摘要
现有的许多研究提出用各种编译器导向的方法来静态或动态地管理SPM(Scratch-Pad Memory)内容,以取得对SPM部件的最佳利用,其关键问题是如何最大化地重复访问SPM中的内容。针对此问题,通过分析程序调用序列,构造低功耗编译优化,使无调用链关联函数的局部变量共享全局存储单元,以降低程序运行时对内存的需求,并提高相应存储单元的利用率。该优化编译可以与现有编译器导向的数据SPM管理策略相结合,提高现有方法的性能和功耗效果。
Many compiler-directed Scratch-Pad-Memory(SPM) management methods using both static and dynamic approaches have been proposed to obtain optimized utilization of SPM components.Important issues in those methods are determining the SPM layouts under limited size and maximizing the reuse of data in SPM.To address these issues,a compiler optimization pass for the data SPM based on call chain analysis was developed.It makes some non-call related functions share storage space for their local variables,which reduc...
出处
《吉林大学学报(工学版)》
EI
CAS
CSCD
北大核心
2009年第1期143-147,共5页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(60673165)
国家发改委高新技术产业化软件专项项目(发改高技[2004]2040)
关键词
计算机软件
低功耗
编译优化
SPM管理
调用链分析
computer software
low power
compile optimization
scratch-pad memory(SPM) management
call chain analysis