At the moment, we employ the standard Trixi implementation that is likely too slow for atmospheric applications. We should consider having as default implementation the turbo version for the volume kernel (see #128) and the combined approach for handling conservative and nonconservative terms all together, especially when running simulations on a cubed-sphere.
The cons of doing that is of course losing readability and a bit of flexibility, since one should define all the combinations of conservative and nonconservative fluxes to retain the same flexibility as in the standard implementation, but different versions/combinations can be implemented as needed.
At the moment, we employ the standard Trixi implementation that is likely too slow for atmospheric applications. We should consider having as default implementation the turbo version for the volume kernel (see #128) and the combined approach for handling conservative and nonconservative terms all together, especially when running simulations on a cubed-sphere.
The cons of doing that is of course losing readability and a bit of flexibility, since one should define all the combinations of conservative and nonconservative fluxes to retain the same flexibility as in the standard implementation, but different versions/combinations can be implemented as needed.