[clang] [VK][HLSL][SPIRV] Implement vk::builtin("PointSize") (PR #231275)

Diego Novillo via cfe-commits cfe-commits at lists.llvm.org
Mon Nov 9 08:57:55 PST 2026


================
@@ -1028,6 +1034,52 @@ static bool isVkPipelineBuiltin(const ASTContext &AstContext, FunctionDecl *FD,
   }
 }
 
+static void validateVkBuiltinUsage(Sema &S, FunctionDecl *FD,
+                                   const DeclaratorDecl *D,
+                                   const HLSLVkBuiltInAttr *Attr,
+                                   IOType CurrentIOType) {
+  const ASTContext &AstContext = S.getASTContext();
+  if (AstContext.getTargetInfo().getTriple().getOS() != llvm::Triple::Vulkan)
+    return;
+
+  StringRef BuiltIn = Attr->getBuiltIn();
+
+  QualType T =
+      (D == FD ? FD->getReturnType() : D->getType()).getNonReferenceType();
+  const auto *ShaderAttr = FD->getAttr<HLSLShaderAttr>();
+  assert(ShaderAttr && "Entry point has no shader attribute");
+  llvm::Triple::EnvironmentType Stage = ShaderAttr->getType();
+  bool IsIn = any(CurrentIOType & IOType::In);
+  bool IsOut = any(CurrentIOType & IOType::Out);
+
+  if (BuiltIn == "PointSize") {
+    if (!T->isFloatingType())
----------------
dnovillo wrote:

`isFloatingType()` also accepts `double`, and `half` (with `-enable-16bit-types`), but Vulkan requires `PointSize` to be a 32-bit float (VUID-PointSize-PointSize-04317). This should only accept 32-bit `float`.

https://github.com/llvm/llvm-project/pull/231275


More information about the cfe-commits mailing list