[X86][SSE] Simplify vector LOAD + EXTEND on pre-SSE41 hardware

Improve vector extension of vectors on hardware without dedicated VSEXT/VZEXT instructions.

We already convert these to SIGN_EXTEND_VECTOR_INREG/ZERO_EXTEND_VECTOR_INREG but can further improve this by using the legalizer instead of prematurely splitting into legal vectors in the combine as this only properly helps for lowering to VSEXT/VZEXT.

Removes a lot of unnecessary any_extend + mask pattern - (Fix for PR25718).

Differential Revision: http://reviews.llvm.org/D17932

llvm-svn: 263303

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index 8ba19f7..c795056 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -653,6 +653,7 @@ private:
   void SplitVecRes_UnaryOp(SDNode *N, SDValue &Lo, SDValue &Hi);
   void SplitVecRes_ExtendOp(SDNode *N, SDValue &Lo, SDValue &Hi);
   void SplitVecRes_InregOp(SDNode *N, SDValue &Lo, SDValue &Hi);
+  void SplitVecRes_ExtVecInRegOp(SDNode *N, SDValue &Lo, SDValue &Hi);
 
   void SplitVecRes_BITCAST(SDNode *N, SDValue &Lo, SDValue &Hi);
   void SplitVecRes_BUILD_VECTOR(SDNode *N, SDValue &Lo, SDValue &Hi);

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index d0187d3..575eb3c 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -621,6 +621,12 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
     SplitVecRes_VECTOR_SHUFFLE(cast<ShuffleVectorSDNode>(N), Lo, Hi);
     break;
 
+  case ISD::ANY_EXTEND_VECTOR_INREG:
+  case ISD::SIGN_EXTEND_VECTOR_INREG:
+  case ISD::ZERO_EXTEND_VECTOR_INREG:
+    SplitVecRes_ExtVecInRegOp(N, Lo, Hi);
+    break;
+
   case ISD::BITREVERSE:
   case ISD::BSWAP:
   case ISD::CONVERT_RNDSAT:
@@ -917,6 +923,39 @@ void DAGTypeLegalizer::SplitVecRes_InregOp(SDNode *N, SDValue &Lo,
                    DAG.getValueType(HiVT));
 }
 
+void DAGTypeLegalizer::SplitVecRes_ExtVecInRegOp(SDNode *N, SDValue &Lo,
+                                                 SDValue &Hi) {
+  unsigned Opcode = N->getOpcode();
+  SDValue N0 = N->getOperand(0);
+
+  SDLoc dl(N);
+  SDValue InLo, InHi;
+  GetSplitVector(N0, InLo, InHi);
+  EVT InLoVT = InLo.getValueType();
+  unsigned InNumElements = InLoVT.getVectorNumElements();
+
+  EVT OutLoVT, OutHiVT;
+  std::tie(OutLoVT, OutHiVT) = DAG.GetSplitDestVTs(N->getValueType(0));
+  unsigned OutNumElements = OutLoVT.getVectorNumElements();
+  assert((2 * OutNumElements) <= InNumElements &&
+         "Illegal extend vector in reg split");
+
+  // *_EXTEND_VECTOR_INREG instructions extend the lowest elements of the
+  // input vector (i.e. we only use InLo):
+  // OutLo will extend the first OutNumElements from InLo.
+  // OutHi will extend the next OutNumElements from InLo.
+
+  // Shuffle the elements from InLo for OutHi into the bottom elements to
+  // create a 'fake' InHi.
+  SmallVector<int, 8> SplitHi(InNumElements, -1);
+  for (unsigned i = 0; i != OutNumElements; ++i)
+    SplitHi[i] = i + OutNumElements;
+  InHi = DAG.getVectorShuffle(InLoVT, dl, InLo, DAG.getUNDEF(InLoVT), SplitHi);
+
+  Lo = DAG.getNode(Opcode, dl, OutLoVT, InLo);
+  Hi = DAG.getNode(Opcode, dl, OutHiVT, InHi);
+}
+
 void DAGTypeLegalizer::SplitVecRes_INSERT_VECTOR_ELT(SDNode *N, SDValue &Lo,
                                                      SDValue &Hi) {
   SDValue Vec = N->getOperand(0);

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 4f00eb3..eac67d1 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -28534,7 +28534,9 @@ static SDValue combineToExtendVectorInReg(SDNode *N, SelectionDAG &DAG,
 
   // If target-size is 128-bits (or 256-bits on AVX2 target), then convert to
   // ISD::*_EXTEND_VECTOR_INREG which ensures lowering to X86ISD::V*EXT.
-  if (VT.is128BitVector() || (VT.is256BitVector() && Subtarget.hasInt256())) {
+  // Also use this if we don't have SSE41 to allow the legalizer do its job.
+  if (!Subtarget.hasSSE41() || VT.is128BitVector() ||
+      (VT.is256BitVector() && Subtarget.hasInt256())) {
     SDValue ExOp = ExtendVecSize(DL, N0, VT.getSizeInBits());
     return Opcode == ISD::SIGN_EXTEND
                ? DAG.getSignExtendVectorInReg(ExOp, DL, VT)

diff --git a/llvm/test/CodeGen/X86/vector-zext.ll b/llvm/test/CodeGen/X86/vector-zext.ll
index ec1a4ec..4427926 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-zext.ll
+++ b/llvm/test/CodeGen/X86/vector-zext.ll
@@ -544,23 +544,20 @@ define <4 x i64> @load_zext_4i8_to_4i64(<4 x i8> *%ptr) {
 ; SSE2-LABEL: load_zext_4i8_to_4i64:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,1,3]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
-; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: load_zext_4i8_to_4i64:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movd {{.*#+}} xmm1 = mem[0],zero,zero,zero
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[4],zero,zero,zero,zero,zero,zero,zero
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[8],zero,zero,zero,zero,zero,zero,zero,xmm1[12],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[2],zero,zero,zero,zero,zero,zero,zero,xmm1[3],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: load_zext_4i8_to_4i64:
@@ -625,22 +622,21 @@ define <8 x i32> @load_zext_8i8_to_8i32(<8 x i8> *%ptr) {
 ; SSE2-LABEL: load_zext_8i8_to_8i32:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
 ; SSE2-NEXT:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: load_zext_8i8_to_8i32:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; SSSE3-NEXT:    pxor %xmm2, %xmm2
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[6],zero,zero,zero
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[8],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[14],zero,zero,zero
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSSE3-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: load_zext_8i8_to_8i32:
@@ -674,34 +670,33 @@ entry:
 define <8 x i64> @load_zext_8i8_to_8i64(<8 x i8> *%ptr) {
 ; SSE2-LABEL: load_zext_8i8_to_8i64:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,1,0,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,6,7]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSE2-NEXT:    pand %xmm4, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[1,1,1,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,7]
-; SSE2-NEXT:    pand %xmm4, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,1,2,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,5,6,7]
-; SSE2-NEXT:    pand %xmm4, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[3,1,3,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,5,6,7]
-; SSE2-NEXT:    pand %xmm4, %xmm3
+; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
+; SSE2-NEXT:    pxor %xmm4, %xmm4
+; SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; SSE2-NEXT:    movdqa %xmm3, %xmm2
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm4[0],xmm2[1],xmm4[1]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm3 = xmm3[2],xmm4[2],xmm3[3],xmm4[3]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: load_zext_8i8_to_8i64:
 ; SSSE3:       # BB#0: # %entry
-; SSSE3-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3],xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
-; SSSE3-NEXT:    movdqa %xmm3, %xmm0
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero
-; SSSE3-NEXT:    movdqa %xmm3, %xmm1
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[4],zero,zero,zero,zero,zero,zero,zero,xmm1[6],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm4 = [0,128,128,128,128,128,128,128,1,128,128,128,128,128,128,128]
+; SSSE3-NEXT:    movdqa %xmm1, %xmm0
+; SSSE3-NEXT:    pshufb %xmm4, %xmm0
+; SSSE3-NEXT:    movdqa {{.*#+}} xmm5 = [2,128,128,128,128,128,128,128,3,128,128,128,128,128,128,128]
+; SSSE3-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
+; SSSE3-NEXT:    pshufb %xmm5, %xmm1
 ; SSSE3-NEXT:    movdqa %xmm3, %xmm2
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm2 = xmm2[8],zero,zero,zero,zero,zero,zero,zero,xmm2[10],zero,zero,zero,zero,zero,zero,zero
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm3 = xmm3[12],zero,zero,zero,zero,zero,zero,zero,xmm3[14],zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pshufb %xmm4, %xmm2
+; SSSE3-NEXT:    pshufb %xmm5, %xmm3
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: load_zext_8i8_to_8i64:
@@ -851,21 +846,21 @@ define <4 x i64> @load_zext_4i16_to_4i64(<4 x i16> *%ptr) {
 ; SSE2-LABEL: load_zext_4i16_to_4i64:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,1,3]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0]
-; SSE2-NEXT:    pand %xmm2, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
-; SSE2-NEXT:    pand %xmm2, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: load_zext_4i16_to_4i64:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; SSSE3-NEXT:    pxor %xmm2, %xmm2
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSSE3-NEXT:    movdqa %xmm1, %xmm0
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0,1],zero,zero,zero,zero,zero,zero,xmm0[4,5],zero,zero,zero,zero,zero,zero
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[8,9],zero,zero,zero,zero,zero,zero,xmm1[12,13],zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: load_zext_4i16_to_4i64: