[Abir Deol (apollo-2006)]

Fixes #9084.

Thanks again for the quick go-ahead on this one. DXIL replaces a pixel shader SV_CullPrimitive input with 0, but the SPIR-V backend mapped it to CullPrimitiveEXT, which needs MeshShadingEXT and isn't allowed in fragment shaders, so the output failed validation.

This mirrors DXIL and uses a null constant of the declared type instead (false for bool, 0 for uint). I also removed the old PSIn builtin path, since nothing reaches it anymore.

I added meshshading.ext.cullprimitive.ps.hlsl, which checks bool and uint without -fcgl so spirv-val runs on the output. It fails without the change, and the CodeGenSPIRV suite passes.

Jonathan Zakharov (@jzakharovnv) as promised, here it is :)

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

🟡 Changes recommended

The regression test does not verify that SV_Target is zero for both bool and uint runs, allowing incorrect true/1 behavior to pass undetected.

1 open issue

• Assert SV_Target is zero in both bool and uint runs · New

What changed in this PR

This PR fixes #9084 by making the SPIR-V backend replace pixel shader SV_CullPrimitive inputs with zero, matching DXIL and avoiding invalid fragment-shader builtins.

Changes:

• Emit a null constant of the declared input type and remove the obsolete pixel-shader builtin path.
• Add bool and uint regression cases with SPIR-V validation enabled.
• Document the behavior fix in the release notes.

File	Description
tools/​clang/​test/​CodeGenSPIRV/​meshshading.ext.cullprimitive.ps.hlsl	Adds bool and uint pixel-shader regression cases.
tools/​clang/​lib/​SPIRV/​DeclResultIdMapper.cpp	Replaces pixel-shader cull inputs with typed null constants.
docs/​ReleaseNotes.md	Documents the DXIL-compatible behavior.

_{Review effort: Balanced}

[Jonathan Zakharov (jzakharovnv)]

Thanks for the quick turnaround. Code looks good, the testing just needs a little more coverage, which I've left some ideas for.

[Abir Deol (apollo-2006)]

Thanks for these, pushed tests for all three:

• the -fcgl runs now check for the OpConstantNull directly (bool, uint, bool2 and bool[2])
• a struct shared by the MS and PS, with SV_CullPrimitive first and not first (locations match, and the 0 lands in the right member both times)

Both fail without the fix. Also, while I was in there I noticed DXIL crashes on a bool[2] PS input (happy to file that separately if it helps) :)

[Copilot]

🟡 Changes recommended

The early return incorrectly replaces explicit Vulkan builtin values with false, breaking existing builtin-override behavior when combined with SV_CullPrimitive.

2 open issues

• Preserve explicit Vulkan builtins for SV_CullPrimitive · New
• Assert SV_Target is zero in both bool and uint runs

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The early return bypasses explicit vk::builtin overrides, silently replacing valid helper-invocation input with false; declarations with VKBuiltInAttr need to retain the existing mapping, with regression coverage.

2 open issues

• Preserve explicit Vulkan builtins for SV_CullPrimitive
• Assert SV_Target is zero in both bool and uint runs

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

No active review issues remain for the SPIR-V SV_CullPrimitive input change.

No open issues 🎉

2 resolved since last review

• Preserve explicit Vulkan builtins for SV_CullPrimitive
• Assert SV_Target is zero in both bool and uint runs

_{Review effort: Balanced}

[Gregory Roth (pow2clk)]

/azp run

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[Copilot]

🟢 Approval recommended

No active review issues were identified for the change aligning SPIR-V pixel-shader SV_CullPrimitive handling with DXIL.

No open issues 🎉

_{Review effort: Balanced}