Semantics-Based Analysis for Optimizing Compilation of Concurrent Programs

Concurrent programming languages provide a paradigm for describing programs with complicated communication networks in a clear way. They support the creation of multiple computational units called processes that run concurrently in a program together with the mechanism for communication between processes, thereby simplifying the programming of synchronization and communication. Among them, concurrent languages that support finegrained processes are suitable for formal specification of concurrent systems. A runtime system of a fine-grained concurrent language, however, tends to suffer from overheads of primitive operations needed for handling concurrency. Although there is a lot of work on static analysis and optimization techniques for removing these overheads, comprehensive research on the safety of applying multiple optimization techniques has been limited in its scope. For instance, no frameworks have existed that can justify the correctness of static scheduling of concurrent processes and memory reuse. The point is that some issues of optimization cannot be fully expressed as program transformation on the original concurrent language. When these issues are not described formally, the optimization must be carried out on the responsibility of implementors of an optimizing compiler and hence there is no proof that the optimization is really correct. The objective of this dissertation is to clarify (a) how to perform semanticsbased runtime system optimization for concurrent programming languages and (b) how to give theoretical justification to such optimization. To this end, we take two concrete target languages—the concurrent logic programming language and the concurrent graph rewriting language LMNtal (pronounced as elemental)—and discuss for these languages several methods to perform static analysis and runtime system optimization based on the analysis. In this dissertation, several semantics-based program analysis techniques are proposed for optimizing compilation. For concurrent logic programs, the following will be explained: the safety of moving synchronization points as semantics-preserving program transformation for process scheduling, sequentiality analysis and code generation as a justified framework of applying multiple optimization techniques, and occurs-check analysis under cooperative