[clang] [llvm] CodeGen: Replace TargetOptions::ThreadModel with module flag (PR #223157)

[Yonah Goldberg via cfe-commits]

================
@@ -457,8 +500,11 @@ bool AtomicExpandImpl::processAtomicInstr(Instruction *I) {
 bool AtomicExpandImpl::run(Function &F,
                            const ModuleLibcallLoweringInfo &LibcallResult,
                            const TargetMachine *TM) {
+  SingleThreaded = F.getParent()->getThreadModel() == ThreadModel::Single;
+
   const auto *Subtarget = TM->getSubtargetImpl(F);
-  if (!Subtarget->enableAtomicExpand())
+  // In a single-threaded environment atomics are lowered to non-atomic form
+  if (!SingleThreaded && !Subtarget->enableAtomicExpand())
----------------
YonahGoldberg wrote:

If we're making lowering atomic -> not atomic part of AtomicExpand it's a little weird to selectively enable it even when the pass is disabled. ARM doesn't ever disable atomic expand so I'm not sure if we need this.

https://github.com/llvm/llvm-project/pull/223157