Speaker
Description
Simulation of fusion plasma often requires representations of non-Maxwellian distributions. A common method is to represent the distribution function as a collection of individual particles where the state of each particle evolves over time. In this talk we describe NESO-Particles (NP) which is a performance portable framework for describing particle based algorithms.
NP provides users with abstractions for describing particle data and looping operations for the fusion plasma use case. These abstractions form a separation of concerns between the description of particle based algorithms and the execution of these algorithms on modern HPC hardware. Our implementation is realised as a MPI+SYCL library that executes on both CPU and GPU hardware.