Rename oneMKL Interface to oneMath

[Rbiessy]

Description

Related RFC: #564 and related spec PR: uxlfoundation/oneAPI-spec#596
As a reminder the plan is to have this PR and the spec PR approved before we move this repository to https://github.com/uxlfoundation and merge the 2 PRs.

• The PR has the following implications on the MKLCPU and MKLGPU backends:
  • Had to introduce src/include/common_onemkl_conversion.hpp to convert oneMath common types to oneMKL types. This wasn't needed before as the backends relied on the 2 projects to use the same namespace. The file also provide helper macros to catch oneMKL exceptions and rethrow them as oneMath exceptions.
  • The DFT MKLGPU backend used to catch and rethrow some but not all oneapi::mkl exceptions when committing a descriptor. We discussed this issue by email recently. With the changes needed by the renaming, we now catch and rethrow all the exceptions as oneapi::math exceptions and I was able to remove a related workaround in the tests (FYI @oneapi-src/onemkl-dft-write). Also see related discussion below: Rename oneMKL Interface to oneMath #602 (comment)
  • The RNG MKLGPU backend relied heavily on the 2 libraries using the same namespace. I introduced onemkl_distribution_conversion.hpp to convert the RNG types and call the backend's free functions such as oneapi::mkl::rng::generate, oneapi::mkl::rng::skip_ahead. @oneapi-src/onemkl-rng-write you may want to carefully review src/rng/backends/mklgpu/ as I am not familiar with the RNG domain.
  • The renaming made it clear that some header files were duplicating function declarations for no reason in the BLAS and LAPACK domains. They are removed in baeb476.
• The PR also remove the domain documentation for BLAS and LAPACK which duplicates the specification (commit b382291). We discussed in the issue [BLAS] Wrong namespace usage in BLAS Documentation #489 that the duplicated documentation was an issue and some of us are keen to remove it. Now is a good time as it becomes even more outdated with the name change. (@oneapi-src/onemkl-blas-write and @oneapi-src/onemkl-lapack-write FYI)
• The changes assume that the teams @oneapi-src/onemkl-* will be renamed (or duplicated) to @uxlfoundation/onemath-* in the new repository (@rscohn2 FYI)
• The changes assume that the onemkl Slack channel will be renamed to onemath (@rodburns FYI)
• Removed USE_MKLREF which was only used in Lapack tests and was not documented. This macro wouldn't work easily with the namespace change anymore (@oneapi-src/onemkl-lapack-write FYI)
• For reference the remaining occurrences of MKL are:
  • Intel backend names MKLCPU and MKLGPU.
  • Usages of oneapi::mkl inside the Intel backends.
  • File cmake/mkl/MKLConfig.cmake (could probably be removed IMO).
  • [email protected] in CODE_OF_CONDUCT.md and CONTRIBUTING.md. Let me know if there is any plan to create [email protected]. If not this can be updated separately.
  • Folder include/oneapi/mkl for deprecated headers and deprecated oneapi::mkl C++ namespace.
  • Deprecated ONEMKL CMake namespace.
  • Some files in the MKLCPU and MKLGPU backends use "mkl" in their name. They could be updated to onemkl if needed.

Checklist

All Submissions

• Do all unit tests pass locally?
  • Local CPU and iGPU with MKLCPU and MKLGPU backends: log_intel_cpu_igpu.txt
  • Intel PVC: log_pvc.txt
    • The BLAS and DFT tests failing are reproduced on the current develop branch, see GitHub issues [BLAS][MKLGPU] Trsv tests can fail on PVC #600 and [DFT][MKLGPU] Tests can fail on PVC #601
  • CUDA backends A100: log_a100.txt
    • The BLAS tests failing are reproduced on the develop branch, see GitHub issue [cuBLAS] Gemm tests using half can fail #599
  • AMD backends MI210: log_mi210.txt
  • BLAS netlib: log_netlib.txt
  • portBLAS and portFFT backends on Intel CPU: log_portblas_portfft.txt
• Have you formatted the code using clang-format? The PR already includes the clang-format changes from [clang-format] Reformat due to transition from v9.0.0 to v19.1.0 and update of style configuration #594 to avoid major conflicts when merging with develop.

[rafbiels]

That's an immense amount of work, well done! I only found a few missed occurrences.

[Rbiessy]

I merged the PR with the develop branch and the clang-format-19.10 changes which reduces the diff.
I have also updated the PR description with more details on the implications to make it work with the MKLCPU and MKLGPU backends (first bullet point).

[anantsrivastava30]

These are a lot of changes for one PR!. Mostly they look fine with what was discussed in the RFC and spce change. However I have a few question in my review.

[Rbiessy]

The conflicts are mostly due to the recent PR for cuSPARSE support. I will work on fixing them but the other domains can still be reviewed.

[andreyfe1]

Thanks for doing this huge work!
I have a general question from my side. Have you used git mv for renaming/duplicating files? This command preserves the file’s history. So, we will be able to see the history of commits for old files.

[Rbiessy]

Thanks for doing this huge work! I have a general question from my side. Have you used git mv for renaming/duplicating files? This command preserves the file’s history. So, we will be able to see the history of commits for old files.

Yes I have used git mv in the second commit: 68d3966
In this commit GitHub shows the movement unfortunately when the changes from the other commits are shown GitHub decides it's best to show some files have been deleted and re-created.
To work around this you could try to compare the diff between the commit after the move and the latest commit if that's easier. You can see it here: https://github.com/Rbiessy/oneMKL/compare/600cd44...2a8a76b?diff=split&w=

[andreyfe1]

ok. Thanks. Just wanted to make sure that we don't accidentally miss important information

[Rbiessy]

@oneapi-src/onemkl-sparse-write I have updated the PR to fix the conflicts with the cuSPARSE PR. It would be useful if you can review the sparse related changes in this PR before the rocSPARSE PR #544!
Log for Intel backends: intel_mklcpu_mklgpu.txt
Log for Nvidia backends: log_a100.txt

[Rbiessy]

I fixed some more recent conflicts.
@oneapi-src/onemkl-blas-write, @oneapi-src/onemkl-lapack-write in the last commit (baeb476) I have removed a couple of header files in blas and lapack which were duplicating function declarations for the MKLCPU and MKLGPU backends. Given that we can simply include the Intel oneMKL headers instead they can be removed here. This was discussed in #606 (comment)
Log testing the MKLCPU and MKLGPU backends with the 2025.0 base toolkit: intel_mklcpu_mklgpu.txt

[andrewtbarker]

BLAS changes look good to me.

[andrewtbarker]

Suggested change

	This page describes building the oneMath with either the Intel(R) oneAPI DPC++
	This page describes building oneMath with either the Intel(R) oneAPI DPC++

[sknepper]

/intelci: run