基于Range Test的交互式数据相关性分析技术

数据相关性分析是开发程序并行性最重要的技术之一 ,但是传统的自动数据相关性分析难以处理实际应用程序的复杂情况 ,而且 ,程序中固有的语义信息也无法为并行化工具所理解。这种状况限制了现有并行化系统的并行效果。一种新技术——交互式数据相关性分析可以改善这一状况 ,其中核心的数据相关性信息提取模块能够将自动数据相关性分析所得的不确定性结果转化成一些简单的问题并提交给用户 ,从而使用户与并行化工具紧密合作 ,改进数据相关性分析的结果 ,进而提高并行化系统的能力。交互式数据相关性分析技术已在清华大学开发的 TIPS并行化系统中实现了。

Interactive data dependence analysis based on range tests

Data dependence analysis is one of the most important techniques for the exploitation of the inherent parallelism in sequential programs, but traditional automatic data dependence analysis can not manipulate the complexities of real applications. In addition, programs have some semantic information that the automatic tools can never know without the user's knowledge. This situation has limited the performance of automatic parallelizing systems. This paper describes a new technique, interactive data dependence analysis, to improve data dependence analysis. A data dependence information extractor is developed to transform the uncertain result of the automatic data dependence analysis into some simple questions submitted to the user, to combine the power of the computer and the user to improve the data dependence analysis and thus, the performance of parallelizing systems. The interactive data dependence analysis has been implemented in TIPS, a parallelizing compiler being developed at Tsinghua University.