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

[Justin Bogner via llvm-commits]

================
@@ -249,16 +255,38 @@ static FunctionType *getDXILOpFunctionType(const OpCodeProperty *Prop,
       ArgTys[0], ArrayRef<Type *>(&ArgTys[1], ArgTys.size() - 1), false);
 }
 
+static uint16_t getValidOverloadMask(const OpCodeProperty *Prop,
+                                     uint32_t SMVer) {
+  uint16_t ValidTyMask = 0;
+  // std::vector Prop->OverloadProp is in ascending order of SM Version
+  // Overloads of highest SM version that is not greater than SMVer
+  // are the ones that are valid for SMVer.
+  for (auto OL : Prop->OverloadProp) {
+    if (OL.ShaderModelVer <= SMVer) {
+      ValidTyMask = OL.ValidTys;
+    } else {
+      break;
+    }
+  }
+  return ValidTyMask;
+}
+
 namespace llvm {
 namespace dxil {
 
-CallInst *DXILOpBuilder::createDXILOpCall(dxil::OpCode OpCode, Type *ReturnTy,
-                                          Type *OverloadTy,
+CallInst *DXILOpBuilder::createDXILOpCall(dxil::OpCode OpCode, uint32_t SMVer,
+                                          Type *ReturnTy, Type *OverloadTy,
                                           SmallVector<Value *> Args) {
   const OpCodeProperty *Prop = getOpCodeProperty(OpCode);
+  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);
+  }
----------------
bogner wrote:

You have this check after every call to `getValidOverloadMask` - should it be inside that function instead? Further, is this actually something we can hit because of user input, or is this a programmer error where we've made a mistake in `DXIL.td`? For the latter, this should be an assert rather than a fatal error.

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