Effect of Unique Table Implementation in the Performance of BDD Packages

This paper presents an evaluation of the data structures that implement the unique tables used to represent the Binary Decision Diagrams (BDD) nodes in the strong canonical form of BDDs. The unique table is an important element in BDDs, as it avoids re-computations and no duplicated nodes are created. The choice of the data structures used in the unique tables strongly affects the performance of BDD packages. Therefore, is important to identify some aspects that may contribute to the performance. We compare several implementations of unique tables intending to find the more efficient implementations for modern BDD packages using integers (as opposed to pointers) to identify nodes. Among the tested implementations of unique tables, the data structure using a hash table merged with BDD nodes has resulted in reductions of 34.6% in runtime and 43.5% in memory usage compared to the second-best for each criterion.