add clubb time-step to the atm.log

[adamrher]

Clubb's time-step is not currently outputted anywhere. As clubb is probably the 2nd most expensive piece of the CAM code, it is important to keep track of its time-step as it relates to model performance.

At present, the clubb timestep, "dtime" is computed in clubb_intr.F90. It internally computes dtime from dt_macmic = dtphys/cld_macmic_num_steps. It solves for the smallest integer "nadv" one needs to divide dt_macmic by such that dtime <= clubb_timestep namelist var. So there is no guarantee that dtime = clubb_timestep namelist var.

[adamrher]

Working on a PR to clean-up some clubb issues and am requesting help from the SE's on how best to output the clubb time-step to atm.log. Looking through the atm.log, It looks like ancillary info is written out in the readnl or init subroutines. How can I pull the physics time-step into the init or readnl subroutines in clubb_intr?

[gold2718]

Isn't this a fun game? I think one confusion is a comment in physpkg:

    real(r8), intent(in) :: ztodt            ! physics time step unless nstep=0

I think this comment is incorrect. The physics timestep is set in stepon_run1 which means that it is not available at namelist read or init time. However, phys_run1 is always called after stepon_run1 so that is the place to add the report. Something like:

if ((is_first_step() .or. is_first_restart_step()) .and. masterproc) then
  write(iulog, *) 'Yeehaw, we know the time step size! ', cld_macmic_ztodt
end if

added to the top of clubb_tend_cam or maybe even in tphysbc after cld_macmic_ztodt is calculated.

Note: is_first_step and is_first_restart_step are from the time_manager module.

[adamrher]

purrfect. FYI I'm running some 5 year simulations to test mods in this issue and #336. If all goes well I plan on opening a PR to close these two issues this evening/tmrw. OK by you? [edit: these mods should not be climate changing]

[adamrher]

"physics time step unless nstep=0" is confusing so might as well just remove the "unless nstep=0" part?

[gold2718]

OK by you?

Opening an issue, easy, getting the SEs to do anything about it, priceless ;)

just remove the "unless nstep=0" part?

Yes please. As far as I can tell, it is not true. I think the history has to do with the fact that tphysbc is called at model initialization time. Perhaps this comment was true when that was first set up but I think it has been fixed long ago (if anyone invents a regression test for comments, they should get the Turing Award).

[adamrher]

I have put a branch together with an initial implementation: branch cam.tstepout at https://github.com/adamrher/CAM/branches

As we've discussed this implementation is fragile and needs to be improved. I put it at the top of phys_run1 for is_first_step():
https://github.com/adamrher/CAM/blob/5a892f6c102ee61860ff8eaab4624e509afaff32/src/physics/cam/physpkg.F90#L1134-L1193

One of the issues is the time-steps used by radiation.F90, micro_pumas_cam.F90 and clubb_intr.F90 are all computed during run time, rather than in the initialization routine. That should be changed with this work.