Blocker: Tile generation isn't working properly at high tile counts (64x64)

[Jan]

Hi,
I can’t seem to generate tiles when I set it to my target configuration (64x64 tiles). I get these errors in the log:

I tested it with a flat constant a little while back to verify that everything works in this version (so we don’t need to go back to 3029 for yet another project :)), and I noticed WM froze but, at least the tiles were generated. I didn’t veryfy that they were correct (well, they were flat).

Now I’m doing a more thorough look at it, with a terrain, and I’m gettin the above errors. The tiles are generated, but they are garbage. I can’t really show that here though.

Thankfully, this can be tested with a simple setup using a perlin noise. When generating the perlin directly to 64x64 tiles, it doesn’t appear to be completely garbage, but some tiles were blank. But when convertng 8x8 tiles to 64x64, the result is garbage:

Repro:

1. Create a blank scene with a main extent 16384m. Set it to generate 8x8 tiles with the size of 4096 (detail is 0.5m).
2. Hook up a perlin to a height output.
3. Generate.

It generates fine. As expected.

Now:

1. Add a tile input and load the tiles you just output.
2. Hook it up to another output node.
3. Create and activate another extent. Same size, but with a 64x64 tile output. Each tile should be 512m (again, detail is 0,5m)
4. Generate.

The expected result is that WM should just spit out the tiles as usual.

But instead, the process stalls almost immediately. The program appears to freeze and eventually error messages appear in the log window.

Look at the generated tiles. They are garbage.

Can you have a look at this asap if possible? I wish I would have tested this more thoroughly earlier (my test scenes were only 8km and I didn’t think larger worlds would be an issue). I would really hate having to go back to 3029 again…

Cheers,
-Jan

ps. I can send you my very simple test-scene if you like. In case there is something else…

EDIT:
After doing some more testing, it seems like the issue might be with the tile input, which doesn’t seem to work with the tile output. In any case, the repro steps should still be valid.

It is also worth mentioning that the application freeze at high tile counts is still a problem, even if it isn’t directly related to the garbled result.

[A_Lingll]

Can you share a minimal reproducible project? I’m trying it too.

[Jan]

I don’t have 4049 installed anymore. I had some issues with scripts and program version so I needed a clean install of 3029 in order to progress with our project.

I think I saved out a project before though:

TileExportTest.tmd (114.9 KB)

It’s basically just a noise generated to tiles. And in the second step, using the other extent, those tiles are converted to more granular tiles. And in the third group, you can evaluate the result. But first you need to set the input/output paths of course

Cheers,
-Jan

[Stephen]

Hi Jan,

I’m guessing this is from the Tiled Input device based on the error message. I’ll take a look at it shortly.

[Stephen]

I’ve confirmed that this issue is caused by the Tiled File Input device. There is a sharing issue with the tile cache across the build threads. I’ll look at getting a fix out soon.

In the meantime, you can use the following workaround for the corrupted tile issue:

• Setting “Tiles at a time” to 1 in Program Settings->Compute tab.

[Jan]

Cheers Stephen, I’ll have a look at it when you have a fix.

There are some other issues with running scripts, while also having WM open, which made it difficult to have several versions installed. So I had to uninstall everything and go with a, for us, production-ready version (3029).

I hope to be able to review things again soon (possibly on my local machine before messing with my project setup) so we can take advantage of the new features.

Thankls again,
-Jan

[Stephen]

Just a note that I have a fix for the TFI issue, as well as the apparent UI freeze, that will be available fairly soon. It is being rolled up with several other fixes