1 files changed, 15 insertions, 0 deletions
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index b3a611d37c9..483e9b171d5 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -11213,6 +11213,14 @@ SDValue PPCTargetLowering::expandVSXLoadForLE(SDNode *N,
   }
 
   MVT VecTy = N->getValueType(0).getSimpleVT();
+
+  // Do not expand to PPCISD::LXVD2X + PPCISD::XXSWAPD when the load is
+  // aligned and the type is a vector with elements up to 4 bytes
+  if (Subtarget.needsSwapsForVSXMemOps() && !(MMO->getAlignment()%16)
+      && VecTy.getScalarSizeInBits() <= 32 ) {
+    return SDValue();
+  }
+
   SDValue LoadOps[] = { Chain, Base };
   SDValue Load = DAG.getMemIntrinsicNode(PPCISD::LXVD2X, dl,
                                          DAG.getVTList(MVT::v2f64, MVT::Other),
@@ -11277,6 +11285,13 @@ SDValue PPCTargetLowering::expandVSXStoreForLE(SDNode *N,
   SDValue Src = N->getOperand(SrcOpnd);
   MVT VecTy = Src.getValueType().getSimpleVT();
 
+  // Do not expand to PPCISD::XXSWAPD and PPCISD::STXVD2X when the load is
+  // aligned and the type is a vector with elements up to 4 bytes
+  if (Subtarget.needsSwapsForVSXMemOps() && !(MMO->getAlignment()%16)
+      && VecTy.getScalarSizeInBits() <= 32 ) {
+    return SDValue();
+  }
+
   // All stores are done as v2f64 and possible bit cast.
   if (VecTy != MVT::v2f64) {
     Src = DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, Src);