我是靠谱客的博主 迷路铃铛,最近开发中收集的这篇文章主要介绍SurfaceFlinger旋转流程分析,觉得挺不错的,现在分享给大家,希望可以做个参考。

概述

chipset: MSM8X25Q

codebase: Android4.1

本文主要对SF(SurfaceFilnger)处理旋转事件的流程做个简单分析。GPU和mdp都可以用来旋转,文中对两者穿插说明。

 系统初始化会调用GraphicPlane::setDisplayHardware,此函数主要判断系统是否默认设置了rotation property值,如果有,则先保存下来。另外,SF是以transform其实就是以矩阵来作旋转计算的,计算方法以线性代数中的知识为依据:

void GraphicPlane::setDisplayHardware(DisplayHardware *hw)
{
mHw = hw;
// initialize the display orientation transform.
// it's a constant that should come from the display driver.
int displayOrientation = ISurfaceComposer::eOrientationDefault;
char property[PROPERTY_VALUE_MAX];
/*读取property*/
if (property_get("ro.sf.hwrotation", property, NULL) > 0) {
//displayOrientation
switch (atoi(property)) {
/*当然,你也可以仿照添加180°的旋转。*/
case 90:
displayOrientation = ISurfaceComposer::eOrientation90;
break;
case 270:
displayOrientation = ISurfaceComposer::eOrientation270;
break;
}
}
const float w = hw->getWidth();
const float h = hw->getHeight();
/*根据宽和高以及orientation生成一个transform*/
GraphicPlane::orientationToTransfrom(displayOrientation, w, h,
&mDisplayTransform);
/*90°或者270°时需要变换宽高。*/
if (displayOrientation & ISurfaceComposer::eOrientationSwapMask) {
mDisplayWidth = h;
mDisplayHeight = w;
} else {
mDisplayWidth = w;
mDisplayHeight = h;
}
/*保存当前全局旋转角度,意味着整个系统的显示都要根据整个值
来得出最后是否要旋转。如果为90°或者270°,那就是横屏,如果
是180°,画面就要反转180°。*/
setOrientation(ISurfaceComposer::eOrientationDefault);
}
status_t GraphicPlane::setOrientation(int orientation)
{
// If the rotation can be handled in hardware, this is where
// the magic should happen.
const DisplayHardware& hw(displayHardware());
const float w = mDisplayWidth;
const float h = mDisplayHeight;
mWidth = int(w);
mHeight = int(h);
Transform orientationTransform;
GraphicPlane::orientationToTransfrom(orientation, w, h,
&orientationTransform);
if (orientation & ISurfaceComposer::eOrientationSwapMask) {
mWidth = int(h);
mHeight = int(w);
}
mOrientation = orientation;
/*将orientation和h/w计算和得出一个全局的transform,这里的
相乘计算方法,就是利用矩阵来实现的。*/
mGlobalTransform = mDisplayTransform * orientationTransform;
return NO_ERROR;
}

由于整个系统一直默认旋转只是流程中一个坐标的特殊处理,不管系统上层是否请求旋转,如横屏游戏,sensor坐标变化,默认的旋转是一直会被处理的。

注意别把这两种旋转混淆了,默认旋转是一般情况下用户看到的显示效果,而上层apk

请求的旋转相对默认旋转是瞬间的。

         现在以上层发生旋转事件为例看下SF处理旋转的流程。

 上层调用SurfaceFlinger::setTransactionState设置当前角度变化了:

void SurfaceFlinger::setTransactionState(const Vector<ComposerState>& state,
int orientation, uint32_t flags) {
Mutex::Autolock _l(mStateLock);
uint32_t transactionFlags = 0;
/*当前坐标和要设置的坐标不相等表明要作旋转了!*/
if (mCurrentState.orientation != orientation) {
if (uint32_t(orientation)<=eOrientation270 || orientation==42) {
mCurrentState.orientation = orientation;
transactionFlags |= eTransactionNeeded;
} else if (orientation != eOrientationUnchanged) {
ALOGW("setTransactionState: ignoring unrecognized orientation: %d",
orientation);
}
}
/*设置旋转标志,SF处理的时候会用到。*/
const size_t count = state.size();
for (size_t i=0 ; i<count ; i++) {
const ComposerState& s(state[i]);
sp<Client> client( static_cast<Client *>(s.client.get()) );
transactionFlags |= setClientStateLocked(client, s.state);
}
if (transactionFlags) {
// this triggers the transaction
setTransactionFlags(transactionFlags);
~~snip
}

SF收到消息后,调用SurfaceFlinger::onMessageReceived进行处理,核心的部分都在这里了。

void SurfaceFlinger::onMessageReceived(int32_t what)
{
ATRACE_CALL();
switch (what) {
case MessageQueue::INVALIDATE:
case MessageQueue::REFRESH: {
//
case MessageQueue::INVALIDATE: {
// if we're in a global transaction, don't do anything.
const uint32_t mask = eTransactionNeeded | eTraversalNeeded;
uint32_t transactionFlags = peekTransactionFlags(mask);
/*处理事务*/
if (CC_UNLIKELY(transactionFlags)) {
handleTransaction(transactionFlags);
}
const uint32_t mask1 = eDelayedTraversalNeeded;
uint32_t transactionFlags1 = peekTransactionFlags(mask1);
if (CC_UNLIKELY(transactionFlags1)) {
handleDelayedTransaction(transactionFlags);
}
/*计算可视区域*/
// post surfaces (if needed)
handlePageFlip();
~~snip
case MessageQueue::REFRESH: {
~~snip
/*处理几何坐标*/
if (CC_UNLIKELY(mHwWorkListDirty)) {
// build the h/w work list
handleWorkList();
}
/*render*/
if (CC_LIKELY(hw.canDraw())) {
// repaint the framebuffer (if needed)
handleRepaint();
// inform the h/w that we're done compositing
hw.compositionComplete();
postFramebuffer();
} else {
// pretend we did the post
hw.compositionComplete();
}
} break;
}
}

下面一一对这几个函数分析。

handleTransaction

void SurfaceFlinger::handleTransaction(uint32_t transactionFlags)
{
~~snip
const uint32_t mask = eTransactionNeeded | eTraversalNeeded;
transactionFlags = getTransactionFlags(mask);
handleTransactionLocked(transactionFlags);
~~snip
}
void SurfaceFlinger::handleTransactionLocked(uint32_t transactionFlags)
{
const LayerVector& currentLayers(mCurrentState.layersSortedByZ);
const size_t count = currentLayers.size();
/*
* Traversal of the children
* (perform the transaction for each of them if needed)
*/
/*判断是否有事务要处理。*/
const bool layersNeedTransaction = transactionFlags & eTraversalNeeded;
if (layersNeedTransaction) {
for (size_t i=0 ; i<count ; i++) {
const sp<LayerBase>& layer = currentLayers[i];
uint32_t trFlags = layer->getTransactionFlags(eTransactionNeeded);
if (!trFlags) continue;
const uint32_t flags = layer->doTransaction(0);
if (flags & Layer::eVisibleRegion)
mVisibleRegionsDirty = true;
}
}
~~snip
if (transactionFlags & eTransactionNeeded) {
/*有旋转角度变化的事务。*/
if (mCurrentState.orientation != mDrawingState.orientation) {
// the orientation has changed, recompute all visible regions
// and invalidate everything.
const int dpy = 0;
const int orientation = mCurrentState.orientation;
// Currently unused: const uint32_t flags = mCurrentState.orientationFlags;
GraphicPlane& plane(graphicPlane(dpy));
/*根据当前旋转的坐标,还有默认的旋转坐标,计算和得出最终全局transform。
此函数前面分析过了。*/
plane.setOrientation(orientation);
const Transform& planeTransform(plane.transform());
// update the shared control block
const DisplayHardware& hw(plane.displayHardware());
volatile display_cblk_t* dcblk = mServerCblk->displays + dpy;
dcblk->orientation = orientation;
dcblk->w = plane.getWidth();
dcblk->h = plane.getHeight();
/*旋转过后当然要重新绘制显示画面了,设置标志。*/
mVisibleRegionsDirty = true;
mDirtyRegion.set(hw.bounds());
}
~~snip
}

可见handleTransaction主要设置全局的transform。

 handlePageFlip

void SurfaceFlinger::handlePageFlip()
{
ATRACE_CALL();
const DisplayHardware& hw = graphicPlane(0).displayHardware();
const Region screenRegion(hw.bounds());
const LayerVector& currentLayers(mDrawingState.layersSortedByZ);
const bool visibleRegions = lockPageFlip(currentLayers);
if (visibleRegions || mVisibleRegionsDirty) {
Region opaqueRegion;
/*有旋转过,所以要重新计算可视区域*/
computeVisibleRegions(currentLayers, mDirtyRegion, opaqueRegion);
~~snip
}
~~snip
}
void SurfaceFlinger::computeVisibleRegions(
const LayerVector& currentLayers, Region& dirtyRegion, Region& opaqueRegion)
{
ATRACE_CALL();
const GraphicPlane& plane(graphicPlane(0));
const Transform& planeTransform(plane.transform());
const DisplayHardware& hw(plane.displayHardware());
const Region screenRegion(hw.bounds());
Region aboveOpaqueLayers;
Region aboveCoveredLayers;
Region dirty;
bool secureFrameBuffer = false;
size_t i = currentLayers.size();
while (i--) {
const sp<LayerBase>& layer = currentLayers[i];
/*将全局的transform传进去,然后调用各个layer作计算。*/
layer->validateVisibility(planeTransform);
~~snip
}
}
void LayerBase::validateVisibility(const Transform& planeTransform)
{
const Layer::State& s(drawingState());
/*全局transform和当前layer的transform相乘计算得到一个transform.*/
const Transform tr(planeTransform * s.transform);
const bool transformed = tr.transformed();
const DisplayHardware& hw(graphicPlane(0).displayHardware());
const uint32_t hw_h = hw.getHeight();
const Rect& crop(s.active.crop);
Rect win(s.active.w, s.active.h);
if (!crop.isEmpty()) {
win.intersect(crop, &win);
}
mNumVertices = 4;
/* mVertices 相当重要,系统根据当前的transform还有当前layer的
区域作计算,当然也包含了全局的transform信息在里面,计算出来的
坐标保存到了mVertices 去。要注意,mVertices 只是给GPU用的,如果
用mdp做旋转,SF后面的流程会对其做相应设置。*/
tr.transform(mVertices[0], win.left,
win.top);
tr.transform(mVertices[1], win.left,
win.bottom);
tr.transform(mVertices[2], win.right, win.bottom);
tr.transform(mVertices[3], win.right, win.top);
for (size_t i=0 ; i<4 ; i++)
mVertices[i][1] = hw_h - mVertices[i][1];
if (CC_UNLIKELY(transformed)) {
// NOTE: here we could also punt if we have too many rectangles
// in the transparent region
/*根据transform对透明区域作处理。*/
if (tr.preserveRects()) {
// transform the transparent region
transparentRegionScreen = tr.transform(s.transparentRegion);
} else {
// transformation too complex, can't do the transparent region
// optimization.
transparentRegionScreen.clear();
}
} else {
transparentRegionScreen = s.transparentRegion;
}
/*保存相应的一些信息到新的参数变量中,以便后面用到。*/
// cache a few things...
mOrientation = tr.getOrientation();
mPlaneOrientation = planeTransform.getOrientation();
mTransform = tr;
mTransformedBounds = tr.transform(win);
}

可见,从handlePageFlip中,我们主要是得到了GPU要处理的坐标,保存到了mVertices数组中。

 handleWorkList

void SurfaceFlinger::handleWorkList()
{
mHwWorkListDirty = false;
HWComposer& hwc(graphicPlane(0).displayHardware().getHwComposer());
if (hwc.initCheck() == NO_ERROR) {
const Vector< sp<LayerBase> >& currentLayers(mVisibleLayersSortedByZ);
const size_t count = currentLayers.size();
hwc.createWorkList(count);
hwc_layer_t* const cur(hwc.getLayers());
for (size_t i=0 ; cur && i<count ; i++) {
/*依次调用各个layer的setGeometry 。*/
currentLayers[i]->setGeometry(&cur[i]);
}
}
}
void Layer::setGeometry(hwc_layer_t* hwcl)
{
/*调用基类函数。*/
LayerBaseClient::setGeometry(hwcl);
hwcl->flags &= ~HWC_SKIP_LAYER;
/*不完全透明时需要GPU处理。*/
// we can't do alpha-fade with the hwc HAL
const State& s(drawingState());
if (s.alpha < 0xFF) {
hwcl->flags = HWC_SKIP_LAYER;
}
/*
* Transformations are applied in this order:
* 1) buffer orientation/flip/mirror
* 2) state transformation (window manager)
* 3) layer orientation (screen orientation)
* mTransform is already the composition of (2) and (3)
* (NOTE: the matrices are multiplied in reverse order)
*/
/*这里就是在根据要求的旋转信息得出最终的transform,
然后赋值给hwcl,hwcl的信息最后会传给mdp处理。*/
const Transform bufferOrientation(mCurrentTransform);
hwcl->sourceTransform = bufferOrientation.getOrientation();
const Transform tr(mTransform * bufferOrientation);
// this gives us only the "orientation" component of the transform
const uint32_t finalTransform = tr.getOrientation();
// we can only handle simple transformation
if (finalTransform & Transform::ROT_INVALID) {
hwcl->flags = HWC_SKIP_LAYER;
} else {
hwcl->transform = finalTransform;
}
/*得到layer区域相对坐标,绝对坐标在基类中计算得到。*/
Rect crop = computeBufferCrop();
hwcl->sourceCrop.left
= crop.left;
hwcl->sourceCrop.top
= crop.top;
hwcl->sourceCrop.right
= crop.right;
hwcl->sourceCrop.bottom = crop.bottom;
}

由此看来handleWorkList主要是服务于mdp的,如果用GPU旋转的话可以不需要这些信息。

 

handleRepaint

void SurfaceFlinger::handleRepaint()
{
~~snip
setupHardwareComposer();
/*对各个layer作render*/
composeSurfaces(mDirtyRegion);
// update the swap region and clear the dirty region
mSwapRegion.orSelf(mDirtyRegion);
mDirtyRegion.clear();
}
void SurfaceFlinger::composeSurfaces(const Region& dirty)
{
const DisplayHardware& hw(graphicPlane(0).displayHardware());
HWComposer& hwc(hw.getHwComposer());
hwc_layer_t* const cur(hwc.getLayers());
const size_t fbLayerCount = hwc.getLayerCount(HWC_FRAMEBUFFER);
if (!cur || fbLayerCount) {
~~snip
const Vector< sp<LayerBase> >& layers(mVisibleLayersSortedByZ);
const size_t count = layers.size();
for (size_t i=0 ; i<count ; i++) {
const sp<LayerBase>& layer(layers[i]);
const Region clip(dirty.intersect(layer->visibleRegionScreen));
~~snip
/*如果不是HWC_FRAMEBUFFER 类型的话不处理,也就是让mdp去做处理。*/
#ifdef QCOMHW
if (cur && (cur[i].compositionType != HWC_FRAMEBUFFER))
continue;
#endif
/*依次调用各种layer的draw函数 */
// render the layer
layer->draw(clip);
}
}
~~snip
}
void LayerBase::draw(const Region& clip) const
{
//Dont draw External-only layers
if (isLayerExternalOnly(getLayer())) {
return;
}
onDraw(clip);
}
void Layer::onDraw(const Region& clip) const
{
~~snip
/*画图*/
drawWithOpenGL(clip);
glDisable(GL_TEXTURE_EXTERNAL_OES);
glDisable(GL_TEXTURE_2D);
}
void LayerBase::drawWithOpenGL(const Region& clip) const
{
const DisplayHardware& hw(graphicPlane(0).displayHardware());
const uint32_t fbHeight = hw.getHeight();
const State& s(drawingState());
~~snip
struct TexCoords {
GLfloat u;
GLfloat v;
};
Rect crop(s.active.w, s.active.h);
if (!s.active.crop.isEmpty()) {
crop = s.active.crop;
}
GLfloat left = GLfloat(crop.left) / GLfloat(s.active.w);
GLfloat top = GLfloat(crop.top) / GLfloat(s.active.h);
GLfloat right = GLfloat(crop.right) / GLfloat(s.active.w);
GLfloat bottom = GLfloat(crop.bottom) / GLfloat(s.active.h);
/*得到顶点坐标*/
TexCoords texCoords[4];
texCoords[0].u = left;
texCoords[0].v = top;
texCoords[1].u = left;
texCoords[1].v = bottom;
texCoords[2].u = right;
texCoords[2].v = bottom;
texCoords[3].u = right;
texCoords[3].v = top;
for (int i = 0; i < 4; i++) {
texCoords[i].v = 1.0f - texCoords[i].v;
}
/*OpenGL利用下面这些函数将根据纹理坐标和顶点坐标
得到最后layer的显示区域以及内容。*/
glEnableClientState(GL_TEXTURE_COORD_ARRAY);
glVertexPointer(2, GL_FLOAT, 0, mVertices);
glTexCoordPointer(2, GL_FLOAT, 0, texCoords);
glDrawArrays(GL_TRIANGLE_FAN, 0, mNumVertices);
glDisableClientState(GL_TEXTURE_COORD_ARRAY);
glDisable(GL_BLEND);
}

所以,composeSurfaces是服务于GPU的。

虽然mdp已经旋转完成了,但是似乎还没有看到如何送给mdp啊?看看postFramebuffer.

  postFramebuffer

void SurfaceFlinger::postFramebuffer()
{
~~snip
const DisplayHardware& hw(graphicPlane(0).displayHardware());
const nsecs_t now = systemTime();
mDebugInSwapBuffers = now;
hw.flip(mSwapRegion);
~~snip
}
void DisplayHardware::flip(const Region& dirty) const
{
~~snip
if (mHwc->initCheck() == NO_ERROR) {
mHwc->commit();
} else {
eglSwapBuffers(dpy, surface);
}
~~snip
}
status_t HWComposer::commit() const {
/*调用了HAL的set,看起来和硬件有关联了!*/
int err = mHwc->set(mHwc, mDpy, mSur, ((mSwapRectOn)?mListDirty:mList));
if (mSwapRectOn && mListDirty) {
mListDirty->flags &= ~HWC_GEOMETRY_CHANGED;
} else if ( mList) {
mList->flags &= ~HWC_GEOMETRY_CHANGED;
}
return (status_t)err;
}
static int hwc_set(hwc_composer_device_t *dev,
hwc_display_t dpy,
hwc_surface_t sur,
hwc_layer_list_t* list)
{
int ret = 0;
hwc_context_t* ctx = (hwc_context_t*)(dev);
if (LIKELY(list)) {
VideoOverlay::draw(ctx, list);
ExtOnly::draw(ctx, list);
CopyBit::draw(ctx, list, (EGLDisplay)dpy, (EGLSurface)sur);
/*调用到mdp的draw了!!!*/
MDPComp::draw(ctx, list);
EGLBoolean sucess = eglSwapBuffers((EGLDisplay)dpy, (EGLSurface)sur);
~~snip
}
return ret;
}

总的下来,流程还是比较清晰的,不过里面很多transform很容易把人给弄晕掉,可以加log看值跟踪下。


2013/02/17




最后

以上就是迷路铃铛为你收集整理的SurfaceFlinger旋转流程分析的全部内容,希望文章能够帮你解决SurfaceFlinger旋转流程分析所遇到的程序开发问题。

如果觉得靠谱客网站的内容还不错,欢迎将靠谱客网站推荐给程序员好友。

本图文内容来源于网友提供,作为学习参考使用,或来自网络收集整理,版权属于原作者所有。
点赞(50)

评论列表共有 0 条评论

立即
投稿
返回
顶部