[llvm] LLVMContext: add getSyncScopeName() to lookup individual scope name (PR #109484)

[via llvm-commits]

================
@@ -16142,11 +16142,8 @@ static bool atomicIgnoresDenormalModeOrFPModeIsFTZ(const AtomicRMWInst *RMW) {
 
 static OptimizationRemark emitAtomicRMWLegalRemark(const AtomicRMWInst *RMW) {
   LLVMContext &Ctx = RMW->getContext();
-  SmallVector<StringRef> SSNs;
-  Ctx.getSyncScopeNames(SSNs);
-  StringRef MemScope = SSNs[RMW->getSyncScopeID()].empty()
-                           ? "system"
-                           : SSNs[RMW->getSyncScopeID()];
+  StringRef SS = Ctx.getSyncScopeName(RMW->getSyncScopeID());
----------------
gonzalobg wrote:

Have updated AMDGPU to map an out of bound `SyncScope::ID` to `""` (system) with `.value_or("")`, but NVPTX maps out of bounds `SyncScope::ID` to errors, so just returning an empty string doesn't work there.

https://github.com/llvm/llvm-project/pull/109484