Technical Report: Region and Effect Inference for Safe Parallelism

In this paper, we present the first full regions-and-effects inference algorithm for explicitly parallel fork-joinprograms. We infer annotations inspired by Deterministic ParallelJava (DPJ) for a type-safe subset of C++. We chose the DPJannotations because they give the strongest safety guarantees ofany existing concurrency-checking approach we know of, staticor dynamic, and it is also the most expressive static checkingsystem we know of that gives strong safety guarantees. Thisexpressiveness, however, makes manual annotation difficult andtedious, which motivates the need for automatic inference, butit also makes the inference problem very challenging: the codemay use region polymorphism, imperative updates with complexaliasing, arbitrary recursion, hierarchical region specifications,and wildcard elements to describe potentially infinite sets ofregions. We express the inference as a constraint satisfactionproblem and develop, implement, and evaluate an algorithm forsolving it. The region and effect annotations inferred by thealgorithm constitute a checkable proof of safe parallelism, and itcan be recorded both for documentation and for fast and modularsafety checking.