[llvm] AMDGPU: Add codegen for atomicrmw operations usub_cond and usub_sat (PR #141068)

[Matt Arsenault via llvm-commits]

================
@@ -16806,10 +16808,10 @@ static bool isV2BF16(Type *Ty) {
 }
 
 /// \return true if atomicrmw integer ops work for the type.
-static bool isAtomicRMWLegalIntTy(Type *Ty) {
+static bool isAtomicRMWLegalIntTy(Type *Ty, bool Allow64 = true) {
   if (auto *IT = dyn_cast<IntegerType>(Ty)) {
     unsigned BW = IT->getBitWidth();
-    return BW == 32 || BW == 64;
+    return BW == 32 || (BW == 64 && Allow64);
   }
----------------
arsenm wrote:

I would just leave this alone and handle the one special case separately 

https://github.com/llvm/llvm-project/pull/141068