-
Notifications
You must be signed in to change notification settings - Fork 37
2023.11.09 Meeting Notes
Philipp Grete edited this page Dec 14, 2023
·
3 revisions
- Individual/group updates
- API forward compatibility (cont. discussion from last sync)
- Next developer/user(?) meeting
- Doc sprint
- Gordon Bell Award
- review non-WIP PRs
JM
- Added hook for
UserWorkBeforeLoop
with one more callback incoming - Cleanup shared ptr mess between
MeshBlockData
andMeshData
- 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?
- sparked more extended discussion on combining problem generators and boundary conditions
- 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
- 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
- desperately needed, though realistically first happening during the dev meeting
- one key missing piece is an onboarding tutorial
- 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)
- 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)