| 摘要: |
| 现代的超标量(superscalar)和超流水线(superpipeline)高速处理器通常都带有二级高速缓冲存储器(cache)，以及较多的快速寄存器(register)．利用程序变换来改善程序性能．是当今编译技术研究的热门课题之一．本文重点讨论超级编译技术中的循环分布、交换、合并和stripmtntng对串行程序的优化应用和效果． |
| 关键词: 数据依赖，循环变换，cache命中率，数据局部性． |
| DOI： |
| 分类号: |
| 基金项目:本课题是国家高技术863资助项目. |
|
| USING SUPERCOMPILING TECHNIQUES TO OPTIMIZE SEQUENTIAL PROGRAMS |
|
Zhang Zhaoqing,Qiao Ruliang
|
| Abstract: |
| Recent superscalar and superpipelined processors commonly have at least two levels caches and a large register set．Using the transformations to improve performance of programs that is one of the research projects of compiling technique．This paper discusses optimization of sequential programs by using do—loop distribution，do_loop interchange，do—loop fusion，and strip—mining used in supercompiler．These tramsformations have been shown to result in dramatic improvements． |
| Key words: Data dependence,loop transformation,cache hit ratio,data locality． |
