NASA Formal Methods

Volume 6617 of the series Lecture Notes in Computer Science pp 131-145

Efficient Predicate Abstraction of Program Summaries

  • Arie GurfinkelAffiliated withCarnegie Mellon University
  • , Sagar ChakiAffiliated withCarnegie Mellon University
  • , Samir SapraAffiliated withCarnegie Mellon University

* Final gross prices may vary according to local VAT.

Get Access


Predicate abstraction is an effective technique for scaling Software Model Checking to real programs. Traditionally, predicate abstraction abstracts each basic block of a program \(\mathcal{P}\) to construct a small finite abstract model – a Boolean program BP, whose state-transition relation is over some chosen (finite) set of predicates. This is called Small-Block Encoding (SBE). A recent advancement is Large-Block Encoding (LBE) where abstraction is applied to a “summarized” program so that the abstract transitions of BP correspond to loop-free fragments of \(\mathcal{P}\). In this paper, we expand on the original notion of LBE to promote flexibility. We explore and describe efficient ways to perform CEGAR bottleneck operations: generating and solving predicate abstraction queries (PAQs). We make the following contributions. First, we define a general notion of program summarization based on loop cutsets. Second, we give a linear time algorithm to construct PAQs for a loop-free fragment of a program. Third, we compare two approaches to solving PAQs: a classical AllSAT-based one, and a new one based on Linear Decision Diagrams (LDDs). The approaches are evaluated on a large benchmark from open-source software. Our results show that the new LDD-based approach significantly outperforms (and complements) the AllSAT one.