Lazy Decision Diagrams for Word-Level Model Manipulation in Software Verification

Word-level predicates involve high-level descriptions of integer variables and can be complex to represent and manipulate with traditional decision diagrams like BDDs (binary decision diagrams) and MDDs (multiple-valued decision diagrams). We propose a new type of decision diagram nodes, called LD-nodes (lazy decision nodes), for word-level inequalities that allow for lazy evaluation. Such nodes can be incorporated in BDDs, MDDs, and CRDs (clock-restriction diagrams). We present algorithms for operations on diagrams with LDD nodes. We then report our experiment of our technology with several benchmarks. A library implementing the approach is available at Source Forge web page for project REDLIB.