Modular Programming of Synchronization and Communication Among Tasks in Parallel Programs

Abstract

Implementing synchronization and communication among tasks in parallel programs is a major challenge. We present a high-level DSL geared toward this challenge, by generalizing the existing protocol language Reo from supporting only a compile-time/statically set number of tasks (unsuitable for parallel programming), to supporting also a run-time/dynamically set number of tasks. Our contribution comprises new syntax, a new compilation/execution approach, and experimental results. Most surprisingly, the new approach can outperform the existing approach, even though the new approach requires more work to be done at run-time.