基于占优关系的MPI并行程序死锁检测

摘要

MPI parallel programs have been applied broadly and their reliability has attracted more and more attention from the software engineering community.A MPI parallel program contains several processes, each of which has a series of statements executed sequentially.The deadlock is a typical error that occurs in parallel programs.This study proposes a method of detecting deadlocks based on dominance relation.Firstly, the dominance relations of communication edges are firstly conducted from those of communication statements.Secondly, the deadlock is detected by checking the dominance relations of communication edges.The proposed method is preliminarily implemented and used for detecting several MPI parallel programs.The experimental results verify the effectiveness of the proposed method.%随着高性能技术的发展,MPI并行程序得到了越来越广泛的应用,其可靠性也得到了软件工程界越来越多的关注.一个MPI并行程序有多个并行执行的进程,每个进程包含一系列串行执行语句.死锁是并行程序的一类典型错误,本文提出一种基于占优关系的死锁检测方法.首先,根据通信语句之间的占优关系,得到每个进程内,通信边之间的占优关系;然后,通过判定通信边占优关系是否相互违背,检测程序中的死锁.基于上述思想,开发了基于占优关系的死锁检测原型工具,用于检测多个典型的MPI并行程序.结果表明,本文方法能够检测出程序中的死锁缺陷.