Getting More From Your Multicore: Exploiting OpenMP for Astronomy

TL;DR

This paper demonstrates how astronomers can leverage OpenMP to automatically parallelize array-based analysis tasks in scripting environments, significantly improving performance on multicore systems without requiring parallel programming expertise.

Contribution

The authors show how to adapt the ISIS astrophysical analysis system to exploit OpenMP for parallelism, enabling efficient multicore utilization in astronomy research workflows.

Findings

Speedups achieved on various hardware and compiler setups.

ISIS can exploit a wide range of parallelism from single to multiple machines.

Automatic parallelization of array loops enhances analysis efficiency.

Abstract

Motivated by the emergence of multicore architectures, and the reality that parallelism is rarely used for analysis in observational astronomy, we demonstrate how general users may employ tightly-coupled multiprocessors in scriptable research calculations while requiring no special knowledge of parallel programming. Our method rests on the observation that much of the appeal of high-level vectorized languages like IDL or MatLab stems from relatively simple internal loops over regular array structures, and that these loops are highly amenable to automatic parallelization with OpenMP. We discuss how ISIS, an open-source astrophysical analysis system embedding the S-Lang numerical language, was easily adapted to exploit this pattern. Drawing from a common astrophysical problem, model fitting, we present beneficial speedups for several machine and compiler configurations. These results complement our previous efforts with PVM, and together lead us to believe that ISIS is the only general purpose spectroscopy system in which such a range of parallelism -- from single processors on multiple machines to multiple processors on single machines -- has been demonstrated.