Make OpenMP, Kokkos, and Cabana hard dependencies

[RudolfWeeber]

find src/core -name '.hpp' -o -name '.cpp' -o -name '*.impl.hpp' |
xargs grep -l ESPRESSO_SHARED_MEMORY_PARALLELISM |
xargs -I{} unifdef -DESPRESSO_SHARED_MEMORY_PARALLELISM -t {} -o /dev/null

[jngrad]

Blocked by espressomd/docker#224

Now that all CI jobs use shared memory, FPE tests can run on the parallel code, and divisions by zero are detected.

[jngrad]

A floating-point exception is triggered in cufftPlan3d. Using ./pypresso --gdb ../testsuite/python/coulomb_tuning.py CoulombCloudWallTune.test_p3m_gpu with a hardcoded mesh=156 reproduces the issue on lama. I'm not exactly sure why CUDA is affected by FPE traps, this is supposed to be a CPU feature.

ScaFaCoS dipoles are also affected, usually in the tuning function. GDB has trouble extracting the source code from my ScaFaCoS library built in Debug mode...

The Verlet test triggers a FPE because we purposefully place 1000 particles at the box origin to check the Verlet list backup algorithm when the neighbor list is too small. This test clearly needs to be skipped when FPE is enabled.

[jngrad]

Blocked by the bugfix PR #5286

[jngrad]

Blocked by PR #5288 due to undersampling in the GCMC tutorial

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[jngrad]

Merge conflict with #5287, I'm working on it.

[jngrad]

Blocked by #5292