Algorithms for software model checking: Predicate abstraction vs. Impact

CEGAR, SMT solving, and Craig interpolation are successful approaches for software model checking. We compare two of the most important algorithms that are based on these techniques: lazy predicate abstraction (as in Blast) and lazy abstraction with interpolants (as in Impact). We unify the algorithms formally (by expressing both in the CPA framework) as well as in practice (by implementing them in the same tool). This allows us to flexibly experiment with new configurations and gain new insights, both about their most important differences and commonalities, as well as about their performance characteristics. We show that the essential contribution of the Impact algorithm is the reduction of the number of refinements, and compare this to another approach for reducing refinement effort: adjustable-block encoding (ABE).