Evaluating Performance Portability with the CMS Heterogeneous Pixel Reconstruction code

EPJ Web of Conferences, vol. 295, pp. 11008

Abstract

In the past years the landscape of tools for expressing parallel algorithms in a portable way across various compute accelerators has continued to evolve significantly. There are many technologies on the market that provide portability between CPU, GPUs from several vendors, and in some cases even FPGAs. These technologies include C++ libraries such as Alpaka and Kokkos, compiler directives such as OpenMP, the SYCL open specification that can be implemented as a library or in a compiler, and standard C++ where the compiler is solely responsible for the offloading. Given this developing landscape, users have to choose the technology that best fits their applications and constraints. For example, in the CMS experiment the experience so far in heterogeneous reconstruction algorithms suggests that the full application contains a large number of relatively short computational kernels and memory transfer operations. In this work we use a stand-alone version of the CMS heterogeneous pixel reconstruction code as a realistic use case of HEP reconstruction software that is capable of leveraging GPUs effectively. We summarize the experience of porting this code base from CUDA to Alpaka, Kokkos, SYCL, std::par, and OpenMP offloading. We compare the event processing throughput achieved by each version on NVIDIA and AMD GPUs as well as on a CPU, and compare those to what a native version of the code achieves on each platform.

Authors 15

1. Nikolaos Andriotis

   Barcelona Supercomputing Center

   Affiliation as printed

   Barcelona Supercomputing Center, Spain

2. Andrea Bocci

   European Organization for Nuclear Research

   Affiliation as printed

   CERN, Geneva, Switzerland

3. Eric Cano

   European Organization for Nuclear Research

   Affiliation as printed

   CERN, Geneva, Switzerland

4. Laura Cappelli

   Istituto Nazionale di Fisica Nucleare, Sezione di Bologna

   Affiliation as printed

   INFN Bologna, Italy

5. Tony Di Pilato

   University of Geneva · Center for Advanced Systems Understanding

   Affiliation as printed

   Center for Advanced Systems Understanding (CASUS), Görlitz, Germany

   University of Geneva, Switzerland

6. Luca Ferragina

   University of Bologna

   Affiliation as printed

   University of Bologna, Italy

7. Gabrielle Hugo

   European Organization for Nuclear Research

   Affiliation as printed

   CERN, Geneva, Switzerland

8. M. J. Kortelainen

   Fermi National Accelerator Laboratory

   Affiliation as printed

   Fermi National Accelerator Laboratory, Batavia, IL, USA

9. Martin Kwok

   Fermi National Accelerator Laboratory

   Affiliation as printed

   Fermi National Accelerator Laboratory, Batavia, IL, USA

10. Juan Jose Olivera Loyola

    Tecnológico de Monterrey

    Affiliation as printed

    Institute of Technology and Higher Studies of Monterrey, Mexico

11. Felice Pantaleo

    European Organization for Nuclear Research

    Affiliation as printed

    CERN, Geneva, Switzerland

12. Aurora Perego

    University of Milano-Bicocca

    Affiliation as printed

    University of Milano Bicocca, Italy

13. W. Redjeb Aachen

    European Organization for Nuclear Research · RWTH Aachen University

    Affiliation as printed

    CERN, Geneva, Switzerland

    RWTH Aachen University, Germany

14. Mark Dewing

    Argonne National Laboratory

    Affiliation as printed

    Argonne National Laboratory, Lemont, IL, USA

15. Julien Esseiva

    Lawrence Berkeley National Laboratory

    Affiliation as printed

    Lawrence Berkeley National Laboratory, Berkeley, CA, USA