diff --git a/lib/Target/ARM/ARMFastISel.cpp b/lib/Target/ARM/ARMFastISel.cpp
index 2af42c9..0f3e6cf 100644
--- a/lib/Target/ARM/ARMFastISel.cpp
+++ b/lib/Target/ARM/ARMFastISel.cpp
@@ -115,6 +115,11 @@
                                      const TargetRegisterClass *RC,
                                      unsigned Op0, bool Op0IsKill,
                                      unsigned Op1, bool Op1IsKill);
+    virtual unsigned FastEmitInst_rrr(unsigned MachineInstOpcode,
+                                      const TargetRegisterClass *RC,
+                                      unsigned Op0, bool Op0IsKill,
+                                      unsigned Op1, bool Op1IsKill,
+                                      unsigned Op2, bool Op2IsKill);
     virtual unsigned FastEmitInst_ri(unsigned MachineInstOpcode,
                                      const TargetRegisterClass *RC,
                                      unsigned Op0, bool Op0IsKill,
@@ -315,6 +320,31 @@
   return ResultReg;
 }
 
+unsigned ARMFastISel::FastEmitInst_rrr(unsigned MachineInstOpcode,
+                                       const TargetRegisterClass *RC,
+                                       unsigned Op0, bool Op0IsKill,
+                                       unsigned Op1, bool Op1IsKill,
+                                       unsigned Op2, bool Op2IsKill) {
+  unsigned ResultReg = createResultReg(RC);
+  const TargetInstrDesc &II = TII.get(MachineInstOpcode);
+
+  if (II.getNumDefs() >= 1)
+    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
+                   .addReg(Op0, Op0IsKill * RegState::Kill)
+                   .addReg(Op1, Op1IsKill * RegState::Kill)
+                   .addReg(Op2, Op2IsKill * RegState::Kill));
+  else {
+    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
+                   .addReg(Op0, Op0IsKill * RegState::Kill)
+                   .addReg(Op1, Op1IsKill * RegState::Kill)
+                   .addReg(Op2, Op2IsKill * RegState::Kill));
+    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+                           TII.get(TargetOpcode::COPY), ResultReg)
+                   .addReg(II.ImplicitDefs[0]));
+  }
+  return ResultReg;
+}
+
 unsigned ARMFastISel::FastEmitInst_ri(unsigned MachineInstOpcode,
                                       const TargetRegisterClass *RC,
                                       unsigned Op0, bool Op0IsKill,
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 9dc103f..11ff6bf 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -866,6 +866,7 @@
   case ARMISD::BFI:           return "ARMISD::BFI";
   case ARMISD::VORRIMM:       return "ARMISD::VORRIMM";
   case ARMISD::VBICIMM:       return "ARMISD::VBICIMM";
+  case ARMISD::VBSL:          return "ARMISD::VBSL";
   case ARMISD::VLD2DUP:       return "ARMISD::VLD2DUP";
   case ARMISD::VLD3DUP:       return "ARMISD::VLD3DUP";
   case ARMISD::VLD4DUP:       return "ARMISD::VLD4DUP";
@@ -5336,6 +5337,37 @@
     }
   }
 
+  SDValue N0 = N->getOperand(0);
+  if (N0.getOpcode() != ISD::AND)
+    return SDValue();
+  SDValue N1 = N->getOperand(1);
+
+  // (or (and B, A), (and C, ~A)) => (VBSL A, B, C) when A is a constant.
+  if (Subtarget->hasNEON() && N1.getOpcode() == ISD::AND && VT.isVector() &&
+      DAG.getTargetLoweringInfo().isTypeLegal(VT)) {
+    APInt SplatUndef;
+    unsigned SplatBitSize;
+    bool HasAnyUndefs;
+
+    BuildVectorSDNode *BVN0 = dyn_cast<BuildVectorSDNode>(N0->getOperand(1));
+    APInt SplatBits0;
+    if (BVN0 && BVN0->isConstantSplat(SplatBits0, SplatUndef, SplatBitSize,
+                                  HasAnyUndefs) && !HasAnyUndefs) {
+      BuildVectorSDNode *BVN1 = dyn_cast<BuildVectorSDNode>(N1->getOperand(1));
+      APInt SplatBits1;
+      if (BVN1 && BVN1->isConstantSplat(SplatBits1, SplatUndef, SplatBitSize,
+                                    HasAnyUndefs) && !HasAnyUndefs &&
+          SplatBits0 == ~SplatBits1) {
+        // Canonicalize the vector type to make instruction selection simpler.
+        EVT CanonicalVT = VT.is128BitVector() ? MVT::v4i32 : MVT::v2i32;
+        SDValue Result = DAG.getNode(ARMISD::VBSL, dl, CanonicalVT,
+                                     N0->getOperand(1), N0->getOperand(0),
+                                     N1->getOperand(1));
+        return DAG.getNode(ISD::BITCAST, dl, VT, Result);
+      }
+    }
+  }
+
   // Try to use the ARM/Thumb2 BFI (bitfield insert) instruction when
   // reasonable.
 
@@ -5343,7 +5375,6 @@
   if (Subtarget->isThumb1Only() || !Subtarget->hasV6T2Ops())
     return SDValue();
 
-  SDValue N0 = N->getOperand(0), N1 = N->getOperand(1);
   DebugLoc DL = N->getDebugLoc();
   // 1) or (and A, mask), val => ARMbfi A, val, mask
   //      iff (val & mask) == val
@@ -5354,8 +5385,6 @@
   //  2b) iff isBitFieldInvertedMask(~mask) && isBitFieldInvertedMask(mask2)
   //          && ~mask == mask2
   //  (i.e., copy a bitfield value into another bitfield of the same width)
-  if (N0.getOpcode() != ISD::AND)
-    return SDValue();
 
   if (VT != MVT::i32)
     return SDValue();
diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h
index e09c1da..e37855d 100644
--- a/lib/Target/ARM/ARMISelLowering.h
+++ b/lib/Target/ARM/ARMISelLowering.h
@@ -179,6 +179,9 @@
       // Vector AND with NOT of immediate
       VBICIMM,
 
+      // Vector bitwise select
+      VBSL,
+
       // Vector load N-element structure to all lanes:
       VLD2DUP = ISD::FIRST_TARGET_MEMORY_OPCODE,
       VLD3DUP,
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 374dac0..c4fb374 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -80,6 +80,12 @@
 def NEONvorrImm   : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
 def NEONvbicImm   : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
 
+def NEONvbsl      : SDNode<"ARMISD::VBSL",
+                           SDTypeProfile<1, 3, [SDTCisVec<0>,
+                                                SDTCisSameAs<0, 1>,
+                                                SDTCisSameAs<0, 2>,
+                                                SDTCisSameAs<0, 3>]>>;
+
 def NEONvdup      : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
 
 // VDUPLANE can produce a quad-register result from a double-register source,
@@ -3767,16 +3773,21 @@
                      (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
                      N3RegFrm, IIC_VCNTiD,
                      "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
-                     [(set DPR:$Vd,
-                       (v2i32 (or (and DPR:$Vn, DPR:$src1),
-                                  (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
+                     [(set DPR:$Vd, (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
+
+def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
+                     (and DPR:$Vm, (vnotd DPR:$Vd)))),
+          (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
+
 def  VBSLq    : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
                      (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
                      N3RegFrm, IIC_VCNTiQ,
                      "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
-                     [(set QPR:$Vd,
-                       (v4i32 (or (and QPR:$Vn, QPR:$src1),
-                                  (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
+                     [(set QPR:$Vd, (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
+
+def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
+                     (and QPR:$Vm, (vnotq QPR:$Vd)))),
+          (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
 
 //   VBIF     : Vector Bitwise Insert if False
 //              like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",