Application of performance portability solutions for GPUs and many-core CPUs to track reconstruction kernels

• DOI: 10.48550/arxiv.2401.14221
• Publication Date: 2024-01-01
• Authors: Ka Hei Martin Kwok; M. J. Kortelainen; G. B. Cerati; Alexei Strelchenko; O. Gutsche; Allison Reinsvold...; S. Lantz; M. Reid; Daniel Riley; S. Berkman; Seyong Lee; Hammad Ather; Boyana Norris; Cong Wang
• Keywords: Software portability; Benchmark (surveying); Code (set theory); Multi-core processor

Next generation High-Energy Physics (HEP) experiments are presented with significant computational challenges, both in terms of data volume and processing power. Using compute accelerators, such as GPUs, is one the promising ways to provide necessary power meet challenge. The current programming models for accelerators often involve using architecture-specific languages promoted by hardware vendors hence limit set platforms that code can run on. Developing software platform restrictions especially unfeasible HEP communities it takes effort convert typical algorithms into ones efficient accelerators. Multiple performance portability solutions have recently emerged an alternative path which allow be executed on from different vendors. We apply several solutions, Kokkos, SYCL, C++17 std::execution::par Alpaka, two mini-apps extracted mkFit project: p2z p2r. These apps include basic kernels a Kalman filter track fit, propagation update parameters, detectors at fixed z or r position, respectively. explore memory layout formats. report development experience well their GPUs many-core CPUs, measured throughput GPU CPU NVIDIA, AMD Intel.