Automatically deriving cost models for structured parallel processes using hylomorphisms

Abstract

Structured parallelism using nested algorithmic skeletons can greatly ease the task of writing parallel software, since common, but hard-to-debug, problems such as race conditions are eliminated by design. However, choosing the best combination of algorithmic skeletons to yield good parallel speedups for a specific program on a specific parallel architecture is still a difficult problem. This paper uses the unifying notion of hylomorphisms, a general recursion pattern, to make it possible to reason about both the functional correctness properties and the extra-functional timing properties of structured parallel programs. We have previously used hylomorphisms to provide a denotational semantics for skeletons, and proved that a given parallel structure for a program satisfies functional correctness. This paper expands on this theme, providing a simple operational semantics for algorithmic skeletons and a cost semantics that can be automatically derived from that operational semantics. We prove that both semantics are sound with respect to our previously defined denotational semantics. This means that we can now automatically and statically choose a provably optimal parallel structure for a given program with respect to a cost model for a (class of) parallel architecture. By deriving an automatic amortised analysis from our cost model, we can also accurately predict parallel runtimes and speedups.