Extracting Time History Data

[dalarlla]

Apologies if this has been covered elsewhere but I couldn't find anything that covered quite what I was looking for.

I am writing a fluid solver based on AmrLevel and would like to provide functionality to extract full time histories at given spatial locations. Based on a previous solver users will in some cases have tens of thousands of output locations. Typically to avoid GPU transfers I would collect say 100 points at a time for each probe before transferring back to write to file. I also want to ensure that the data I collect at each position is from the finest level that covers it at each timestep.

I have seen the thread https://github.com/AMReX-Codes/amrex/discussions/1936#discussion-3310786 however the use of a parallel reduction seems to suggest that I would need to have every probe on every core when running over MPI.

Ideally I want to avoid replicating these across all cores so my first thought was to use particles to do this Is this something that has been implemented previously/does it seem like the best approach?

[atmyers]

Hi @dalarlla - using particles was going to be my suggestion. You could check out FieldProbe and FieldProbeParticleContainer in WarpX for a reference implementation: https://github.com/BLAST-WarpX/warpx/tree/7f0db07bd9a75839566e0a61bfc49db329c45832/Source/Diagnostics/ReducedDiags