Enhancing OpenMP Tasking Model: Performance and Portability
The AMPERE consortium recently presented a groundbreaking paper titled 'Enhancing OpenMP Tasking Model: Performance and Portability' at the International Workshop on OpenMP (IWOMP) 2021. This event marks a significant milestone for the AMPERE project, as it partially covers the consortium's ongoing work in providing recommendations to specification committees, particularly the OpenMP Architecture Review Board (ARB).
The paper, published on SpringerLink, focuses on OpenMP, the de-facto standard programming model in symmetric multiprocessing for High-Performance Computing (HPC). The authors propose a new feature in the OpenMP specification to allow users to define taskgraph regions, i.e., regions of an OpenMP task-based program that can be implemented more efficiently. This enhancement is substantiated on the Task Dependency Graph (TDG) used to represent the execution of a task-based program (or region).
The proposed approach can reduce the runtime overhead related to task management and offers a higher abstraction level than previously presented methods. This makes it easier for existing OpenMP implementations to integrate taskgraph, thus benefiting from several layers of optimization. The paper also presents preliminary results on the benefits that can be extracted from this new feature, considering performance gain and interoperability provided by the TDG.
The presentation of this paper at IWOMP 2021 is an important step for the AMPERE project. It demonstrates the consortium's commitment to enhancing the performance and portability of the OpenMP tasking model, contributing to the advancement of HPC and parallel computing.
The AMPERE project is funded by the European Union’s Horizon 2020 research and innovation programme under grant agreement No 871669.