[llvm] [CodeGen][AMDGPU] Support type-aware intrinsic feature checks (PR #205536)

[Matt Arsenault via llvm-commits]

================
@@ -1133,18 +1133,22 @@ class LLVM_ABI DiagnosticInfoUnsupportedTargetIntrinsic
     : public DiagnosticInfoWithLocationBase {
 private:
   unsigned IntrinsicID;
+  std::string IntrinsicName;
   StringRef RequiredFeatures;
 
 public:
-  DiagnosticInfoUnsupportedTargetIntrinsic(
-      const Function &Fn, unsigned IntrinsicID,
-      const DiagnosticLocation &Loc = DiagnosticLocation());
+  DiagnosticInfoUnsupportedTargetIntrinsic(const Function &Fn,
+                                           unsigned IntrinsicID,
+                                           StringRef IntrinsicName,
----------------
arsenm wrote:

Shouldn't need the intrinsic name, that's computable from the ID 

https://github.com/llvm/llvm-project/pull/205536