2023.11.09 Meeting Notes

Jump to bottom

Philipp Grete edited this page Dec 14, 2023 · 3 revisions

Agenda

JM

Added hook for UserWorkBeforeLoop with one more callback incoming
Cleanup shared ptr mess between MeshBlockData and MeshData
Working on downstream 4th-order method
- after couple of iteration the current design looks quite satisfactory
- could eventually go upstream, might be part of a larger effort on how to handle these kind of changes

JD

added more automated profiling
added new tracing capability
- logs start and stop times for regions and are stored in a global object
- target is timer based load balancing
- WIP PR is open

LR

fixed MG MPI bug (now works and already merged)
open PR for boundary conditions by package (and problem generators)
- sparked more extended discussion on combining problem generators and boundary conditions
  - e.g., wrt to selecting pgen
  - custom BC (e.g., for subset of vars)
  - solution: make pgens packages?
added performance fixes for MG
- boundary conditions came up more severely
- traced back to not caching pack descriptors

PM

designing support for multiple instances of low storage integrators

PG

updated histogram PR, discussion finalized, should be ready for merge early next week
would like to go forward with current "analysis restart" design if there are no objections
two INCITE grants for 2024 with AthenaPK

Next developer/user(?) meeting

overall consensus is to plan one next summer in Hamburg
PG will send poll to determine the week
Should remain mostly dev focus though local (i.e., Germany/Europe) users should be invited, too (similar to the last meeting in the US)
will start collecting ideas (also from past syncs) for the meeting agenda

Doc sprint

desperately needed, though realistically first happening during the dev meeting
one key missing piece is an onboarding tutorial

Gordon Bell Award

we should prepare a submission highlighting unique features (at corresponding performance)
first brainstorming for ideas (involving sparse variables)
needs to be "science" case so not pure Parthenon capabilities
- e.g., incite type AGN feedback sims with many passive, sparse tracers
need to double check if export control codes an be used (and can be run on corresponding machines)

API forward compatibility (cont. discussion from last sync)

we don't guarantee a stable API just yet
will specifically highlight API breaking changes in the CHANGELOG
implement that change with the next release (PG will handle next week)

Next meeting 30 Nov