Some bathymetry grid sizes are not immediately friendly to XBGPU

[davidbenncsiro]

We have found it necessary to sometimes change the bathymetry grid in order for XGBPU to not yield unexpected results, e.g. by cutting the first row and column.

It would be good for XBGPU to be able to detect and correct for this automatically.

[hot007]

Hi @CyprienBosserelle , I don't want to pressure you on this issue at all, just wondering what your projected time to address this issue might be? It affects whether we assume it will be done in the next 6mo, say, or we add warnings into our conversion code when the input grids might not work with XB-GPU.

[CyprienBosserelle]

I definitely want to fix that issue and there are two ways of dealing with this.

• Enforce the factor of 16 at the model level. (relatively easy and requires only changes in the front part of the model) but will lead to output that are somewhat different to the input...
• Search and destroy the bugs that makes the issue... that is preferable in my view but will take some time.

XBGPU does need a bit of attention but I'm flat out for the next 4-6 month so I don't think I can do the second option any time soon. If extending grids as part of the conversion is not acceptable I can try to squeeze option 1.

[davidbenncsiro]

We may yet be able to provide an intermediate workaround by including the ability to modify the bathymetry file via the conversion script if necessary.

[davidbenncsiro]

A somewhat related issue is the apparent need to double up the last rows and columns in the bathy file, at least in some scenarios.