[clang] [llvm] [HLSL][DirectX] Move handling of resource element types into the frontend (PR #75674)

[Justin Bogner via cfe-commits]

================
@@ -216,15 +214,62 @@ void CGHLSLRuntime::addBufferResourceAnnotation(llvm::GlobalVariable *GV,
     assert(false && "Unsupported buffer type!");
     return;
   }
-
   assert(ResourceMD != nullptr &&
          "ResourceMD must have been set by the switch above.");
 
   llvm::hlsl::FrontendResource Res(
-      GV, TyName, RK, IsROV, Binding.Reg.value_or(UINT_MAX), Binding.Space);
+      GV, RK, ET, IsROV, Binding.Reg.value_or(UINT_MAX), Binding.Space);
   ResourceMD->addOperand(Res.getMetadata());
 }
 
+static llvm::hlsl::ElementType
+calculateElementType(const ASTContext &Context, const clang::Type *ResourceTy) {
+  using llvm::hlsl::ElementType;
+
+  // TODO: We may need to update this when we add things like ByteAddressBuffer
+  // that don't have a template parameter (or, indeed, an element type).
+  const auto *TST = ResourceTy->getAs<TemplateSpecializationType>();
+  assert(TST && "Resource types must be template specializations");
+  ArrayRef<TemplateArgument> Args = TST->template_arguments();
+  assert(!Args.empty() && "Resource has no element type");
+
+  // At this point we have a resource with an element type, so we can assume
+  // that it's valid or we would have diagnosed the error earlier.
+  QualType ElTy = Args[0].getAsType();
+
+  // We should either have a basic type or a vector of a basic type.
----------------
bogner wrote:

I can't really find good documentation on what element types are really supposed to be allowed for resources, and dxc doesn't do a great job of diagnosing these things. However, it doesn't look like matrix types are supported here, at least in practice: https://hlsl.godbolt.org/z/aaze4ceYs

https://github.com/llvm/llvm-project/pull/75674