                               getBE().compositionInfo.mBuffer->getHeight());
        mTexture.setFiltering(useFiltering);
        mTexture.setMatrix(textureMatrix);

        engine.setupLayerTexturing(mTexture);
    } else {
        engine.setupLayerBlackedOut();
    }
    drawWithOpenGL(renderArea, useIdentityTransform);
    engine.disableTexturing();
}

textureMatrix是在GLConsumer::computeTransformMatrix中计算的,感兴趣的可以去看看。

  • 用OpenGL绘制
    主要通过drawWithOpenGL函数完成:
void BufferLayer::drawWithOpenGL(const RenderArea& renderArea, bool useIdentityTransform) const {
    const State& s(getDrawingState());

     //计算区域边界,获取Mesh
    computeGeometry(renderArea, getBE().mMesh, useIdentityTransform);

    const Rect bounds{computeBounds()}; // Rounds from FloatRect

    Transform t = getTransform();
    Rect win = bounds;
    if (!s.finalCrop.isEmpty()) {
         ... ... //处理finalCrop
    }

    float left = float(win.left) / float(s.active.w);
    float top = float(win.top) / float(s.active.h);
    float right = float(win.right) / float(s.active.w);
    float bottom = float(win.bottom) / float(s.active.h);

    // 计算Texture的坐标顶点
    Mesh::VertexArray<vec2> texCoords(getBE().mMesh.getTexCoordArray<vec2>());
    texCoords[0] = vec2(left, 1.0f - top);
    texCoords[1] = vec2(left, 1.0f - bottom);
    texCoords[2] = vec2(right, 1.0f - bottom);
    texCoords[3] = vec2(right, 1.0f - top);

    RenderEngine& engine(mFlinger->getRenderEngine());
    engine.setupLayerBlending(mPremultipliedAlpha, isOpaque(s), false /* disableTexture */,
                              getColor());
    engine.setSourceDataSpace(mCurrentState.dataSpace);
    engine.drawMesh(getBE().mMesh);
    engine.disableBlending();
}

setupLayerBlending处理Alpha的Blend:

void GLES20RenderEngine::setupLayerBlending(bool premultipliedAlpha, bool opaque,
                                            bool disableTexture, const half4& color) {
    mState.setPremultipliedAlpha(premultipliedAlpha);
    mState.setOpaque(opaque);
    mState.setColor(color);

    if (disableTexture) {
        mState.disableTexture();
    }

    if (color.a < 1.0f || !opaque) {
        glEnable(GL_BLEND);
        glBlendFunc(premultipliedAlpha ? GL_ONE : GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
    } else {
        glDisable(GL_BLEND);
    }
}

drawMesh绘制内容:

void GLES20RenderEngine::drawMesh(const Mesh& mesh) {
    if (mesh.getTexCoordsSize()) {
        glEnableVertexAttribArray(Program::texCoords);
        glVertexAttribPointer(Program::texCoords, mesh.getTexCoordsSize(), GL_FLOAT, GL_FALSE,
                              mesh.getByteStride(), mesh.getTexCoords());
    }

    glVertexAttribPointer(Program::position, mesh.getVertexSize(), GL_FLOAT, GL_FALSE,
                          mesh.getByteStride(), mesh.getPositions());

    if (usesWideColor()) {
        Description wideColorState = mState;
        if (mDataSpace != HAL_DATASPACE_DISPLAY_P3) {
            ... ...
        }
        ProgramCache::getInstance().useProgram(wideColorState);

        glDrawArrays(mesh.getPrimitive(), 0, mesh.getVertexCount());

        if (outputDebugPPMs) {
            ... ...
        }
    } else {
        ProgramCache::getInstance().useProgram(mState);

        glDrawArrays(mesh.getPrimitive(), 0, mesh.getVertexCount());
    }

    if (mesh.getTexCoordsSize()) {
        glDisableVertexAttribArray(Program::texCoords);
    }
}

glDrawArrays绘制~

所有Layer都绘制完成后,swapBuffer

6.交换Buffer
Surface交换Buffer

void Surface::swapBuffers() const {
    if (!eglSwapBuffers(mEGLDisplay, mEGLSurface)) {
        ... ...
    }
}

eglSwapBuffers 将交换GPU处理的Buffer,处理完的Buffer,也就是包含Layer合成数据后的Buffer将被queue到BufferQueue中。

前面已经说过advanceFrame时,将acquireBuffer,通过setClientTarget给HWC设置Client端的合成结果,传给底层进行显示。

以上就是Client端的合成处理。

推荐阅读更多精彩内容