Bevy 0.12 panics when loading a skinned GLTF mesh with shadow-enabled directional light in WASM #10509
Comments
idanarye
added
C-Bug
|
I think on webgl2 we had to add a workaround for the base instance index not being part of the instance index value and we had to get the base instance index into the shader via a push constant. I think the bug is something to do with that but can’t look into it right now. See instance_index.wgsl and its get_instance_index function. |
JMS55
added
A-Rendering
|
Just repro-ed with the |
|
It looks like Also discussion about possible workaround on discord: https://discord.com/channels/691052431525675048/743663924229963868/1174881166705315880 |
|
Following up: sounds like this is only broken on some platform configurations. Still broken, but so far only reproducible on linux + nvidia. |
and linux + integrated radeon graphics 👍 |
|
A summary of the problem:
@cwfitzgerald let us know that one way to work around this is to ensure that every shader uses the push constant. Sure enough, by adding this to #ifdef VERTEX_OUTPUT_INSTANCE_INDEX
out.instance_index = get_instance_index(vertex_no_morph.instance_index);
#endif
// Hack: adding this line ensures the push constant is always used, as get_instance_index accesses the push constant
out.position.x += f32(get_instance_index(0u)) * 0.00001;
return out;
}However, this breaks down the second a shader is used that doesn't use the push constant. Therefore it is not a "real" fix, although it does make simple examples work / any game that sticks with the built in materials + shaders. Therefore as a short term solution I think we should roll with this. One medium term-solution is to cut out the "push constant" entirely and use a uniform buffer. Given that this is what wgpu's WebGL2 push constant emulation is doing anyway, thats actually not too big of a deal. This will take a bit of engineering work to update our data flow. The long-term solution is to wait for the next wgpu release so we can use instance_index everywhere. |
This isn't actually true, we emulate them using gl uniforms which are closer to push constants than a proper buffer. If this change is made, once 0.19 comes out, it probably should be rolled back. |
|
Makes sense. Thanks! |
…10706) # Objective Kind of helps #10509 ## Solution Add a line to `prepass.wgsl` that ensure the `instance_index` push constant is always used on WebGL 2. This is not a full fix, as the _second_ a custom shader is used that doesn't use the push constant, the breakage will resurface. We have satisfying medium term and long term solutions. This is just a short term hack for 0.12.1 that will make more cases work. See #10509 for more details.
|
Moving to 0.13 as we have kind of mitigated this for 0.12.1 with #10706 |
…10706) # Objective Kind of helps #10509 ## Solution Add a line to `prepass.wgsl` that ensure the `instance_index` push constant is always used on WebGL 2. This is not a full fix, as the _second_ a custom shader is used that doesn't use the push constant, the breakage will resurface. We have satisfying medium term and long term solutions. This is just a short term hack for 0.12.1 that will make more cases work. See #10509 for more details.
|
Did wgpu 0.19 fix this? |
|
Seems like the entire mechanism for working around this was removed in #11280, and my code that previously used the workaround seems to work fine. |
Bevy version
Bevy v0.12.0
[Optional] Relevant system information
cargo --version)I get this both when building locally and
$ uname -srvmpio Linux 6.5.9-arch2-1 #1 SMP PREEMPT_DYNAMIC Thu, 26 Oct 2023 00:52:20 +0000 x86_64 unknown unknown GNU/LinuxIf your bug is rendering-related, copy the adapter info that appears when you run Bevy.
Formatted for readability:
What you did
Note that this is a minimal example. If I don't add an armature - the bug does not replicate. If I don't enable shadows on the directional lights - the bug does not replicate. I have no idea how these two are related...
Here is a repository that demonstrates the problem: https://github.com/idanarye/demonstrate-bevy-0.12-wasm-shadows-bug
What went wrong
The text was updated successfully, but these errors were encountered: