Skip to content

2020.12.02 Meeting Notes

Jump to bottom
Andrew Gaspar edited this page Dec 7, 2020 · 1 revision

Agenda

Individual/group updates

LANL CS

Andrew is working on Sparse variables. Trying to get in a PR to refactor SparseVariable to be collapsed into CellVariable.

Josh Brown has been doing some fixes for CI. Currently working on code for publishing logs and performance information from behind LANL's firewall.

LANL Physics

Josh Dolence is making it possible to spawn new MeshBlockData objects are are a subset of variables. Generalized Data and MeshBlockData objects for summation.

Jonah is working on a change to have private variables.

Ben is still working on particles communication.

KAthena

Phil added tons of new performance regions.

Also working on pack-in-one features. Running in AMR, too. Total performance with small mesh blocks increased by a factor of 10 with AMR! Still losing to skylake.

Update on Continuous Benchmarking

What performance metrics we want:

  • Start with AMR advection test
  • Whole runtime is helpful
  • Could use individual profiling timers, too.
  • Phil thinks that the performance test execution probably needs to take place outside ctest

NOTE: We should consider using an existing performance dashboard rather than rolling our own.

For example, potentially use Ginkgo: https://ginkgo-project.github.io/gpe/

https://github.com/lanl/parthenon/issues/248#issuecomment-736028277

Changes to CellVariable and SparseVariable

General agreement that CellVariable + SparseVariable is worthy of refactor. Phil agreed that he ran into this issue.

AMR performance, see https://github.com/lanl/parthenon/pull/359#issuecomment-737369316

We need to do something about small allocations.

@AndrewGaspar will schedule a meeting for next Wednesday to discuss performance measurement.

Clone this wiki locally