Minisymposium Presentation

Gyselalib++ is a portable, GPU-accelerated C++ library designed for high-performance gyrokinetic semi-Lagrangian simulations. It uses Kokkos to ensure performance portability across diverse hardware architectures, including modern multi-core CPUs and GPUs, making it well-suited for exascale computing. Additionally, Gyselalib++ makes use of DDC (Discrete Domain Computation library), a library that provides a framework for strongly typing mathematical concepts. Gyselalib++ is the result of a rewriting of GYSELA, a Gyrokinetic Semi-Lagrangian code written in Fortran. While the original code was highly optimised to run petascale simulations, the lack of modularity makes it difficult to add non-trivial extensions, such as X-point geometries, to the code. It was also difficult to optimise for new GPU architectures. This talk will introduce the design and capabilities of Gyselalib++, including its approach to parallelism, memory management, and performance optimisation for large-scale gyrokinetic modelling. We will present benchmarking results on a 4D simulation and discuss ongoing work to extend the capabilities of the library. In particular, we will highlight early developments toward a patch-based approach for handling complex magnetic field geometries, such as those found near the X-point in fusion devices.