MAPPING DVMH-PROGRAMS WITH REGULAR DEPENDENCIES ONTO CLUSTERS WITH GPU

Authors

  • Vladimir A. Bakhtin Keldysh Institute of Applied Mathematics Russian Academy of Sciences
  • Aleksander S. Kolganov Keldysh Institute of Applied Mathematics Russian Academy of Sciences
  • Victor A. Krukov Keldysh Institute of Applied Mathematics Russian Academy of Sciences
  • Natalia V. Podderyugina Keldysh Institute of Applied Mathematics Russian Academy of Sciences
  • Mikhail N. Pritula Keldysh Institute of Applied Mathematics Russian Academy of Sciences

DOI:

https://doi.org/10.14529/cmse130404

Keywords:

DVM for Heterogeneous systems, Fortran DVMH, hybrid computational systems with accelerators, GPU, CUDA

Abstract

In the 2011 year DVMH programming model for new heterogeneous and hybrid supercomputer systems (or DVM for Heterogeneous systems) was introduced in the Keldysh Institute for Applied Mathematics of RAS. The developed high-level programming languages were based on standard Fortran and C programming languages, but extended with the directives for mapping the program onto a parallel computer. The directives are represented as special comments (or pragmas). The paper describes problems and methods for mapping loops, which have dependencies to the GPU. Efficiency of the developed Fortran DVMH parallel programs with regular dependencies is demonstrated.

Author Biographies

Vladimir A. Bakhtin, Keldysh Institute of Applied Mathematics Russian Academy of Sciences

кандидат  физико-математических  наук

Victor A. Krukov, Keldysh Institute of Applied Mathematics Russian Academy of Sciences

доктор  физико-математических  наук,  профессор

Natalia V. Podderyugina, Keldysh Institute of Applied Mathematics Russian Academy of Sciences

кандидат  физико-математических  наук

References

Top500 List – November 2012 TOP500 Supercomputer Sites. URL: http://top500.org/list/2012/11/ (дата обращения 01.12.2012).

High Performance Fortran. URL: http://hpff.rice.edu/ (дата обращения 01.12.2012).

Параллельное программирование в системе DVM. Языки Fortran-DVM и C-DVM / Н.А. Коновалов, В.А. Крюков, А.А. Погребцов и др. // Труды Международной конференции «Параллельные вычисления и задачи управления» (PACO’2001). — Москва, 2001. — С. 140–154.

Fortran DVM – язык разработки мобильных параллельных программ / Н.А. Коновалов, В.А. Крюков, С.Н. Михайлов, А.А. Погребцов // Программирование. — 1995. — № 1. — С. 49–54.

Коновалов, Н.А. C-DVM – язык разработки мобильных параллельных программ / Н.А. Коновалов, В.А. Крюков, Ю.Л. Сазанов // Программирование. — 1999. — № 1. — С. 54–65.

Dolbeau, R. HMPP™: A Hybrid Multi-core Parallel Programming Environment / R. Dolbeau, S. Bihan, F. Bodin. URL: http://www.caps-entreprise.com/wp-

content/uploads/2012/08/caps-hmpp-gpgpu-Boston-Workshop-Oct-2007.pdf (дата обращения 02.12.2012).

The Portland Group. PGI Accelerator Programming Model for Fortran & C. URL: http://www.pgroup.com/lit/whitepapers/pgi_accel_prog_model_1.3.pdf (дата обращения 02.12.2012).

OpenACC. URL: http://www.openacc-standard.org/ (дата обращения 01.12.2012).

Han, T.D. hiCUDA: High-Level GPGPU Programming / T.D. Han, T.S.Abdelrahman.// IEEE Transactions on Parallel and Distributed Systems. — 2011. — Vol.22, No. 3 — P. 78–90.

Расширение DVM-модели параллельного программирования для кластеров с гетерогенными узлами / В.А. Бахтин, М.С. Клинов, В.А. Крюков и др. // Вестник Южно-Уральского государственного университета, серия «Математическое моделирование

и программирование». — Челябинск: Издательский центр ЮУрГУ, 2012. — Вып. 12 — № 18 (277). — С. 82–92.

Pennycook, S.J. Performance Analysis of a Hybrid MPI/CUDA Implementation of the NAS-LU Benchmark / S.J. Pennycook, S.D. Hammond, S.A. Jarvis, G.R. Mudalige // ACM SIGMETRICS Performance Evaluation Review – Special issue on the 1st international workshop on performance modeling, benchmarking and simulation of high performance computing systems (PMBS 10). — 2011. — Vol. 38, Issue 4. — P. 23–29.

Seo, S. Performance Characterization of the NAS Parallel Benchmarks in OpenCL / S. Seo, G. Jo, J. Lee // 2011 IEEE International Symposium on. Workload Characterization (IISWC). — 2011. — P. 137–148.

Published

2014-04-23

Issue

Section

Informatics, Computers and Control