基于指令聚类与指令调度的嵌入式软件功耗优化研究  

Software Power Optimization for Embedded Systems Based on Instruction Clustering and Instruction Rescheduling

作　　者：陈嘉[1] 董渊[1] 杨阳[1] 戴桂兰[1] 王生原[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《小型微型计算机系统》2006年第1期175-179,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60083004)资助;国家"八六三"高技术项目(2002AA1Z030)资助.

摘　　要：选用指令级能耗评估模型,提出和验证了一种基于指令聚类与指令调度的功耗优化方案.该方案采用深度优先算法搜索局部最优解,挑选出能耗较小的一种指令序列.又兼顾测试工作量与精确度,将能耗相似的指令归入同类,有效降低了获取相邻指令切换能耗参数的工作量过大这一问题.通过分析基于SimpleScalar/Wattch模拟器的实验结果,指出仅用指令调度技术进行指令级功耗优化,其效果有限,为了提高优化效率,必须进行更高级别的功耗评估与优化.Proposed a scheme to lower the run time software power consumption based on instruction clustering and instruction rescheduling. To pick out an instruction sequence with relatively less power, a DFS （depth first search） algorism is used. To reduce the workload of measuring switching power between instruction pairs, the scheme classifies those instructions with similar basic power into a same cluster. The software power optimization scheme was experimentally validated on a wide range of embedded software routines. By analyzing the experimental results, some limits of the scheme were shown. It is also pointed out that higher level power estimation models and optimization techniques should be used to get better optimization results.

关 键 词：软件功耗 指令调度 功耗优化 深度优先搜索 聚类 

分 类 号：TP333[自动化与计算机技术—计算机系统结构]