simplify imex ark #212
simplify imex ark #212
Conversation
210: Add post callbacks r=charleskawczynski a=charleskawczynski This PR adds `post_explicit!` and `post_implicit!` callback hooks to ClimaODEFunction, which do nothing by default, and calls to functions were added in the ARS343`step_u!`. Should we bother calling these functions in the generic callback? cc `@simonbyrne` I spoke with `@simonbyrne` about this, and he'll add these into #212 once we are happy with a version that is simpler to reason about. Co-authored-by: Charles Kawczynski <kawczynski.charles@gmail.com>
| T_lim = ntuple(i -> zero(u0), Nstage_exp) | ||
| T_exp = ntuple(i -> zero(u0), Nstage_exp) | ||
| T_imp = ntuple(i -> zero(u0), Nstage_imp) |
There was a problem hiding this comment.
Just a heads up: we previously had tuples and this resulted in massive compilation times (especially for the edmf cases). That's why we collect above.
There was a problem hiding this comment.
that's probably the same issue as CliMA/ClimaCore.jl#1467
There was a problem hiding this comment.
so you use vectors?
Yes
that's probably the same issue as CliMA/ClimaCore.jl#1467
That seems likely but considering how many fields we have in the solver cache, it's a bit strange that this ntuple, which only has a few, was so expensive. Maybe we were getting hit by more things that operate on the state?
|
Can/should we close this? ( it should probably at least be rebased ) |
|
I like the idea of simplifying along this direction, but I see a logical bug: I'm going to close, and open an issue instead. |
Purpose
To-do
Content