search

NOAA-GFDL / icebergs

GFDL Climate Model Icebergs
Other
5 stars 19 forks source link

303 bergs cgrid memory errors #27

Closed nichannah closed 8 years ago

nichannah commented 8 years ago

There is a discussion of the issue here: https://github.com/NOAA-GFDL/MOM6/issues/303

Valgrind was producing a kind of false-positive: unitialised halo data was being used but later masked.

adcroft commented 8 years ago

Thanks @nicjhan . I'm pulling in your code changes but using a zero fill value to get "sensible" stresses in the globally southernmost cells. @wfcooke encountered the same code lines in a related way where staggered taux/y on the southern edge of the global domain are not being provided to the icebergs. Most often the ocean/sea-ice/icebergs are shielded from the southern edge by masked land points and so bad values are not felt in the physical domain but when a southernmost cell is an ocean point we can pick up these undefined stresses. The real fix would be to modify the coupler to provide stresses on a domain wider than the ocean. Alternatively, we could never allow ocean points in the southern most row but we already have configurations like this so the zero work around is the least onerous for now.