For MS2.8, most of the linear-algebra intensive work was handled by a hand-coded abstraction layer, with specific calls to vendor specific libraries at the lowest level. Need to explore how far Kokkos::Kernels can eliminate a lot of this, and compare performance with the current implementation.
For MS2.8, most of the linear-algebra intensive work was handled by a hand-coded abstraction layer, with specific calls to vendor specific libraries at the lowest level. Need to explore how far Kokkos::Kernels can eliminate a lot of this, and compare performance with the current implementation.