Dynamic Symbolic Verification of MPI Programs

The success of dynamic verification techniques for Message Passing Interface (MPI) programs rests on their ability to address communication nondeterminism. As the number of processes in the program grows, the dynamic verification techniques suffer from the problem of exponential growth in the size of the reachable state space. In this work, we provide a hybrid verification technique for message passing programs that combines explicit-state dynamic verification with symbolic analysis. The dynamic verification component deterministically replays the execution runs of the program, while the symbolic component encodes a set of interleavings of the observed run of the program in a quantifier-free first order logic formula and verifies it for communication deadlocks. In the absence of property violations, it performs analysis to generate a different run of the program that does not fall in the set of already verified runs. We demonstrate the effectiveness of our approach, which is sound and complete, using our prototype tool Hermes. Our evaluation indicates that Hermes performs significantly better than the state-of-the-art verification tools for multi-path MPI programs.