Combined multi-patch and AMR output

[eschnett]

I have a simulation that splits the domain into multiple (non-overlapping) patches, each with its own (distorted but topologically Cartesian) coordinate system. Each such patch has its own AMR hierarchy. How do I best map this to a Silo file?

So far I have been outputting data for a single patch with its AMR hierarchy, and am creating an Mrgvar to represent the hierarchy, and then create a MultiMesh. This is working fine.

How do I extend this to multiple patches? Would I e.g. just use separate Mrgvars to represent this? Shouldn't I somehow tell Silo that there multiple patches that belong together? Is there an example for this?

A related, secondary question is: So far I am declaring both the wmrgtree_ijkExts and wmrgtree_xyzExts of the AMR hierarchy. When the coordinates are distorted in each patch, would I then omit the wmrgtree_xyzExts?

[markcmiller86]

Hmmm...good questions. I think this may be an opportunity for you to help define how best to support this kind of hierarchy in Silo's MRG tree conventions. So far, conventions have been developed only for regular gridded situations.

Might you have any example Silo files with some what you want already in them we could work from to develop support for this?

I assume your target is also being able to visualize them in something like VisIt.

One other thing to consider may be Conduit/Blueprint. I would be surprised if Conduit/Blueprint has direct support for this kind of hierarchy already. But, it may be closer than Silo presently is.

[eschnett]

Thanks for the quick answer!

I have some output files in Dropbox. They are 8 MByte combined.

This setup consists of 7 patches, a central cubic patch (that has AMR) and 6 surrounding patches turning the whole domain into a "cubed sphere".

I tried loading these data into VisIt. (Yes, I want to use VisIt to visualize them.) VisIt recognizes them, but presents them as 7 separate variables instead of combining them.

I'll have a closer look at Conduit/Blueprint. According to the documentation this should already be supported, but I need to verify the details.

[markcmiller86]

Thanks for sharing this dataset! 💪

Ok, so I just tried seeing how VisIt presently handles the data and it doesn't plot anything. Well, every plot I attempt gives the Plot yielded no data warning.

In your description, you say there are 7 patches. So, I am expecting to see a single multimesh object with 7 pieces and then an mrgTree that describes how those pieces interrelate. But, I am seeing 7 multimesh objects.

I tried looking at just the domain file (the one without the multi-block objects) and VisIt's mesh menu comes up gray'd out. I am not currently sure why. It should show all 7 quadmeshes that are defined in that file. That said, I do see some oddities there.

For example, when I look at File->File information and look at metadata for the mesh named box_m00_rl00_c00000000, the spatial extents seem to be zero in x and y dimensions,...

        Spatial extents are: ((0, 2.83053e-09), (0, 0), (1, 5.2))

Can you maybe share some more details about how you are trying to structure things?

[eschnett]

Thanks for looking into this. It seems I need to change the output to use a single multimesh. I I assume I would then add a single mrgTree to describe how these connect. That is, I assume I need to specify the AMR relations in the mrgTree, and just have the coordinate of the grid points of all meshes set up properly.

Not sure about the spatial extents. The 2e-9 also looks wrong.

Should I even define the spatial extent? How is it defined for a distorted mesh – I assume it's just a box that encloses the whole mesh?

[eschnett]

I looked at Conduit. It can describe my desired setup without problems. I'm not sure whether VisIt would be able to read it, though.

[JustinPrivitera]

VisIt may be able to read it.

[JustinPrivitera]

Let me correct my statement. VisIt likely cannot read this kind of data from the Blueprint reader at this time, but we could look to add support in the near future.

[eschnett]

That would be amazing. It doesn't matter whether VisIt can read the respective Silo data, or Blueprint-formatted HDF5 data – we're flexible there. We're currently testing respective simulations, and we're looking for a visualization tool that can beat "use Python to slurp in 3d data patch-by-patch, interpolate, and then use matplotlib" speed-wise. The bar isn't very high.

[JustinPrivitera]

FYI: visit-dav/visit#20265

[eschnett]

I updated my writer according to your suggestions. New output is here. VisIt now displays something, but when I create a slice, I'm seeing a square instead of a disk.

[markcmiller86]

I will try to take a look later today. Thanks for patience.

[eschnett]

I've been playing around with small changes. I think I've figured it out. The current problem is that VisIt displays things, but starts out zoomed into the centre of the domain instead of scaling outwards so that the whole domain is shown. I think the problem is that the coordinate extents of the individual patches are wrong.

[eschnett]

Thanks for your help. I think I figured everything out now.

[markcmiller86]

Ok, great work 💪. Sorry I was delayed in getting back to you but glad you got it all working.