摘要： 针对当前并发程序的不确定性和复杂性，以及程序原子性质获取困难的问题，提出一种并发程序原子性质提取方法。将并发程序中的同步区域转化为与并发操作相关的并发操作图后，采用频繁子图挖掘算法自动提取程序中的原子图，使其能刻画并发程序的原子性质，包括并发操作以及操作之间的控制依赖关系。实验结果证明，该方法能以较低的误测率有效提取并发程序的原子性质。

关键词: 并发程序, 原子图, 原子性质, 多线程, 频繁子图挖掘

Abstract: Because of the complexity and uncertainty of parallel behaviors, currently it is difficult to get atom rules from concurrent programs. Aiming at the problem, this paper presents a new atom property extraction method in concurrent program. A technique to automatically extract atom property from programs forms graph structure by using a frequent subgraph mining algorithm. It targets at inferring atomicity of both low level read-write operations as well as high level functional operations on multiple variables with rich structure information. Experimental result shows that this method manages to find atom property with a relatively high accuracy.

Key words: concurrent program, atom graph, atom property, multi-threading, frequency subgraph mining

中图分类号: 

• TP12