Skip to main content

SIAM CSE 23 SYCL

SYCL’s impact on algorithms, data structures and implementations

Tom Deakin and Tobias Weinzierl

Computational Science and Engineering software will need to embrace GPU accelerated systems as they prepare for Exascale. GPU accelerated systems dominate the top tier supercomputers. However, with multiple vendors offering competitive solutions, it is not yet clear which programming interface the applications should use as they update their codes for GPUs. With scientific software being used for many years, far beyond the lifetime of any one supercomputer, this investment in software needs to continue to flourish beyond just the next system. SYCL is an open standard that promises portable and performant heterogeneous parallel programming using modern C++. SYCL portability allows programs to run on GPU accelerated systems from all vendors. In this minisymposium, we bring together diverse groups that have recently ported part of their simulation software to SYCL. We ask them to share how the transition to SYCL and GPUs has motivated them to redesign their algorithms and numerical methods, and the implementations of those methods in software. Speakers will share whether the choice of programming model affected the algorithmic and numerical design for their scientific and engineering domains. We will also discuss how SYCL codes characteristically differ from comparable codes on the CPU, directive-based implementations such as OpenMP, or those written for the GPU using CUDA.

Programme

  • Tom Deakin: Vision and Scope of the SYCL Minisymposium
  • Igor Baratta: Performance-portable matrix-free finite element solvers with SYCL
  • Hatem Ltaief: Making HiCMA Hardware-Agnostic with SYCL
  • Ravil Dorozhinskii: Performance-portable earthquake simulation with SeisSol and SYCL
  • Daniel Arndt: Implementing a SYCL Backend for Kokkos
  • Amanda Dufek: A summary of the parallel implementation processes of Milc-Dslash benchmark using SYCL
  • Will R. Saunders: Exploration of Performance-Portability in the ExCALIBUR Fusion Use Case
  • Nisha Patel: Intel Developer Tools for Serious Sycl