A Performance Portable, Fully Implicit Landau Collision Operator with Batched Linear Solvers

Modern accelerators use hierarchical parallel programming models that enable massive multithreading within a processing element (PE), with multiple PEs per device driven by traditional processes. Batching is a technique for exposing PE-level parallelism in algorithms that have traditionally run on MPI processes or multiple threads within a single process. Opportunities for batching arise in, for example, kinetic discretizations of magnetized plasmas where collisions are advanced in velocity space at each spatial point independently. This paper builds on previous work on a high-performance, fully nonlinear, Landau collision operator by batching the linear solver, as well as batching the spatial point problems and adding new support for multiple grids for multiscale, multispecies problems. An anisotropic relaxation verification test that agrees well with previously published results and analytical models is presented. The performance results from NVIDIA A100 and AMD MI250X nodes are presented with hardware utilization analysis for each architecture. Finally, the entire implicit Landau operator time advance is implemented in Kokkos for performance portability, running entirely on the device and is available in the PETSc numerical library.. Authors: Adams, Mark F. [Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)] (ORCID:0000000345042047); Wang, Peng [NVIDIA Corporation, Santa Clara, CA (United States)]; Merson, Jacob [Rensselaer Polytechnic Inst., Troy, NY (United States)]; Huck, Kevin [Univ. of Oregon, Eugene, OR (United States)]; Knepley, Matthew G. [Univ. of Buffalo, NY (United States)] (ORCID:0000000222920735). DOE Contract: AC02-05CH11231; SC0021299. Subjects: 97 MATHEMATICS AND COMPUTING; GPU finite elements; Landau collision operator; batch solvers; kinetic methods