摘要
将经典的数据重用理论扩充到并行领域,分别提出了面向OpenMP和OpenTM应用的并行数据重用理论.针对重用在线程、事务中的关系,系统地讨论了并行应用中重用的分类、判定和求解方法.同时,应用这一理论研究了OpenTM循环的优化技术,以降低事务被回退的风险.最后,使用并行数据重用理论分析和统计了SPEComp2001中的数据重用.并行数据重用理论可以用于指导面向多核存储共享结构的并行程序分析和编译优化技术研究.
This paper extends the famous data reuse theory to a parallel domain and proposes parallel data reuse theories for OpenMP and OpenTM applications, respectively. Through studying the relationships between threads and transactions, the parallel data reuse theories systemically discuss how to classify, judge, and compute reuses in parallel programs. Meanwhile, the optimization framework for reducing OpenTM transactions rolled back is studied. Finally, the data reuses in SPECornp2001 benchmarks are analyzed. The parallel data reuse theories can be used to direct the analysis of parallel applications and the research of compiler optimization techniques on multi-core shared memory architecture.
出处
《软件学报》
EI
CSCD
北大核心
2010年第12期3011-3028,共18页
Journal of Software
基金
国家自然科学基金Nos.60921062
60873014
61003082~~