X86: remove temporary atomicrmw used during lowering.

We construct a temporary "atomicrmw xchg" instruction when lowering atomic
stores for widths that aren't supported natively. This isn't on the top-level
worklist though, so it won't be removed automatically and we have to do it
ourselves once that itself has been lowered.

Thanks Saleem for pointing this out!

llvm-svn: 212948

1 files changed, 5 insertions, 2 deletions

diff --git a/llvm/lib/Target/X86/X86AtomicExpandPass.cpp b/llvm/lib/Target/X86/X86AtomicExpandPass.cpp
index 1637b55b6d3..61eefbbf75b 100644
--- a/llvm/lib/Target/X86/X86AtomicExpandPass.cpp
+++ b/llvm/lib/Target/X86/X86AtomicExpandPass.cpp
@@ -277,8 +277,11 @@ bool X86AtomicExpandPass::expandAtomicStore(StoreInst *SI) {
                               SI->getValueOperand(), Order);
 
   // Now we have an appropriate swap instruction, lower it as usual.
-  if (shouldExpandAtomicRMW(AI))
-    return expandAtomicRMW(AI);
+  if (shouldExpandAtomicRMW(AI)) {
+    expandAtomicRMW(AI);
+    AI->eraseFromParent();
+    return true;
+  }
 
   return AI;
 }