Compute stat file diagnostics on device for fields currently on the device

[kynan]

See discussion in #32

[gmarkall]

(This is a slightly premature response but I understand it's going to be discussed at the meeting in about 15 minutes).

I suspect we need a new function, compute_stat_() in the generated code that generates the stats on the device for each field - I imagine the cuda kernels that it uses will not need to be generated, but will just need to be invoked with a pointer to the field data. I expect compute_stat_() can iterate over the fields map.

The fluidity call calculate_diagnostic_variables tries to compute them for all fields, so for the fields that are only up to date on the device, the values are going to be wrong - we need some way to either stop calculate_diagnostic_variables from computing these values, and instead use them from elsewhere, but I don't know how to do this - maybe we can look at its source to see if there is some easy way to do this?

[dham]

My plan on this is that fluidity fields are going to hold a pointer to their corresponding OP_DAT. Fluidity can also know if the fluidity version of the field is current. Using these it should be easy during calculate_diagnostic_variables to choose the current code or to launch an OP2 version.

I also think we don't need to fix this yet.