hPIC2: A Parallelized, Unstructured Particle-in-Cell Code Dynamically Coupled to Plasma-Material Interaction Models
Logan Meredith, University of Illinois at Urbana-Champaign
The emergence of exascale systems in high-performance computing is expanding the capabilities of plasma physics codes to include ever larger and more complex multiphysics simulations. However, the majority of projected exascale supercomputers use heterogeneous computing architectures that rely on device acceleration, especially through the use of general-purpose graphics processing units. Such architectures require careful algorithmic optimization for proper utilization. To satisfy the need for a general-purpose plasma modeling code that runs on any system from laptops to supercomputers, we have developed hPIC2, a hybrid, electrostatic particle-in-cell code. hPIC2 is built atop the Kokkos framework, which allows a single source code to be deployable to a range of conventional and heterogeneous computing architectures; MFEM, a finite element method library, to enable simulation of plasmas in almost arbitrarily-shaped domains; and RustBCA, which models the interactions between plasmas and materials. Furthermore, hPIC2 explores the use of multiple plasma models simultaneously. With these capabilities, hPIC2 exhibits good performance while simulating the effects of plasma-material interactions on large plasmas in nontrivial domains.
Abstract Author(s): L.T. Meredith, M. Rezazadeh, M.F. Huq, J. Drobny, D. Curreli