Volume plot regression with ParallelPerRank resampling

[biagas]

Describe the bug

Recent volume plot regressions after PR #19642.

Parallel results of the volume_33 test, which is testing ParallelPerRank resampling option on multi_ucd3d.silo.

https://visit-dav.github.io/dashboard/2024-07-12-22:00/pascal_trunk_parallel/volume_33.html

I believe it is showing the per-rank division.

I ran the test with the DUMP option turned on in avtVisItVTKRenderer.C and avtVisItVTKRenderFilter.C of the volume plot, to get intermediate images and image data.

This image shows a different view of the same test, along with the per-rank intermediate image data that was dumped.

I have tried to determine why the results after the PR are showing what appears to be lack of correct compositing between the two ranks. Best I can determine based on -dump of this particular test before the PR and after: ghost-zone communication isn't happening. Before PR (I have a checkout of develop that I haven't updated in awhile) , I see gdb.xxx.vtk files with CommunicateGhost in the name. After PR, no files with CommunicateGhost in the name get dumped.

I haven't been able to figure out how/why the ghosting is different between the pre-PR and post-PR.

Helpful additional information

• Did you get wrong results: (?)

To Reproduce

Run VisIt with a parallel engine with at least 2 procs. Create Volume plot of multi_ucd3d.silo, with Parallel rendering type and ParallelPerRank resampling option. Draw.

You will get different results depending on the number of processors on the engine.

Desktop

• OS and version: Linux
• VisIt Version: 3.4RC (3.4.2), develop

[biagas]

Will need to remove volume_X from skip list when this is fixed.

[cyrush]

ghost zones should help with the resampling operation before raytracing in either case. So if ghosts are missing, we should dig into why.