VisIt still fails to save large images when shadows are enabled in the advanced rendering settings

[JustinPrivitera]

3.4.2

Even after this work #20043

Running on Tuolumne

4000x4000 does not work
4096x4096 does work
5120x5120 does not work
6144x6144 does not work
7168x7168 does not work
8000x8000 does not work
8192x8192 does not work

data + session files + script for running is here: /p/lustre5/justin/llnl_logo_viz

I think lustre5 is only accessible from tuolumne

Edit: see this comment for the actual issue that needs to be solved: #20146 (comment)

[JustinPrivitera]

save settings used:

I see the same behavior with the GUI and with the CLI

[markcmiller86]

Can you elaborate on what you mean by does not work. Does it fail entirely to save any file? Does it save a file of correct size/shape but contains garbage or maybe not garbage but also not correct image? Does it segv?

[JustinPrivitera]

It creates an all black image.

Same issue for jpeg and png.

[JustinPrivitera]

like cyrus was seeing here: #19430

[markcmiller86]

Ok, FWIW, I am getting all black viewer windows on occasion when I run on poodle. When I drag to resize, the plots and window briefly appear normal. Sometimes I get real luck and the plots and window will stay normal after I finish the drag/resize event. Other times, I wind up having to restart VisIt to make it go away. This literally just happened a few mins ago when looking at @biagas VTK-GEOS data.

I wonder if these things could be related.

[JustinPrivitera]

A user has given me another session file for which 8k images are able to be saved. There is something very fishy here. 🐟

[markcmiller86]

A user has given me another session file for which 8k images are able to be saved. There is something very fishy here. 🐟

Do you think it could be those fish that can drive?

[markcmiller86]

@JustinPrivitera out of curiosity, I know its not helpful, but did you try same running with -nowin? It could help us isolate cause. And, is the problem only on Toulumne? I just saved 8000 x 8000 on poodle for example (in -nowin mode).

[JustinPrivitera]

Hi @markcmiller86; I didn't try with -nowin just yet. I am currently narrowing down the cause of this by using the fresh session file from the user and iteratively making changes with python that lead towards the state of my session file, and saving out images each step of the way. I've narrowed it down to some change in the rendering attributes, but that was as far as I got today. I'll pick it back up again tomorrow.

This issue probably exists elsewhere but I am running into it with data on p/lustre5, which I believe is only accessible from tuolumne. It has something to do with the rendering settings. My guess is anti-aliasing or specular light but I need to see where things start to not work.

[JustinPrivitera]

The issue is the shadowing advanced rendering setting. Turning it on causes things to go haywire. With it off I can save images at large resolutions. Shadowing also causes a .depth.Z file to be written every time a .png is written. We need to investigate the cause and fix it.

[JustinPrivitera]

I am still seeing the .depth.Z files even without shadowing turned on. I am not sure what causes them to be written or what purpose they serve.

[brugger1]

The .depth.Z file is written out if there is Z buffer data available at the time the image is saved. It looks like this is independent of the setting in the GUI. It was added in VisIt 3.0.0. One case where this is true is when using scalable rendering and combining opaque and transparent plots.

[JustinPrivitera]

Thanks @brugger1. What can it be used for?

[brugger1]

If you read the png file back into VisIt it looks for the depth.Z file and you can display the Z value if it finds it. I'm not sure what else it can be used for.

The current behavior is probably not ideal. It should only output it if the user asks for it. This would probably require that the Z value be forced to be kept around so that it can be saved out since obviously its not available most of the time or else we'd always get the file.

[markcmiller86]

Can't the depth.Z file be used as part of a background image annotation? I mean, I thought I recall @BradWhitlock doing that in some of his work (perhaps a long while back now). Where he had a scene rendered with VisIt and he wanted that scene to be used as a background image in a new visualization but then have it integrate with the objects in that new visualization. I dunno...I thought I recall something like that but maybe I am just blowing smoke.

[brugger1]

That seems like a reasonable use case.