Single precision suite-4 from NEPTUNE -- branch jm-nrl-32bitfp-24cc09e -> main

[michalakes]

Changes that allows CCPP physics to be compiled and run in the
NRL NEPTUNE model. These changes are relative to CCPP repository
hash 24cc09e (Thu May 27 13:02:47 2021 -0600).

Note that some of the changes here are NRL-specific and may not
relate to the changes for 32-bit floating point. Rather than
take these out, I erred on the side of caution and left them in
to avoid introducing errors.

A few highlights:

  - Autopromotion using -r8 is turned off except for
        mersenne_twister when SINGLE_PREC is defined

      - A number of source files that contained bare subroutines
        and functions are turned to modules and use-associated where
        needed. This is valuable for checking dummy/actual argument
        lists for precision type-agreement

      - Conditional guard-rails added where needed to prevent div-zero
        and neg args to log, sqrt, etc.

Status is that this compiles and runs in NEPTUNE for 150 steps
(75 minutes) without floating point exceptions or NaNs using CCPP
physics suite four.  More careful meteorological V&V is a TODO.
On Narwhal, an AMD Epyc Rome system at NAVO, these changes halved
the time spent in physics.

    J. Michalakes, 20211107

Changes to be committed:
modified: CMakeLists.txt
modified: physics/GFS_MP_generic.F90
modified: physics/GFS_rad_time_vary.fv3.F90
modified: physics/GFS_rad_time_vary.fv3.meta
modified: physics/GFS_rrtmg_post.meta
modified: physics/GFS_rrtmg_pre.F90
modified: physics/GFS_rrtmg_pre.meta
modified: physics/GFS_rrtmg_setup.F90
modified: physics/GFS_rrtmg_setup.meta
modified: physics/calpreciptype.f90
modified: physics/cires_orowam2017.f
modified: physics/cires_ugwp.F90
modified: physics/cires_ugwp_triggers.F90
modified: physics/cires_ugwpv1_sporo.F90
modified: physics/funcphys.f90
modified: physics/gcycle.F90
modified: physics/h2ophys.f
modified: physics/iounitdef.f
modified: physics/machine.F
modified: physics/mfpbl.f
modified: physics/mfpblt.f
modified: physics/mfpbltq.f
modified: physics/mfscu.f
modified: physics/mfscuq.f
modified: physics/module_SGSCloud_RadPre.F90
modified: physics/module_bl_mynn.F90
modified: physics/module_mp_thompson.F90
modified: physics/moninedmf.f
modified: physics/physcons.F90
modified: physics/radiation_astronomy.f
modified: physics/radlw_main.F90
modified: physics/radsw_main.F90
modified: physics/samfdeepcnv.f
modified: physics/sascnvn.F
modified: physics/sfc_diag_post.F90
modified: physics/sfc_drv.f
modified: physics/sfc_drv_ruc.meta
modified: physics/sflx.f
modified: physics/surface_perturbation.F90
modified: physics/tridi.f
modified: physics/ugwp_driver_v0.F

[michalakes]

I have been porting this version to the Cray compiler, which has found a number if instances where variables or arrays are being used before being defined. I have flagged these with TODO comments. There is also one reported for cu_gf_deep.F90 that I haven't added to the PR yet because I'm checking with Georg Grell.

[grantfirl]

#797 and #798 contain changes from this PR on top of the latest main. This one can likely be closed.

[areinecke]

Are there any plans to proceed with this PR?

[grantfirl]

Are there any plans to proceed with this PR?

@areinecke I'm going to close this particular PR. The work contained within was split into 2 different parts: one that contained the 32-bit changes (which were merged in #918) and one for everything else (#798 ). Some of the changes in #798 have also already been merged, so I'm just about to close that one and submit a "fresh" one with the remaining un-merged changes.

[grantfirl]

The changes in this PR were separated into separate PRs for purposes of merging (#918 and #952)