Sub-domain Selection Strategies For Floating Point Constraint Systems

Heytem Zitoun (1, 2) , Claude Michel (1, 2) , Michel Rueher (1, 2) , Laurent Michel (3)

1 COMUE UCA - COMUE Université Côte d'Azur (2015-2019)
2 I3S - Laboratoire d'Informatique, Signaux, et Systèmes de Sophia Antipolis
3 UCONN - University of Connecticut

Program verification is a key issue for critical applications such as aviation, aerospace, or embedded systems. Bounded model checking (BMC) and constraint programming (CBMC, CBPV, ...) approaches are based on counterexamples that violate a property of the program to verify. Searching for such counterexamples can be very long and costly when the programs to check contains floating point computations. This stems from the fact that existing search strategies have been designed for discrete domains and, to a lesser extent, continuous domains. In [12], we have introduced a set of variable choice strategies that take advantages of the specificities of the floats, e.g., domain density, cancellation and absorption phenomena. In this paper we introduce new sub-domain selection strategies targeting domains involved in absorption and using techniques derived from higher order consistencies. Preliminary experiments on a significant set of benchmarks are very promising.

Domaines: Informatique [cs]

Heytem Zitoun, Claude Michel, Michel Rueher, Laurent Michel. Sub-domain Selection Strategies For Floating Point Constraint Systems. 2018. ⟨hal-01956094⟩