【作者】 梁晓；

【导师】 陈天洲； 施青松；

【作者基本信息】 浙江大学 ， 计算机应用， 2007， 硕士

【摘要】 近年来，随着半导体和芯片技术的快速发展，集成了越来越多的硬件、软件和应用程序的系统级芯片变得越来越重要。随之而来的，是可移动嵌入式设备，如手机(Cell-Phone)，个人掌上电脑(PDA)等设备的性能大幅度地提升，带来对电池的需求不断呈上升趋势。电池的容量与使用时间已经是制约便携嵌入式设备发展的一个重要问题。使用合理的硬件和软件的办法，降低应用程序和系统的功耗，成为嵌入式系统设计的一个重要课题。当前的节能的方面的研究，在操作系统调度方面取得了很多的成果。通过操作系统调度和硬件结合的办法，通常是对历史记录进行判断，准确度成为一个问题。本文认为通过编译器和硬件技术相结合，可以获得更准确的调度信息。本文结合硬件上的动态电压缩放技术(Dynamic Voltage Scaling，DVS)和编译技术，提出了基于DVS的节能编译框架，利用编译器对代码的了解和修改能力，采用时间与功耗判断的算法，为不同的代码段设定不同的电压，从而达到节能的目的。本框架分为静态编译和动态编译两种，通过静态节能编译框架验证编译器和DVS技术结合的有效，通过动态节能编译框架，考虑到系统运行过程中多程序并行的实际情况，以及系统的电压限制和性能要求，让节能框架更加准确和实用。通过在SimpleScalar软件模拟平台和Intel XScale硬件平台上分别了实现静态和动态节能编译软件。能耗测试结果表明，静态节能编译器在代码性能下降5％的可允许范围以内，能够获得最高37％的功耗节省；动态节能编译器在实验平台上上可得到最高28％平均23％的功耗节省。本文提出的基于DVS的节能编译框架，经验证确实可以达到平均23％的节能效果，并保持性能不下降。更多还原

【Abstract】 With the fast development of semiconductor and the fast growth of capacity of chips, more and more new technique is available on the embedded systems, especially the mobile embedded systems. The result of fast development of the application on the embedded system is that the battery of become the limitation of the system. The life and capacity of the battery is now a big problem for the reason that these applications consume more and more energy. As the development of Chemistry is on the bottle neck, to reduce the energy consumption by software and hardware become the main issue of design of embedded system.Lots of research on reducing energy in the embedded systems has been done. Most of these research focuses on scheduling of tasks in the OS level. Due to the limitation of OS, the schedule is not accurate for the reason that the basis of the schedule is the trace of the tasks in the OS, without inner information from the instructions of the tasks. To schedule the tasks more accurately than OS level, there is some research on the compiler and DVS.In this paper we present a compiler framework to reduce the energy consumption on embedded systems. Firstly the Dynamic Voltage Scaling (DVS) technology was applied on the static compiler and then it was integrated in the dynamic compiler. The Framework of Runtime Dynamic Compiler based DVS (RDCD) contains 3 steps, firstly it selects the candidate block which may suitable for DVS and then the runtime judging algorithm select the suitable blocks, finally the RDCD insert the DVS instruction at the beginning and end of these blocks.The framework is realized in a simulator and a real-system. Simulation results on the simple scalar with a new energy testing module show that CPU energy can save to 13%-15% for the Intel Xscale PXA270 benchmark with a time performance penalty of at most 5%. And the result from the real system shows that the dynamic compiler with DVS can save at most 23%.更多还原