Axel Huebl

Just for the record, a recommendation by @ahelm https://www.youtube.com/watch?v=Oc9khbXBes8

Probably, just pointing to officially used and known namings for such algorithms :) Did you use a reference for your namings (which)?

good idea, one could hide that from the user by handling it within `init()`. Nevertheless, providing an interface to call that internal state is sometimes useful: as an example, imagine...

I think that would be super awesome for hacking PIConGPU: init configurations, new solvers/methods/... or live visualisation with matplotlib/paraview

Thank you, @RTSandberg ! For reproducibility, can you please add the OS you used, versions of Python, pyAMReX, PyTorch, MPI flavor and version, and mpi4py version?

If we can reduce this problem to a pure `mpi4py` + `PyTorch` issue, then we could also report this upstream in `PyTorch`: https://github.com/pytorch/pytorch/issues

The issue seems to be more general, also showing sometimes later, after model load: https://github.com/ECP-WarpX/impactx/issues/773#issuecomment-2585043099 Could also be a mixing of OpenMP libraries (gomp and llvm omp)... Especially when downloading...

Thank you for the details @pnorbert ! Checking the code, it looks like it is auto-enabled when SQLLite is found, https://github.com/ornladios/ADIOS2/blob/v2.10.0/cmake/DetectOptions.cmake#L578-L583 which probably was the case on the system I...

Sounds good - yes active opt-in at runtime is what we need.

Fixed in v2.10.1+ via #4174