Refactor Fluxes, change data layout.

[sriharshakandala]

Purpose

Refactor Fluxes, change data layout.
Use (ncol, nlev) ordering instead of (nlev, ncol).

---

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

[sriharshakandala]

Prior to this PR: https://buildkite.com/clima/rrtmgp-clima-a100-pipeline/builds/46#019581be-ba33-4bf2-a011

-a6771bcbf5d2

[sriharshakandala]

With this PR: https://buildkite.com/clima/rrtmgp-clima-a100-pipeline/builds/47#01958270-fefc-435a-bcbc-9e1240db87b3

[szy21]

Please make necessary changes in ClimaAtmos and ClimaCoupler to make sure this change works correctly in atmos runs and AMIP, thanks!

[sriharshakandala]

Please make necessary changes in ClimaAtmos and ClimaCoupler to make sure this change works correctly in atmos runs and AMIP, thanks!

@szy21 : Would you prefer making changes when we make the next breaking release or would you like them changed after each PR? I believe there is a set of about 5 reverted PRs we need to merge back!

[szy21]

Please make necessary changes in ClimaAtmos and ClimaCoupler to make sure this change works correctly in atmos runs and AMIP, thanks!

@szy21 : Would you prefer making changes when we make the next breaking release or would you like them changed after each PR? I believe there is a set of about 5 reverted PRs we need to merge back!

Will all the 5 PRs have breaking changes? In any case, I would prefer a change after this PR as the data layout change seems like a pretty important change. More practically, merging breaking PRs will break this pipeline: https://buildkite.com/clima/climacoupler-coarse-nightly-amip, which many people use for debugging other things now, so it would be good for other people to know what changes in atmos and coupler are needed at least.

[sriharshakandala]

Please make necessary changes in ClimaAtmos and ClimaCoupler to make sure this change works correctly in atmos runs and AMIP, thanks!

@szy21 : Would you prefer making changes when we make the next breaking release or would you like them changed after each PR? I believe there is a set of about 5 reverted PRs we need to merge back!

Will all the 5 PRs have breaking changes? In any case, I would prefer a change after this PR as the data layout change seems like a pretty important change. More practically, merging breaking PRs will break this pipeline: https://buildkite.com/clima/climacoupler-coarse-nightly-amip, which many people use for debugging other things now, so it would be good for other people to know what changes in atmos and coupler are needed at least.

I started a draft PR here: CliMA/ClimaAtmos.jl#3690

[charleskawczynski]

Can we not realize this performance gain by changing the threeading pattern?

[sriharshakandala]

Can we not realize this performance gain by changing the threeading pattern?

RRTMGP distributes work by assigning one thread per each column. Updating data layouts for some variables is the easiest way to realize performance gains.