Cartesian Partial-Order Reduction

Verifying concurrent programs is challenging since the number of thread interleavings that need to be explored can be huge even for moderate programs. We present a cartesian semantics that reduces the amount of nondeterminism in concurrent programs by delaying unnecessary context switches. Using this semantics, we construct a novel dynamic partial-order reduction algorithm. The cartesian semantics can be used to create other partial-order reduction algorithms and can also be used as a basis for abstract interpretation. We have implemented our algorithm and evaluate it on a small set of benchmarks. Our preliminary experimental results show a significant potential saving in the number of explored states and transitions.