An algebraic framework for optimizing parallel programs

The paper proposes a theoretical framework for verifying and deriving code optimizations for programs written in parallel programming languages. The key idea of this framework is to formalize code optimizations as compositional transformation rules for programs presented as terms of an enriched process calculus. The rules are formulated on the basis of an algebraic order relation between two programs which state that they are behaviorally equivalent and one of them is faster than the other. The correctness and effectiveness of optimized programs derived from the rules can be ensured in all circumstances. The framework is unique among other existing works in being able to quantitatively analyze the temporal costs of synchronizations among parallel programs. The paper presents basic ideas and definitions of the framework with several examples