Optimize context-sensitive Andersen-style points-to analysis by method summarization and cycle-elimination

Abstract

This paper presents an efficient context-sensitive, field-based Andersen-style points-to analysis algorithm for Java programs. This algorithm first summarizes methods of the program under analysis using directed graphs. Then it performs local circle elimination on these summary graphs to reduce their sizes. The main analysis algorithm uses these graphs to construct the main points-to graph. Topological sort and cycle-elimination is performed on the nodes of both main points-to graphs and summary graphs to speed up the transitive closure computation on the main points-to graph. A suite of Java program benchmarks are used to demonstrate the efficiency of our algorithm. © 2010 Springer-Verlag.