Skip to content

2022.11.09 Meeting Notes

Jump to bottom
Philipp Grete edited this page Nov 10, 2022 · 3 revisions

Parthenon paper

  • accepted!
  • need short bio from everyone before it can go into production

Mailing list

  • reminded people to sign up
  • there should be a "join" button now so that everyone can join without invitation

Individual/group updates

AJ

  • background, use Parthenon as motivating example for load balancing test
  • looking into monitoring barriers in real time (even at the network level)
  • should be implemented as a separate "service" that eventually may feed back/provide info to applications like Parthenon for load balancing

BP

  • several small PRs (output numbering, general output logic, removing signal handling)
  • discussion on automated extended CI on PRs (see below)

JD

  • working on making all variables Variables, e.g., removing .fluxes and make those Variables with Face property
  • PR almost ready for review, need to fix some final tests
  • Added simple performance benchmark (based on Burgers eqn + number of scalar cons. laws) to be used in new performance metric for future machine procurement
    • use outer/inner pattern (matching patterns used in Riot)

LR

  • Sparse var ready for review (should get higher prio) containing couple of bug fixes
  • Looking into requirements for downstream dev of FLD, e.g., flux correction during BiCGSTab
  • Started to think about more flexible "userspace prolong", e.g., for vertex data or accessing other fields. Posted example in PR discuss.

JS

  • need flux corr for E-fields at the same level, also need to average not just taking one or the other
  • currently working on _a_mr in AthenaK

FG

  • Renaming of coord system done (moved to templated). Seems like all downstream users are happy.
  • Reviewing (and testing yt) new output

PM

  • got some eye candy with 10 levels of refinement for accretion flow
  • prolong/restrict on conserved vars seems to do fine in GRMHD (which is a tough test case)

PG

  • lots of testing/debugging (downstream) and reviewing various PRs
  • added multi-dim limiting to prolongation (will open PR soon)
  • confirmed bug on MI250x on a different (non-OLCF) node -- still no clue what's going on

Extended CI tests for PRs

  • A PR got merged that made the extended test fail again.
  • This happened in the past, too.
  • Will now require extended CI to pass before merging (PG will enable) but need to be triggered manually (to keep load on CI machine low)

Next meeting

Already next week (as there's thanksgiving in two weeks). Will talk about leftover agenda items next week.

Clone this wiki locally