-
Notifications
You must be signed in to change notification settings - Fork 37
2020.08.26 Meeting Notes
-
Individual/group updates
-
CMake machine specific (auto) configuration (Philipp)
-
Restart branch conundrum (Philipp)
-
Reducing CI time (Jonah)
-
User-modifiable params/variables? (Jonah)
-
Quick run-through of outstanding non-WIP pull request reviews
Andrew's been working on integration with our Eulerian code. Have a target calculate PI example working with adaptive refinement.
Andrew's giving a presentation on Parthenon to another partner. Will share slides when they've been approved for release.
Josh fixed the code coverage labels, added some checks for python modules used in tests, some additional work on CI
Carola has been fixing some build issues. Met with Phil and Jonah about doing some initial performance benchmarking on mesh block packing.
Sriram has been fixing errors in the restart capability - mpi and GPU are wrong, but test passes, which is strange. Additionally the LANL physics code fails restart due to (suspected) boundary condition issues.
Josh Dolence has been continuing to push on the generalized task list. The open Q is about how to launch a single task list over multiple mesh blocks. Added a "region" concept that encapsulates multiple "mesh blocks".
Jonah fixed the issues with the mesh modernization PR and got that merged. Has been integrating the mesh block packing with Josh's task list sharing change. Approach is promising.
Sriram is working on a fix for an issue with restart.
Phil has been reviewing a lot of small PRs. Put together a more realistic test for mesh block packing https://github.com/lanl/parthenon/pull/263#issuecomment-681064111
Basically - much better across-the-board performance, performance for static mesh decomposed into small mesh blocks vs. large mesh blocks is incredibly similar. Very low overhead for smaller mesh blocks.
Forrest has been mostly busy with other projects. Been trying to put together a test of Views of Views of Views.
Define two variables: a variable for specifying a "known" system (e.g. "Summit", "DarwinPower9", etc.) and a variable for specifying a "user configuration file" - path to a CMake script that gets included.
Sriram will delete restart branch that was weird.
We're going to move the expensive tests to a CI that runs either on push to develop or nightly. Push to develop is desirable since it tells us the PR that failed, without blocking the merge.
Jonah will open a PR demonstrating what he has in mind.