Tiling/halo bug in AG

[AndyHoggANU]

I'm now running some new cases with the revised AG scheme. There are a few issues to deal with here -- and they may overlap. But I'll list them all as separate issues for now. The first one is shown in the picture below. This is cell thickness at some depth:

You can see six clear meridional lines, about 4 grid points wide ... i.e., exactly where the halos are! I think this indicates a meridional halo bug, right @angus-g??

[angus-g]

Yeah, I'd agree with that. Possibly a result of the latest smoothing addition!

[angus-g]

I've just pushed https://github.com/angus-g/MOM6/commit/754496e219c6ec7e77c7d1a03504a0d97c6753e8 which should hopefully deal with this and #13 (maybe also #12 if it's somehow gaining mass through bogus interface movement...). It's available at /g/data/x77/ahg157/exes/MOM6/AG-774f4402.

I don't think this is the "best" way to fix the issue: I've essentially cut off the filtering stencil at the halo, so we don't have to push them out even further for the calculation. I suppose it would be possible to do a halo exchange of the local dz_s fields before the stencil, so that we have the full halo available, but that would require some restructuring of the loops.

Because the smoothing doesn't see across the halo, I suppose this won't reproduce properly across layouts, but I wonder if we'll still see a discontinuity in this case?

[AndyHoggANU]

OK, will test it out.

[AndyHoggANU]

Think this is fixed with the above update.