Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model

Matt Martineau; Simon McIntosh-Smith; Wayne Gaudin

doi:10.1109/IPDPSW.2016.70

Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model

Matt Martineau, Simon McIntosh-Smith, Wayne Gaudin

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

35 Citations (Scopus)

436 Downloads (Pure)

Abstract

Although the OpenMP 4.0 standard has been available since 2013, support for GPUs has been absent up until very recently, with only a handful of experimental compilers available. In this work we evaluate the performance of Cray's new NVIDIA GPU targeting implementation of OpenMP 4.0, with the mini-apps TeaLeaf, CloverLeaf and BUDE. We successfully port each of the applications, using a simple and consistent design throughout, and achieve performance on an NVIDIA K20X that is comparable to Cray's OpenACC in all cases. BUDE, a compute bound code, required 2.2x the runtime of an equivalently optimised CUDA code, which we believe is caused by an inflated frequency of control flow operations and less efficient arithmetic optimisation. Impressively, both TeaLeaf and CloverLeaf, memory bandwidth bound codes, only required 1.3x the runtime of hand-optimised CUDA implementations. Overall, we find that OpenMP 4.0 is a highly usable open standard capable of performant heterogeneous execution, making it a promising option for scientific application developers.

Original language	English
Title of host publication	Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016
Publisher	Institute of Electrical and Electronics Engineers (IEEE)
Pages	338-347
Number of pages	10
ISBN (Electronic)	9781509021406
DOIs	https://doi.org/10.1109/IPDPSW.2016.70
Publication status	Published - 4 Aug 2016
Event	30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016 - Chicago, United States Duration: 23 May 2016 → 27 May 2016

Conference

Conference	30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016
Country/Territory	United States
City	Chicago
Period	23/05/16 → 27/05/16

Keywords

Application programming interfaces
High performance computing
OpenMP
Parallel computing
Performance portability

Access to Document

10.1109/IPDPSW.2016.70

Full-text PDF (accepted author manuscript)
This is the author accepted manuscript (AAM). The final published version (version of record) is available online via IEEE at http://ieeexplore.ieee.org/document/7529889/?arnumber=7529889.
Accepted author manuscript, 267 KB

Handle.net

Persistent link

Cite this

Martineau, M., McIntosh-Smith, S., & Gaudin, W. (2016). Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model. In Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016 (pp. 338-347). Article 7529889 Institute of Electrical and Electronics Engineers (IEEE). https://doi.org/10.1109/IPDPSW.2016.70

@inproceedings{1ca3fc1149f14fa4a818d742568c9a4d,

title = "Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model",

abstract = "Although the OpenMP 4.0 standard has been available since 2013, support for GPUs has been absent up until very recently, with only a handful of experimental compilers available. In this work we evaluate the performance of Cray's new NVIDIA GPU targeting implementation of OpenMP 4.0, with the mini-apps TeaLeaf, CloverLeaf and BUDE. We successfully port each of the applications, using a simple and consistent design throughout, and achieve performance on an NVIDIA K20X that is comparable to Cray's OpenACC in all cases. BUDE, a compute bound code, required 2.2x the runtime of an equivalently optimised CUDA code, which we believe is caused by an inflated frequency of control flow operations and less efficient arithmetic optimisation. Impressively, both TeaLeaf and CloverLeaf, memory bandwidth bound codes, only required 1.3x the runtime of hand-optimised CUDA implementations. Overall, we find that OpenMP 4.0 is a highly usable open standard capable of performant heterogeneous execution, making it a promising option for scientific application developers.",

keywords = "Application programming interfaces, High performance computing, OpenMP, Parallel computing, Performance portability",

author = "Matt Martineau and Simon McIntosh-Smith and Wayne Gaudin",

year = "2016",

month = aug,

day = "4",

doi = "10.1109/IPDPSW.2016.70",

language = "English",

pages = "338--347",

booktitle = "Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016",

publisher = "Institute of Electrical and Electronics Engineers (IEEE)",

address = "United States",

note = "30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016 ; Conference date: 23-05-2016 Through 27-05-2016",

}

Martineau, M, McIntosh-Smith, S & Gaudin, W 2016, Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model. in Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016., 7529889, Institute of Electrical and Electronics Engineers (IEEE), pp. 338-347, 30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016, Chicago, United States, 23/05/16. https://doi.org/10.1109/IPDPSW.2016.70

Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model. / Martineau, Matt; McIntosh-Smith, Simon; Gaudin, Wayne.
Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016. Institute of Electrical and Electronics Engineers (IEEE), 2016. p. 338-347 7529889.

Research output: Chapter in Book/Report/Conference proceeding › Conference Contribution (Conference Proceeding)

TY - GEN

T1 - Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model

AU - Martineau, Matt

AU - McIntosh-Smith, Simon

AU - Gaudin, Wayne

PY - 2016/8/4

Y1 - 2016/8/4

N2 - Although the OpenMP 4.0 standard has been available since 2013, support for GPUs has been absent up until very recently, with only a handful of experimental compilers available. In this work we evaluate the performance of Cray's new NVIDIA GPU targeting implementation of OpenMP 4.0, with the mini-apps TeaLeaf, CloverLeaf and BUDE. We successfully port each of the applications, using a simple and consistent design throughout, and achieve performance on an NVIDIA K20X that is comparable to Cray's OpenACC in all cases. BUDE, a compute bound code, required 2.2x the runtime of an equivalently optimised CUDA code, which we believe is caused by an inflated frequency of control flow operations and less efficient arithmetic optimisation. Impressively, both TeaLeaf and CloverLeaf, memory bandwidth bound codes, only required 1.3x the runtime of hand-optimised CUDA implementations. Overall, we find that OpenMP 4.0 is a highly usable open standard capable of performant heterogeneous execution, making it a promising option for scientific application developers.

AB - Although the OpenMP 4.0 standard has been available since 2013, support for GPUs has been absent up until very recently, with only a handful of experimental compilers available. In this work we evaluate the performance of Cray's new NVIDIA GPU targeting implementation of OpenMP 4.0, with the mini-apps TeaLeaf, CloverLeaf and BUDE. We successfully port each of the applications, using a simple and consistent design throughout, and achieve performance on an NVIDIA K20X that is comparable to Cray's OpenACC in all cases. BUDE, a compute bound code, required 2.2x the runtime of an equivalently optimised CUDA code, which we believe is caused by an inflated frequency of control flow operations and less efficient arithmetic optimisation. Impressively, both TeaLeaf and CloverLeaf, memory bandwidth bound codes, only required 1.3x the runtime of hand-optimised CUDA implementations. Overall, we find that OpenMP 4.0 is a highly usable open standard capable of performant heterogeneous execution, making it a promising option for scientific application developers.

KW - Application programming interfaces

KW - High performance computing

KW - OpenMP

KW - Parallel computing

KW - Performance portability

UR - http://www.scopus.com/inward/record.url?scp=84991577081&partnerID=8YFLogxK

U2 - 10.1109/IPDPSW.2016.70

DO - 10.1109/IPDPSW.2016.70

M3 - Conference Contribution (Conference Proceeding)

AN - SCOPUS:84991577081

SP - 338

EP - 347

BT - Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016

PB - Institute of Electrical and Electronics Engineers (IEEE)

T2 - 30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016

Y2 - 23 May 2016 through 27 May 2016

ER -

Evaluating OpenMP 4.0's Effectiveness as a Heterogeneous Parallel Programming Model

Abstract

Conference

Keywords

Access to Document

Handle.net

Other files and links

Fingerprint

Cite this