Interpolation-Based Function Summaries in Bounded Model Checking

During model checking of software against various specifications, it is often the case that the same parts of the program have to be modeled/verified multiple times. To reduce the overall verification effort, this paper proposes a new technique that extracts function summaries after the initial successful verification run, and then uses them for more efficient subsequent analysis of the other specifications. Function summaries are computed as over-approximations using Craig interpolation, a mechanism which is well-known to preserve the most relevant information, and thus tend to be a good substitute for the functions that were examined in the previous verification runs. In our summarization-based verification approach, the spurious behaviors introduced as a side effect of the over-approximation, are ruled out automatically by means of the counter-example guided refinement of the function summaries. We implemented interpolation-based summarization in our FunFrog tool, and compared it with several state-of-the-art software model checking tools. Our experiments demonstrate the feasibility of the new technique and confirm its advantages on the large programs.