Composing Parallel Runtime Systems: A Case Study in How to Compose the Julia and OpenCilk Runtimes

Abstract

Julia [5] [15] is a high-level computing language used by many developers for its performance and ease of use. Julia operates on tasks that are run concurrently on threads. In its current state, however, Julia is not able to effectively employ fine-grained parallelism. OpenCilk [9] is an open-source implementation of the Cilk concurrency platform designed to utilize fine-grain parallelism. The Cilk runtime system, based on Cheetah [12], offers provably efficient parallel scheduling whose performance is borne out in theory and practice. I propose a combination of the Julia and OpenCilk runtimes through the integration of multiple components. One contribution of this thesis is a novel algorithm for combining C/C++ memory allocations with Julia’s precise garbage collector. Composing the parallelism of OpenCilk and Julia enables programmers to write efficient multithreaded code. Additionally, this work is a case study of combining the high levels of parallelism present in Cilk with a high-level language.