Sriharsha Kandala

The code seems to be working fine, irrespective of planet radius, for an aspect ration

A local (instead of global), cell dependent, tolerance needs to be specified if we need to be more thorough. However, in our case, the GCM simulation domain is very thin...

I believe tuning single GPU performance is still work in progress! cc: @charleskawczynski

We can try ``` cartidx = CartesianIndices(dest)[gidx] ```

Can we apply operators by `element` or even a group of elements, instead of `by slab`. Using one block per each slab would result in about 16 to 25 threads...

> @sriharshakandala would you be able to help me understand/debug the errors now in `dss_2d!`? DSS buffer sizing can be different from the number of fields. The sizing is conservative...

@charleskawczynski , @szy21 : Is this issue still active?

bors try

bors try

bors try