Migrate CA column ops to ClimaCore

[charleskawczynski]

This PR moves the column indefinite integrals and column mapreduce from ClimaAtmos to ClimaCore. These are needed for a few different parameterizations (e.g., radiation, gravity wave, turbconv etc.).

@dennisYatunin, I didn't see tests for this in ClimaAtmos, can you please take this PR over and add some tests?

[dennisYatunin]

I just made the following updates to this PR:

• Add unit tests for column_integral_definite!, column_indefinite_integral!, and column_mapreduce! that check for correctness, type stability, and allocations. Run the tests on both CPUs and GPUs.
• Use RecursiveApply in column_integral_definite! and column_indefinite_integral! to make them compatible with Tuples and NamedTuples. Check that RecursiveApply is used correctly in the unit tests.
• Restrict column_integral_definite! to only accept cell-center fields as inputs because it gives incorrect results for cell-face fields. This is due to the fact that Δz_field(face_field) does not contain information about the half-widths of the topmost and bottommost cells. Only Fields.local_geometry_field(face_field).WJ contains this information, but we cannot use WJ to compute vertical integrals in 2D or 3D spaces because it is proportional to the cell volume, rather than the cell height. ClimaAtmos is currently only using column_integral_definite! on cell-center fields, so we can safely make this restriction.
• Rewrite column_mapreduce! to reduce its allocations. Unfortunately, I was not able to completely get rid of the allocations, but I was able to bring them down. I think the remaining allocations have something to do with mapping a function over a tuple of fields (or, more precisely, a Vararg of fields). The function contains a captured variable (in this case, the level at which the fields are being accessed), so maybe the allocations are related to that.
• Reformat column_integral_definite!, column_indefinite_integral!, and column_mapreduce! so that they have similar variable names, docstrings, etc.
• Add the new docstrings to the Operators documentation page.

@charleskawczynski @simonbyrne Could you take a look at the tests and other changes?

[dennisYatunin]

bors try

[bors]

try

Build failed:

• buildkite/climacore-ci

[dennisYatunin]

bors try

[bors]

try

Build failed:

• buildkite/climacore-ci

[charleskawczynski]

bors try

[bors]

try

Build failed:

• buildkite/climacore-ci

[charleskawczynski]

I've updated the names to match the existing ones. The tests look good to me, I think this is ready to merge.

[charleskawczynski]

bors r+

[bors]

Build failed:

• buildkite/climacore-ci

[bors]

try

Build failed:

• buildkite/climacore-ci

[charleskawczynski]

It turned out that the gpu tests were not exercised in the original commit, and some changes were needed. I got the column_mapreduce! working in the column case, but it's failing compilation on the gpu for the extruded space case.

That function is only currently needed / used in the old EDMF dycoms case, however, it's a very useful function and we should get it working at some point (e.g., it will come in handy in the gravity wave parameterization). So, I'm included to leave it as a broken test for now. I think this PR is ready to go, I'm happy to squash if that's preferred.

[dennisYatunin]

New changes look good to me!

[charleskawczynski]

bors r+

[bors]

Build failed:

• buildkite/climacore-ci

[charleskawczynski]

bors try

[bors]

try

Build failed:

• buildkite/climacore-ci

[charleskawczynski]

bors r+

[bors]

Build succeeded!

The publicly hosted instance of bors-ng is deprecated and will go away soon.

If you want to self-host your own instance, instructions are here.
For more help, visit the forum.

If you want to switch to GitHub's built-in merge queue, visit their help page.

• buildkite/climacore-ci
• docbuild
• format
• lib-climacore-makie
• lib-climacore-plots
• lib-climacore-tempestremap
• lib-climacore-vtk
• test-os (ubuntu-20.04)
• test-os (windows-latest)