| 摘要: |
| 传统的基于静态编译的数据预取大多针对数组访问,现在的应用程序中大量出现由指针构成的链式数据结构,依赖传统的编译方法难以进行预取优化.反馈式数据预取优化技术是当前高性能计算技术中前沿的一种编译优化手段,可以很好地解决链式结构的预取问题.在研究ORC编译器反馈式编译优化技术的基础上,针对Alpha结构的特点,对针对链式结构的反馈式数据预取进行了优化.SPEC2000测试表明,平均性能提高了4.1%. |
| 关键词: 反馈式编译优化 数据预取 跨度反馈 指针追逐 链式数据结构 |
| DOI： |
| 分类号: |
| 基金项目:Supported by the National Basic Research Program of China under Grant No.2007CB310907 (国家重点基础研究发展计划(973)) |
|
| Feedback Directed Prefetching Optimization for Linked Data Structure |
|
QI Feng-Bin,WANG Fei,LI Zhong-Sheng
|
| Abstract: |
| Traditional data prefetching based on static compiler analysis mainly aims at array access. Now, linked data structures due to pointers abound in applications, which is difficult to prefetch by traditional technique. Feedback directed data prefetching optimization has become one of the advanced compiler optimization techniques. it can do well in linked data prefetching. This paper researches the profile guided optimization in ORC compiler, improves the prefetching algorithm according to features of Alpha architecture. SPEC2000 performance test shows 4.1% speedup. |
| Key words: feedback directed optimization data prefetching stride profiling pointer chasing linked data structure |
