Meeting Minutes – February 6, 2025
Present: Erik, Lucas, Roland, Liwei, Zach, Steve, Swapnil, Jay, Helvi, Allen
Comparing performance between Carpet and CarpetX.
Findings:
One refinement level test shows similar speeds.
With six levels, CarpetX is about 20% slower (with and without subcycling).
Removing additional synchronization calls for interprocess boundaries (ghost-zones) and keeping sync calls only for prolongation during the substep makes the speed comparable to Carpet.
Further optimizations are in progress.
Erik’s suggestion: Introduce a new grid point type to track prolongation boundaries and set a flag to check if they are synced.
Another potential issue: Using the PointDesc object (p.I) in CarpetX causes minor slowdowns compared to accessing indices in Carpet.
Erik’s input: The compiler should optimize this; Liwei will share an example .o file for Erik to analyze.
Z4c requires a test case.
Currently working on it and may seek Steve’s review later.
Testing hybrid Runge-Kutta integrators with BBH simulations.
Successfully set up a multipatch system with two puncture data and extracted multipole data.
May need assistance with Z4c.
Updating the robust stability test parfile and will open a PR. Currently experimenting with different boundary conditions.
Reviewing PRs for NewRadX in SpacetimeX and CarpetX repositories.
Zach’s question: Any tests beyond scalar wave for multipatch?
Lucas is working on a BBH test case and aims for completion before the May release.
Helvi’s question: Status of multipatch implementation in CanudaX?
Lucas is still working on it.
Liwei’s question: Can the multipatch code be used directly?
Lucas explained that evolution codes must be modified to compute derivatives and project them onto the multipatch system.
Suggests adding an option to enable/disable multipatch rather than maintaining two code versions.
Thursday, February 13, 2025 @ 10 AM CDT / 11 AM EST