[clang] [llvm] [AMDGPU] Implement AsyncMark Stages (PR #220442)

[Sameer Sahasrabuddhe via cfe-commits]

================
@@ -5251,6 +5264,28 @@ static Value *upgradeAMDGCNIntrinsicCall(StringRef Name, CallBase *CI,
     return NewCall;
   };
 
+  // Legacy asyncmark intrinsics missed the stage mask operand. Append an empty
----------------
ssahasra wrote:

What happens to code generators that directly construct calls in memory, without going through the parser?

https://github.com/llvm/llvm-project/pull/220442