A General Framework for Analyzing Shared-Memory Parallel Programs

Explicit parallelism greatly complicates the program semantics, especially if concurrent activities are allowed to interact with each other through shared variables. Existing compiler analysis and optimization techniques for sequential programs must be carefully examined before they can be safely applied to parallel programs. In this paper, we present a general framework for analyzing programs with cobegin parallelism, which is based on state space exploration. Reducing the explosion of state space is necessary to make compile-time analysis feasible. This state space explosion problem is relieved first by eliminating redundant interleaving, through the stubborn set theory. Then, abstract interpretation techniques are employed which provide systematic methods for folding related states for further state space reduction. With this framework, we have developed static analysis for obtaining program properties, such as side effects, data dependences, and object lifetimes. The information obtained facilitates program optimization, restructuring, and memory management.