Towards Efficient Parameterized Synthesis

Parameterized synthesis was recently proposed as a way to circumvent the poor scalability of current synthesis tools. The method uses cut-off results in token rings to reduce the problem to bounded distributed synthesis, and thus ultimately to a sequence of SMT problems. This solves the problem of scalability in the size of the architecture, but experiments show that the size of the specification is still a major issue. In this paper we propose several optimizations of the approach. First, we tailor the SMT encoding to systems with isomorphic processes and tokenring architecture. Second, we extend the cut-off results for token rings and refine the reduction, using modularity and abstraction techniques. Some of our optimizations also apply to isomorphic or distributed synthesis in arbitrary architectures. To evaluate these optimizations, we developed the first completely automatic implementation of parameterized synthesis. Experiments show a speed-up of several orders of magnitude, compared to the original method.