Support mpifort in addtion to and/or to replace mpif90

[matthewrmshin]

Are there any linked Issues or Pull Requests?

• MetOffice/mo-bundle#859 (Private)

Brief description

In https://github.com/MetOffice/lfric_core/tree/main/infrastructure/build/fortran, we have support for mpif90 but not mpifort. The former is legacy.

Further details of the issue.

• https://github.com/MetOffice/mo-bundle/pull/859#discussion_r2728183798 (Private)

[andrewcoughtrie]

I'm guessing just duplicating mpif90.mk to create an mpifort.mk would solve your issue.

[matthewrmshin]

Possibly, as I don't fully understand the logic in mpif90.mk and its relations with the other *.mk files.

[yaswant]

I'm guessing just duplicating mpif90.mk to create an mpifort.mk would solve your issue.

And replacing the MPIF90 environments for consistency.

[andrewcoughtrie]

I'm guessing just duplicating mpif90.mk to create an mpifort.mk would solve your issue.

And replacing the MPIF90 environments for consistency.

We aren't actually using the mpif90 wrapper, everything is done manually rather than using the mpi wrappers. The mpif90.mk file is just extracting the underlying compiler from the wrapper and then all the linking steps are done manually. The inclusion of mpif90.mk comes from here: infrastructure/build/fortran.mk. So there isn't really anything to change, it is all just logic based on what the FC environment variable is set to. We pick the contents of FC up from the compiler module when it's loaded (on AZSPICE); the Crays have different magic but it works in a similar way.

Since all of this will be replaced by fab there isn't really much point doing extensive changes.

[matthewrmshin]

Andrew Coughtrie (@andrewcoughtrie) Sounds like a good plan. Let's go with the least effort solution for now.

[andrewcoughtrie]

One thing we could change in our environments is the LDMPI environment variable which is currently set to mpif90. This just tells the build system which linker to use when building with MPI. I don't see much point in doing so though as it currently works and will be replaced soon enough. It is probably worth Matt Shin (@matthewrmshin) being aware of this though (in case he isn't already) for the mo-bundle MPI wrapper migration, that and FPP which defines the pre-processor command.

All of this stuff has become more complex over time and is well due for a rewrite (bring on fab) so it's good that this is happening and I'm sure careful thought is being put into how this all works for the new build system.

[matthewrmshin]

[like] Matthew Shin reacted to your message:

…

________________________________ From: Andrew Coughtrie ***@***.***> Sent: Tuesday, January 27, 2026 4:30:30 PM To: MetOffice/lfric_core ***@***.***> Cc: Matthew Shin ***@***.***>; Mention ***@***.***> Subject: Re: [MetOffice/lfric_core] Support mpifort in addtion to and/or to replace mpif90 (Issue #245) [https://avatars.githubusercontent.com/u/24609575?s=20&v=4]andrewcoughtrie left a comment (MetOffice/lfric_core#245)<#245 (comment)> One thing we could change in our environments is the LDMPI environment variable which is currently set to mpif90. This just tells the build system which linker to use when building with MPI. I don't see much point in doing so though as it currently works and will be replaced soon enough. It is probably worth @matthewrmshin<https://github.com/matthewrmshin> being aware of this though (in case he isn't already) for the mo-bundle MPI wrapper migration, that and FPP which defines the pre-processor command. All of this stuff has become more complex over time and is well due for a rewrite (bring on fab) so it's good that this is happening and I'm sure careful thought is being put into how this all works for the new build system. — Reply to this email directly, view it on GitHub<#245 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AAG52RLNK5E3F7M6QBBULQ34I6HCNAVCNFSM6AAAAACS5VYH2OVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQMBWGE4DKNJYGI>. You are receiving this because you were mentioned.Message ID: ***@***.***>