Code Level Context Sensitivity Exploration Algorithm

Abstract

The context sensitivity is an important property in program analysis which can improve the analysis accuracy of the function call context of the source codes and helps to improve the abilities of the compiler optimization and verification procedures. The automata model based on the function call of the program source code is an effective method of modeling the behaviors of the current software, but this method does not consider the context sensitivity of the software function calls. To solve this problem, on the base of high-performance open source compiler Open64, we achieve a source code level context sensitivity exploration algorithm to analyze the relationships between the points of the function calls. Using the iterator, by the circulating way, the algorithm traverses the calling relationships between functions and obtains the context sensitive information analysis results. With this method, it can guide the source code modeling of using the single context sensitivity or the mixed context sensitivity models. In the benchmark sets SPEC2006 and NPB3.3.1, we verify the correctness of the algorithm, and the results show that the proposed algorithm can detect the context sensitivity of the source code level, and the hybrid context sensitivity modeling approach which we used, comparing with other modeling methods, has had a higher accuracy.