Update GFS_rrtmg_setup.F90

[michalakes]

Move statement testing the is_initialized variable until after control variables in the argument list have been set. This is needed to make sure that the control variables are set for multiple instances of physics. This is related to the previously addressed issue #999 "Allow multiple instances of CCPP physics in single executable for ensemble DA".

The GFS_rrtmg_setup had not been a problem when #999 was being addressed, even though it has a heap-stored latch variable, because iaerflg andother control flags were also stored as module data in the physparam module and use-associated through the GFS_rrtmg_setup module. Since that time, GFS_rrtmg_setup was changed and these control variables are now passed in through the argument list of GFS_rrtmg_setup_init, where the are set only after the test to see if is_initialized is .true.

The error showed up in testing as NRL updated to the newer version of ccpp-physics.

[dustinswales]

@michalakes This looks fine to me.
Would you mind adding the same change to GFS_rrtmgp_setup.F90?

[michalakes]

@michalakes This looks fine to me. Would you mind adding the same change to GFS_rrtmgp_setup.F90?

I made the same change to GFS_rrtmgp_setup.F90

[michalakes]

One additional comment. This fix will work as long as each instance of the physics wants the same radiation flags set. That's probably okay for us at NRL. If we want to allow for different instances to have different radiation flags set, it'll be necessary to do what we've done with other physics that has an initialization latch variable: make is_initialized an array and index by ccpp_instance.

[dustinswales]

@michalakes Thanks for making that change.
(Sometime in the not so distant future I plan on cleaning up and unifying the G/GP radiation GFS interstitials. As part of a larger cleanup effort to make the schemes "stateless" and host-agnostic)

[grantfirl]

@michalakes I'm pulling these changes into #1037 to save testing time on our part.