A Study on the Performance Portability of the Finite Element Assembly Process Within the Albany Land Ice Solver

This paper presents a performance analysis of the finite element assembly process of the Albany Land Ice solver. The analysis shows that a speedup over traditional MPI-only simulations is achieved on multiple architectures including Intel Haswell CPUs, Intel Xeon Phi Knights Landing and IBM POWER8/NVIDIA P100 platforms. A scalability study also shows that performance remains reasonably close among all architectures. These results are obtained on a single codebase without architecture-dependent code optimizations by utilizing abstractions in shared memory parallelism from the Kokkos library and is part of an ongoing process of achieving performance portability for the Albany Land Ice code.