[llvm] [DXIL] Model DXIL Class and Shader Model association of DXIL Ops in DXIL.td (PR #87803)

[S. Bharadwaj Yadavalli via llvm-commits]

================
@@ -276,14 +304,22 @@ CallInst *DXILOpBuilder::createDXILOpCall(dxil::OpCode OpCode, Type *ReturnTy,
   return B.CreateCall(DXILFn, Args);
 }
 
-Type *DXILOpBuilder::getOverloadTy(dxil::OpCode OpCode, FunctionType *FT) {
+Type *DXILOpBuilder::getOverloadTy(dxil::OpCode OpCode, uint32_t SMVer,
+                                   FunctionType *FT) {
 
   const OpCodeProperty *Prop = getOpCodeProperty(OpCode);
   // If DXIL Op has no overload parameter, just return the
   // precise return type specified.
   if (Prop->OverloadParamIndex < 0) {
     auto &Ctx = FT->getContext();
-    switch (Prop->OverloadTys) {
+    uint16_t ValidTyMask = getValidOverloadMask(Prop, SMVer);
+    if (ValidTyMask == 0) {
+      report_fatal_error(StringRef(std::to_string(SMVer).append(
+                             ": Unhandled Shader Model Version")),
+                         /*gen_crash_diag*/ false);
+    }
+
+    switch (ValidTyMask) {
----------------
bharadwajy wrote:

> Is this right? You have something that's called a mask but you're checking if it exactly matches specific bits?

This is in the body of the if-condition (`Prop->OverloadParamIndex < 0`) and thus the opcode has no valid overload types. Overload mask encodes the valid precise (single) type for such opcodes with no overloads. So, switching on the `ValidTyMask` with precise value is correct.

https://github.com/llvm/llvm-project/pull/87803