Post

MHD Debugging & Compute Sanitizer

Posted
By Robert Caddy 1 min read

MHD Debugging

It looks like the weird edge effect issues I was seeing might just be oscillations due to the higher order Van Leer method as it’s present in hydro builds as well.

It turns out that the issues I was having with the Sod shock tube under MHD were actually fixed in PR #238 when I fixed the missing boundary conditions in the parameters file, I just forgot to update the data to do a proper comparison. This has been fixed as of PR #249.

I’ve been working now on debugging the Ryu & Jones 1a shock tube (details here as it has a visually simple structure with no sharp spikes. It’s still throwing a non-zero divergence errors after the first time step. I think I’ve narrowed this issue (or at least one issue) down to something weird going on at the X boundary conditions, though I’m not sure yet what exactly is causing the issue.

NVIDIA Compute Sanitizer

I raised Issue #247 on how we’re using the CUDA compute sanitizer and if we should add it to the automated tests. I spent some time experimenting with it and believe that we can eliminate the uninitialized memory warnings by clever changes to our thread guards. That might also let us reduce GPU memory usage if we do some clever indexing.

Other

MHD
Debugging MHD HLLD Boundary Conditions Compute Sanitizer
This post is licensed under CC BY 4.0 by the author.