Keynote Speaker:  Virginie Grandgirard , Research Director at CEA.

Title: Gysela-X++ – A new C++ Kokkos-based code to meet the exascale challenge of gyrokinetic plasma turbulence simulations

Bio:

Dr. Virginie Grandgirard – (F.) Virginie Grandgirard is presently research director at CEA, France. She received the PhD degree in mathematics and applications from Besancon University, France, in 1999. She then obtained the Habilitation à Diriger des Recherches in 2016. For 20 years, she was one of the main developers of the 5D GYSELA non-linear gyrokinetic semi-Lagrangian code used for plasma turbulence simulations. https://gyselax.github.io/ . This Fortran code is highly parallelized up to hundreds of thousands cores. For the past two years, she has been in charge of the Gysela-X++ project, which aims to develop a new C++ Kokkos-based code for future exascale simulations of ITER plasmas. Her research interests focus on numerical methods for Vlasov equations, high performance computing and tokamak plasma turbulence and more recently on Physic-based machine learning.

She  is involved in both national (Moonshot CExA https://cexa-project.org/, PEPR NUMPEX https://numpex.org/) and european projects (EoCoE-III https://www.eocoe.eu/eocoe-iii-on-the-enea-news-website/). She actively participates in the joint lab NTU/CEA SAFE (Singapore Alliance with France for Fusion Energy) and regularly visits NTU university in Singapore. She has co-authored more than 100 publications in peer-reviewed journals.

Abstract:

Controlled fusion offers the promise of sustainable and safe energy production on Earth, free of greenhouse gas emissions. In magnetic fusion devices, the power gain increases nonlinearly with the energy confinement time. The quality of the plasma energy confinement thus largely determines the size and therefore the cost of a fusion reactor. Unfortunately, small-scale turbulence limits the quality of confinement in most fusion devices. Hence modelling of turbulent transport is mandatory to find routes towards improved confinement regimes. Numerical simulations are based on a kinetic description of the plasma that can only be performed on most powerful High Performance Computers(HPC). The gyrokinetic GYSELA code [1] has been developed for 20 years to this aim. The code runs efficiently on several hundred thousand cores on current standard architectures, and already makes use of petascale computing capacities. However full description of electron dynamics for ITER plasma requires exascale capabilities.
In this context, we present the new Gysela-X++ code [2], a modern C++ rewrite of the original Fortran code launched two years ago and currently under construction. Its ultimate goal is not only to extend the code’s physical modelling capabilities, but also to ensure better portability on exascale architectures. It is built on top of the DDC library [3, 4] which provides associative arrays with labeled dimensions, making the source code both safer and more intuitive.
Additionally, Gysela-X++ is built on the Kokkos performance portability framework, ensuring efficient execution both on CPU and GPU (AMD and NVIDIA) architectures. As part of this rewrite, we developed Gyselalib++ [5,6], an open-source library providing all essential simulating a distribution function discretized in phase space on a structured grid. computational kernels for constructing kinetic or gyrokinetic plasma simulation codes in C++, Gyselalib++ is not only useful for the fusion community but also addresses a broader audience, as it has been specifically designed to serve as a testbed for developing, benchmarking, and rapidly prototyping new numerical schemes.
We will share our initial experiences with this modular, C++ Kokkos-based approach, focusing on its impact on code readability, portability, and scalability across accelerated architectures, as well as its ability to address exascale challenges.

[1] V. Grandgirard et al., Computer Physics Communications 207 (2016) 35–68.
[2] https://gyselax.github.io/
[3] https://github.com/CExA-project/ddc ;
[4] T. Padioleau et al. J0SS 2025, https://joss.theoj.org/papers/10.21105/joss.09122
[5] https://github.com/gyselax/gyselalibxx
[6] E. Bourne et al., JOSS 2025, https://doi.org/10.21105/joss.08582