CI: Add Ubuntu on ARM runner to CI

[mmuetzel]

GitHub started hosting runners with Ubuntu on ARM64 processors for open source projects for free:
https://github.blog/changelog/2025-01-16-linux-arm64-hosted-runners-now-available-for-free-in-public-repositories-public-preview/

Add one configuration that is using these runners to the build matrix.

According to their blog post, the arm64 runners are more efficient and potentially faster than the x86_64 runners. (But it is still in a preview phase and maybe it will take some time for them to better scale and balance the load.) If that turns out to be true, it should be easy to switch more configurations in that workflow (or the one using the root CMake file) to the arm64 runners (and maybe keep only one or two running on x86_64).

Since there is now the possibility to run tests for this target natively, remove that target from the build matrix of the workflow that emulates it with qemu.

There is an open PR for the Alpine action that would allow running the armv7 tests natively on aarch64. But it hasn't been merged yet:
jirutka/setup-alpine#22
So, keep emulating armv7 on x86_64 for the time being.

[mmuetzel]

Hu? Why are the runners that are using Clang on Ubuntu failing? I didn't touch anything in the rules about them...

Looking back on the actions that ran in the last couple of weeks: That isn't new with the changes from here. These runners started failing without any change in this repository.
Might be related to the fact that the ubuntu-latest label was moved from ubuntu-22.04 to ubuntu-24.04 by GitHub recently. That came with a newer version of Clang...

A few tests for LAGraphX are failing with a segfault now.
@DrTimothyAldenDavis: Do you see a pattern for the failing tests?

  The following tests FAILED:
  	 39 - LAGraphX_BF (SEGFAULT)
  	 40 - LAGraphX_FastGraphletTransform (SEGFAULT)
  	 50 - LAGraphX_SquareClustering (SEGFAULT)
  	 55 - LAGraphX_msf (Failed)
  Errors while running CTest
  make: *** [Makefile:74: test] Error 8

[mmuetzel]

Sure enough, those workflows succeed if they are reverted to the Ubuntu 22.04 runners.
That could be a work-around for the time being. But it might also be worth finding out why it is failing on Ubuntu 24.04. (Fwiw, GitHub - at least until now - provides runners for the last two LTS versions of Ubuntu. So, the Ubuntu 22.04 runners will probably be available until some time in 2026.)

What I've missed before: The runners that build for Windows MSVC using Clang as the compiler are also failing the tests for LAGraph.
It might be that the compatibility with newer versions of LLVM Clang is broken for some reason (independent on the OS -- Linux or Windows).

[DrTimothyAldenDavis]

Thanks. yes, I've noticed the Windows clang failures. I was considering just deleting those runners.

[mmuetzel]

I tried to enable CUDA on the Ubuntu on ARM runner. But CMake couldn't detect a working CUDA compiler (albeit installing the same packages that work for Ubuntu on x86_64).

  -- Looking for a CUDA compiler
  -- Looking for a CUDA compiler - NOTFOUND
  -- CUDA:             not found
  CMake Error at cmake_modules/SuiteSparsePolicy.cmake:367 (message):
    CUDA required for SuiteSparse but not found
  Call Stack (most recent call first):
    CMakeLists.txt:46 (include)
  
  
  -- CUDA:             not enabled
  -- Configuring incomplete, errors occurred!
  Error: Process completed with exit code 1.

I don't know how to install a working CUDA toolchain on ARM or how to tell CMake to use it. I also can't test that locally (no ARM hardware here). So, I opted for disabling CUDA on the ARM runner.

In the last run for the root CMakeLists.txt file on Ubuntu on ARM, checking out the repository failed apparently for some reason.
@DrTimothyAldenDavis: When you come around to it, could you please restart that runner?

Other than that, this should be ready for review now.