Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs

Joshua Hoke Davis; Christopher Daley; Swaroop Pophale; Thomas Huber; Sunita Chandrasekaran; Nicholas J. Wright

doi:10.1007/978-3-030-74224-9_2

Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs

Joshua Hoke Davis, Christopher Daley, Swaroop Pophale, Thomas Huber, Sunita Chandrasekaran, Nicholas J. Wright

Programming Systems

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

10 Scopus citations

Abstract

Heterogeneous systems are becoming increasingly prevalent. In order to exploit the rich compute resources of such systems, robust programming models are needed for application developers to seamlessly migrate legacy code from today’s systems to tomorrow’s. Over the past decade and more, directives have been established as one of the promising paths to tackle programmatic challenges on emerging systems. This work focuses on applying and demonstrating OpenMP offloading directives on five proxy applications. We observe that the performance varies widely from one compiler to the other; a crucial aspect of our work is reporting best practices to application developers who use OpenMP offloading compilers. While some issues can be worked around by the developer, there are other issues that must be reported to the compiler vendors. By restructuring OpenMP offloading directives, we gain an 18x speedup for the su3 proxy application on NERSC’s Cori system when using the Clang compiler, and a 15.7x speedup by switching max reductions to add reductions in the laplace mini-app when using the Cray-llvm compiler on Cori.

Original language	English
Title of host publication	Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings
Editors	Sridutt Bhalachandra, Sandra Wienke, Sunita Chandrasekaran, Guido Juckeland
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	25-44
Number of pages	20
ISBN (Print)	9783030742232
DOIs	https://doi.org/10.1007/978-3-030-74224-9_2
State	Published - 2021
Event	7th International Workshop on Accelerator Programming using Directives, WACCPD 2020 - Virtual, Online Duration: Nov 20 2020 → Nov 20 2020

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12655 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	7th International Workshop on Accelerator Programming using Directives, WACCPD 2020
City	Virtual, Online
Period	11/20/20 → 11/20/20

Funding

Acknowledgements. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility operated under Contract No. DE-AC02-05CH11231. This research also used resources of the Oak Ridge Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC05-00OR22725. The authors would like to thank Doug Doerfler and Rahul Gayatri for helpful discussion about the su3 benchmark and useful research directions for this project.

Keywords

Directive-based programming
GPU
Heterogeneous systems
NVIDIA
OpenMP
Performance portability
V100

Access to Document

10.1007/978-3-030-74224-9_2

Cite this

Davis, J. H., Daley, C., Pophale, S., Huber, T., Chandrasekaran, S., & Wright, N. J. (2021). Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs. In S. Bhalachandra, S. Wienke, S. Chandrasekaran, & G. Juckeland (Eds.), Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings (pp. 25-44). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12655 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-74224-9_2

Davis, Joshua Hoke ; Daley, Christopher ; Pophale, Swaroop et al. / Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs. Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings. editor / Sridutt Bhalachandra ; Sandra Wienke ; Sunita Chandrasekaran ; Guido Juckeland. Springer Science and Business Media Deutschland GmbH, 2021. pp. 25-44 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{ccf3bf79d1064f16b2a1ef3a7883613b,

title = "Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs",

abstract = "Heterogeneous systems are becoming increasingly prevalent. In order to exploit the rich compute resources of such systems, robust programming models are needed for application developers to seamlessly migrate legacy code from today{\textquoteright}s systems to tomorrow{\textquoteright}s. Over the past decade and more, directives have been established as one of the promising paths to tackle programmatic challenges on emerging systems. This work focuses on applying and demonstrating OpenMP offloading directives on five proxy applications. We observe that the performance varies widely from one compiler to the other; a crucial aspect of our work is reporting best practices to application developers who use OpenMP offloading compilers. While some issues can be worked around by the developer, there are other issues that must be reported to the compiler vendors. By restructuring OpenMP offloading directives, we gain an 18x speedup for the su3 proxy application on NERSC{\textquoteright}s Cori system when using the Clang compiler, and a 15.7x speedup by switching max reductions to add reductions in the laplace mini-app when using the Cray-llvm compiler on Cori.",

keywords = "Directive-based programming, GPU, Heterogeneous systems, NVIDIA, OpenMP, Performance portability, V100",

author = "Davis, {Joshua Hoke} and Christopher Daley and Swaroop Pophale and Thomas Huber and Sunita Chandrasekaran and Wright, {Nicholas J.}",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 7th International Workshop on Accelerator Programming using Directives, WACCPD 2020 ; Conference date: 20-11-2020 Through 20-11-2020",

year = "2021",

doi = "10.1007/978-3-030-74224-9_2",

language = "English",

isbn = "9783030742232",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "25--44",

editor = "Sridutt Bhalachandra and Sandra Wienke and Sunita Chandrasekaran and Guido Juckeland",

booktitle = "Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings",

}

Davis, JH, Daley, C, Pophale, S, Huber, T, Chandrasekaran, S & Wright, NJ 2021, Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs. in S Bhalachandra, S Wienke, S Chandrasekaran & G Juckeland (eds), Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12655 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 25-44, 7th International Workshop on Accelerator Programming using Directives, WACCPD 2020, Virtual, Online, 11/20/20. https://doi.org/10.1007/978-3-030-74224-9_2

Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs. / Davis, Joshua Hoke; Daley, Christopher; Pophale, Swaroop et al.
Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings. ed. / Sridutt Bhalachandra; Sandra Wienke; Sunita Chandrasekaran; Guido Juckeland. Springer Science and Business Media Deutschland GmbH, 2021. p. 25-44 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12655 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs

AU - Davis, Joshua Hoke

AU - Daley, Christopher

AU - Pophale, Swaroop

AU - Huber, Thomas

AU - Chandrasekaran, Sunita

AU - Wright, Nicholas J.

PY - 2021

Y1 - 2021

N2 - Heterogeneous systems are becoming increasingly prevalent. In order to exploit the rich compute resources of such systems, robust programming models are needed for application developers to seamlessly migrate legacy code from today’s systems to tomorrow’s. Over the past decade and more, directives have been established as one of the promising paths to tackle programmatic challenges on emerging systems. This work focuses on applying and demonstrating OpenMP offloading directives on five proxy applications. We observe that the performance varies widely from one compiler to the other; a crucial aspect of our work is reporting best practices to application developers who use OpenMP offloading compilers. While some issues can be worked around by the developer, there are other issues that must be reported to the compiler vendors. By restructuring OpenMP offloading directives, we gain an 18x speedup for the su3 proxy application on NERSC’s Cori system when using the Clang compiler, and a 15.7x speedup by switching max reductions to add reductions in the laplace mini-app when using the Cray-llvm compiler on Cori.

AB - Heterogeneous systems are becoming increasingly prevalent. In order to exploit the rich compute resources of such systems, robust programming models are needed for application developers to seamlessly migrate legacy code from today’s systems to tomorrow’s. Over the past decade and more, directives have been established as one of the promising paths to tackle programmatic challenges on emerging systems. This work focuses on applying and demonstrating OpenMP offloading directives on five proxy applications. We observe that the performance varies widely from one compiler to the other; a crucial aspect of our work is reporting best practices to application developers who use OpenMP offloading compilers. While some issues can be worked around by the developer, there are other issues that must be reported to the compiler vendors. By restructuring OpenMP offloading directives, we gain an 18x speedup for the su3 proxy application on NERSC’s Cori system when using the Clang compiler, and a 15.7x speedup by switching max reductions to add reductions in the laplace mini-app when using the Cray-llvm compiler on Cori.

KW - Directive-based programming

KW - GPU

KW - Heterogeneous systems

KW - NVIDIA

KW - OpenMP

KW - Performance portability

KW - V100

UR - http://www.scopus.com/inward/record.url?scp=85105965004&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-74224-9_2

DO - 10.1007/978-3-030-74224-9_2

M3 - Conference contribution

AN - SCOPUS:85105965004

SN - 9783030742232

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 25

EP - 44

BT - Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings

A2 - Bhalachandra, Sridutt

A2 - Wienke, Sandra

A2 - Chandrasekaran, Sunita

A2 - Juckeland, Guido

PB - Springer Science and Business Media Deutschland GmbH

T2 - 7th International Workshop on Accelerator Programming using Directives, WACCPD 2020

Y2 - 20 November 2020 through 20 November 2020

ER -

Davis JH, Daley C, Pophale S, Huber T, Chandrasekaran S, Wright NJ. Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs. In Bhalachandra S, Wienke S, Chandrasekaran S, Juckeland G, editors, Accelerator Programming Using Directives - 7th International Workshop, WACCPD 2020, Proceedings. Springer Science and Business Media Deutschland GmbH. 2021. p. 25-44. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-74224-9_2

Performance Assessment of OpenMP Compilers Targeting NVIDIA V100 GPUs

Abstract

Publication series

Conference

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this