Build SU2 and run regressions on self hosted aarch64 runner scheduled

[martin-g]

Proposed Changes

Add workflow_call event to regression.yml Github Actions workflow. This way it is possible to re-use it from another workflow.

Add a new workflow - scheduled-arm64.yml. It is a scheduled workflow that just executes regressions.yml on a self-hosted runner.

Related Work

su2code/Docker-Builds#1 - A PR that updates the build and test Docker images to Ubuntu 20.04 so that they could be used on both AMD64 and ARM64.

PR Checklist

• I am submitting my contribution to the develop branch.
• My contribution generates no new compiler warnings (try with the '-Wall -Wextra -Wno-unused-parameter -Wno-empty-body' compiler flags, or simply --warnlevel=2 when using meson).
• My contribution is commented and consistent with SU2 style.
• I have added a test case that demonstrates my contribution, if necessary.
• I have updated appropriate documentation (Tutorials, Docs Page, config_template.cpp) , if necessary.

[martin-g]

The lines will be uncommented later to lock the caller only to scheduled-arm64.yml

[martin-g]

Temporarily reduced the config_set to reduce the build time. Once it works I will re-enable the complete list again.

[martin-g]

Related to su2code/Docker-Builds#1. I need to use the Docker image on the self-hosted Linux ARM64 runner and this requires updating the base image to Ubuntu 20.04 that is multi-platform.

The url to the Docker image will be updated to an official one once su2code/Docker-Builds#1 is merged.

[martin-g]

Same as with the build-su2 image.

[martin-g]

Every 2 minutes to be able to see the result quickly. Once everything works the schedule will be once per day or something similar.

[martin-g]

Initially I tried to work on this task in my personal fork - https://github.com/martin-g/SU2/pull/1.
The problem there was that compileSU2.sh clones https://github.com/su2code/SU2 and tries to find the PR ref in it and fails.
Now this problem should be gone but the expected issue is that the self-hosted runner won't be found because it is not yet registered for this Github repository.
@pcarruscag Please send me the preferred hardware specs (CPU cores, RAM, disk, ...) at martin.grigorov at gmail dot com and I will reserve a VM!

[martin-g]

@pcarruscag Could you please enable the Github Actions checks for this PR ?
They are disabled because it is my first PR in this repository.
Thank you!

[martin-g]

The Github Actions checks passed after making two changes:

1. efe98fe
   I needed to lower the warnlevel due to problems in externals/cgns/hdf5.
   Maybe the HDF5 files should be updated to a newer version ?! But it is not clear to me which version has been used in Add HDF5 backend support for CGNS in SU2 #1500.
   @MicK7 Do you have an idea how to fix this ?

2. 2614022
   A workaround recommended at Issue building python wrapper with CGNS support (problems with static/dynamic HDF5 dependencies) #1568 (comment)

3. The regression tests fail because the new test-su2 Docker image is private. I will need someone with higher privileges than me to make it public.

[martin-g]

@pcarruscag The build passes now after updating the regression tests expectations!

Should I continue working on this PR for the ARM64 part or should we merge it (without the new regressions-arm64.yml) and do the rest in another PR ?

Also what is your opinion on the change for the warnlevel ? Without it it is not possible to compile HDF5

[pcarruscag]

If you can continue on this one for bit more while we wrap up #1560 (which also had some changes in some regressions) it would help us.
You'll need to disable specific CGNS or HDF5 warnings in the respective meson.build files (in externals/cgns/...), it took some time to get our code warning-free and we shouldn't go back on that.

[martin-g]

The regression tests fail on Ubuntu 20.04 aarch64:

==================================================================
Summary of the parallel tests
python version: 3.8.10 (default, Mar 15 2022, 12:22:08) 
[GCC 9.4.0]
  passed - discadj_naca0012
  passed - discadj_cylinder3D
  passed - discadj_arina2k
  passed - ea_naca64206
  passed - discadj_rans_naca0012_sa
  passed - discadj_rans_naca0012_sst
  passed - discadj_incomp_NACA0012
  passed - discadj_incomp_cylinder
  passed - discadj_incomp_turb_NACA0012_sa
  passed - discadj_incomp_turb_NACA0012_sst
  passed - discadj_axisymmetric_rans
  passed - unsteady_cylinder
  passed - unsteady_cylinder_windowed_average_AD
* FAILED - unsteady_cylinder_windowed_average
  passed - unsteady_cylinder_DT_1ST
  passed - pitchingNACA0012
  passed - transonic_stator
* FAILED - discadj_fea
  passed - discadj_heat
  passed - discadj_fsi
* FAILED - discadj_fsi_airfoil
  passed - discadj_cht
  passed - da_sp_pinArray_cht_2d_dp_hf
  passed - da_sp_pinArray_cht_2d_mf
  passed - da_unsteadyCHT_cylinder
  passed - discadj_topol_optim
  passed - unsteady_NACA0012_restart_adjoint
  passed - unsteady_NACA0012_adjoint_only
  passed - restart_shape_optimization
* FAILED - dyn_discadj_fsi
* FAILED - grad_smooth_sob
Tests failed

It is the same version of Python and GCC as on the Github Actions hosted runner.
We will need logic to differentiate the expected results on x86_64 and aarch64 :-/

[martin-g]

@pcarruscag I've reverted the warnlevel back to =3 and added some exclusions for errors and warnings in externals/**!

But there are also some problems in SU2's code (we should see them as build errors in the new checks).

With

diff --git a/meson.build b/meson.build
index 4c95a97047..1a34b8ac1d 100644
--- a/meson.build
+++ b/meson.build
@@ -22,6 +22,12 @@ if build_machine.system() != 'windows'
     default_warning_flags += ['-Wno-empty-body']
   endif
   default_warning_flags += ['-Wno-unused-parameter',
+    '-Wno-error=cast-function-type',
+    '-Wno-cast-function-type',
+    '-Wno-error=stringop-truncation',
+    '-Wno-stringop-truncation',
+    '-Wno-error=missing-field-initializers',
+    '-Wno-missing-field-initializers',
                             '-Wno-deprecated-declarations']

the build passes locally but I guess those should be fixed instead of suppressed.

[martin-g]

@pcarruscag The PR is ready for review!
There are two regression tests which always fail - https://github.com/su2code/SU2/runs/7019382870?check_suite_focus=true
polar_naca0012 in serial_regression.py and parallel_regression.py. The computed values are different in each run :-/

[martin-g]

Since I used the Github UI to merge the latest updates in develop branch now it is not possible to rebase+squash the commits in this branch.
Unless you have a better idea I plan to just copy the contents of the modified files in this PR to a new branch and open a new PR.

[pcarruscag]

Usually we dont squash-merge code PRs, so it's fine.