Combine ip and ip2 libraries #54
Conversation
ip and ip2 were combined using modern Fortran to abstract away the grib and grib2 specific portions into a unified interface. Instead of each function taking grib1/grib2 descriptors directly, a derived type ip_grid_descriptor is used and then each interpolation operates on that.
Combined ip libraries warrant a bump in major version
…ip into feature/combined-ip
…ip into feature/combined-ip
kgerheiser
requested review from
edwardhartnett,
Hang-Lei-NOAA and
GeorgeGayno-NOAA
|
@kgerheiser Can you build it on Hera or Dell? I can test it with UFS_UTILS. |
|
Try this |
|
@GeorgeGayno-NOAA I ran the UFS_UTILS also passes all the Github workflows with the updated library. |
What UFS_UTILS branch are you using? |
The 'orog' code uses IPLIB. |
|
@GeorgeGayno-NOAA is that included in one of the consistency tests? |
The only tests we have for the 'orog' code currently are the 'grid_gen' consistency tests. |
|
|
|
The corresponding UFS_UTILS issue is: ufs-community/UFS_UTILS#242 |
|
@kgerheiser I checked out your UFS_UTILS feature/combined-ip branch on Hera. It compiled, but was pointing to the current stack version of IPLIB. Am I looking at the wrong branch? I recall having to make source code changes to get the combined iplib to work with |
|
Oh, I didn't commit that change. I removed |
| @@ -1,3 +1,6 @@ | |||
| !> @file | |||
| !! @brief Gaussian grid | |||
|
|
|||
There was a problem hiding this comment.
Author tag at file level as well...
| !! @param[inout] self The grid to initialize | ||
| !! @param[in] g1_desc A grib1_descriptor | ||
| !! | ||
| !! @author Kyle Gerheiser |
There was a problem hiding this comment.
Add a @Date tag as well.
| @@ -110,128 +124,128 @@ subroutine init_grib2(self, g2_desc) | |||
|
|
|||
| end subroutine init_grib2 | |||
|
|
|||
| !$$$ SUBPROGRAM DOCUMENTATION BLOCK | |||
There was a problem hiding this comment.
This whole comment block must also be doxygenated.
| implicit none | ||
|
|
||
| !> @param Constant to choose BILINEAR interpolation method |
There was a problem hiding this comment.
The param statements all go into the doxygen header at the top of the subroutine.
There was a problem hiding this comment.
It's not a subroutine. @param can be used to document module variables and constants.
There was a problem hiding this comment.
OK, interesting. Can we get this merged and then discuss doxygen afterwards?
| case(BICUBIC_INTERP_ID) | ||
| CALL interpolate_bicubic(IPOPT,grid_in,grid_out,MI,MO,KM,IBI,LI,GI,NO,RLAT,RLON,IBO,LO,GO,IRET) | ||
| CALL interpolate_bicubic(IPOPT,grid_in,grid_out,MI,MO,KM,IBI& |
There was a problem hiding this comment.
For easy review, what I suggest is to do all whitespace only reformatting as a separate PR, with no code changes. So merge this PR, then reformat the code however you like, and then do a PR with that.
That way we can easily scan the whitespace changes to ensure no code was accidently changed.
I think this PR is finally ready.
This PR combines the ip and ip2 libraries into a single library with modern interfaces and modules. Internally, instead of passing around the complete grib1/grib2 spec, an abstract
ip_grid_descriptorobject is used as a wrapper. Then, all the interpolation routines can re-use the majority of the code. Spectral interpolation is a special case and under the hood there are still two separate routines for grib1/grib2 because there's a lot of case specific things going on that were difficult to generalize.The code has been tested with a series of unit tests (Intel and GCC, with 66%+ coverage), UFS_UTILS, and wgrib2.
To use the code
use ip_modwhich provides access to the main routinesipolates,ipolatev, andgdswzd. The specific routine can also be called by appending_grib1or_grib2.To Do: Finish the Doxygen documentation. A lot of it is complete, but still some more to go.