Use oneapi compiler

[anton-seaice]

For ongoing support and better optimisation on new hardware, we need to move to using the Intel Oneapi / LLVM compilers instead of the current Intel Classic compilers. Initial discussion is here

We expect this change to not be bit for bit.

To-do:

• Add latest compiler to spack-config - Add latest version of the Intel OneAPI compilers on gadi spack-config#52
• Confirm if oneapi matches against when=%intel in spack ? e.g. https://github.com/ACCESS-NRI/spack-packages/blob/288ef380159af4536137b40360b2a87c043b7ce1/packages/access-om3-nuopc/package.py#L58 and
  https://github.com/ACCESS-NRI/spack-packages/blob/288ef380159af4536137b40360b2a87c043b7ce1/packages/access-om3-nuopc/package.py#L65
• Confirm if oneapi matches against (CMAKE_Fortran_COMPILER_ID MATCHES "Intel")
• (Optional ?) - check compiler flags used in CMakeLists ?
• Update docs

It doesn't really matter if we do this before or after #34 / #35 which uses the new build system (we just need to make sure CMake and spack changes, if required are done in all the related repositories / packages.

I suggest we do this seperately from code changes (e.g. 2025.01.0 could be build with intel-classic and 2025.01.1 could be oneapi with the same source code). This makes it easier to examine any differences in the results.

I may have added extra steps here - let me know if you have done some of these already @micaeljtoliveira

[micaeljtoliveira]

Confirm if oneapi matches against when=%intel in spack ?

No, it doesn't.

[micaeljtoliveira]

Confirm if oneapi matches against (CMAKE_Fortran_COMPILER_ID MATCHES "Intel")

Yes, this matches correctly.

[chrisb13]

Just seen this new issue, great to see some of the things we discussed here, already being addressed!

[anton-seaice]

Confirm if oneapi matches against when=%intel in spack

I had a quick look through the components, the only place I found this needed setting is in access-om3-nuopc package?

https://github.com/ACCESS-NRI/spack-packages/blob/3bfca290c24b48c410503b1bf8228f4cabd6e857/packages/access-om3-nuopc/package.py#L58

I can make a PR to add '-fp-model precise' to the esmf build when %oneapi (The flag name is unchanged according to oneAPI Programming Guide)

I looked in the fms and parallelio packages and there was nothing set specific for %intel, so I think we are in the clear otherwise ?

[anton-seaice]

setting -fp-model=precise in ACCESS-NRI/access-spack-packages#203

[anton-seaice]

Compiling FMS with %oneapi gives an internal compiler error:

/scratch/tm70/as2285/tmp/ifx0728833253Ii9FFe/ifx1jSaRa.i90: error #5633: **Internal compiler error: segmentation violation signal raised** Please report this error along with the circumstances in which it occurred in a Software Problem Report.  Note: File and line given may not be explicit cause of this error.

As reported here:

https://community.intel.com/t5/Intel-Fortran-Compiler/Internal-compiler-error-segmentation-violation-signal-raised/td-p/1590217

This should have been fixed in oneapi@2024.2 (so don't use %oneapi@2024.0.2)

[harshula]

Hi, This is how I'm building ACCESS-OM2 with oneapi@2025.0.4, spack.yaml contains:

    gcc-runtime:
      require:
        - '%gcc'
    all:
      require:
        - '%oneapi@2025.0.4'
        - 'target=x86_64'