You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Routines offloaded to the GPU now take the 1d base state mapped to a 3d MultiFab. This mapping is often done within the C++ wrapper function of these routines, which is likely to lead to lots of duplication, the same quantities being mapped repeatedly by different functions. It might be more efficient to do the mapping once (e.g. just after the base state is changed) and carrying this around rather than computing it on the fly every time it's needed.
The text was updated successfully, but these errors were encountered:
One tricky part is that some of the X_cart temporary multifabs are filled with things like "rho0", where rho0 is passed in to a function and could be rho0_old, rho0_new, or some temporarily created rho_nph.
That's true. I wasn't thinking that this should be done for all of the X_carts, but I certainly noticed that there are an awful lot of w0_carts created, which seems rather inefficient
I finished the first part of cleaning up w0_cart. w0_cart is now part of the MAESTRO class. the next step is to eliminate the calls that fill in w0_cart from w0. I think the idea is to fill in w0_cart from w0 immediately after make_w0 is called, and only then.
Routines offloaded to the GPU now take the 1d base state mapped to a 3d MultiFab. This mapping is often done within the C++ wrapper function of these routines, which is likely to lead to lots of duplication, the same quantities being mapped repeatedly by different functions. It might be more efficient to do the mapping once (e.g. just after the base state is changed) and carrying this around rather than computing it on the fly every time it's needed.
The text was updated successfully, but these errors were encountered: