[clang] [llvm] [AMDGPU] Add an async barrier arrival workaround (PR #238539)

Yaxun Liu via cfe-commits cfe-commits at lists.llvm.org
Sat Nov 7 05:54:20 PST 2026


================
@@ -2033,6 +2034,7 @@ void GCNPassConfig::addPreEmitPass() {
   addPass(createSIMemoryLegalizerPass());
   if (getOptLevel() > CodeGenOptLevel::None)
     addPass(&SIPostRA16BitMovFoldingLegacyID);
+  addPass(createSIInsertAsyncBarrierArriveWorkaroundPass());
----------------
yxsamliu wrote:

Both pipelines now check the subarchitecture. I kept the function check for explicit feature disables and mixed-CPU legacy modules.

https://github.com/llvm/llvm-project/pull/238539


More information about the cfe-commits mailing list