[Enhancement] Possible performance improvement by using Kokkos Teams rather than `parallel::for()`

[PhilipOesterlePekrun]

Description

There may be some places where the unbalanced nature of the parallel kernels perform better with Kokkos Teams, e.g.:

// TODO: For potentially better performance, try using teams instead of MDRangePolicy
      Kokkos::parallel_for(
          Kokkos::MDRangePolicy<Kokkos::Rank<2>>({0, 0}, {systemsize, systemsize}),
          KOKKOS_LAMBDA(const int i, const int j) {
            if (j >= i) return;
            const double tmp1 = xv0(j) - xv0(i);
            const double tmp2 = yv0(j) - yv0(i);
            const double r = sqrt(tmp1 * tmp1 + tmp2 * tmp2);
            const double tmp3 = C * asin(frac_GridSize_2 / r);
            H(i, j) = tmp3;
            H(j, i) = tmp3;
          });

See https://kokkos.org/kokkos-core-wiki/ProgrammingGuide/HierarchicalParallelism.html#thread-teams for info.

Related Issues and Merge Requests

• Related to Kokkos integration and performance improvements #117