Accelerating extended-MHD simulation for the CETOP SciDAC project

Architectural changes in HPC systems, specifically GPU accelerators, necessitate algorithmic changes to extended-MHD codes. In addition, algorithms are being extended to accommodate multispecies MHD simulation. This extension is critical to fusion-energy applications such as core-edge integration with a radiating detached divertor; disruption mitigation through impurity injection; and burning-plasma physics with multiple fuel species and fusion by-products. Progress to accelerate the NIMROD [Sovinec, JCP, 195, 355-386 (2004)] extended-MHD code through the Center for Edge of Tokamak OPtimization SciDAC are presented. Broadly, this work covers three categories: (1) algorithms to support preconditioning operations that exploit massive concurrency such as parallel LU decomposition with threshold-based drop tolerances as implemented in the Ginkgo code, (2) using the capability to perform batched ODE integration of local atomic rate effects with the SUNDIALS code and the Kokkos backend for performance portability, and (3) improvements to the extended-MHD codes themselves. Progress towards these efforts in open-source, open-access repositories [https://gitlab.com/NIMRODteam/open] will be discussed. Finally, we discuss the status of the accelerated version of the full NIMROD MHD code.