The compute_textures.py example segfaults on CI

[almarklein]

All examples passed fine until recently, but now the compute_textures.py example segfaults consistently. I saw it first in #787, so I figured it was a bug in the latest wgpu-native. But it also happens in main. I am trying to fix it in #791, but no luck yet. 😢

[almarklein]

I can reproduce locally on Linux using force_fallback_adapter=True (force lavapipe).

[almarklein]

I'm pretty sure there's something in the compute shader's wgsl that trips lavapipe. Could well be a bug in Lavapipe ...

[Vipitis]

I got the example to run in the browser yesterday which sorta is an alternative backend and it did run without making any changes to the functional bits of the example.

There have been quite a lot of flaky CI failures in upstream wgpu-core too so partially it could be blamed on GH? although if it consistently fails locally it might be something else.

[almarklein]

The compute shader in the example writes to two textures (one for luminance, one for chroma). If I disable either of these for-loops, the example runs. But if both are enabled it crashes. I tried limiting the indices to make sure it's not something going out of bounds, but that did not help.

And then there's the fact that it worked earlier, so this looks a lot like a regression in Lavapipe. I will sleep on it, and if I feel the same way tomorrow, I'll just disable the example for CI 🤷

The flaky CI failures are something different since they're not consistent. But also very annoying, and it looks like they suddenly appeared, as if (again) it's a regression in Lavapipe.