gromacs: add v2024.4, apple-clang openmp support for 2025+

[mathomp4]

This is a "by-hand" transfer of spack/spack#47931 to spack-packages. The migrate-package-prs script by @haampie didn't work for this because of a conflict.

---

First up, I mention the maintainers of Gromacs:

• @eirrgang
• @junghans
• @mabraham

I was trying to build Gromacs on my Mac today and when I tried I got the:

        if(CMAKE_CXX_COMPILER_ID MATCHES "AppleClang")
            message(FATAL_ERROR "The compiler you are using does not support OpenMP parallelism, "
                "Apple has unfortunately explicitly disabled OpenMP in their clang-derived compiler. "
                "You can disable OpenMP in Gromacs with -DGMX_OPENMP=OFF, but instead "
                "we recommend installing the unsupported library distributed by the R "
                "project from https://mac.r-project.org/openmp/ - or switch to gcc.")

error. This is because +openmp is the default variant.

Now, I've seen similar in the past on projects I work on and the solution has been to depend on llvm-openmp. So I added that.

Further build testing showed, though, that the latest version here, 2024.3, was not yet happy as it failed in build with:

  >> 397    clang++: error: unknown argument: '-Xclang -fopenmp'

The issue was that in the verbose output it had in the flags:

...
-mmacosx-version-min=15.0 -fPIC "-Xclang -fopenmp" -Xclang -fopenmp -MD -MT
...

So, I did a bit more testing and did some spack checksum as well and found that gromacs@2024.4 did not work, but gromacs@2025.0-beta did and so did gromacs@main. As there isn't yet an official 2025 version, this is sort of a...PR in waiting.

Now, the one issue is, I don't really know how to run Gromacs¹ let alone test if OpenMP is working. So I'm hoping one of the maintainers can try out this change and see if it lets the macOS version of Gromacs use OpenMP.

Echoing what @mabraham did in #47935, I've added a conflict if the version is 2024 or older.

I've also added 2024.4 as...it wasn't there.

Footnotes

1. I know, sounds odd, but I saw this Macinchem post in my RSS feeds asking about Gromacs and Apple Silicon and, well, spack was right there on my laptop so... 😄 ↩

[tldahlgren]

The syntax has changed. See the suggested fix.

Apologies, I thought I was in another tab 😞

Usually, we don't recommend supporting openmp parallelism when using apple-clang. Trying to maintain support for a feature that was deliberately removed from the compiler might have a high maintenance burden.

[mabraham]

I'm a listed maintainer, why can't I approve this?

[mabraham]

@alalazo Thanks for the tip. If openmp and apple-clang causes trouble, I'll revert it. But since we're only supporting the latest of everything maybe it will be ok...

[mathomp4]

@alalazo I can close this if you like, in re openmp on apple-clang.

[tldahlgren]

Will defer to @alalazo for final decision on this PR being merged.

[tldahlgren]

Will defer to @alalazo for final decision on this PR being merged.

ping @alalazo

[alalazo]

Apologies. I missed this ping, but feel free to merge if this seems ready. I really entered this review by accident, thinking I was approving another PR 🙂

The comment above #61 (comment) was really just an "fyi". I didn't meant to block (otherwise I would have used "require changes")