diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 32e41c36802..85656d80914 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -7371,7 +7371,11 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
   if (V1IsUndef && V2IsUndef)
     return DAG.getUNDEF(VT);
 
-  assert(!V1IsUndef && "Op 1 of shuffle should not be undef");
+  // When we create a shuffle node we put the UNDEF node to second operand,
+  // but in some cases the first operand may be transformed to UNDEF.
+  // In this case we should just commute the node.
+  if (V1IsUndef)
+    return CommuteVectorShuffle(SVOp, DAG);
 
   // Vector shuffle lowering takes 3 steps:
   //
diff --git a/test/CodeGen/X86/avx-shuffle.ll b/test/CodeGen/X86/avx-shuffle.ll
index 0956361c7e3..02aa617c56c 100644
--- a/test/CodeGen/X86/avx-shuffle.ll
+++ b/test/CodeGen/X86/avx-shuffle.ll
@@ -297,3 +297,12 @@ entry:
 }
 declare <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double>, i8) nounwind readnone
 declare <4 x double> @llvm.x86.avx.vinsertf128.pd.256(<4 x double>, <2 x double>, i8) nounwind readnone
+
+; this test case just should not fail
+define void @test20() {
+  %a0 = insertelement <3 x double> <double 0.000000e+00, double 0.000000e+00, double undef>, double 0.000000e+00, i32 2
+  store <3 x double> %a0, <3 x double>* undef, align 1
+  %a1 = insertelement <3 x double> <double 0.000000e+00, double 0.000000e+00, double undef>, double undef, i32 2
+  store <3 x double> %a1, <3 x double>* undef, align 1
+  ret void
+}