[clang] [Clang][AMDGPU] Stop defaulting to `one-as` for all atomic scopes (PR #120095)

[Matt Arsenault via cfe-commits]

================
@@ -537,7 +537,8 @@ AMDGPUTargetCodeGenInfo::getLLVMSyncScopeID(const LangOptions &LangOpts,
     break;
   }
 
-  if (Ordering != llvm::AtomicOrdering::SequentiallyConsistent) {
+  if (LangOpts.OpenCL &&
----------------
arsenm wrote:

The amdgcn fence builtin also looks like it already takes optional arguments to append MMRAs. If OpenCL is actually using this builtin anywhere (which I doubt it is), it can easily migrate to __builtin_amdgcn_fence 

https://github.com/llvm/llvm-project/pull/120095