[WIP]: Cell Count Example

[Yurlungur]

PR Summary

I recently wanted to scope out a simulation cost, and thought "why not let Parthenon do the work for me?" That led me to developing this cute little example, which I figured I'd push upstream. The idea here is to have Parthenon generate a mesh tree/hierarchy based on some estimated needed resolution (say 100m resolution for a neutron star) in some region (say a sphere of some radius) without actually allocating any variables. Then have Parthenon report how big the mesh is.

The way I implement this is I create a package with no variables, but a custom mesh refinement check that checks to see if the desired resolution has been reached in the desired region. (Just for simplicity, this is hardcoded to a sphere. I'm not trying to create a full-featured capability here. If you want to spec something out other than a single isolated star, I think you could use this example as a jumping off point pretty easily.)

The example has no driver. I just init packages and mesh, run a loop through the mesh to count total number of cells heuristically, and print the results. Results can also be saved to a text file.

To visualize the mesh structure, I also wrote a little script that can read a phdf file and plot the meshblocks as rectangles, similar to the movie2d script. I initialize outputs in my little example (no driver so it must be done manually) and spit out an hdf5 file if requested. I found that by far the slowest part the calculation was outputting the mesh---in particular, the coordinates grids. So I added a new hdf5 output type, which I call gridh5. This is the standard phdf format, but instead of outputting full cell and face coordinates, it just outputs the minimum and maximum face coordinate values for each meshblocks. To do so I had to make a few one-line changes (and some indentation) to the parthenon_hdf5 output.

Outputs visualized with my little script, which I called plot_mesh.py look like this:

Those boxes are meshblocks, not cells. This particular grid had 32^3 meshblocks. This is an X3=0 slice. Grids with cell counts of 100s of millions of cells can be generated and visualized in less than a minute.

I hooked this machinery into regression tests, but because I didn't want to deal with a new gold file release, I didn't hook the gridh5 into the regression tests, just a cell count.

PR Checklist

• Code passes cpplint
• New features are documented.
• Adds a test for any bugs fixed. Adds tests for new features.
• Code is formatted
• Changes are summarized in CHANGELOG.md
• CI has been triggered on Darwin for performance regression tests.
• (@lanl.gov employees) Update copyright on changed files

[Yurlungur]

Josh Brown has moved on so his email should probably not be here. That said, should we have an official Parthenon email?

[Yurlungur]

@pgrete just a ping on this. I think there's a higher-level discussion to be had about this capability.