Comment (by Gabriele Bozzola):

I implemented this and other improvements to Lean here (and here for Proca). These are the changes:

41be06c * origin/master LeanBSSNMoL:: split the three evolution methods
61698a8 * LeanBSSNMoL:: Optionally compute RHS after initial data
e9c355b * LeanBSSNMoL: Rename CCTK_BOOL to BOOLEAN
084825a * LeanBSSNMoL, NPScalars: Add ParamCheck
9713bb1 * LeanBSSNMoL: Calculate constraints every N iterations
bb3f2d1 * NPScalars: Compute NP scalars every N iterations
70f508b * LeanBSSNMoL: Compute matter terms only if needed
c59c4b7 * LeanBSSNMoL: Add OpenMP parallelization to adm_vs_bssn
be0f263 * NPScalars: Add OpenMP parallelism

It wasn’t wise to completely decouple the three evolutions, so the rhss are still computed in the same for loop. My expectation is that this leads to a performance penalty compared to the current version (due to the additional branching in hot loops), but I’d bet that the effect is overall negligible.

I’ve run some medium size simulations with this version, and everything seems fine. Of course, this doesn’t mean that I haven’t introduced new problems.

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2487/_evolution_method-leanbssnmol-in-lean-does