hPIC2: a GPU-accelerated, hybrid particle-in-cell code for plasma-material interactions in complex geometries

Many of the primary exascale supercomputers under construction rely on graphics processing unit (GPU) acceleration in order to reach their performance goals. While the particle-in-cell (PIC) method is an "embarrassingly parallelizable" algorithm for plasma simulation and well-suited to GPU acceleration, the range of device architectures to be implemented in future supercomputers often demands the use of unique parallel programming paradigms, threatening code portability. hPIC2 is a hybrid, electrostatic particle-in-cell code designed ab initio with shared-memory parallelism using the Kokkos performance portability framework, which allows for a single source code to be deployable to a diverse set of massively parallel architectures. hPIC2 is therefore scalable and performant on computing architectures ranging from a single core to many-core systems with GPU accelerators from all three major manufacturers. In addition, we have begun a number of preliminary explorations: parallelizing hybrid plasma simulation algorithms, which combine fluid and kinetic methods for a single species; modeling complex geometries, such as tokamak tile gaps, using implicit, non-uniform meshes; and coupling to RustBCA, a binary collision approximation (BCA) code for arbitrary ion-material interactions.