diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index f0870fbcf39..ba7827a11e0 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -4107,14 +4107,60 @@ AArch64TargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
       // just use DUPLANE. We can only do this if the lane being extracted
       // is at a constant index, as the DUP from lane instructions only have
       // constant-index forms.
+      //
+      // If there is a TRUNCATE between EXTRACT_VECTOR_ELT and DUP, we can
+      // remove TRUNCATE for DUPLANE by apdating the source vector to
+      // appropriate vector type and lane index.
+      //
       // FIXME: for now we have v1i8, v1i16, v1i32 legal vector types, if they
       // are not legal any more, no need to check the type size in bits should
       // be large than 64.
-      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
-          isa<ConstantSDNode>(Value->getOperand(1)) &&
-          Value->getOperand(0).getValueType().getSizeInBits() >= 64) {
-          N = DAG.getNode(AArch64ISD::NEON_VDUPLANE, DL, VT,
-                        Value->getOperand(0), Value->getOperand(1));
+      SDValue V = Value;
+      if (Value->getOpcode() == ISD::TRUNCATE)
+        V = Value->getOperand(0);
+      if (V->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
+          isa<ConstantSDNode>(V->getOperand(1)) &&
+          V->getOperand(0).getValueType().getSizeInBits() >= 64) {
+
+        // If the element size of source vector is larger than DUPLANE
+        // element size, we can do transformation by,
+        // 1) bitcasting source register to smaller element vector
+        // 2) mutiplying the lane index by SrcEltSize/ResEltSize
+        // For example, we can lower
+        //     "v8i16 vdup_lane(v4i32, 1)"
+        // to be
+        //     "v8i16 vdup_lane(v8i16 bitcast(v4i32), 2)".
+        SDValue SrcVec = V->getOperand(0);
+        unsigned SrcEltSize =
+            SrcVec.getValueType().getVectorElementType().getSizeInBits();
+        unsigned ResEltSize = VT.getVectorElementType().getSizeInBits();
+        if (SrcEltSize > ResEltSize) {
+          assert((SrcEltSize % ResEltSize == 0) && "Invalid element size");
+          SDValue BitCast;
+          unsigned SrcSize = SrcVec.getValueType().getSizeInBits();
+          unsigned ResSize = VT.getSizeInBits();
+
+          if (SrcSize > ResSize) {
+            assert((SrcSize % ResSize == 0) && "Invalid vector size");
+            EVT CastVT =
+                EVT::getVectorVT(*DAG.getContext(), VT.getVectorElementType(),
+                                 SrcSize / ResEltSize);
+            BitCast = DAG.getNode(ISD::BITCAST, DL, CastVT, SrcVec);
+          } else {
+            assert((SrcSize == ResSize) && "Invalid vector size of source vec");
+            BitCast = DAG.getNode(ISD::BITCAST, DL, VT, SrcVec);
+          }
+
+          unsigned LaneIdx = V->getConstantOperandVal(1);
+          SDValue Lane =
+              DAG.getConstant((SrcEltSize / ResEltSize) * LaneIdx, MVT::i64);
+          N = DAG.getNode(AArch64ISD::NEON_VDUPLANE, DL, VT, BitCast, Lane);
+        } else {
+          assert((SrcEltSize == ResEltSize) &&
+                 "Invalid element size of source vec");
+          N = DAG.getNode(AArch64ISD::NEON_VDUPLANE, DL, VT, V->getOperand(0),
+                          V->getOperand(1));
+        }
       } else
         N = DAG.getNode(AArch64ISD::NEON_VDUP, DL, VT, Value);
 
diff --git a/test/CodeGen/AArch64/neon-copy.ll b/test/CodeGen/AArch64/neon-copy.ll
index 881a858bc19..c4fb5076406 100644
--- a/test/CodeGen/AArch64/neon-copy.ll
+++ b/test/CodeGen/AArch64/neon-copy.ll
@@ -726,3 +726,125 @@ define <8 x i8> @getl(<16 x i8> %x) #0 {
   %vecinit14 = insertelement <8 x i8> %vecinit12, i8 %vecext13, i32 7
   ret <8 x i8> %vecinit14
 }
+
+define <4 x i16> @test_dup_v2i32_v4i16(<2 x i32> %a) {
+; CHECK-LABEL: test_dup_v2i32_v4i16:
+; CHECK: dup v0.4h, v0.h[2]
+entry:
+  %x = extractelement <2 x i32> %a, i32 1
+  %vget_lane = trunc i32 %x to i16
+  %vecinit.i = insertelement <4 x i16> undef, i16 %vget_lane, i32 0
+  %vecinit1.i = insertelement <4 x i16> %vecinit.i, i16 %vget_lane, i32 1
+  %vecinit2.i = insertelement <4 x i16> %vecinit1.i, i16 %vget_lane, i32 2
+  %vecinit3.i = insertelement <4 x i16> %vecinit2.i, i16 %vget_lane, i32 3
+  ret <4 x i16> %vecinit3.i
+}
+
+define <8 x i16> @test_dup_v4i32_v8i16(<4 x i32> %a) {
+; CHECK-LABEL: test_dup_v4i32_v8i16:
+; CHECK: dup v0.8h, v0.h[6]
+entry:
+  %x = extractelement <4 x i32> %a, i32 3
+  %vget_lane = trunc i32 %x to i16
+  %vecinit.i = insertelement <8 x i16> undef, i16 %vget_lane, i32 0
+  %vecinit1.i = insertelement <8 x i16> %vecinit.i, i16 %vget_lane, i32 1
+  %vecinit2.i = insertelement <8 x i16> %vecinit1.i, i16 %vget_lane, i32 2
+  %vecinit3.i = insertelement <8 x i16> %vecinit2.i, i16 %vget_lane, i32 3
+  %vecinit4.i = insertelement <8 x i16> %vecinit3.i, i16 %vget_lane, i32 4
+  %vecinit5.i = insertelement <8 x i16> %vecinit4.i, i16 %vget_lane, i32 5
+  %vecinit6.i = insertelement <8 x i16> %vecinit5.i, i16 %vget_lane, i32 6
+  %vecinit7.i = insertelement <8 x i16> %vecinit6.i, i16 %vget_lane, i32 7
+  ret <8 x i16> %vecinit7.i
+}
+
+define <4 x i16> @test_dup_v1i64_v4i16(<1 x i64> %a) {
+; CHECK-LABEL: test_dup_v1i64_v4i16:
+; CHECK: dup v0.4h, v0.h[0]
+entry:
+  %x = extractelement <1 x i64> %a, i32 0
+  %vget_lane = trunc i64 %x to i16
+  %vecinit.i = insertelement <4 x i16> undef, i16 %vget_lane, i32 0
+  %vecinit1.i = insertelement <4 x i16> %vecinit.i, i16 %vget_lane, i32 1
+  %vecinit2.i = insertelement <4 x i16> %vecinit1.i, i16 %vget_lane, i32 2
+  %vecinit3.i = insertelement <4 x i16> %vecinit2.i, i16 %vget_lane, i32 3
+  ret <4 x i16> %vecinit3.i
+}
+
+define <2 x i32> @test_dup_v1i64_v2i32(<1 x i64> %a) {
+; CHECK-LABEL: test_dup_v1i64_v2i32:
+; CHECK: dup v0.2s, v0.s[0]
+entry:
+  %x = extractelement <1 x i64> %a, i32 0
+  %vget_lane = trunc i64 %x to i32
+  %vecinit.i = insertelement <2 x i32> undef, i32 %vget_lane, i32 0
+  %vecinit1.i = insertelement <2 x i32> %vecinit.i, i32 %vget_lane, i32 1
+  ret <2 x i32> %vecinit1.i
+}
+
+define <8 x i16> @test_dup_v2i64_v8i16(<2 x i64> %a) {
+; CHECK-LABEL: test_dup_v2i64_v8i16:
+; CHECK: dup v0.8h, v0.h[4]
+entry:
+  %x = extractelement <2 x i64> %a, i32 1
+  %vget_lane = trunc i64 %x to i16
+  %vecinit.i = insertelement <8 x i16> undef, i16 %vget_lane, i32 0
+  %vecinit1.i = insertelement <8 x i16> %vecinit.i, i16 %vget_lane, i32 1
+  %vecinit2.i = insertelement <8 x i16> %vecinit1.i, i16 %vget_lane, i32 2
+  %vecinit3.i = insertelement <8 x i16> %vecinit2.i, i16 %vget_lane, i32 3
+  %vecinit4.i = insertelement <8 x i16> %vecinit3.i, i16 %vget_lane, i32 4
+  %vecinit5.i = insertelement <8 x i16> %vecinit4.i, i16 %vget_lane, i32 5
+  %vecinit6.i = insertelement <8 x i16> %vecinit5.i, i16 %vget_lane, i32 6
+  %vecinit7.i = insertelement <8 x i16> %vecinit6.i, i16 %vget_lane, i32 7
+  ret <8 x i16> %vecinit7.i
+}
+
+define <4 x i32> @test_dup_v2i64_v4i32(<2 x i64> %a) {
+; CHECK-LABEL: test_dup_v2i64_v4i32:
+; CHECK: dup v0.4s, v0.s[2]
+entry:
+  %x = extractelement <2 x i64> %a, i32 1
+  %vget_lane = trunc i64 %x to i32
+  %vecinit.i = insertelement <4 x i32> undef, i32 %vget_lane, i32 0
+  %vecinit1.i = insertelement <4 x i32> %vecinit.i, i32 %vget_lane, i32 1
+  %vecinit2.i = insertelement <4 x i32> %vecinit1.i, i32 %vget_lane, i32 2
+  %vecinit3.i = insertelement <4 x i32> %vecinit2.i, i32 %vget_lane, i32 3
+  ret <4 x i32> %vecinit3.i
+}
+
+define <4 x i16> @test_dup_v4i32_v4i16(<4 x i32> %a) {
+; CHECK-LABEL: test_dup_v4i32_v4i16:
+; CHECK: dup v0.4h, v0.h[2]
+entry:
+  %x = extractelement <4 x i32> %a, i32 1
+  %vget_lane = trunc i32 %x to i16
+  %vecinit.i = insertelement <4 x i16> undef, i16 %vget_lane, i32 0
+  %vecinit1.i = insertelement <4 x i16> %vecinit.i, i16 %vget_lane, i32 1
+  %vecinit2.i = insertelement <4 x i16> %vecinit1.i, i16 %vget_lane, i32 2
+  %vecinit3.i = insertelement <4 x i16> %vecinit2.i, i16 %vget_lane, i32 3
+  ret <4 x i16> %vecinit3.i
+}
+
+define <4 x i16> @test_dup_v2i64_v4i16(<2 x i64> %a) {
+; CHECK-LABEL: test_dup_v2i64_v4i16:
+; CHECK: dup v0.4h, v0.h[0]
+entry:
+  %x = extractelement <2 x i64> %a, i32 0
+  %vget_lane = trunc i64 %x to i16
+  %vecinit.i = insertelement <4 x i16> undef, i16 %vget_lane, i32 0
+  %vecinit1.i = insertelement <4 x i16> %vecinit.i, i16 %vget_lane, i32 1
+  %vecinit2.i = insertelement <4 x i16> %vecinit1.i, i16 %vget_lane, i32 2
+  %vecinit3.i = insertelement <4 x i16> %vecinit2.i, i16 %vget_lane, i32 3
+  ret <4 x i16> %vecinit3.i
+}
+
+define <2 x i32> @test_dup_v2i64_v2i32(<2 x i64> %a) {
+; CHECK-LABEL: test_dup_v2i64_v2i32:
+; CHECK: dup v0.2s, v0.s[0]
+entry:
+  %x = extractelement <2 x i64> %a, i32 0
+  %vget_lane = trunc i64 %x to i32
+  %vecinit.i = insertelement <2 x i32> undef, i32 %vget_lane, i32 0
+  %vecinit1.i = insertelement <2 x i32> %vecinit.i, i32 %vget_lane, i32 1
+  ret <2 x i32> %vecinit1.i
+}
+