Section: New Results
Participants : Philippe Helluy, Nhung Pham, Michel Massaro, Pierre Gerhard, David Coulette, Laura Mendoza, Conrad Hillairet.
In order to harness hybrid computers architecture, we have developed software and algorithms that are well adapted to CPU/GPU computing. For instance we have applied a task-graph approach for computing electromagnetic waves (https://hal.archives-ouvertes.fr/hal-01134222). We have also used an OpenCL-based GPU version of schnaps for computing a drift-kinetic plasma model (Nhung Pham's PhD). Recently, we have also developed a new implementation of the Discontinuous Galerkin solver into schnaps. We now use the StarPU runtime (http://starpu.gforge.inria.fr) for addressing automatically CPUs or GPUs available on the computational node. This development has been applied to the MHD equations (thesis of Michel Massaro). The new development will now be applied to kinetic acoustic simulations (Pierre Gerhard's PhD), gyrokinetic plasma simulations (Laura Mendoza's Postdoc) and implicit MHD simulations (Conrad Hillairet's PhD).