feat(dx12, hlsl-out) Support for SHADER_EARLY_DEPTH_TEST on DX12 (only non-MSAA framebuffers)

[DouglasDwyer]

Connections

#3961 - Support for conservative depth
#10625 - feat(metal, msl-out) Support for SHADER_EARLY_DEPTH_TEST on Metal
gpuweb/gpuweb#5349 - [WGSL fragment depth] (less, greater, any) portability

Description

Right now, the early_depth_test intrinsic is supported on Vulkan, but not DX12 or Metal. WGSL code like this fails to compile:

@fragment
@early_depth_test(less_equal)
public fn foo(input: VertexOutput) -> FragmentOutput {
    ...
}

DX12 is capable of supporting conservative depth. In HLSL, instead of writing to SV_Depth, you can write to SV_DepthGreaterEqual or SV_DepthLessEqual. There is no equivalent for @early_depth_test(unchanged).

This PR makes Naga emit SV_DepthGreaterEqual/SV_DepthLessEqual in the appropriate cases. For @early_depth_test(unchanged), Naga makes sure that its output never actually modifies the real SV_Depth - instead, all depth writes are redirected to a private variable.

This PR does not support MSAA render pipelines yet

When conservative depth testing is enabled, DXC requires that the SV_Position fragment input be interpolated as either:

• Centroid
• Per-sample

Notably, pixel center is not allowed as an option. This is troublesome because the WebGPU spec always defines @builtin(position) to be at the pixel center, at least according to my reading. So choosing either of these options, for an MSAA framebuffer, would be incorrect.

As such, this PR will only emit conservative depth attributes when compiling for a 1-sample framebuffer. In other cases, it falls back to SV_Depth. But I'm open to reconsidering this approach if centroid/per-sample interpolation turn out to be allowed within the spec.

Testing

I have tested this in my game engine and it worked - there was a measurable improvement in frame times with @early_depth_test(less_equal) in one of my shaders.

Additionally, the following unit tests are added:

• Added snapshot tests (naga/tests/in/wgsl/early-depth-test-*) that cover every @early_depth_test setting for HLSL:
  • force, greater_equal, less_equal, and unchanged on a bare frag_depth result.
  • A depth output in a struct shared by several entry points, both when they agree and when they don't.
  • unchanged with a color-plus-depth struct, a depth-only struct, and an unrestricted entry point sharing the struct.

Squash or Rebase?

Squash

LLM Use?

Yes - I used Claude to implement and test these changes.

Checklist

• I self-reviewed and fully understand this PR.
• CHANGELOG.md entries for the user-facing effects of this change are present.
• The PR is minimal, and doesn't make sense to land as multiple PRs.
• Commits are logically scoped and individually reviewable.
• The PR description has enough context to understand the motivation and solution implemented.
• (If applicable) WebGPU implementations built with wgpu may be affected behaviorally.
• (If applicable) Validation and feature gates are in place to confine behavioral changes.
• (If applicable) Tests demonstrate the validation and altered logic works.