Scalable Session Programming for Heterogeneous High-Performance Systems.

Abstract. This paper introduces a programming framework based on the theory of session types for safe and scalable parallel designs. Sessionbased languages can offer a clear and tractable framework to describe communications between parallel components and guarantee communicationsafety and deadlock-freedom by compile-time type checking and parallel MPI code generation. Many representative communication topologies such as ring or scatter-gather can be programmed and verified in session-based programming languages. We use a case study involving N-body simulation, dense and sparse matrix multiplication to illustrate the session-based programming style. Finally, we outline a proposal to integrate session programming with heterogeneous systems for efficient and communication-safe parallel applications by a combination of code generation and type checking