面向国产Unicore架构的Android图形系统优化

Optimization for Android graphics system based on domestic Unicore architecture

采用基于国产Unicore内核的嵌入式处理器平台,进行了Android系统移植.首先,移植实现了本地框架层的物理内存模块(PMEM)、图形显示内存分配模块(Gralloc)、块拷贝模块(Copybit)和图形处理单元驱动(GPU驱动)模块.然后,通过分析传统GPU(graphic processingunit)驱动调用过程中Linux内核态与用户态的数据交互方式,提出了一种新的访存机制,增加了Skia库到GPU驱动的访问路径,使内核态可以直接访问用户态数据,克服了传统模式中必须将用户态数据拷贝到内核态才能被内核态访问的弊端,降低了访存开销.最后,对Skia图形库的热点函数进行了优化.实验结果表明:与GPU驱动优化前相比,优化后的图形绘制性能平均加速3.08倍;Skia图形库中矩形搬运函数的运行时长缩短至原来的1/9;相比于无GPU的情况,GPU驱动优化后的图形绘制性能平均提升7.46倍.

Based on the embedded processor platform of the domestic Unicore,the Android system is transplanted.Firstly,four critical function modules,PMEM（physical memory）,Gralloc,Copybit and GPU（graphic processing unit） driver,in the local frame of Android graphics system,are designed and implemented.Then,the data interaction process between the Linux kernel and the user mode in the GPU driver is analyzed and a novel memory access mechanism is put forward.A memory access path from Skia lib to GPU driver is introduced,and the kernel mode can directly access user data,avoiding the drawbacks of copying data from the user mode to the kernel mode in the traditional mode and reducing the memory access overhead.Finally,the hotspot functions in Skia lib are optimized.The experimental results show that after optimization the graphics rendering performance can obtain 3.08 times speedup on average,and the execution time of rectangle transporting function is compressed to 1/9 of that before optimization for Skia lib.The graphics rendering performance of the systems with optimized GPU driver is accelerated by 7.46 times on average than that of the systems without GPU.