Split implicit and explicit precomputed quantities

[dennisYatunin]

Purpose

This PR splits the precomputed quantities that need to be handled implicitly (i.e., updated on each Newton iteration) from those that can be handled explicitly (i.e., held constant throughout each Newton solve).

This change should improve performance for all simulations with more than one Newton iteration, since we are currently spending a significant chunk of time updating precomputed quantities that can be handled explicitly. Specifically, this should reduce the performance penalty that we have observed in #3662 from increasing the number of Newton iterations for configurations with implicit diffusion.

---

• I have read and checked the items on the review checklist.

[szy21]

tagging @costachris because there are prognostic edmf changes

[szy21]

Is it expected to change MSE?

[charleskawczynski]

Overall the changes look fine, but I'm also concerned that there are mse changes. Any idea what is wrong there?

[dennisYatunin]

Yes, we do expect MSE changes for simulations with implicit diffusion. We were previously updating the entire cache before computing the implicit tendency, but now we are only updating quantities related to velocity and thermodynamics. In particular, we are no longer updating the diffusivity, so its value in the implicit tendency is now reused from the previous explicit tendency. Since we are computing the implicit tendency with a different diffusivity, we expect the results to change. However, the diffusivity is mostly constant with respect to the state, so we only have small MSE changes and no meaningful differences in the summary plots.

I am going to add a simple heat equation test with fixed diffusivity in a separate PR, and then I will be able to confirm that these changes have no effect on implicit diffusion when diffusivity is constant.

[szy21]

@costachris Could you take a look at the prognostic edmf relevant changes in this PR and see if it makes sense to you and/or make sure you understand the change?