An Empirical Performance Study of Chapel Programming Language

Abstract

In this paper we evaluate the performance of the Chapel programming language from the perspective of its language primitives and features, where the microbenchmarks are synthesized from our lessons learned in developing molecular dynamics simulation programs in Chapel. Experimental results show that most language building blocks have comparable performance to corresponding hand-written C code, while the complex applications can achieve up to 70% of the performance of C implementation. We identify several causes of overhead that can be further optimized by Chapel compiler. This work not only helps Chapel users understand the performance implication of using Chapel, but also provides useful feedbacks for Chapel developers to make a better compiler.