update mpif.h to use mpi

[sebastianbeyer]

Locally compiles, let's see what the tests say. Also, do we test compiling with e.g. ifs_interface enabled? Otherwise we might be missing something ;)
I will try use mpi_f08 as well if this works and also try to build on levante and lumi...

[sebastianbeyer]

I could build and run on levante and I also noticed that mpi_topology_module.F90 already contains use mpi in one place ;)
use mpi_f08 takes more effort (for some reason the general mpi routines can not be found, but if I call everything with _f08 postfix it does find it...) so I would suggest to merge this now and look at f08 later again.

[JanStreffing]

Standalone looks good. I assume you tested this with IFS interface by now? Our test suit checks if FESOM builds as a library, but obviously it can't check if it runs as a library in IFS.

[sebastianbeyer]

I am testing with IFS now, but it takes a little, because we are behind main quite a bit and now I need to add all of the namelist groups that did not exist before to the config on the HPCs...

[sebastianbeyer]

okay, now I tested on levante with IFS-FESOM and it still works. So from my point of view this can be merged

[ackerlar]

I got the following error during compilation:

/home/a/a270124/model_codes/awiesm-2.1-paleodyn-2.5/fesom-2.5/src/cpl_driver.F90(354): error #6285: There is no matching specific subroutine for this generic subroutine call.   [MPI_COMM_RANK]
    CALL MPI_Comm_Rank ( MPI_COMM_WORLD, commRank, ierror )
---------^
compilation aborted for /home/a/a270124/model_codes/awiesm-2.1-paleodyn-2.5/fesom-2.5/src/cpl_driver.F90 (code 1)

#634 solves the issue for me

[sebastianbeyer]

You need to fix the call, probably you need to use the correct type for MPI_COMM_WORLD

[sebastianbeyer]

ah, no it's not that...
looks like this file is not always compiled. Seems like it does call MPI subroutines, but there is no use mpi nor include mpif.h anywhere. strange.

[sebastianbeyer]

Okay, it looks like the file cpl_driver.F90 does not contain either use mpi nor include mpif.h. This was never correct, but before it did work, because when that file included another module which did have include mpif.h it would also get a copy of the subroutines. So the solution is to rather add use mpi to the respective modules in cpl_driver.F90.

[sebastianbeyer]

and also update the CI, so that this file is also built during in the CI builds.

[ackerlar]

Unfortunately, I still get the same error during compilation

[sebastianbeyer]

Then please give me a way to replicate thatAm 09.10.2024 um 13:56 schrieb Lars Ackermann ***@***.***>: Unfortunately, I still get the same error during compilation —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were assigned.Message ID: ***@***.***>

[JanStreffing]

You can install awi-esm2 with esm_tools. that should give you the error. @ackerlar, which version does Sebastian have to try and install?

[ackerlar]

The version is "awiesm-2.1-paleodyn-2.5"

[sebastianbeyer]

But that requires access to https://git.geomar.de/foci/src/oasis3mct.git if I understand correctly. And I don't have that...

[JanStreffing]

That doesn't sound right, awi-esm2 should not use geomar oasis... should be from https://gitlab.dkrz.de/modular_esm/oasis3-mct i think

[ackerlar]

Apparently, oasis and echam come from the AWI gitlab:

https://gitlab.awi.de/paleodyn/Models/oasis3-mct.git
https://gitlab.awi.de/paleodyn/Models/echam6.git

@mandresm or @pgierz could give you access

[sebastianbeyer]

ah, I did not see the other possible repositories.

Okay, but this is too tedious, sorry

[JanStreffing]

I downloaded the source codes, you can copy from here /work/ab0246/a270092/model_codes/awiesm-2.1-paleodyn-2.5 and do esm_master comp-awiesm-2.1-paleodyn-2.5 in the copy you copy it to.

[mandresm]

You should have access now @sebastianbeyer

[sebastianbeyer]

Okay, the line that you show: CALL MPI_Comm_Rank ( MPI_COMM_WORLD, commRank, ierror )
has commRank and that is a logical! but MPI_COMM_RANK expects an integer here! that is the reason to do use mpi instead of include mpif.h! Type safety!
What is the purpose of commRank? what should be its type??

[ackerlar]

It seems to me that commRank is only used for few write statements. I changed its type and adapted the if statements (I don't know who included this part but I guess it should read ioerrer .eq. oasis_ok) and now the models compiles. I a doing a quick test run and push the changes. Thanks @sebastianbeyer