基于Trace合并和寄存器分配的Dalvik优化  

Optimization of Dalvik Based on Trace-combination and Register Allocation

作　　者：余超君 李春强[1] 尚云海 张培勇[1] 

机构地区：[1]浙江大学超大规模集成电路设计研究所,杭州310027

出　　处：《计算机工程》2014年第10期61-65,70,共6页Computer Engineering

基　　金：国家自然科学基金资助项目(61204111);"核高基"重大专项(2010ZX01030-001-001-002)

摘　　要：Dalvik虚拟机作为Android系统上运行所有应用程序的基础,其性能瓶颈一直制约着Android系统的用户体验。通过研究Android系统中的Dalvik架构,分析其解释器和JIT模块的工作原理,发现热Trace选择过程中短Trace编译损耗大以及即时编译过程中寄存器分配不合理的情况。结合Java虚拟机技术和编译器技术,在现有热Trace选择和寄存器分配机制的基础上,提出基于Trace合并和寄存器分配的优化算法,在国产高性能嵌入式CPU CSKY体系下移植Dalvik虚拟机并实现了上述优化算法。通过实验证明优化后Dalvik执行Java程序的性能提高了近10%。As the basics of running application on Android system,performance of Dalvik virtual machine restricts the Android’ s user experience. By researching Dalvik architecture in the Android system and analyzing some key techniques of the interpreter and Just in Time（ JIT） module,it finds that short Trace’ s compiler dissipation is large and there are some irrational situations on register allocation in JIT. Combining nowadays JVM technology with modern compiler technology,and based on the Trace selection strategy and register allocation mechanism of Dalvik, this paper proposes algorithms of combining Trace and optimizing strategy of register allocation. These algorithms are implemented in high performance embedded CPU CSKY architecture. The experiments prove that this Dalvik can improve the performance by about 10%.

关 键 词：DALVIK虚拟机 JIT技术 性能优化 Trace合并 寄存器分配 生命周期 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]