Machines: rename oneapi to intel everywhere by bartgol · Pull Request #8652 · E3SM-Project/E3SM

bartgol · 2026-08-20T21:04:58Z

The oneapi-ifx compiler becomes intel, while oneapi-ifxgpu becomes intelgpu. If conflicting, intel becomes intel-classic.

A few comments:

gnugpu.cmake was virtually the same as gnu.cmake, and simillarly for intelgpu.cmake vs intel.cmake. So I made the gpu ones simply include the non-gpu ones, possibly adding a couple of new settings. This saves code duplication.
In intel.cmake, I pick the MPI compilers names based on the MPILIB var.
machines that have 'intel' compiler but are actually using intel-classic compilers will continue to work, provided that they have a $machine_intel.cmake macro file, where they set the compiler names to the old intel compilers ($mach_$compiler.cmake is parsed AFTER $compiler.cmake).

@rljacob This will break nightlies on all machines that use oneapi-ifx, as that compiler no longer exists. I will take care of changing the testing scripts repo once this is merged.

Side q: should this PR be marked BFB or non-BFB? It will break baselines tests (and scripts), but it technically does not change answers.

ndkeen · 2026-08-20T21:08:40Z

I see some odd changes in cime_config/machines/cmake_macros/intel.cmake -- can we just ask why you are wanting to make so many changes to these core files?

bartgol · 2026-08-20T21:12:28Z

I see some odd changes in cime_config/machines/cmake_macros/intel.cmake -- can we just ask why you are wanting to make so many changes to these core files?

Because intel.cmake was defining as intel compilers the classic compilers. Since we want to use the name "intel" for the new (oneapi) compilers, this file must define the correct compilers names. The flags were relatively similar between oneapi-ifx.cmake and (the old) intel.cmake but, in the spirit of "intel means oneapi", i basically copied oneapi-ifx.cmake to intel.cmake (after moving the old intel.cmake to intel-classic.cmake). Unfortunately gh does not show that move, as intel.cmake already existed, so it lists diffs instead.

ndkeen · 2026-08-20T21:14:54Z

Why do you need to make changes to CICE builds? Maybe just explain what you need to do in the first place? What is not working for you now?

bartgol · 2026-08-20T21:21:47Z

I am not sure what you mean with changing CICE builds.

The key point in today's call seemed (to me, I may have musunderstood) that E3SM officially supported machines should use "intel" to mean "the new oneapi compilers series", and the old compilers should be referred to as "intel-classic". This PR addresses that by:

in config machines: rename oneapi-ifx to intel and oneapi-ifxgpu to intelgpu
in cmake_macros:
- the intel.cmake file has been moved to intel-classic.cmake
- the old oneapi-ifx.cmake has been moved to intel.cmake
- the XYZgpu.cmake files simply include XYZ.cmake (to avoid repeating the same flags), possibly adding extra stuff (e.g., intelgpu.cmake adds the --offload-compress CXX flags that only makes sense when building for GPUs). Any additional setting that only made sense for oneapi-ifxgpu (like set (USE_SYCL "TRUE")) has been moved to individual $mach_intelgpu.cmake files).

bartgol · 2026-08-20T21:29:29Z

-set(SFC "ifort")

-# Intel has to link fortran mains with fortran
+if (MPILIB STREQUAL impi)


I added this bit, since it seems silly that the base compiler cmake file dictates the MPI compilers without checking what MPI lib to use. The original oneapi-ifx did not do this distinction.

The change should not be consequential, since all the machines that had oneapi-ifx are using mpich, not impi.

rljacob · 2026-08-21T03:42:49Z

      <env name="ZFP_ROOT">$SHELL{if [ -z "$ZFP_ROOT" ]; then echo /lcrc/soft/climate/zfp/1.0.1/gcc-11.2.0; else echo "$ZFP_ROOT"; fi}</env>
    </environment_variables>
-    <environment_variables compiler="oneapi-ifx" mpilib="openmpi">
+    <environment_variables compiler="intel" mpilib="openmpi">


This seems like its left over from when we changed oneapi to intel on chrysalis. Just delete this and the MOAB_ROOT setting. Right @vijaysm ?

Yeah, I can reinstall to "intel" folder if we want to make that the oneapi default now, and use "intel-classic" for the older one? Do we need the classic version?

I also see line 2854 repeats this. Need to check this more.

will moving the folder work? A symlink also works probably.

rljacob · 2026-08-21T03:45:18Z

    </resource_limits>
  </machine>

  <machine MACH="jlse">


@amametjanov do we use this machine?

yes, it's an experimental cluster

I know but are we interested in running on it?

We may consider to remove the JLSE. THe intel stack is outdated and hardware is a copy of what we have on sunspot and aurora. It wont be updated as frequently as other places. I am not sure from reading the config-machines, which GPU hardware are we targetting in JLSE. (because it has machines from all 3 major vendors)

That was the first place Intel hardware landed I think which is why we once supported it.

rljacob · 2026-08-21T03:54:14Z

    </environment_variables>
  </machine>

  <machine MACH="sunspot">


@abagusetty can you check these changes to sunspot?

@amametjanov We might have to redo the sunspot setup, paths, modules are not valid. The only filesystem mounted is tegu so the paths with gila is not valid. So as the modules. Lets work this offline ?

/tegu/E3SMinput/soft

and the modules:

Currently Loaded Modules: 1) gcc/14.3.0 6) cray-libpals/1.8.0 2) oneapi/release/2026.1.0 7) cmake/3.31.11 3) mpich/prd/5.0.0.aurora_test.87e2045 8) gcc-runtime/14.3.0-pzdp6un (H) 4) libfabric/2.3.1 9) intel-oneapi-runtime/2026.1.0-sesoxjw (H) 5) cray-pals/1.8.0 10) kokkos/5.1.1-sycl Where: H: Hidden Module

rljacob · 2026-08-21T04:01:30Z

The only machines reporting to cdash this should disrupt is aurora and ghci-snl-oneapi.

rljacob · 2026-08-21T04:07:03Z

This should be tested on chrysalis, pm-cpu and pm-gpu before merging to next.

jgfouca

I like this change, LGTM!

Copilot

Pull request overview

Standardizes Intel oneAPI compiler names and consolidates shared compiler
configuration.

Changes:

Renames oneapi-ifx/oneapi-ifxgpu to intel/intelgpu.
Introduces intel-classic for the legacy Intel toolchain.
Deduplicates GNU and Intel GPU macro configuration.

Reviewed changes

Copilot reviewed 10 out of 12 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
`cime_config/machines/config_machines.xml`	Renames compiler selections.
`cime_config/machines/cmake_macros/sunspot_intelgpu.cmake`	Enables SYCL on Sunspot.
`cime_config/machines/cmake_macros/oneapi-ifxgpu.cmake`	Removes the old GPU macro.
`cime_config/machines/cmake_macros/oneapi-ifx.cmake`	Removes the old CPU macro.
`cime_config/machines/cmake_macros/intelgpu.cmake`	Reuses Intel configuration.
`cime_config/machines/cmake_macros/intel.cmake`	Configures the oneAPI toolchain.
`cime_config/machines/cmake_macros/intel-classic.cmake`	Preserves classic Intel settings.
`cime_config/machines/cmake_macros/gnugpu.cmake`	Reuses GNU configuration.
`cime_config/machines/cmake_macros/gnu.cmake`	Corrects flag spacing and comments.
`cime_config/machines/cmake_macros/dane_intel.cmake`	Adds Dane library RPATHs.
`cime_config/machines/cmake_macros/aurora_intelgpu.cmake`	Enables SYCL on Aurora.
`cime_config/machines/cmake_macros/aurora_intel.cmake`	Adds Aurora CPU linker settings.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

bartgol · 2026-08-21T23:28:11Z

+if (MPILIB STREQUAL impi)
+  set(MPICC "mpiicc")
+  set(MPICXX "mpiicpc")
+  set(MPIFC "mpiifort")


Copilot is right here. Since the old oneapi-ifx.cmake script set only the mpicxx/mpicc/mpifort wrappers (without checking the MPILIB name), I'm rolling this back to match that.

bartgol · 2026-08-21T23:29:21Z

    <NODENAME_REGEX>jlse.*</NODENAME_REGEX>
    <OS>LINUX</OS>
-    <COMPILERS>oneapi-ifx,oneapi-ifxgpu,gnu</COMPILERS>
+    <COMPILERS>intel,intelgpu,gnu</COMPILERS>


@abagusetty I'm going to ignore this, since from what you said we may remove this machine altogether.

bartgol · 2026-08-21T23:30:50Z

    <NODENAME_REGEX>jlse.*</NODENAME_REGEX>
    <OS>LINUX</OS>
-    <COMPILERS>oneapi-ifx,oneapi-ifxgpu,gnu</COMPILERS>
+    <COMPILERS>intel,intelgpu,gnu</COMPILERS>


bartgol · 2026-08-21T23:33:43Z

@rljacob I will test this next week on pm and chrys next week.

@abagusetty As you said, the current state of the jlse configuration is already questionable, so I am not going to make any further mods there and leave jlse in a limbo state. I will let you decide whether to remove it later or fix it. Also, I don't have access to it to test any mod.

bartgol requested review from jgfouca and rljacob August 20, 2026 21:04

bartgol self-assigned this Aug 20, 2026

bartgol added the Machine Files label Aug 20, 2026

bartgol force-pushed the bartgol/machines/intel-mods branch from eb5c49a to 95d9f6e Compare August 20, 2026 21:14

bartgol force-pushed the bartgol/machines/intel-mods branch from 95d9f6e to 8c30dd3 Compare August 20, 2026 21:16

bartgol force-pushed the bartgol/machines/intel-mods branch from 8c30dd3 to e48e5dc Compare August 20, 2026 21:26

bartgol commented Aug 20, 2026

View reviewed changes

rljacob requested a review from amametjanov August 21, 2026 03:27

rljacob reviewed Aug 21, 2026

View reviewed changes

jgfouca requested a balanced review from Copilot August 21, 2026 19:23

Copilot started reviewing on behalf of jgfouca August 21, 2026 19:23 View session

jgfouca approved these changes Aug 21, 2026

View reviewed changes

Copilot AI reviewed Aug 21, 2026

View reviewed changes

bartgol force-pushed the bartgol/machines/intel-mods branch from e48e5dc to b30051e Compare August 21, 2026 23:30

bartgol force-pushed the bartgol/machines/intel-mods branch from b30051e to 54a9ac0 Compare August 25, 2026 21:09

abagusetty approved these changes Aug 25, 2026

View reviewed changes

bartgol added 3 commits August 28, 2026 09:49

Machines: rename oneapi to intel

5252835

Machines: disable fortran linking on aurora/sunspot intelgpu machines

b4cfefe

EAMxx: adjust compiler name in machine specs for aurora

c662a11

bartgol force-pushed the bartgol/machines/intel-mods branch from 54a9ac0 to c662a11 Compare August 28, 2026 15:49

Uh oh!

Conversation

bartgol commented Aug 20, 2026

Uh oh!

ndkeen commented Aug 20, 2026

Uh oh!

bartgol commented Aug 20, 2026

Uh oh!

ndkeen commented Aug 20, 2026

Uh oh!

bartgol commented Aug 20, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rljacob Aug 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

abagusetty Aug 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rljacob Aug 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rljacob commented Aug 21, 2026

Uh oh!

rljacob commented Aug 21, 2026

Uh oh!

jgfouca left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

bartgol commented Aug 21, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants

rljacob Aug 21, 2026 •

edited

Loading

abagusetty Aug 21, 2026 •

edited

Loading

rljacob Aug 21, 2026 •

edited

Loading