Skip to content

2021.07.08 Meeting Notes

Philipp Grete edited this page Jul 8, 2021 · 2 revisions

Agenda

  • Individual/group updates
  • Schedule communication patterns discussion
  • Next release
  • Paper
  • Review non-WIP PR

Individual/group updates

LANL/External CS

Jonas Lippuner

  • going back to original pack interface in sparse PR (majority vote)
  • currently tracking down a regression while reverting
  • looking at communication patterns to figure out signaling of sparse variable communication
  • will add new regression test with "sparse advection"

Tanner Broaddus

  • going through Athena++ method paper to get more familiar with block structure AMR as implemented in Parthenon

LANL Physics

Josh Dolence

  • working on implementing support for iterative solvers in the task infrastructure
  • there existed some while workarounds/hacks but they didn't allow for interleaved tasks
  • will have new AddIterativeTask call with existing interfaces untouched
  • will add WIP PR soon
  • may potentially add a AddMeshTask (with support for reduction), which is also sth already implemented as a workaround in AthenaPK

AthenaPK

Philipp Grete

  • reviewed sparse PR
  • almost done implementing parthenon-hydro miniapp (right now ~1.7K LOC - trying to get it down to <1K)
  • still working on AMR performance regression tests

Agenda

Schedule communication patterns discussion

  • target date the week after next week
  • PG will also write the text for the Parthenon paper by then
  • 11am EST Mon 19 Jul -- JL will send webex

Next release

  • merge sparse in, then make next release
  • afterwards remove old comm interfaces and rename StateDescriptor to Package

Paper

  • goal to have some "very basic" text in all the method sections with the goal to get a better sense of how the pieces fit together and may still be missing/not well connected.

AMRFillDerived

  • Phoebus uses complex geometries that are expensive to calc on the fly
  • Caching of geometry helped a lot in performance
  • However, geometry only changes after mesh refinement
  • Added hack to "FillDerived" on first call after refinement
  • Will try to reintroduce InitUserMeshBlockData (which is still present in Parthenon but the function call is commented out)
Clone this wiki locally