On the performance portability of structured grid codes on many-core computer architectures

Simon McIntosh-Smith; Michael Boulton; Daniel Curran; James Price

doi:10.1007/978-3-319-07518-1_4

On the performance portability of structured grid codes on many-core computer architectures

Simon McIntosh-Smith, Michael Boulton, Daniel Curran, James Price

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

30 Citations (Scopus)

Abstract

With the advent of many-core computer architectures such as GPGPUs from NVIDIA and AMD, and more recently Intel's Xeon Phi, ensuring performance portability of HPC codes is potentially becoming more complex. In this work we have focused on one important application area - structured grid codes - and investigated techniques for ensuring performance portability across a diverse range of different, high-end many-core architectures. We chose three codes to investigate: a 3D lattice Boltzmann code (D3Q19 BGK), the CloverLeaf hydrodynamics mini application from Sandia's Mantevo benchmark suite, and ROTORSIM, a production-quality structured grid, multiblock, compressible finite-volume CFD code. We have developed OpenCL versions of these codes in order to provide cross-platform functional portability, and compared the performance of the OpenCL versions of these structured grid codes to optimized versions on each platform, including hybrid OpenMP/MPI/AVX versions on CPUs and Xeon Phi, and CUDA versions on NVIDIA GPUs. Our results show that, contrary to conventional wisdom, using OpenCL it is possible to achieve a high degree of performance portability, at least for structured grid applications, using a set of straightforward techniques. The performance portable code in OpenCL is also highly competitive with the best performance using the native parallel programming models on each platform.

Original language	English
Title of host publication	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Publisher	Springer Verlag
Pages	53-75
Number of pages	23
Volume	8488 LNCS
ISBN (Print)	9783319075174
DOIs	https://doi.org/10.1007/978-3-319-07518-1_4
Publication status	Published - 1 Jan 2014
Event	29th International Supercomputing Conference, ISC 2014 - Leipzig, United Kingdom Duration: 22 Jun 2014 → 26 Jun 2014

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	8488 LNCS
ISSN (Print)	03029743
ISSN (Electronic)	16113349

Conference

Conference	29th International Supercomputing Conference, ISC 2014
Country/Territory	United Kingdom
City	Leipzig
Period	22/06/14 → 26/06/14

Bibliographical note

ISC is one of the top HPC conferences world-wide, second only to IEEE/ACM SuperComputing in the US.

This paper is one of the first to demonstrate highly performance portable code by exploiting OpenCL.

Keywords

GPU
heterogeneous
lattice Boltzmann
Many-core
multi-grid multi-block
structured grid
Xeon Phi

Access to Document

10.1007/978-3-319-07518-1_4

http://link.springer.com/chapter/10.1007/978-3-319-07518-1_4

Handle.net

Persistent link

Future-proof massively-parallel execution of multi-block applications
McIntosh-Smith, S. N. (Principal Investigator)
1/10/13 → 31/03/17
Project: Research

HPC (High Performance Computing) and HTC (High Throughput Computing) Facilities
Alam, S. R. (Manager), Eccleston, P. E. (Other), Williams, D. A. G. (Manager) & Atack, S. H. (Other)
Facility/equipment: Facility

Professor Simon N McIntosh-Smith
- S.McIntosh-Smithbristol.acuk
- School of Computer Science - Professor in High Performance Computing
- Microelectronics
Person: Academic , Group lead

Cite this

McIntosh-Smith, S., Boulton, M., Curran, D., & Price, J. (2014). On the performance portability of structured grid codes on many-core computer architectures. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8488 LNCS, pp. 53-75). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8488 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-07518-1_4

McIntosh-Smith, Simon ; Boulton, Michael ; Curran, Daniel et al. / On the performance portability of structured grid codes on many-core computer architectures. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 8488 LNCS Springer Verlag, 2014. pp. 53-75 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{d257f78005de4dc0ac93a5ebbd48faf1,

title = "On the performance portability of structured grid codes on many-core computer architectures",

abstract = "With the advent of many-core computer architectures such as GPGPUs from NVIDIA and AMD, and more recently Intel's Xeon Phi, ensuring performance portability of HPC codes is potentially becoming more complex. In this work we have focused on one important application area - structured grid codes - and investigated techniques for ensuring performance portability across a diverse range of different, high-end many-core architectures. We chose three codes to investigate: a 3D lattice Boltzmann code (D3Q19 BGK), the CloverLeaf hydrodynamics mini application from Sandia's Mantevo benchmark suite, and ROTORSIM, a production-quality structured grid, multiblock, compressible finite-volume CFD code. We have developed OpenCL versions of these codes in order to provide cross-platform functional portability, and compared the performance of the OpenCL versions of these structured grid codes to optimized versions on each platform, including hybrid OpenMP/MPI/AVX versions on CPUs and Xeon Phi, and CUDA versions on NVIDIA GPUs. Our results show that, contrary to conventional wisdom, using OpenCL it is possible to achieve a high degree of performance portability, at least for structured grid applications, using a set of straightforward techniques. The performance portable code in OpenCL is also highly competitive with the best performance using the native parallel programming models on each platform.",

keywords = "GPU, heterogeneous, lattice Boltzmann, Many-core, multi-grid multi-block, structured grid, Xeon Phi",

author = "Simon McIntosh-Smith and Michael Boulton and Daniel Curran and James Price",

note = "ISC is one of the top HPC conferences world-wide, second only to IEEE/ACM SuperComputing in the US. This paper is one of the first to demonstrate highly performance portable code by exploiting OpenCL.; 29th International Supercomputing Conference, ISC 2014 ; Conference date: 22-06-2014 Through 26-06-2014",

year = "2014",

month = jan,

day = "1",

doi = "10.1007/978-3-319-07518-1_4",

language = "English",

isbn = "9783319075174",

volume = "8488 LNCS",

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

publisher = "Springer Verlag",

pages = "53--75",

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

address = "Germany",

}

McIntosh-Smith, S, Boulton, M, Curran, D & Price, J 2014, On the performance portability of structured grid codes on many-core computer architectures. in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). vol. 8488 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 8488 LNCS, Springer Verlag, pp. 53-75, 29th International Supercomputing Conference, ISC 2014, Leipzig, United Kingdom, 22/06/14. https://doi.org/10.1007/978-3-319-07518-1_4

On the performance portability of structured grid codes on many-core computer architectures. / McIntosh-Smith, Simon; Boulton, Michael; Curran, Daniel et al.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 8488 LNCS Springer Verlag, 2014. p. 53-75 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 8488 LNCS).

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

TY - GEN

T1 - On the performance portability of structured grid codes on many-core computer architectures

AU - McIntosh-Smith, Simon

AU - Boulton, Michael

AU - Curran, Daniel

AU - Price, James

N1 - ISC is one of the top HPC conferences world-wide, second only to IEEE/ACM SuperComputing in the US. This paper is one of the first to demonstrate highly performance portable code by exploiting OpenCL.

PY - 2014/1/1

Y1 - 2014/1/1

N2 - With the advent of many-core computer architectures such as GPGPUs from NVIDIA and AMD, and more recently Intel's Xeon Phi, ensuring performance portability of HPC codes is potentially becoming more complex. In this work we have focused on one important application area - structured grid codes - and investigated techniques for ensuring performance portability across a diverse range of different, high-end many-core architectures. We chose three codes to investigate: a 3D lattice Boltzmann code (D3Q19 BGK), the CloverLeaf hydrodynamics mini application from Sandia's Mantevo benchmark suite, and ROTORSIM, a production-quality structured grid, multiblock, compressible finite-volume CFD code. We have developed OpenCL versions of these codes in order to provide cross-platform functional portability, and compared the performance of the OpenCL versions of these structured grid codes to optimized versions on each platform, including hybrid OpenMP/MPI/AVX versions on CPUs and Xeon Phi, and CUDA versions on NVIDIA GPUs. Our results show that, contrary to conventional wisdom, using OpenCL it is possible to achieve a high degree of performance portability, at least for structured grid applications, using a set of straightforward techniques. The performance portable code in OpenCL is also highly competitive with the best performance using the native parallel programming models on each platform.

AB - With the advent of many-core computer architectures such as GPGPUs from NVIDIA and AMD, and more recently Intel's Xeon Phi, ensuring performance portability of HPC codes is potentially becoming more complex. In this work we have focused on one important application area - structured grid codes - and investigated techniques for ensuring performance portability across a diverse range of different, high-end many-core architectures. We chose three codes to investigate: a 3D lattice Boltzmann code (D3Q19 BGK), the CloverLeaf hydrodynamics mini application from Sandia's Mantevo benchmark suite, and ROTORSIM, a production-quality structured grid, multiblock, compressible finite-volume CFD code. We have developed OpenCL versions of these codes in order to provide cross-platform functional portability, and compared the performance of the OpenCL versions of these structured grid codes to optimized versions on each platform, including hybrid OpenMP/MPI/AVX versions on CPUs and Xeon Phi, and CUDA versions on NVIDIA GPUs. Our results show that, contrary to conventional wisdom, using OpenCL it is possible to achieve a high degree of performance portability, at least for structured grid applications, using a set of straightforward techniques. The performance portable code in OpenCL is also highly competitive with the best performance using the native parallel programming models on each platform.

KW - GPU

KW - heterogeneous

KW - lattice Boltzmann

KW - Many-core

KW - multi-grid multi-block

KW - structured grid

KW - Xeon Phi

U2 - 10.1007/978-3-319-07518-1_4

DO - 10.1007/978-3-319-07518-1_4

M3 - Conference Contribution (Conference Proceeding)

AN - SCOPUS:84903700700

SN - 9783319075174

VL - 8488 LNCS

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

SP - 53

EP - 75

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

PB - Springer Verlag

T2 - 29th International Supercomputing Conference, ISC 2014

Y2 - 22 June 2014 through 26 June 2014

ER -

McIntosh-Smith S, Boulton M, Curran D, Price J. On the performance portability of structured grid codes on many-core computer architectures. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Vol. 8488 LNCS. Springer Verlag. 2014. p. 53-75. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-07518-1_4

On the performance portability of structured grid codes on many-core computer architectures

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Handle.net

Fingerprint

Projects

Future-proof massively-parallel execution of multi-block applications

Equipment

HPC (High Performance Computing) and HTC (High Throughput Computing) Facilities

Profiles

Professor Simon N McIntosh-Smith

Cite this