Teach CodeGenPrep to look past bitcast when it's duplicating return instruction
into predecessor blocks to enable tail call optimization.

rdar://11958338


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160894 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Transforms/Scalar/CodeGenPrepare.cpp b/lib/Transforms/Scalar/CodeGenPrepare.cpp
index f425cd0..4b4a8c5 100644
--- a/lib/Transforms/Scalar/CodeGenPrepare.cpp
+++ b/lib/Transforms/Scalar/CodeGenPrepare.cpp
@@ -645,10 +645,18 @@
   if (!TLI)
     return false;
 
+  PHINode *PN = 0;
+  BitCastInst *BCI = 0;
   Value *V = RI->getReturnValue();
-  PHINode *PN = V ? dyn_cast<PHINode>(V) : NULL;
-  if (V && !PN)
-    return false;
+  if (V) {
+    BCI = dyn_cast<BitCastInst>(V);
+    if (BCI)
+      V = BCI->getOperand(0);
+
+    PN = dyn_cast<PHINode>(V);
+    if (!PN)
+      return false;
+  }
 
   BasicBlock *BB = RI->getParent();
   if (PN && PN->getParent() != BB)
@@ -666,6 +674,9 @@
   if (PN) {
     BasicBlock::iterator BI = BB->begin();
     do { ++BI; } while (isa<DbgInfoIntrinsic>(BI));
+    if (&*BI == BCI)
+      // Also skip over the bitcast.
+      ++BI;
     if (&*BI != RI)
       return false;
   } else {