NRL Neptune model 32-bit physics support

[SamuelTrahanNOAA]

This adds changes from NRL to add 32-bit physics support to their model. The changes are a subset of #772 and it replaces the prior attempt at this PR, #797

[grantfirl]

This looks familiar and OK to me, although it looks like it needs a merge with main given the conflict. The status of this is that it compiles in single precision and runs RTs successfully in double precision?

[SamuelTrahanNOAA]

I did not apply the changes needed at the ufs-weather-model or fv3atm level to get it to compile with 32 bits. That would be easy to add. Should I?

[grantfirl]

I did not apply the changes needed at the ufs-weather-model or fv3atm level to get it to compile with 32 bits. That would be easy to add. Should I?

I don't know that we need to merge changes in parent repos in order to enable compiling, but I'm guessing that this code should be able to compile before we merge it (this was Dom's requirement). Has it complied successfully in single precision?

[SamuelTrahanNOAA]

You'd need pretty much everything from my other 32-bit PR. It would just be a non-draft version of that PR.

Ligia's instructions were to do a PR that only compiles 64 bits, but contains all of the CCPP 32-bit changes. That is this PR. She said the next step is to get it to actually run, which is the purpose of the other PR. At this point, the other PR will compile 32 bit but not run, and this one won't compile or run 32 bits.

[grantfirl]

If it can compile in 32 bit mode using changes in the other draft PR, then that is good enough for me.

[SamuelTrahanNOAA]

As of this posting, this PR is up-to-date with all submodules and has passed intel compiler tests (on jet) and gnu compiler tests (on hera).

[DusanJovic-NOAA]

I would rewrite this big #ifndef / #else / #endif block as:

      integer, parameter :: kind_io4  = 4, kind_io8  = 8 , kind_ior = 8 &
     &,                     kind_evod = 8, kind_dbl_prec = 8            &
     &,                     kind_sngl_prec = 4
# ifdef __PGI
     &,                     kind_qdt_prec = 8                           &
# else
     &,                     kind_qdt_prec = 16                          &
# endif
     &,                     kind_LOGICAL = 4                            &
     &,                     kind_INTEGER = 4                             ! -,,-

#ifdef SINGLE_PREC
      integer, parameter :: kind_rad  = kind_sngl_prec                  &
     &,                     kind_phys = kind_sngl_prec                  &
     &,                     kind_taum = kind_sngl_prec                  &
     &,                     kind_grid = kind_dbl_prec                   &! atmos_cubed_sphere requres kind_grid=8
     &,                     kind_REAL = kind_sngl_prec                   ! used in cmp_comm
#else
      integer, parameter :: kind_rad  = kind_dbl_prec                   &
     &,                     kind_phys = kind_dbl_prec                   &
     &,                     kind_taum = kind_dbl_prec                   &
     &,                     kind_grid = kind_dbl_prec                   &! atmos_cubed_sphere requres kind_grid=8
     &,                     kind_REAL = kind_dbl_prec                    ! used in cmp_comm
#endif

avoiding unnecessary (repeated) definition of kinds that are equal in both cases.

[DusanJovic-NOAA]

Also, I do not see where kind_qdt_prec is used. I think it can be removed if it's not used anywhere.

[SamuelTrahanNOAA]

I think kind_qdt_prec is used in the Neptune fork. I'd rather leave it since it's not causing any troubles now.

[DusanJovic-NOAA]

I think kind_qdt_prec is used in the Neptune fork. I'd rather leave it since it's not causing any troubles now.

Neptune's fork of ccpp-physics? If they are using it in the code that's not in this repo it should be defined in their code, not here. I do not think this repo should 'collect' random definitions used elsewhere.

[DusanJovic-NOAA]

What about kind_taum?

[climbfuji]

I see what you mean. Let's wait to hear from NRL, given what you just said you are probably correct that it isn't used and can be removed. machine.F was ported from IPD as-is a long time ago and basically not updated since then.

[DusanJovic-NOAA]

The machine.F is used outside ccpp, so it is possible the Neptune dynamical core is using kind_taum or kind_qdt_prec

If this is the case, than that's even bigger reason to have definitions of kinds ONLY used by the code in this repo, and no any other kind, since code that depends on ccpp physics should only use definitions defined here to interface their code with the ccpp physics, ie. to declare variables used in the ccpp-physics public API. This creates unnecessary dependency.

[matusmartini]

@climbfuji @SamuelTrahanNOAA @DusanJovic-NOAA Thank you for including us in this discussion. At NRL, we only use kind_phys, kind_dbl_prec, and kind_sngl_prec from machine.F. @michalakes added the single precision kind and can comment more.

[michalakes]

The dbl and sngl types are necessary for the single precision option. But I gather from back-reading the thread that those are not at issue. With regard to kind_qdt_prec, I don't know where that comes from. It's not used in the NEPTUNE version of the code, except where it's defined in machine.F . But as Sam suggested, it doesn't do any harm whatever it might be for. I don't have an opinion whether to keep it or not.

[SamuelTrahanNOAA]

I've cleaned up that section of the code in a similar manner to what Dusan suggested. I've removed kind_taum and kind_qdt_prec since they don't seem to serve a purpose for UFS or NRL at the moment.

[SamuelTrahanNOAA]

I'm hoping @michalakes can review this and comment on the types. He has no repository access yet though, so I'll have to add him and hope he answers soon.

[BrianCurtis-NOAA]

@SamuelTrahanNOAA @grantfirl UFS RT passed. Ready for merge.