Multiregion tests take a lot of time

[simone-silvestri]

We can probably reduce the multi-region tests. In particular, a lot of time is taken by

Oceananigans.jl/test/test_multi_region_cubed_sphere.jl

Line 754 in dbc13ce

@testset "Testing simulation on conformal and immersed conformal cubed sphere grids" begin

These test loops over float types; we probably do not need to do that since the simulation should not necessarily change depending on the float type. We could test only on Float64, and then introduce more unit tests on the grid size for Float32. This would slash the time required for these tests by 50%.

What do you think @siddharthabishnu?

[glwagner]

Be aware that there can be errors based on float type. This can happen when Float64 is hardcoded. We've recently found one in TKEDissipationVerticalDiffusivity.

[ali-ramadhan]

I also noticed this. We should probably move it up to the top of the list of test groups to run in Buildkite so it doesn't bottleneck test build times. Although looking at https://buildkite.com/clima/oceananigans/builds/27214/waterfall it might only shave a minute or two right now.

Are the tests taking longer because the scope of the test group has grown, or does it just take longer to compile the same old tests?

[siddharthabishnu]

@simone-silvestri Yes, I agree. Testing on Float64 should be sufficient here. When I originally wrote these tests, I was just following the general pattern of testing across all float types and architectures.

@ali-ramadhan The main bottleneck is the final test, which involves generating the immersed cubed sphere grid and running a basic simulation on it.