MHD Linear Waves Tests

Linear Waves

I need to add MHD linear wave tests and initial conditions at some point so I decided to do it this week to help with debugging. I already added support for generalized linear wave initial conditions but the support in the system testing class was lacking. I added a new method for running the waves that computes the L1 error comparing the final and initial conditions. I also added linear wave tests for the fast, slow, Alfven, and contact waves.

AMD Software vs. Hardware Atomics

Compared hardware vs software performance atomics on AMD MI250X in Cholla. There’s no difference for us, probably since the kernel is bandwidth bound with relatively few atomics.

In the process of testing this I also modified our atomics to work with AMD and found an issue with the ROCm occupancy API, the issue is detailed in ROCm issue #1842

Alfven Wave Debugging

Now that I’ve got linear waves set up I started debugging the Alfven wave. I’ve confirmed that there are issues by the end of the first timestep update but I haven’t exactly figured out where yet. Possibly in the HLLD solver as the interface states look ok. Next week I plan on setting up pure hydro linear waves to make sure they pass when run under MHD.

Other

• Attended OLCF User Meeting