[X86][SSE] Match zero/any extension shuffles that don't start from the first element

[RKSimon]

Summary

This patch generalizes the lowering of shuffles as zero extensions to allow extensions that don't start from the first element. It now recognises extensions starting anywhere in the lower 128-bits or at the start of any higher 128-bit lane.

The motivation was to reduce the number of high cost pshufb calls, but it also improves the SSE2 case as well.

[RKSimon]

Updated SSE2 zero extension by UNPCK - only shuffle odd offsets by a single lane to allow extension to multiple 128-bit vectors to share shuffles.

[RKSimon]

ping

[RKSimon]

Lower with pmovzx for offset and scale=2 if its for a vector size greater than 128-bits.

[qcolombet]

Hi Simon,

This looks mostly good to me, though it seems a couple of tests are regressing, aren’t they?

I have annotated them, could you comment on why this is beneficial or at least neutral?

Thanks,
-Quentin

lib/Target/X86/X86ISelLowering.cpp

76

Could you explain why in the comment?

76

Explain why, I am sure it won’t be obvious when we look at that line in a couple of months :).

77

NumEltsPerLane?

77

What is “-ve"?

78

Document those preconditions on the function prototype as well.

82

I wonder if this lambda makes sense.
Unless I am mistaken, this value is an invariant of the lambda and can be hoisted outside of the only loop where it is used.
I.e., we can just make the test: OffsetLane == Idx/NumLaneElements at the right place.

test/CodeGen/X86/vec_cast2.ll

5

Isn’t the new sequence less efficient than the previous one?

test/CodeGen/X86/vector-shuffle-256-v32.ll

6

Ditto?

6

Ditto?

test/CodeGen/X86/vector-zext.ll

73

Ditto?

73

Ditto?

[RKSimon]

Updated based on feedback

[RKSimon]

Regarding the tests - the over use of PSHUFB can be a performance issue - you often have to load the shuffle mask (>5cy) and many targets (AMD + Intel) have a poor latency/throughput executing it (>3cy). In comparison PSHUFD/PSRLDQ as well as PMOVZX are pretty light instructions (1-2cy each).

lib/Target/X86/X86ISelLowering.cpp

7191

I've pulled the 'OffsetLane' constant out - I'd prefer to keep the lambda though as its used in quite a few places and avoids cluttering the code.

7212

This is definitely a borderline case - without the earlyout most of the time we are just replacing a XOR (zero)+PUNCKH with a PSHUFD+PMOVZX. There's next to nothing in it so I went with avoiding a change.

[qcolombet]

Regarding the tests - the over use of PSHUFB can be a performance issue - you often have to load the shuffle mask (>5cy) and many targets (AMD + Intel) have a poor latency/throughput executing it (>icy).

Ah, that’s the catch, loading the mask! From the diff that wasn’t clear since we do not see the actual operands.

Alright, LGTM then!

Thanks,
-Quentin

lib/Target/X86/X86ISelLowering.cpp

static cl::opt<bool> ExperimentalVectorWideningLegalization(
     cl::Hidden);
 
 X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
                                      const X86Subtarget &STI)
     : TargetLowering(TM), Subtarget(&STI) {
   X86ScalarSSEf64 = Subtarget->hasSSE2();
   X86ScalarSSEf32 = Subtarget->hasSSE1();
   MVT PtrVT = MVT::getIntegerVT(8 * TM.getPointerSize());
 
   // Set up the TargetLowering object.
   static const MVT IntVTs[] = { MVT::i8, MVT::i16, MVT::i32, MVT::i64 };
 
   // X86 is weird. It always uses i8 for shift amounts and setcc results.
   setBooleanContents(ZeroOrOneBooleanContent);
   // X86-SSE is even stranger. It uses -1 or 0 for vector masks.
   setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
 
   // For 64-bit, since we have so many registers, use the ILP scheduler.
   // For 32-bit, use the register pressure specific scheduling.
   // For Atom, always use ILP scheduling.
   if (Subtarget->isAtom())
     setSchedulingPreference(Sched::ILP);
   else if (Subtarget->is64Bit())

 }
 
 /// \brief Lower a vector shuffle as a zero or any extension.
 ///
 /// Given a specific number of elements, element bit width, and extension
 /// stride, produce either a zero or any extension based on the available
 /// features of the subtarget.
 static SDValue lowerVectorShuffleAsSpecificZeroOrAnyExtend(
-    SDLoc DL, MVT VT, int Scale, bool AnyExt, SDValue InputV,
+    SDLoc DL, MVT VT, int Scale, int Offset, bool AnyExt, SDValue InputV,
     ArrayRef<int> Mask, const X86Subtarget *Subtarget, SelectionDAG &DAG) {
   assert(Scale > 1 && "Need a scale to extend.");
-  int NumElements = VT.getVectorNumElements();
   int EltBits = VT.getScalarSizeInBits();
+  int NumElements = VT.getVectorNumElements();
+  int NumLaneElements = 128 / EltBits;
   assert((EltBits == 8 || EltBits == 16 || EltBits == 32) &&
          "Only 8, 16, and 32 bit elements can be extended.");
   assert(Scale * EltBits <= 64 && "Cannot zero extend past 64 bits.");
+  assert(0 <= Offset && "Extension offset must be positive.");
+  assert((Offset < NumLaneElements || Offset % NumLaneElements == 0) &&
+         "Extension offset must be in the first lane or start an upper lane.");
+
+  // Check that an offseted index is in same lane as the base offset.
+  auto SafeOffset = [&](int Idx) {
+    int OffsetLane = Offset / NumLaneElements;
+    int IdxLane = Idx / NumLaneElements;
+    return OffsetLane == IdxLane;
+  };
+
+  // Shift along an input so that the offset base moves to the first element.
+  auto ShuffleOffset = [&](SDValue V) {
+    if (!Offset)
+      return V;
+
+    SmallVector<int, 8> ShMask((unsigned)NumElements, -1);
+    for (int i = 0; i * Scale < NumElements; ++i) {
+      int SrcIdx = i + Offset;
+      ShMask[i] = SafeOffset(SrcIdx) ? SrcIdx : -1;
+    }
+    return DAG.getVectorShuffle(VT, DL, V, DAG.getUNDEF(VT), ShMask);
+  };
 
   // Found a valid zext mask! Try various lowering strategies based on the
   // input type and available ISA extensions.
   if (Subtarget->hasSSE41()) {
+    // Not worth offseting if scale == 2.
+    if (Offset && Scale == 2)
+      return SDValue();
     MVT ExtVT = MVT::getVectorVT(MVT::getIntegerVT(EltBits * Scale),
                                  NumElements / Scale);
-    return DAG.getBitcast(VT, DAG.getNode(X86ISD::VZEXT, DL, ExtVT, InputV));
+    InputV = DAG.getNode(X86ISD::VZEXT, DL, ExtVT, ShuffleOffset(InputV));
+    return DAG.getBitcast(VT, InputV);
   }
 
+  assert(VT.getSizeInBits() == 128 && "Only 128-bit vectors can be extended.");
+
   // For any extends we can cheat for larger element sizes and use shuffle
   // instructions that can fold with a load and/or copy.
   if (AnyExt && EltBits == 32) {
-    int PSHUFDMask[4] = {0, -1, 1, -1};
+    int PSHUFDMask[4] = {Offset, -1, SafeOffset(Offset + 1) ? Offset + 1 : -1,
+                         -1};
     return DAG.getBitcast(
         VT, DAG.getNode(X86ISD::PSHUFD, DL, MVT::v4i32,
                         DAG.getBitcast(MVT::v4i32, InputV),
                         getV4X86ShuffleImm8ForMask(PSHUFDMask, DL, DAG)));
   }
   if (AnyExt && EltBits == 16 && Scale > 2) {
-    int PSHUFDMask[4] = {0, -1, 0, -1};
+    int PSHUFDMask[4] = {Offset / 2, -1,
+                         SafeOffset(Offset + 1) ? (Offset + 1) / 2 : -1, -1};
     InputV = DAG.getNode(X86ISD::PSHUFD, DL, MVT::v4i32,
                          DAG.getBitcast(MVT::v4i32, InputV),
                          getV4X86ShuffleImm8ForMask(PSHUFDMask, DL, DAG));
-    int PSHUFHWMask[4] = {1, -1, -1, -1};
+    int PSHUFWMask[4] = {1, -1, -1, -1};
+    unsigned OddEvenOp = (Offset & 1 ? X86ISD::PSHUFLW : X86ISD::PSHUFHW);
     return DAG.getBitcast(
-        VT, DAG.getNode(X86ISD::PSHUFHW, DL, MVT::v8i16,
+        VT, DAG.getNode(OddEvenOp, DL, MVT::v8i16,
                         DAG.getBitcast(MVT::v8i16, InputV),
-                        getV4X86ShuffleImm8ForMask(PSHUFHWMask, DL, DAG)));
+                        getV4X86ShuffleImm8ForMask(PSHUFWMask, DL, DAG)));
   }
 
   // The SSE4A EXTRQ instruction can efficiently extend the first 2 lanes
   // to 64-bits.
   if ((Scale * EltBits) == 64 && EltBits < 32 && Subtarget->hasSSE4A()) {
     assert(NumElements == (int)Mask.size() && "Unexpected shuffle mask size!");
     assert(VT.getSizeInBits() == 128 && "Unexpected vector width!");
 
+    int LoIdx = Offset * EltBits;
     SDValue Lo = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64,
                              DAG.getNode(X86ISD::EXTRQI, DL, VT, InputV,
                                          DAG.getConstant(EltBits, DL, MVT::i8),
-                                         DAG.getConstant(0, DL, MVT::i8)));
-    if (isUndefInRange(Mask, NumElements/2, NumElements/2))
+                                         DAG.getConstant(LoIdx, DL, MVT::i8)));
+
+    if (isUndefInRange(Mask, NumElements / 2, NumElements / 2) ||
+        !SafeOffset(Offset + 1))
       return DAG.getNode(ISD::BITCAST, DL, VT, Lo);
 
-    SDValue Hi =
-        DAG.getNode(ISD::BITCAST, DL, MVT::v2i64,
+    int HiIdx = (Offset + 1) * EltBits;
+    SDValue Hi = DAG.getNode(ISD::BITCAST, DL, MVT::v2i64,
                              DAG.getNode(X86ISD::EXTRQI, DL, VT, InputV,
                                          DAG.getConstant(EltBits, DL, MVT::i8),
-                                DAG.getConstant(EltBits, DL, MVT::i8)));
+                                         DAG.getConstant(HiIdx, DL, MVT::i8)));
     return DAG.getNode(ISD::BITCAST, DL, VT,
                        DAG.getNode(X86ISD::UNPCKL, DL, MVT::v2i64, Lo, Hi));
   }
 
   // If this would require more than 2 unpack instructions to expand, use
   // pshufb when available. We can only use more than 2 unpack instructions
   // when zero extending i8 elements which also makes it easier to use pshufb.
   if (Scale > 4 && EltBits == 8 && Subtarget->hasSSSE3()) {
     assert(NumElements == 16 && "Unexpected byte vector width!");
     SDValue PSHUFBMask[16];
-    for (int i = 0; i < 16; ++i)
-      PSHUFBMask[i] =
-          DAG.getConstant((i % Scale == 0) ? i / Scale : 0x80, DL, MVT::i8);
+    for (int i = 0; i < 16; ++i) {
+      int Idx = Offset + (i / Scale);
+      PSHUFBMask[i] = DAG.getConstant(
+          (i % Scale == 0 && SafeOffset(Idx)) ? Idx : 0x80, DL, MVT::i8);
+    }
     InputV = DAG.getBitcast(MVT::v16i8, InputV);
     return DAG.getBitcast(VT,
                           DAG.getNode(X86ISD::PSHUFB, DL, MVT::v16i8, InputV,
                                       DAG.getNode(ISD::BUILD_VECTOR, DL,
                                                   MVT::v16i8, PSHUFBMask)));
   }
 
+  // If we are extending from an (odd)offset, shuffle them to the start
+  // before unpacking.
+  if (Offset & 1) {
+    InputV = ShuffleOffset(InputV);
+    Offset = 0;
+  }
+
   // Otherwise emit a sequence of unpacks.
   do {
+    unsigned UnpackLoHi = X86ISD::UNPCKL;
+    if (Offset >= (NumElements / 2)) {
+      UnpackLoHi = X86ISD::UNPCKH;
+      Offset -= (NumElements / 2);
+    }
+
     MVT InputVT = MVT::getVectorVT(MVT::getIntegerVT(EltBits), NumElements);
     SDValue Ext = AnyExt ? DAG.getUNDEF(InputVT)
                          : getZeroVector(InputVT, Subtarget, DAG, DL);
     InputV = DAG.getBitcast(InputVT, InputV);
-    InputV = DAG.getNode(X86ISD::UNPCKL, DL, InputVT, InputV, Ext);
+    InputV = DAG.getNode(UnpackLoHi, DL, InputVT, InputV, Ext);
     Scale /= 2;
     EltBits *= 2;
     NumElements /= 2;
   } while (Scale > 1);
   return DAG.getBitcast(VT, InputV);
 }
 
 /// \brief Try to lower a vector shuffle as a zero extension on any microarch.

 /// The reason we have dedicated lowering for zext-style shuffles is that they
 /// are both incredibly common and often quite performance sensitive.
 static SDValue lowerVectorShuffleAsZeroOrAnyExtend(
     SDLoc DL, MVT VT, SDValue V1, SDValue V2, ArrayRef<int> Mask,
     const X86Subtarget *Subtarget, SelectionDAG &DAG) {
   SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2);
 
   int Bits = VT.getSizeInBits();
+  int NumLanes = Bits / 128;
   int NumElements = VT.getVectorNumElements();
+  int NumLaneElements = NumElements / NumLanes;
   assert(VT.getScalarSizeInBits() <= 32 &&
          "Exceeds 32-bit integer zero extension limit");
   assert((int)Mask.size() == NumElements && "Unexpected shuffle mask size");
 
   // Define a helper function to check a particular ext-scale and lower to it if
   // valid.
   auto Lower = [&](int Scale) -> SDValue {
     SDValue InputV;
     bool AnyExt = true;
+    int Offset = 0;
+    int Matches = 0;
     for (int i = 0; i < NumElements; ++i) {
-      if (Mask[i] == -1)
+      int M = Mask[i];
+      if (M == -1)
         continue; // Valid anywhere but doesn't tell us anything.
       if (i % Scale != 0) {
         // Each of the extended elements need to be zeroable.
         if (!Zeroable[i])
           return SDValue();
 
         // We no longer are in the anyext case.
         AnyExt = false;
         continue;
       }
 
       // Each of the base elements needs to be consecutive indices into the
       // same input vector.
-      SDValue V = Mask[i] < NumElements ? V1 : V2;
-      if (!InputV)
+      SDValue V = M < NumElements ? V1 : V2;
+      M = M % NumElements;
+      if (!InputV) {
         InputV = V;
-      else if (InputV != V)
+        Offset = M - (i / Scale);
+      } else if (InputV != V)
         return SDValue(); // Flip-flopping inputs.
 
-      if (Mask[i] % NumElements != i / Scale)
+      // Offset must start in the lowest 128-bit lane or at the start of an
+      // upper lane.
+      // FIXME: Is it ever worth allowing a -ve base offset?
+      if (!((0 <= Offset && Offset < NumLaneElements) ||
+            (Offset % NumLaneElements) == 0))
+        return SDValue();
+
+      // If we are offsetting, all referenced entries must come from the same
+      // lane.
+      if (Offset && (Offset / NumLaneElements) != (M / NumLaneElements))
+        return SDValue();
+
+      if ((M % NumElements) != (Offset + (i / Scale)))
         return SDValue(); // Non-consecutive strided elements.
+      Matches++;
     }
 
     // If we fail to find an input, we have a zero-shuffle which should always
     // have already been handled.
     // FIXME: Maybe handle this here in case during blending we end up with one?
     if (!InputV)
       return SDValue();
 
+    // If we are offsetting, don't extend if we only match a single input.
+    if (Offset != 0 && Matches < 2)
+      return SDValue();
+
     return lowerVectorShuffleAsSpecificZeroOrAnyExtend(
-        DL, VT, Scale, AnyExt, InputV, Mask, Subtarget, DAG);
+        DL, VT, Scale, Offset, AnyExt, InputV, Mask, Subtarget, DAG);
   };
 
   // The widest scale possible for extending is to a 64-bit integer.
   assert(Bits % 64 == 0 &&
          "The number of bits in a vector must be divisible by 64 on x86!");
   int NumExtElements = Bits / 64;
 
   // Each iteration, try extending the elements half as much, but into twice as
if ((IdxVal == OpVT.getVectorNumElements() / 2) &&
       Vec.getOpcode() == ISD::INSERT_SUBVECTOR &&
       OpVT.is256BitVector() && SubVecVT.is128BitVector()) {
     auto *Idx2 = dyn_cast<ConstantSDNode>(Vec.getOperand(2));
     if (Idx2 && Idx2->getZExtValue() == 0) {
       SDValue SubVec2 = Vec.getOperand(1);
       // If needed, look through a bitcast to get to the load.
       if (SubVec2.getNode() && SubVec2.getOpcode() == ISD::BITCAST)
         SubVec2 = SubVec2.getOperand(0);
-      
       if (auto *FirstLd = dyn_cast<LoadSDNode>(SubVec2)) {
         bool Fast;
         unsigned Alignment = FirstLd->getAlignment();
         unsigned AS = FirstLd->getAddressSpace();
         const X86TargetLowering *TLI = Subtarget->getTargetLowering();
         if (TLI->allowsMemoryAccess(*DAG.getContext(), DAG.getDataLayout(),
                                     OpVT, AS, Alignment, &Fast) && Fast) {
           SDValue Ops[] = { SubVec2, SubVec };

test/CodeGen/X86/machine-cp.ll

 
 ; Check that copy propagation does not kill thing like:
 ; dst = copy src <-- do not kill that.
 ; ... = op1 dst<undef>
 ; ... = op2 dst <-- this is used here.
 ;
 ; CHECK-LABEL: foo:
 ; CHECK: psllw $7,
-; CHECK: psllw $7,
-; CHECK-NEXT: pand
-; CHECK-NEXT: pcmpgtb 
-; CHECK-NEXT: pand %xmm{{[0-9]+}}, [[SRC:%xmm[0-9]+]]
-; Machine propagation used to delete the first copy as the
-; first few uses were <undef>.
-; CHECK-NEXT: movdqa [[SRC]], [[CPY1:%xmm[0-9]+]]
-; CHECK-NEXT: movdqa [[SRC]], [[CPY2:%xmm[0-9]+]]
-; CHECK-NEXT: punpckhbw [[SRC]],
-; Check that CPY1 is not redefined.
-; CHECK-NOT: , [[CPY1]]
-; undef use, we do not care.
-; CHECK: punpcklwd [[CPY1]],
-; Check that CPY1 is not redefined.
-; CHECK-NOT: , [[CPY1]]
-; CHECK: punpcklbw [[CPY2]], [[CPY2]]
-; CHECK-NEXT: punpckhwd [[CPY2]], [[CPY2]]
-; CHECK-NEXT: pslld $31, [[CPY2]]
-; Check that CPY1 is not redefined.
-; CHECK-NOT: , [[CPY1]]
-; CHECK: punpcklbw [[CPY1]], [[CPY1]]
-; CHECK-NEXT: punpcklwd [[CPY1]], [[CPY1]]
-; CHECK-NEXT: pslld $31, [[CPY1]]
+; CHECK: movdqa {{.*}}(%rip), [[CST1:%xmm[0-9]+]]
+; CHECK: movdqa {{.*}}(%rip), [[CST2:%xmm[0-9]+]]
+; CHECK: psllw $7, [[SRC1:%xmm[0-9]+]]
+; CHECK-NEXT: pand {{.*}}(%rip), [[SRC1]]
+; CHECK-NEXT: pcmpgtb [[SRC1]], [[SRC2:%xmm[0-9]+]]
+; CHECK-NEXT: pand %xmm{{[0-9]+}}, [[SRC2]]
+; CHECK-NEXT: movdqa [[SRC2]], [[CPY1:%xmm[0-9]+]]
+; CHECK-NEXT: punpckhbw [[CST2]], [[SRC2]]
+; CHECK-NEXT: movdqa [[SRC2]], [[CPY2:%xmm[0-9]+]]
+; CHECK-NEXT: punpckhwd [[CST2]], [[SRC2]]
+; CHECK-NEXT: pslld $31, [[SRC2]]
+; CHECK-NEXT: psrad $31, [[SRC2]]
+; CHECK-NEXT: pxor [[CST1]], [[SRC2]]
+; CHECK-NEXT: punpcklbw [[CST2]], [[CPY1]]
+; CHECK-NEXT: punpcklwd [[CST2]], [[CPY2]]
+; Check that CPY2 is not redefined.
+; CHECK-NOT:  , [[CPY2]]
+; CHECK:      pslld $31, [[CPY2]]
+; CHECK-NEXT: psrad $31, [[CPY2]]
+; CHECK:      pslld $31, [[CPY1]]
+; CHECK-NEXT: psrad $31, [[CPY1]]
 define <16 x float> @foo(<16 x float> %x) {
 bb:
   %v3 = icmp slt <16 x i32> undef, zeroinitializer
   %v14 = zext <16 x i1> %v3 to <16 x i32>
   %v16 = fcmp olt <16 x float> %x, zeroinitializer
   %v17 = sext <16 x i1> %v16 to <16 x i32>
   %v18 = zext <16 x i1> %v16 to <16 x i32>
   %v19 = xor <16 x i32> %v14, %v18

test/CodeGen/X86/vec_cast2.ll

 ; RUN: llc < %s -mtriple=i386-apple-darwin10 -mcpu=corei7-avx -mattr=+avx | FileCheck %s
 ; RUN: llc < %s -mtriple=i386-apple-darwin10 -mcpu=corei7-avx -mattr=+avx -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=CHECK-WIDE
 
 define <8 x float> @foo1_8(<8 x i8> %src) {
 ; CHECK-LABEL: foo1_8:
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm0[4,4,5,5,6,6,7,7]
 ; CHECK-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; CHECK-NEXT:    vpslld $24, %xmm0, %xmm0
 ; CHECK-NEXT:    vpsrad $24, %xmm0, %xmm0
 ; CHECK-NEXT:    vpslld $24, %xmm1, %xmm1
 ; CHECK-NEXT:    vpsrad $24, %xmm1, %xmm1
 ; CHECK-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0

 ; CHECK-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; CHECK-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; CHECK-NEXT:    vcvtdq2ps %ymm0, %ymm0
 ; CHECK-NEXT:    retl
 ;
 ; CHECK-WIDE-LABEL: foo2_8:
 ; CHECK-WIDE:       ## BB#0:
 ; CHECK-WIDE-NEXT:    vpmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; CHECK-WIDE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; CHECK-WIDE-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; CHECK-WIDE-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; CHECK-WIDE-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; CHECK-WIDE-NEXT:    vcvtdq2ps %ymm0, %ymm0
 ; CHECK-WIDE-NEXT:    retl
   %res = uitofp <8 x i8> %src to <8 x float>
   ret <8 x float> %res
 }
 
 define <4 x float> @foo2_4(<4 x i8> %src) {

test/CodeGen/X86/vec_int_to_fp.ll

; AVX2-NEXT:    retq
   %cvt = uitofp <8 x i16> %a to <8 x float>
   ret <8 x float> %cvt
 }
 
 define <8 x float> @uitofp_8i8_to_8f32(<16 x i8> %a) {
 ; SSE-LABEL: uitofp_8i8_to_8f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; SSE-NEXT:    movdqa %xmm0, %xmm2
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
 ; SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
 ; SSE-NEXT:    cvtdq2ps %xmm2, %xmm2
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; SSE-NEXT:    cvtdq2ps %xmm0, %xmm1
 ; SSE-NEXT:    movaps %xmm2, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: uitofp_8i8_to_8f32:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
; AVX2-NEXT:    retq
   %cvt = uitofp <8 x i8> %shuf to <8 x float>
   ret <8 x float> %cvt
 }
 
 define <8 x float> @uitofp_16i8_to_8f32(<16 x i8> %a) {
 ; SSE-LABEL: uitofp_16i8_to_8f32:
 ; SSE:       # BB#0:
 ; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; SSE-NEXT:    movdqa %xmm0, %xmm2
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
 ; SSE-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
 ; SSE-NEXT:    cvtdq2ps %xmm2, %xmm2
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; SSE-NEXT:    cvtdq2ps %xmm0, %xmm1
 ; SSE-NEXT:    movaps %xmm2, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: uitofp_16i8_to_8f32:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2

test/CodeGen/X86/vector-shuffle-256-v16.ll

 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> zeroinitializer, <16 x i32> <i32 2, i32 3, i32 16, i32 16, i32 6, i32 7, i32 16, i32 16, i32 10, i32 11, i32 16, i32 16, i32 14, i32 15, i32 16, i32 16>
   ret <16 x i16> %shuffle
 }
 
 define <16 x i16> @shuffle_v16i16_16_zz_zz_zz_17_zz_zz_zz_18_zz_zz_zz_19_zz_zz_zz(<16 x i16> %a) {
 ; AVX1-LABEL: shuffle_v16i16_16_zz_zz_zz_17_zz_zz_zz_18_zz_zz_zz_19_zz_zz_zz:
 ; AVX1:       # BB#0:
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,5,2,3,4,5,6,7,6,7,10,11,4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
 ; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_16_zz_zz_zz_17_zz_zz_zz_18_zz_zz_zz_19_zz_zz_zz:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <16 x i16> zeroinitializer, <16 x i16> %a, <16 x i32> <i32 16, i32 0, i32 0, i32 0, i32 17, i32 0, i32 0, i32 0, i32 18, i32 0, i32 0, i32 0, i32 19, i32 0, i32 0, i32 0>
   ret <16 x i16> %shuffle

test/CodeGen/X86/vector-shuffle-256-v32.ll

 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX --check-prefix=AVX2
 
 target triple = "x86_64-unknown-unknown"
 
 define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00(<32 x i8> %a, <32 x i8> %b) {
 ; AVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX2-NEXT:    retq
   %shuffle = shufflevector <32 x i8> %a, <32 x i8> zeroinitializer, <32 x i32> <i32 7, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 15, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 23, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>
   ret <32 x i8> %shuffle
 }
 
 define <32 x i8> @shuffle_v32i8_32_zz_zz_zz_zz_zz_zz_zz_33_zz_zz_zz_zz_zz_zz_zz_34_zz_zz_zz_zz_zz_zz_zz_35_zz_zz_zz_zz_zz_zz_zz(<32 x i8> %a) {
 ; AVX1-LABEL: shuffle_v32i8_32_zz_zz_zz_zz_zz_zz_zz_33_zz_zz_zz_zz_zz_zz_zz_34_zz_zz_zz_zz_zz_zz_zz_35_zz_zz_zz_zz_zz_zz_zz:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrld $16, %xmm0, %xmm0
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_32_zz_zz_zz_zz_zz_zz_zz_33_zz_zz_zz_zz_zz_zz_zz_34_zz_zz_zz_zz_zz_zz_zz_35_zz_zz_zz_zz_zz_zz_zz:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    retq
 
   %shuffle = shufflevector <32 x i8> zeroinitializer, <32 x i8> %a, <32 x i32> <i32 32, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 33, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 34, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 35, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <32 x i8> %shuffle
 }
 
 define <32 x i8> @shuffle_v32i8_32_zz_zz_zz_33_zz_zz_zz_34_zz_zz_zz_35_zz_zz_zz_36_zz_zz_zz_37_zz_zz_zz_38_zz_zz_zz_39_zz_zz_zz(<32 x i8> %a) {
 ; AVX1-LABEL: shuffle_v32i8_32_zz_zz_zz_33_zz_zz_zz_34_zz_zz_zz_35_zz_zz_zz_36_zz_zz_zz_37_zz_zz_zz_38_zz_zz_zz_39_zz_zz_zz:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_32_zz_zz_zz_33_zz_zz_zz_34_zz_zz_zz_35_zz_zz_zz_36_zz_zz_zz_37_zz_zz_zz_38_zz_zz_zz_39_zz_zz_zz:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
 ; AVX2-NEXT:    retq
   %shuffle = shufflevector <32 x i8> zeroinitializer, <32 x i8> %a, <32 x i32> <i32 32, i32 0, i32 0, i32 0, i32 33, i32 0, i32 0, i32 0, i32 34, i32 0, i32 0, i32 0, i32 35, i32 0, i32 0, i32 0, i32 36, i32 0, i32 0, i32 0, i32 37, i32 0, i32 0, i32 0, i32 38, i32 0, i32 0, i32 0, i32 39, i32 0, i32 0, i32 0>

test/CodeGen/X86/vector-shuffle-sse4a.ll

 ; BTVER2-LABEL: shuf_0zzzzzzz1zzzzzzz:
 ; BTVER2:       # BB#0:
 ; BTVER2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
 ; BTVER2-NEXT:    retq
   %s = shufflevector <16 x i8> %a0, <16 x i8> zeroinitializer, <16 x i32> <i32 0, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 1, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
   ret <16 x i8> %s
 }
 
+define <16 x i8> @shuf_2zzzzzzz3zzzzzzz(<16 x i8> %a0) {
+; BTVER1-LABEL: shuf_2zzzzzzz3zzzzzzz:
+; BTVER1:       # BB#0:
+; BTVER1-NEXT:    movaps %xmm0, %xmm1
+; BTVER1-NEXT:    extrq {{.*#+}} xmm1 = xmm1[3],zero,zero,zero,zero,zero,zero,zero,xmm1[u,u,u,u,u,u,u,u]
+; BTVER1-NEXT:    extrq {{.*#+}} xmm0 = xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[u,u,u,u,u,u,u,u]
+; BTVER1-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; BTVER1-NEXT:    retq
+;
+; BTVER2-LABEL: shuf_2zzzzzzz3zzzzzzz:
+; BTVER2:       # BB#0:
+; BTVER2-NEXT:    vpsrld $16, %xmm0, %xmm0
+; BTVER2-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; BTVER2-NEXT:    retq
+  %s = shufflevector <16 x i8> %a0, <16 x i8> zeroinitializer, <16 x i32> <i32 2, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 3, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
+  ret <16 x i8> %s
+}
+
 define <16 x i8> @shuf_01zzuuuuuuuuuuuu(<16 x i8> %a0) {
 ; BTVER1-LABEL: shuf_01zzuuuuuuuuuuuu:
 ; BTVER1:       # BB#0:
 ; BTVER1-NEXT:    extrq {{.*#+}} xmm0 = xmm0[0,1],zero,zero,zero,zero,zero,zero,xmm0[u,u,u,u,u,u,u,u]
 ; BTVER1-NEXT:    retq
 ;
 ; BTVER2-LABEL: shuf_01zzuuuuuuuuuuuu:
 ; BTVER2:       # BB#0:

test/CodeGen/X86/vector-zext.ll

 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: zext_16i8_to_16i16:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero
 ; AVX2-NEXT:    retq
 entry:
   %B = zext <16 x i8> %A to <16 x i16>
   ret <16 x i16> %B
 }
 
 define <4 x i32> @zext_16i8_to_4i32(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: zext_16i8_to_4i32:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    pxor %xmm1, %xmm1
entry:
   ret <4 x i32> %C
 }
 
 define <8 x i32> @zext_16i8_to_8i32(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: zext_16i8_to_8i32:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; 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],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; 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: zext_16i8_to_8i32:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; 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:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
+; 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: zext_16i8_to_8i32:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm0 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
+; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: zext_16i8_to_8i32:
 ; AVX1:       # BB#0: # %entry
 ; AVX1-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
 ; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,255,0,0,0,255,0,0,0,255,0,0,0]
 ; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
entry:
   %B = shufflevector <16 x i8> %A, <16 x i8> undef, <2 x i32> <i32 0, i32 1>
   %C = zext <2 x i8> %B to <2 x i64>
   ret <2 x i64> %C
 }
 
 define <4 x i64> @zext_16i8_to_4i64(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: zext_16i8_to_4i64:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    movdqa %xmm0, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,7,5,6,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %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: zext_16i8_to_4i64:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; 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: zext_16i8_to_4i64:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[2],zero,zero,zero,zero,zero,zero,zero,xmm1[3],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    psrld $16, %xmm0
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: zext_16i8_to_4i64:
 ; AVX1:       # BB#0: # %entry
 ; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
 ; AVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
entry:
   %B = shufflevector <8 x i16> %A, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
   %C = zext <2 x i16> %B to <2 x i64>
   ret <2 x i64> %C
 }
 
 define <4 x i64> @zext_8i16_to_4i64(<8 x i16> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: zext_8i16_to_4i64:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,2,1]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,5,6,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; SSE2-NEXT:    movdqa %xmm0, %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: zext_8i16_to_4i64:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    pxor %xmm2, %xmm2
-; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; 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:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[4,5],zero,zero,zero,zero,zero,zero,xmm1[6,7],zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm1 = xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: zext_8i16_to_4i64:
 ; SSE41:       # BB#0: # %entry
 ; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[4,5,2,3,4,5,6,7,6,7,10,11,4,5,6,7]
-; SSE41-NEXT:    pxor %xmm1, %xmm1
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: zext_8i16_to_4i64:
 ; AVX1:       # BB#0: # %entry
 ; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2

 ; 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[2,2,3,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
 ; SSE2-NEXT:    pand %xmm2, %xmm1
 ; 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]

 ; 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:    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[4,4,5,5,6,6,7,7]
+; 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:    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:    movdqa %xmm1, %xmm0

 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[2,2,3,3]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
 ; SSE2-NEXT:    pand %xmm2, %xmm1
 ; 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:    movdqa %xmm1, %xmm0

 
 define <8 x i32> @shuf_zext_8i8_to_8i32(<8 x i8> %A) {
 ; SSE2-LABEL: shuf_zext_8i8_to_8i32:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1
 ; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
 ; SSE2-NEXT:    packuswb %xmm1, %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:    movdqa %xmm1, %xmm0
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    punpckhwd {{.*#+}} xmm1 = xmm1[4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm1
+; 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: shuf_zext_8i8_to_8i32:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; 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:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
 ; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
+; 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: shuf_zext_8i8_to_8i32:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm0 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero
+; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
+; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm2 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: shuf_zext_8i8_to_8i32:
 ; AVX1:       # BB#0: # %entry
 ; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; AVX1-NEXT:    vpmovzxbd {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuf_zext_8i8_to_8i32:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vpmovzxbd {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero
 ; AVX2-NEXT:    retq
 entry:
   %B = shufflevector <8 x i8> %A, <8 x i8> zeroinitializer, <32 x i32> <i32 0, i32 8, i32 8, i32 8, i32 1, i32 8, i32 8, i32 8, i32 2, i32 8, i32 8, i32 8, i32 3, i32 8, i32 8, i32 8, i32 4, i32 8, i32 8, i32 8, i32 5, i32 8, i32 8, i32 8, i32 6, i32 8, i32 8, i32 8, i32 7, i32 8, i32 8, i32 8>
   %Z = bitcast <32 x i8> %B to <8 x i32>
   ret <8 x i32> %Z
 }
 
 define <2 x i64> @shuf_zext_16i8_to_2i64_offset6(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: shuf_zext_16i8_to_2i64_offset6:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE2-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuf_zext_16i8_to_2i64_offset6:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuf_zext_16i8_to_2i64_offset6:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    psrlq $48, %xmm0
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: shuf_zext_16i8_to_2i64_offset6:
 ; AVX:       # BB#0: # %entry
-; AVX-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[6],zero,zero,zero,zero,zero,zero,zero,xmm0[7],zero,zero,zero,zero,zero,zero,zero
+; AVX-NEXT:    vpsrlq $48, %xmm0, %xmm0
+; AVX-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX-NEXT:    retq
 entry:
   %B = shufflevector <16 x i8> %A, <16 x i8> zeroinitializer, <16 x i32> <i32 6, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 7, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
   %Z = bitcast <16 x i8> %B to <2 x i64>
   ret <2 x i64> %Z
 }
 
 define <4 x i64> @shuf_zext_16i8_to_4i64_offset11(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: shuf_zext_16i8_to_4i64_offset11:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm2 = xmm0[3,1,2,3,4,5,6,7]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,0,0,0,0,0,0,255,0,0,0,0,0,0,0]
-; SSE2-NEXT:    pand %xmm3, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,6,5,6,7]
-; SSE2-NEXT:    pand %xmm3, %xmm1
-; SSE2-NEXT:    movdqa %xmm2, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; 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:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuf_zext_16i8_to_4i64_offset11:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[11],zero,zero,zero,zero,zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[13],zero,zero,zero,zero,zero,zero,zero,xmm1[14],zero,zero,zero,zero,zero,zero,zero
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuf_zext_16i8_to_4i64_offset11:
 ; SSE41:       # BB#0: # %entry
 ; SSE41-NEXT:    movdqa %xmm0, %xmm1
-; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[11],zero,zero,zero,zero,zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero
-; SSE41-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[13],zero,zero,zero,zero,zero,zero,zero,xmm1[14],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    pmovzxbq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: shuf_zext_16i8_to_4i64_offset11:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[11],zero,zero,zero,zero,zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[13],zero,zero,zero,zero,zero,zero,zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm1 = xmm0[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX1-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuf_zext_16i8_to_4i64_offset11:
 ; AVX2:       # BB#0: # %entry
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[11],zero,zero,zero,zero,zero,zero,zero,xmm0[12],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[13],zero,zero,zero,zero,zero,zero,zero,xmm0[14],zero,zero,zero,zero,zero,zero,zero
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpsrldq {{.*#+}} ymm0 = ymm0[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[27,28,29,30,31],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-NEXT:    vpmovzxbq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero,xmm0[2],zero,zero,zero,zero,zero,zero,zero,xmm0[3],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    retq
 entry:
   %B = shufflevector <16 x i8> %A, <16 x i8> zeroinitializer, <32 x i32> <i32 11, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 12, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 13, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 14, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
   %Z = bitcast <32 x i8> %B to <4 x i64>
   ret <4 x i64> %Z
 }
 
 define <2 x i64> @shuf_zext_8i16_to_2i64_offset6(<8 x i16> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: shuf_zext_8i16_to_2i64_offset6:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pand {{.*}}(%rip), %xmm0
+; SSE2-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuf_zext_8i16_to_2i64_offset6:
 ; SSSE3:       # BB#0: # %entry
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[6,7],zero,zero,zero,zero,zero,zero,xmm0[8,9],zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pxor %xmm1, %xmm1
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSSE3-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuf_zext_8i16_to_2i64_offset6:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[3,1,2,3,4,5,6,7]
-; SSE41-NEXT:    pxor %xmm0, %xmm0
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2,3],xmm1[4],xmm0[5,6,7]
+; SSE41-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; SSE41-NEXT:    retq
 ;
 ; AVX-LABEL: shuf_zext_8i16_to_2i64_offset6:
 ; AVX:       # BB#0: # %entry
-; AVX-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; AVX-NEXT:    retq
 entry:
   %B = shufflevector <8 x i16> %A, <8 x i16> zeroinitializer, <8 x i32> <i32 3, i32 8, i32 8, i32 8, i32 4, i32 8, i32 8, i32 8>
   %Z = bitcast <8 x i16> %B to <2 x i64>
   ret <2 x i64> %Z
 }
 
 define <4 x i64> @shuf_zext_8i16_to_4i64_offset2(<8 x i16> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: shuf_zext_8i16_to_4i64_offset2:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,1,2,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,7]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [65535,0,0,0,65535,0,0,0]
-; SSE2-NEXT:    pand %xmm2, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,7]
-; SSE2-NEXT:    pand %xmm2, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    punpckhdq {{.*#+}} xmm0 = 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:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: shuf_zext_8i16_to_4i64_offset2:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movdqa %xmm0, %xmm1
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[4,5],zero,zero,zero,zero,zero,zero,xmm0[6,7],zero,zero,zero,zero,zero,zero
-; SSSE3-NEXT:    pshufb {{.*#+}} xmm1 = xmm1[8,9],zero,zero,zero,zero,zero,zero,xmm1[10,11],zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pxor %xmm2, %xmm2
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSSE3-NEXT:    punpckhdq {{.*#+}} xmm0 = 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:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: shuf_zext_8i16_to_4i64_offset2:
 ; SSE41:       # BB#0: # %entry
-; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,1,2,3]
-; SSE41-NEXT:    pshufb {{.*#+}} xmm0 = xmm0[4,5,2,3,4,5,6,7,6,7,10,11,4,5,6,7]
-; SSE41-NEXT:    pxor %xmm2, %xmm2
-; SSE41-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
-; SSE41-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,7]
-; SSE41-NEXT:    pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; SSE41-NEXT:    pmovzxwq {{.*#+}} xmm1 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
+; SSE41-NEXT:    movdqa %xmm2, %xmm0
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: shuf_zext_8i16_to_4i64_offset2:
 ; AVX1:       # BB#0: # %entry
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,5,2,3,4,5,6,7,6,7,10,11,4,5,6,7]
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
-; AVX1-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1,2,3],xmm0[4,5,6,7]
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,0,1,0,1,0,1,10,11,0,1,0,1,0,1]
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; AVX1-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuf_zext_8i16_to_4i64_offset2:
 ; AVX2:       # BB#0: # %entry
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[4,5,2,3,4,5,6,7,6,7,10,11,4,5,6,7]
-; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
-; AVX2-NEXT:    vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3]
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[8,9,0,1,0,1,0,1,10,11,0,1,0,1,0,1]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[1,2,2,3,5,6,6,7]
+; AVX2-NEXT:    vpmovzxwq {{.*#+}} ymm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero
 ; AVX2-NEXT:    retq
 entry:
   %B = shufflevector <8 x i16> %A, <8 x i16> zeroinitializer, <16 x i32> <i32 2, i32 8, i32 8, i32 8, i32 3, i32 8, i32 8, i32 8, i32 4, i32 8, i32 8, i32 8, i32 5, i32 8, i32 8, i32 8>
   %Z = bitcast <16 x i16> %B to <4 x i64>
   ret <4 x i64> %Z
 }
 
 define <4 x i32> @shuf_zext_8i16_to_4i32_offset1(<8 x i16> %A) nounwind uwtable readnone ssp {

 ; AVX-NEXT:    retq
 entry:
   %B = shufflevector <8 x i16> %A, <8 x i16> zeroinitializer, <8 x i32> <i32 1, i32 8, i32 2, i32 8, i32 3, i32 8, i32 4, i32 8>
   %Z = bitcast <8 x i16> %B to <4 x i32>
   ret <4 x i32> %Z
 }
 
 define <8 x i32> @shuf_zext_8i16_to_8i32_offset3(<8 x i16> %A) nounwind uwtable readnone ssp {
-; SSE-LABEL: shuf_zext_8i16_to_8i32_offset3:
-; SSE:       # BB#0: # %entry
-; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
-; SSE-NEXT:    pxor %xmm2, %xmm2
-; SSE-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT:    retq
+; SSE2-LABEL: shuf_zext_8i16_to_8i32_offset3:
+; SSE2:       # BB#0: # %entry
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSE2-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE2-NEXT:    retq
+;
+; SSSE3-LABEL: shuf_zext_8i16_to_8i32_offset3:
+; SSSE3:       # BB#0: # %entry
+; SSSE3-NEXT:    movdqa %xmm0, %xmm1
+; SSSE3-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    pxor %xmm2, %xmm2
+; SSSE3-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
+; SSSE3-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSSE3-NEXT:    retq
+;
+; SSE41-LABEL: shuf_zext_8i16_to_8i32_offset3:
+; SSE41:       # BB#0: # %entry
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE41-NEXT:    pxor %xmm2, %xmm2
+; SSE41-NEXT:    punpckhwd {{.*#+}} xmm0 = xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
+; SSE41-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: shuf_zext_8i16_to_8i32_offset3:
 ; AVX1:       # BB#0: # %entry
 ; AVX1-NEXT:    vpslldq {{.*#+}} xmm1 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
 ; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX1-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
 ; AVX1-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuf_zext_8i16_to_8i32_offset3:
 ; AVX2:       # BB#0: # %entry
 ; AVX2-NEXT:    vpslldq {{.*#+}} xmm1 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
 ; AVX2-NEXT:    vpsrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 entry:
   %B = shufflevector <8 x i16> %A, <8 x i16> zeroinitializer, <16 x i32> <i32 3, i32 8, i32 4, i32 8, i32 5, i32 8, i32 6, i32 8, i32 7, i32 8, i32 undef, i32 8, i32 undef, i32 8, i32 undef, i32 8>
   %Z = bitcast <16 x i16> %B to <8 x i32>
   ret <8 x i32> %Z
 }
 
+define <8 x i32> @shuf_zext_16i16_to_8i32_offset8(<16 x i16> %A) nounwind uwtable readnone ssp {
+; SSE2-LABEL: shuf_zext_16i16_to_8i32_offset8:
+; SSE2:       # BB#0: # %entry
+; SSE2-NEXT:    pxor %xmm2, %xmm2
+; SSE2-NEXT:    movdqa %xmm1, %xmm0
+; 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: shuf_zext_16i16_to_8i32_offset8:
+; SSSE3:       # BB#0: # %entry
+; SSSE3-NEXT:    pxor %xmm2, %xmm2
+; SSSE3-NEXT:    movdqa %xmm1, %xmm0
+; 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: shuf_zext_16i16_to_8i32_offset8:
+; SSE41:       # BB#0: # %entry
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,2,3,3]
+; SSE41-NEXT:    pxor %xmm2, %xmm2
+; SSE41-NEXT:    pblendw {{.*#+}} xmm2 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7]
+; SSE41-NEXT:    pmovzxwd {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
+; SSE41-NEXT:    movdqa %xmm2, %xmm1
+; SSE41-NEXT:    retq
+;
+; AVX1-LABEL: shuf_zext_16i16_to_8i32_offset8:
+; AVX1:       # BB#0: # %entry
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
+; AVX1-NEXT:    vpxor %xmm2, %xmm2
+; AVX1-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
+; AVX1-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: shuf_zext_16i16_to_8i32_offset8:
+; AVX2:       # BB#0: # %entry
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm0
+; AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX2-NEXT:    vpxor %ymm1, %ymm1, %ymm1
+; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7],ymm0[8],ymm1[9],ymm0[10],ymm1[11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
+; AVX2-NEXT:    retq
+entry:
+  %B = shufflevector <16 x i16> %A, <16 x i16> zeroinitializer, <16 x i32> <i32 8, i32 16, i32 9, i32 16, i32 10, i32 16, i32 11, i32 16, i32 12, i32 16, i32 undef, i32 16, i32 14, i32 16, i32 undef, i32 16>
+  %Z = bitcast <16 x i16> %B to <8 x i32>
+  ret <8 x i32> %Z
+}
+
 define <2 x i64> @shuf_zext_4i32_to_2i64_offset2(<4 x i32> %A) nounwind uwtable readnone ssp {
 ; SSE-LABEL: shuf_zext_4i32_to_2i64_offset2:
 ; SSE:       # BB#0: # %entry
 ; SSE-NEXT:    pxor %xmm1, %xmm1
 ; SSE-NEXT:    punpckhdq {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: shuf_zext_4i32_to_2i64_offset2: