Ternary Propagation-Based Local Search for more Bit-Precise Reasoning

Abstract

Current state of the art for reasoning about quantifier-free bit-vector constraints in Satisfiability Modulo Theories (SMT) is a technique called bit-blasting, an eager translation into propositional logic (SAT). While efficient in practice, it may not scale for large bit-widths when the input size cannot be sufficiently reduced with preprocessing techniques. A recent propagation-based local search procedure was shown to be effective on hard satisfiable instances, in particular in combination with bit-blasting in a sequential portfolio setting. However, a major weakness of this approach is its obliviousness to bits that can be simplified to constant values. In this paper, we generalize propagation-based local search with respect to such constant bits to ternary values. We further extend the procedure to handle more bit-vector operators, and introduce heuristics for more precise inverse value computation via bound tightening for inequality constraints. We provide an extensive experimental evaluation and show that the presented techniques yield a considerable improvement in performance.