Validation fails to reject out-of-range fragment shader output locations

[jimblandy]

There is no language in the WebGPU API or WGSL specs that would reject a fragment shader that has an output whose location exceeds maxColorAttachments.

For example, suppose that maxColorAttachments is 8 and we have the following fragment shader:

struct FragmentOut {
    @location(0) color: vec4<f32>,
    @location(8) slurve: vec4<f32>,
}

@fragment
fn blue() -> FragmentOut {
   return FragmentOut(vec4<f32>(0.0, 0.0, 1.0, 1.0), vec4<f32>());
}

Chrome refuses to construct a render pipeline:

Uncaptured WebGPU error: [EntryPoint "blue"] infringes limits:
 - Fragment output variable "<retval>.slurve" has a location (8) that exceeds the maximum (8).

 - While validating fragment stage ([ShaderModule (unlabeled)], entryPoint: "blue").
 - While validating fragment state.
 - While calling [Device].CreateRenderPipeline([RenderPipelineDescriptor]).

But as far as I can see, there is no language in the WebGPU spec that supports Chrome in producing this error. Although every non-null member in the GPUFragmentState::targets array is checked against the fragment shader's outputs, there is no requirement to inspect fragment shader outputs that do not have targets.

I'm also a little surprised that WebGPU allows fragment shaders to write to targets that aren't specified; is this to allow reuse of fragment shaders with different subsets of the possible targets?

[kainino0x]

I think you're right this is missing from the spec. We probably thought this validation was covered by descriptor.targets.size must be ≤ device.[[limits]].maxColorAttachments but in the end this wasn't true because we allowed shaders to write to targets that don't exist in the pipeline.

I'm also a little surprised that WebGPU allows fragment shaders to write to targets that aren't specified; is this to allow reuse of fragment shaders with different subsets of the possible targets?

And yes, I think that's why. #1250 is related but I couldn't find confirmation there. (#1900 is for the reverse.)

[Kangz]

GPU Web WG 2025-10-08 Atlantic-time

• CW: never validate that what WGSL exposes is less than the limit.
• JB: Kai confirms it's just an omission.
• CW: will just add the validation to the spec.
• MW: any concerns with eventually trying to precompile shaders? Or ship intermediate representations of shaders? Value might not be known at that stage.
• CW: no plan for precompiled shaders at all. Not sure we'll get there ever. (Validation is always required for incoming shaders.)
• MW: OK, thanks.