2020.10.21 Meeting Notes
- Individual/group updates
- Discuss large-scale testing for AMR at smaller mesh block sizes -- @pgrete/@gshipman (?)
- Public headers vs. private headers
- Review non-WIP PRs
Andrew has been working on getting project space set up for Darwin. Hoping to have a PR in that has the machine config for that today.
Andrew has started on a prototype for Dense of Block. I've documented some of the potential interactions for dense on block in this GitHub issue: https://github.com/lanl/parthenon/issues/322
Josh's CI change for LANL is basically ready to go - just wants @pgrete to take a look.
Carola will be looking into doing performance analysis on parthenon with tau. @pgrete requests documentation on using tau with parthenon. https://github.com/lanl/parthenon/wiki/Profiling
Sriram has been trying to debug an issue with restart causing deterministic, low precision errors seemingly related to a boundary condition.
Josh has been working on multi-block containers. https://github.com/lanl/parthenon/pull/328
Jonah has been working on doing GPU work on internal Parthenon code.
Ben got particles merged in. Looking into writing a regression test on particles.
Phil's looking at mesh block container caching from advection point of view.
Phil is doing performance analysis.
Jim's been working on a toy MPI code.
Some discussion of bad nvcc optimization.
Demonstrated that speed-up is possible with GPUs, but the big impediment is load balancing. PackInOne has only been optimized for static grids - need to update it for prolongation and restriction.
With 32^3 and increasing vector sizes, demonstrate meaningful speedup over CPUs.
Let's defer to a separate discussion.
- Reduce kernel launch overhead
- variable packing (Jonah)
- container-block pack (Joshua D.)
- Reduce infrastructure overhead
- AMR (?)
- Load Balancing (?)
- Reduce memory overhead
- Dense on Block (Andrew)