The MTSA Program Analysis System

This whitepaper describes the basics of using and interpreting the results of the MTSA analysis tool. It also contains the results of several detailed case studies describing the results produced by the analysis for a number of interesting benchmarks, some of the features of the benchmarks that make them particularly interesting problems and examples of how the information provided by the MTSA analysis can be used to perform various program optimizations. The MTSA analysis framework is a shape analysis tool that is designed as a general purpose heap analysis which is capable of providing heap information that is applicable to a wide range of optimization applications (including thread and instruction level parallelization, vectorization, garbage collection, object allocation, scheduling and redundant code elimination). The version described in this paper is based on the Summer 2008 snapshot which supports the majority of the Java programming language (including most of the classes defined in java.util, java.io and java.lang), is capable of efficiently analyzing smaller java programs (up to about 7KLOC in less than a minute), and is able to produce precise and useful information for a wide range of heap structures and algorithms that operate on them (recursive structures, mutation and non-trivial sharing).