[llvm] [AMDGPU][Attributor] Add `ThinOrFullLTOPhase` as an argument (PR #123994)

[Simon Pilgrim via llvm-commits]

================
@@ -86,6 +86,23 @@ enum class ThinOrFullLTOPhase {
   FullLTOPostLink
 };
 
+#ifndef NDEBUG
+static inline const char *to_string(ThinOrFullLTOPhase Phase) {
+  switch (Phase) {
+  case ThinOrFullLTOPhase::None:
+    return "None";
+  case ThinOrFullLTOPhase::ThinLTOPreLink:
+    return "ThinLTOPreLink";
+  case ThinOrFullLTOPhase::ThinLTOPostLink:
+    return "ThinLTOPostLink";
+  case ThinOrFullLTOPhase::FullLTOPreLink:
+    return "FullLTOPreLink";
+  case ThinOrFullLTOPhase::FullLTOPostLink:
+    return "FullLTOPostLink";
+  }
+}
----------------
RKSimon wrote:

MSVC is complaining :) Can you just move the implementation to Pass.cpp (and drop constexpr)?

https://github.com/llvm/llvm-project/pull/123994