Unified Memory Analysis

The ability to accurately model the state of program memory and how it evolves during program execution is critical to many optimization and verification techniques. Current memory analysis techniques either provide very accurate information but run prohibitively slowly or run in an acceptable time but produce very conservative results. This paper presents an analysis method that is capable of accurately modeling many important program properties (aliasing, shape, logical data structures, sharing of data elements in collections) while maintaining an acceptable level of performance. Using several examples we show how our abstract model is able to provide detailed information on the properties of interest in the concrete domain. We demonstrate that the model is a safe approximation of the concrete heap and outline the required data flow operations. Finally, we show that the asymptotic runtime of this method is polynomial in the size of the abstract heap and that in practice it is efficient enough to be used in an optimizing compiler.