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

[Matt Arsenault via cfe-commits]

================
@@ -537,7 +537,11 @@ AMDGPUTargetCodeGenInfo::getLLVMSyncScopeID(const LangOptions &LangOpts,
     break;
   }
 
-  if (Ordering != llvm::AtomicOrdering::SequentiallyConsistent) {
+  // OpenCL assumes by default that atomic scopes are per-address space for
+  // non-sequentially consistent operations.
+  if (Scope >= SyncScope::OpenCLWorkGroup &&
----------------
arsenm wrote:

The primary concern here would be performance loss

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