Characterizing the Performance of Task Reductions in OpenMP 5.X Implementations
Pages 35 - 49
Abstract
OpenMP 5.0 added support for reductions over explicit tasks. This expands the previous reduction support that was limited primarily to worksharing and parallel constructs. While the scope of a reduction operation in a worksharing construct is the scope of the construct itself, the scope of a task reduction can vary. This difference requires syntactical means to define the scope of reductions, e.g., the task_reduction clause, and to associate participating tasks, e.g., the in_reduction clause. Furthermore, the disassociation of the number of threads and the number of tasks creates space for different implementations in the OpenMP runtime. In this work, we provide insights into the behavior and performance of task reduction implementations in GCC/g++ and LLVM/Clang. Our results indicate that task reductions are well supported by both compilers, but their performance differs in some cases and is often determined by the efficiency of the underlying task management.
References
[1]
Balart, J., Duran, A., Gonzàlez, M., Martorell, X., Ayguadé, E., Labarta, J.: Nanos Mercurium: a research compiler for OpenMP. In: European Workshop on OpenMP (EWOMP 2004), pp. 103–109 (2004)
[2]
Ciesko J et al. DeRose L, de Supinski BR, Olivier SL, Chapman BM, Müller MS, et al. Task-parallel reductions in OpenMP and OmpSs Using and Improving OpenMP for Devices, Tasks, and More 2014 Cham Springer 1-15
[3]
Ciesko J, Mateo S, Teruel X, Martorell X, Ayguadé E, and Labarta J Maruyama N, de Supinski BR, and Wahib M Supporting adaptive privatization techniques for irregular array reductions in task-parallel programming models OpenMP: Memory, Devices, and Tasks 2016 Cham Springer 336-349
[4]
Ciesko J et al. Terboven C, de Supinski BR, Reble P, Chapman BM, Müller MS, et al. Towards task-parallel reductions in OpenMP OpenMP: Heterogenous Execution and Data Movements 2015 Cham Springer 189-201
[5]
Duran A et al. OmpSs: a proposal for programming heterogeneous multi-core architectures Parallel Process. Lett. 2011 21 02 173-193
[6]
Duran A, Ferrer R, Klemm M, de Supinski BR, and Ayguadé E Sato M, Hanawa T, Müller MS, Chapman BM, and de Supinski BR A proposal for user-defined reductions in OpenMP Beyond Loop Level Parallelism in OpenMP: Accelerators, Tasking and More 2010 Heidelberg Springer 43-55
[7]
Frigo, M., Halpern, P., Leiserson, C.E., Lewin-Berlin, S.: Reducers and other Cilk++ hyperobjects. In: Proceedings of the Twenty-First Annual Symposium on Parallelism in Algorithms and Architectures (SPAA 2009), pp. 79–90. ACM, New York (2009)
[8]
OpenMP Architecture Review Board: OpenMP Application Programming Interface Version 5.0, November 2018. https://www.openmp.org/wp-content/uploads/OpenMP-API-Specification-5.0.pdf
[9]
Pirkelbauer P, Wilson A, Peterson C, and Dechev D Blaze-Tasks: A framework for computing parallel reductions over tasks ACM Trans. Archit. Code Optim. 2019 15 4 1-25
[10]
Shirako J, Cavé V, Zhao J, and Sarkar V Kasahara H and Kimura K Finish accumulators: an efficient reduction construct for dynamic task parallelism Languages and Compilers for Parallel Computing 2013 Heidelberg Springer 264-265
[11]
Shirako, J., Peixotto, D.M., Sarkar, V., Scherer, W.N.: Phaser accumulators: a new reduction construct for dynamic parallelism. In: IEEE International Symposium on Parallel and Distributed Processing (IPDPS 2009), Rome, Italy, pp. 1–12. IEEE (2009)
Index Terms
- Characterizing the Performance of Task Reductions in OpenMP 5.X Implementations
Index terms have been assigned to the content through auto-classification.
Recommendations
X-OpenMP — eXtreme fine-grained tasking using lock-less work stealing
AbstractProcessors with 100s of threads of execution are among the state-of-the-art in high-end computing systems. This transition to many-core computing has required the community to develop new algorithms to overcome significant latency bottlenecks ...
Highlights- The overhead of synchronization methods is quantified on modern hardware with hundreds of cores.
- A novel approach for building a low-overhead parallel runtime system using lockless concurrent queuing and lockless workstealing is ...
Characterizing and mitigating work time inflation in task parallel programs
Selected Papers from Super Computing 2012Task parallelism raises the level of abstraction in shared memory parallel programming to simplify the development of complex applications. However, task parallel applications can exhibit poor performance due to thread idleness, scheduling overheads, ...
Performance comparison of MPI and OpenMP on shared memory multiprocessors: Research Articles
When using a shared memory multiprocessor, the programmer faces the issue of selecting the portable programming model which will provide the best performance. Even if they restricts their choice to the standard programming environments (MPI and OpenMP), ...
Comments
Please enable JavaScript to view thecomments powered by Disqus.Information & Contributors
Information
Published In
Sep 2022
177 pages
ISBN:978-3-031-15921-3
DOI:10.1007/978-3-031-15922-0
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022.
Publisher
Springer-Verlag
Berlin, Heidelberg
Publication History
Published: 27 September 2022
Author Tags
Qualifiers
- Article
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 0Total Downloads
- Downloads (Last 12 months)0
- Downloads (Last 6 weeks)0
Reflects downloads up to 14 Nov 2024
Other Metrics
Citations
View Options
View options
Get Access
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in