Bugfix in computing absolute value with cupy datatype

[brownbaerchen]

Seems I was too eagerly deleting stuff in #469 ... sorry about that! It's a shame we can't get the GPU tests with multiple GPUs to work.

[pancetta]

What's the matter with multi-GPU tests?

[brownbaerchen]

What's the matter with multi-GPU tests?

I don't remember exactly, maybe @jakob-fritz does. Maybe it was related to job submission?

[jakob-fritz]

I don't recall, we tried multi-GPU at all. So I don't know if (and what) issues arise with multi-GPU tests. I know, that we tried the single-GPU test on HPC, but have no insight in any multi-GPU attempts.

[pancetta]

Could you @brownbaerchen set up a multi-GPU test, then? Not here, but in another PR?

[brownbaerchen]

Could you @brownbaerchen set up a multi-GPU test, then? Not here, but in another PR?

We have such a test here, but we skip it because we don't know how to run it.

[pancetta]

Not sure I understand the problem..? Why is it difficult to run it if you are on JUWELS?

[brownbaerchen]

Currently, we use this script to launch pytest on JUWELS. Usually, we do MPI based tests by executing shell commands from within python scripts. But I am not sure that we can launch a larger job from within the job. I believe we tried posting srun commands, but I am not totally sure. Also, the first job that runs pytest will collect the coverage report, how does that work when the code is run by a different job on a different node? I am sure it's possible but we decided to postpone it. Maybe we can try again now that we are happy with the rest of the GitLab integration.

[pancetta]

Well, we do have multiple pytest calls in the pipeline anyway, so I guess this should be possible within a single step? @jakob-fritz ideas are welcome.