Detrimental task execution patterns in mainstream OpenMP runtimes

TL;DR

This paper identifies performance issues caused by inconsistent implementation of OpenMP task annotations and proposes prescriptive clauses to improve runtime behavior, demonstrated through astrophysical simulation code analysis.

Contribution

It introduces prescriptive clauses for OpenMP task annotations to mitigate performance flaws caused by quasi-standard behaviors in mainstream runtimes.

Findings

Quasi-standard behaviors can cause performance flaws in OpenMP tasks.

Proposed clauses can constrain implementations to improve performance.

Incremental changes in runtimes can realize these clauses.

Abstract

The OpenMP API offers both task-based and data-parallel concepts to scientific computing. While it provides descriptive and prescriptive annotations, it is in many places deliberately unspecific how to implement its annotations. As the predominant OpenMP implementations share design rationales, they introduce "quasi-standards how certain annotations behave. By means of a task-based astrophysical simulation code, we highlight situations where this "quasi-standard" reference behaviour introduces performance flaws. Therefore, we propose prescriptive clauses to constrain the OpenMP implementations. Simulated task traces uncover the clauses' potential, while a discussion of their realization highlights that they would manifest in rather incremental changes to any OpenMP runtime supporting task priorities.