背景
最近好几个网友都来问我HWC是根据什么条件来决定client合成还是device,作为之前一直"吹牛逼",说自己如何好学,如何积极回答网友问题的博主,这波必须安排,看完这篇文章,绝对让你明白。
本文参考实现来源于external/drm_hwcomposer
,我相信别的厂家的思路也大差不差。
一、获取参与SF合成的所有layers
假设我们的所有可见的layers按照z轴,图中有7个layer这样子排列。
二、根据该屏幕支持的Usable Planes数量算出client合成的layer数量
假设该屏幕支持4个Planes,但是由于layer的数量大于4,因为需要保留一个Planes,用于client合成,也就是Framebuffersurface。
HWC只支持4-1=3
个layer为device合成,需要找到连续
的7-3=4
个layers作为client合成(代码分析2_1
)
2.1 选择像素点之和最少的连续4个layers
如图,我们有了四个备选方案,我们要选择像素点个数之和最少的连续4个layers,我觉得这样子的策略,应该是减少GPU合成的工作量(代码分析2_2
)
2.2 最后的结果
假如第二个方案中,像素点个数之和最少,HWC最后返回的策略如下图。
小结
这个就是最简单的策略,但是接下来要在这个基础,引申出更加复杂的策略。
三、假如有layer指定要client合成
layer是可以指定client合成的,例如layer有圆角,当然还有很多情况,后面讲代码的时候会讲(代码分析3_0
)。我们先假设指定client合成的layer为2,4。
第一个指定的client layer和最后一个指定的client layer之间的 layer 必须是变成client了(
代码分析3_1
),如下图:3就很惨的也被设置成client合成了四、再次考虑Usable Planes数量算出client合成的layer的数量
按照二
中说的一样,我们还是需要找到连续的4个layer作为client合成,但是我们还的还得加个条件,包含三
中提到的2~4的三个layer
4.1 得到备选方案
这次我们只有两个备选方案了,因为2.1
中提到的第1个和第4个方案都没有同时包含2~4的三个layer
4.2 选择像素点之和最少的连续4个layer的方案。
还是和二
中的策略一样,假如这次是2,3,4,5的像素点之和最少,最后得到的结果如下。
小结
到这里应该看明白了,其实图解策略就很简单了,再去结合代码看就比较容易了。
代码分析
代码量不多,我就全部贴上来了external/drm_hwcomposer/backend/Backend.cpp
,文章中代码分析X_X
对应下面的部分注释,关键是看懂这个函数的实现
//关键代码,返回client layer的start index和size。
std::tie(client_start, client_size) = GetClientLayers(display, layers);
带着我前面讲的策略去跟这个函数。
/*
* Copyright (C) 2020 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "Backend.h"
#include <climits>
#include "BackendManager.h"
#include "bufferinfo/BufferInfoGetter.h"
namespace android {
HWC2::Error Backend::ValidateDisplay(HwcDisplay *display, uint32_t *num_types,
uint32_t *num_requests) {
*num_types = 0;
*num_requests = 0;
auto layers = display->GetOrderLayersByZPos();
int client_start = -1;
size_t client_size = 0;
if (display->ProcessClientFlatteningState(layers.size() <= 1)) {
display->total_stats().frames_flattened_++;
client_start = 0;
client_size = layers.size();
MarkValidated(layers, client_start, client_size);
} else {
std::tie(client_start, client_size) = GetClientLayers(display, layers);//关键代码,返回client layer的start id和size。
MarkValidated(layers, client_start, client_size);
bool testing_needed = !(client_start == 0 && client_size == layers.size());
AtomicCommitArgs a_args = {.test_only = true};
if (testing_needed &&
display->CreateComposition(a_args) != HWC2::Error::None) {
++display->total_stats().failed_kms_validate_;
client_start = 0;
client_size = layers.size();
MarkValidated(layers, 0, client_size);
}
}
*num_types = client_size;
display->total_stats().gpu_pixops_ += CalcPixOps(layers, client_start,
client_size);
display->total_stats().total_pixops_ += CalcPixOps(layers, 0, layers.size());
return *num_types != 0 ? HWC2::Error::HasChanges : HWC2::Error::None;
}
std::tuple<int, size_t> Backend::GetClientLayers(
HwcDisplay *display, const std::vector<HwcLayer *> &layers) {
int client_start = -1;
size_t client_size = 0;
for (size_t z_order = 0; z_order < layers.size(); ++z_order) {
if (IsClientLayer(display, layers[z_order])) {
if (client_start < 0)
client_start = (int)z_order;
client_size = (z_order - client_start) + 1;
}
}//代码分析3_1
return GetExtraClientRange(display, layers, client_start, client_size);//扩大额外的需要为client的layers
}
//代码分析3_0
//判断layer是不是必须为client合成或者已经指定了client合成
bool Backend::IsClientLayer(HwcDisplay *display, HwcLayer *layer) {
return !HardwareSupportsLayerType(layer->GetSfType()) ||
!BufferInfoGetter::GetInstance()->IsHandleUsable(layer->GetBuffer()) ||
display->color_transform_hint() != HAL_COLOR_TRANSFORM_IDENTITY ||
(layer->RequireScalingOrPhasing() &&
display->GetHwc2()->GetResMan().ForcedScalingWithGpu());
}
bool Backend::HardwareSupportsLayerType(HWC2::Composition comp_type) {
return comp_type == HWC2::Composition::Device ||
comp_type == HWC2::Composition::Cursor;
}
//计算连续几个layer的像素点之和
uint32_t Backend::CalcPixOps(const std::vector<HwcLayer *> &layers,
size_t first_z, size_t size) {
uint32_t pixops = 0;
for (size_t z_order = 0; z_order < layers.size(); ++z_order) {
if (z_order >= first_z && z_order < first_z + size) {
hwc_rect_t df = layers[z_order]->GetDisplayFrame();
pixops += (df.right - df.left) * (df.bottom - df.top);
}
}
return pixops;
}
//标记每个layer的合成策略,最后sf拿到的就是这个标记
void Backend::MarkValidated(std::vector<HwcLayer *> &layers,
size_t client_first_z, size_t client_size) {
for (size_t z_order = 0; z_order < layers.size(); ++z_order) {
if (z_order >= client_first_z && z_order < client_first_z + client_size)
layers[z_order]->SetValidatedType(HWC2::Composition::Client);
else
layers[z_order]->SetValidatedType(HWC2::Composition::Device);
}
}
std::tuple<int, int> Backend::GetExtraClientRange(
HwcDisplay *display, const std::vector<HwcLayer *> &layers,
int client_start, size_t client_size) {
auto planes = display->GetPipe().GetUsablePlanes();
size_t avail_planes = planes.size();//获取整个drm支持的planes的数量
/*
* If more layers then planes, save one plane
* for client composited layers
*/
if (avail_planes < display->layers().size())
avail_planes--;//保留一个用于存放所有client合成的layer
int extra_client = int(layers.size() - client_size) - int(avail_planes);//计算剩余需要成为client的数量
//上面是代码分析2_1
if (extra_client > 0) {
int start = 0;
size_t steps = 0;
if (client_size != 0) {//有layer指定了client合成
int prepend = std::min(client_start, extra_client);
int append = std::min(int(layers.size()) -
int(client_start + client_size),
extra_client);
start = client_start - (int)prepend;
client_size += extra_client;
steps = 1 + std::min(std::min(append, prepend),
int(layers.size()) - int(start + client_size));
} else {//没有layer指定了client合成。代码分析
client_size = extra_client;//剩余的都是client
steps = 1 + layers.size() - extra_client;//计算硬件合成步伐
}
//选择像素之和最少的连续layers
uint32_t gpu_pixops = UINT32_MAX;
for (size_t i = 0; i < steps; i++) {
uint32_t po = CalcPixOps(layers, start + i, client_size);
if (po < gpu_pixops) {
gpu_pixops = po;
client_start = start + int(i);
}
} //代码分析2_2
}
return std::make_tuple(client_start, client_size);
}
// clang-format off
// NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables, cert-err58-cpp)
REGISTER_BACKEND("generic", Backend);
// clang-format on
} // namespace android