Update Omega to C++20 by mwarusz · Pull Request #405 · E3SM-Project/Omega

mwarusz · 2026-05-12T15:40:37Z

This draft PR updates Omega C++ standard to C++20. E3SM is in the process to upgrading to Kokkos 5, which requires C++20. This PR has the necessary fixes on the Omega side.

Checklist

grnydawn

I reviewed the code visually, and it looks good to me. All tests are passed on Frontier(craygnu-mphipcc, craycray-mphipcc, and craygnu), PM-CPU(gnu), PM-GPU(gnugpu), and Chrysalis(gnu). Approved.

grnydawn · 2026-05-27T13:55:20Z

Due to replacing the spdlog submodule with a new version and location, the changes to the logging module in this PR conflict with those in the develop branch.

sbrus89 · 2026-08-13T21:58:46Z

@mwarusz, does E3SM-Project#8559 getting merged mean it is time to do another E3SM/master merge into Omega and then merge this PR?

mwarusz · 2026-08-13T22:19:46Z

@sbrus89

Yes, that would be good, as long as we don't run into any problems. I still need to rebase this PR. I can try to do a local merge after that to see if everything works as expected and, if there are no issues, we can go ahead with the switch. How does that sound ?

grnydawn

The updates in this PR are a subset of those in the previous PR. Previously, all tests passed on Frontier, PM-CPU, PM-GPU, and Chrysalis. The current PR also passed all CTests on PM-GPU and Chrysalis. Approved.

sbrus89 · 2026-08-18T18:13:03Z

@mwarusz, will this PR fix #368, or are more changes needed in another PR?

mwarusz · 2026-08-18T18:33:07Z

@sbrus89

No. At first, I misread E3SM-Project#8559. That PR only changes the standard and doesn't update Kokkos. The PR we need to fix #368 is E3SM-Project#8632, which hasn't been merged into E3SM yet.

I am not sure if we want to merge this PR and do a master merge now, or do we want to wait for E3SM-Project#8632. I will leave this decision up to you.

sbrus89 · 2026-08-18T18:43:34Z

Thanks, that's what I had suspected. Let's wait for E3SM-Project#8632 before we merge #521. Given the level of testing associated with merging master into develop, I think it makes sense to do it once rather than twice. It's good to know all the pieces are almost in place to fix #368. Thanks for helping with this!

sbrus89 · 2026-09-24T19:58:56Z

@mwarusz, I'm testing this in conjunction with #566, and am seeing compile errors related to spdlog on Frontier with craygnu-mphipcc:

/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:440:3: error: reference to __host__ function 'assert_fail' in __host__ __device__ function
  440 |   FMT_ASSERT(std::is_unsigned<Int>::value || value >= 0, "negative value");
      |   ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:397:19: note: expanded from macro 'FMT_ASSERT'
  397 |          : ::fmt::assert_fail(__FILE__, __LINE__, (message)))
      |                   ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:1859:20: note: called by 'append<char>'
 1859 |       auto count = to_unsigned(end - begin);
      |                    ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:354:27: note: 'assert_fail' declared here
  354 | FMT_NORETURN FMT_API void assert_fail(const char* file, int line,
      |                           ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:440:3: error: reference to __host__ function 'assert_fail' in __host__ __device__ function
  440 |   FMT_ASSERT(std::is_unsigned<Int>::value || value >= 0, "negative value");
      |   ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:397:19: note: expanded from macro 'FMT_ASSERT'
  397 |          : ::fmt::assert_fail(__FILE__, __LINE__, (message)))
      |                   ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:1859:20: note: called by 'append<wchar_t>'
 1859 |       auto count = to_unsigned(end - begin);
      |                    ^
/autofs/nccs-svm1_proj/cli115/sbrus/polaris_worktrees/testing/e3sm_submodules/Omega/components/omega/external/spdlog/include/spdlog/fmt/bundled/base.h:354:27: note: 'assert_fail' declared here
  354 | FMT_NORETURN FMT_API void assert_fail(const char* file, int line,
      |                           ^
2 errors generated when compiling for gfx90a.

mwarusz · 2026-09-28T14:59:29Z

@sbrus89

I rebased and pushed a workaround. I also made a small change to the definition of INNER_LAMBDA in hierarchical parallelism to avoid one million warnings that we otherwise get in C++20 mode on frontier.

sbrus89 · 2026-10-02T20:44:59Z

Local merge passes CTests and omega_pr suite on Frontier with craygnu and craygnu-mphipcc. Also passes omega_pr on Aurora with oneapi-ifxgpu and e3sm_omega_developer in pm-cpu with gnu:

Polaris `omega_pr` suite

Baseline workdir: /ccs/home/brus/run/polaris_baseline_omega_pr_omega_craygnu_mpich/
Baseline build: /ccs/home/brus/run/polaris_baseline_omega_pr_omega_craygnu_mpich/ocean_build
PR build: /ccs/home/brus/run/polaris_cxx20_omega_pr_omega_craygnu_mpich/ocean_build
PR workdir: /ccs/home/brus/run/polaris_cxx20_omega_pr_omega_craygnu_mpich
Machine: frontier
Partition: batch
Compiler: craygnu
Build type: Release
Log: /ccs/home/brus/run/polaris_cxx20_omega_pr_omega_craygnu_mpich/polaris_omega_pr.o5574597
Result: All tests passed

Polaris `omega_pr` suite

Baseline workdir: /ccs/home/brus/run/polaris_baseline_omega_pr_omega_craygnu-mphipcc_mpich/
Baseline build: /ccs/home/brus/run/polaris_baseline_omega_pr_omega_craygnu-mphipcc_mpich/ocean_build
PR build: /ccs/home/brus/run/polaris_cxx20_omega_pr_omega_craygnu-mphipcc_mpich/ocean_build
PR workdir: /ccs/home/brus/run/polaris_cxx20_omega_pr_omega_craygnu-mphipcc_mpich
Machine: frontier
Partition: batch
Compiler: craygnu-mphipcc
Build type: Release
Log: /ccs/home/brus/run/polaris_cxx20_omega_pr_omega_craygnu-mphipcc_mpich/polaris_omega_pr.o5575057
Result: All tests passed

Polaris `omega_pr` suite

Baseline workdir: /home/brus/run/polaris_omega_pr_baseline_oneapi-ifxgpu_mpich/
Baseline build: /home/brus/run/polaris_omega_pr_baseline_oneapi-ifxgpu_mpich/ocean_build
PR build: /home/brus/run/polaris_cxx20_omega_pr_omega_oneapi-ifxgpu_mpich/ocean_build
PR workdir: /home/brus/run/polaris_cxx20_omega_pr_omega_oneapi-ifxgpu_mpich
Machine: aurora
Compiler: oneapi-ifxgpu
Build type: Release
Log: /home/brus/run/polaris_cxx20_omega_pr_omega_oneapi-ifxgpu_mpich/polaris_omega_pr.o8898746
Result: All tests passed

The changes to Omega's build system to support C++20, brought in by #405, only affected standalone builds. After #566, we are using Kokkos5 and need C++20 in coupled mode (in addition to standalone). This PR moves setting the CMAKE_CXX_STANDARD and CMAKE_CXX_STANDARD_REQUIRED into the common macro, so that they apply to both couple and standalone builds.

grnydawn self-requested a review May 14, 2026 18:16

grnydawn approved these changes May 14, 2026

View reviewed changes

grnydawn added the Omega label May 14, 2026

grnydawn assigned mark-petersen May 14, 2026

grnydawn self-requested a review May 27, 2026 13:52

mwarusz force-pushed the omega/cxx20 branch from 6370d04 to 379589d Compare August 14, 2026 18:19

grnydawn approved these changes Aug 14, 2026

View reviewed changes

mwarusz marked this pull request as ready for review August 14, 2026 19:46

sbrus89 mentioned this pull request Aug 18, 2026

Merge E3SM master into Omega develop #521

Merged

12 of 20 tasks

mwarusz added 5 commits September 28, 2026 08:54

Switch to C++20

0c658a4

Disable C++20 modules in cpptrace

2370a3a

Don't use deprecated is_hostspace

757c8bb

Capture by reference in INNER_LAMBDA

3a2686b

Add a workaround for spdlog compilation issue with craygnu-mphipcc

4315119

mwarusz force-pushed the omega/cxx20 branch from 379589d to 4315119 Compare September 28, 2026 14:55

sbrus89 merged commit fc9eaef into E3SM-Project:develop Oct 2, 2026
1 check passed

This was referenced Oct 5, 2026

Give analysis output fields the units, standard name and cell methods of the field they reduce #548

Merged

Add Rayleigh damping #587

Merged

andrewdnolan mentioned this pull request Oct 12, 2026

Use C++20 in both standalone and coupled builds #601

Merged

4 of 13 tasks

Navigation Menu

Repository navigation

Update Omega to C++20 - #405

Update Omega to C++20#405
sbrus89 merged 5 commits into
E3SM-Project:developfrom
mwarusz:omega/cxx20

mwarusz commented May 12, 2026

Uh oh!

grnydawn left a comment •

edited

Loading

Uh oh!

grnydawn commented May 27, 2026

Uh oh!

sbrus89 commented Aug 13, 2026

Uh oh!

mwarusz commented Aug 13, 2026

Uh oh!

grnydawn left a comment

Uh oh!

sbrus89 commented Aug 18, 2026

Uh oh!

mwarusz commented Aug 18, 2026

Uh oh!

sbrus89 commented Aug 18, 2026

Uh oh!

sbrus89 commented Sep 24, 2026

Uh oh!

mwarusz commented Sep 28, 2026 •

edited

Loading

Uh oh!

sbrus89 commented Oct 2, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

mwarusz commented May 12, 2026

Uh oh!

grnydawn left a comment • 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!

grnydawn commented May 27, 2026

Uh oh!

sbrus89 commented Aug 13, 2026

Uh oh!

mwarusz commented Aug 13, 2026

Uh oh!

grnydawn left a comment

Choose a reason for hiding this comment

Uh oh!

sbrus89 commented Aug 18, 2026

Uh oh!

mwarusz commented Aug 18, 2026

Uh oh!

sbrus89 commented Aug 18, 2026

Uh oh!

sbrus89 commented Sep 24, 2026

Uh oh!

mwarusz commented Sep 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sbrus89 commented Oct 2, 2026

Polaris omega_pr suite

Polaris omega_pr suite

Polaris omega_pr suite

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

grnydawn left a comment •

edited

Loading

mwarusz commented Sep 28, 2026 •

edited

Loading

Polaris `omega_pr` suite

Polaris `omega_pr` suite

Polaris `omega_pr` suite