[X86] Enable shuffle combining for AVX512 unless the root is used by a vselect

[craig.topper]

Summary

A lot of vectorized code doesn't use masks so we shouldn't penalize them by not doing shuffle combining on avx512 targets.

I've added support for VALIGNQ/VALIGND and 512-bit SHUF128 to prevent some regressions. I also prevented recombining 256-bit SHUF128 to PERM2X128. We may not need to add 256-bit SHUF128 support, but I don't think I found any cases requiring that in my testing.

[RKSimon]

Thanks for doing this - I've been putting it off for a long time!

llvm/lib/Target/X86/X86ISelLowering.cpp

33845

// Attempt to match against VALIGN element rotate.

Also - why not merge the 32/64 bit checks?

33849

Use the isAnyZero helper? (It's recent so isn't used everywhere it could be yet).

34071

Do we need to peek through bitcasts at all?

34163

It might be better to use BaseMask directly - its already dealt with the widening so the mask matching should be easier?

llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll

240

Maybe rename the test? combine_vpermi2var_32i16_as_rotate ? If you want to you could adjust the masks so this (or a new test) still combines to pshufb.

[craig.topper]

llvm/lib/Target/X86/X86ISelLowering.cpp

34071

Maybe, but then element count checks at some of the uses of IsMaskedShuffle would also need to be adjusted? Since NumRootElts wouldn't reflect the bitcast.

34163

That would simplify things if we were looking for a 256-bit wide "element". But we're only looking for repeated lane so there's no guarantee the mask can widened to 2 elements.

[craig.topper]

Address comments

[RKSimon]

LGTM

llvm/lib/Target/X86/X86ISelLowering.cpp

 static bool matchBinaryPermuteShuffle(
     MVT MaskVT, ArrayRef<int> Mask, const APInt &Zeroable,
     bool AllowFloatDomain, bool AllowIntDomain, SDValue &V1, SDValue &V2,
     const SDLoc &DL, SelectionDAG &DAG, const X86Subtarget &Subtarget,
     unsigned &Shuffle, MVT &ShuffleVT, unsigned &PermuteImm) {
   unsigned NumMaskElts = Mask.size();
   unsigned EltSizeInBits = MaskVT.getScalarSizeInBits();
 
+  // Attempt to match against VALIGND/VALIGNQ rotate.
+  if (AllowIntDomain && (EltSizeInBits == 64 || EltSizeInBits == 32) &&
+      ((MaskVT.is128BitVector() && Subtarget.hasVLX()) ||
+       (MaskVT.is256BitVector() && Subtarget.hasVLX()) ||
+       (MaskVT.is512BitVector() && Subtarget.hasAVX512()))) {
+    if (!isAnyZero(Mask)) {
+      int Rotation = matchShuffleAsElementRotate(V1, V2, Mask);
+      if (0 < Rotation) {
+        Shuffle = X86ISD::VALIGN;
+        if (EltSizeInBits == 64)
+          ShuffleVT = MVT::getVectorVT(MVT::i64, MaskVT.getSizeInBits() / 64);
+        else
+          ShuffleVT = MVT::getVectorVT(MVT::i32, MaskVT.getSizeInBits() / 32);
+        PermuteImm = Rotation;
+        return true;
+      }
+    }
+  }
+
   // Attempt to match against PALIGNR byte rotate.
   if (AllowIntDomain && ((MaskVT.is128BitVector() && Subtarget.hasSSSE3()) ||
                          (MaskVT.is256BitVector() && Subtarget.hasAVX2()) ||
                          (MaskVT.is512BitVector() && Subtarget.hasBWI()))) {
     int ByteRotation = matchShuffleAsByteRotate(MaskVT, V1, V2, Mask);
     if (0 < ByteRotation) {
       Shuffle = X86ISD::PALIGNR;
       ShuffleVT = MVT::getVectorVT(MVT::i8, MaskVT.getSizeInBits() / 8);

   bool FloatDomain = VT1.isFloatingPoint() || VT2.isFloatingPoint() ||
                      (RootVT.isFloatingPoint() && Depth >= 1) ||
                      (RootVT.is256BitVector() && !Subtarget.hasAVX2());
 
   // Don't combine if we are a AVX512/EVEX target and the mask element size
   // is different from the root element size - this would prevent writemasks
   // from being reused.
   // TODO - this currently prevents all lane shuffles from occurring.
-  // TODO - check for writemasks usage instead of always preventing combining.
   // TODO - attempt to narrow Mask back to writemask size.
-  bool IsEVEXShuffle =
-      RootSizeInBits == 512 || (Subtarget.hasVLX() && RootSizeInBits >= 128);
+  bool IsMaskedShuffle = false;
+  if (RootSizeInBits == 512 || (Subtarget.hasVLX() && RootSizeInBits >= 128)) {
+    if (Root.hasOneUse() && Root->use_begin()->getOpcode() == ISD::VSELECT &&
+        Root->use_begin()->getOperand(0).getScalarValueSizeInBits() == 1) {
+      IsMaskedShuffle = true;
+    }
+  }
 
   // Attempt to match a subvector broadcast.
   // shuffle(insert_subvector(undef, sub, 0), undef, 0, 0, 0, 0)
   if (UnaryShuffle &&
       (BaseMaskEltSizeInBits == 128 || BaseMaskEltSizeInBits == 256)) {
     SmallVector<int, 64> BroadcastMask(NumBaseMaskElts, 0);
     if (isTargetShuffleEquivalent(BaseMask, BroadcastMask)) {
       SDValue Src = Inputs[0];
       if (Src.getOpcode() == ISD::INSERT_SUBVECTOR &&
           Src.getOperand(0).isUndef() &&
           Src.getOperand(1).getValueSizeInBits() == BaseMaskEltSizeInBits &&
           MayFoldLoad(Src.getOperand(1)) && isNullConstant(Src.getOperand(2))) {
         return DAG.getBitcast(RootVT, DAG.getNode(X86ISD::SUBV_BROADCAST, DL,
                                                   Src.getValueType(),
                                                   Src.getOperand(1)));

       Res = DAG.getBitcast(ShuffleVT, V1);
       unsigned SubIdx = BaseMask[0] * (8 / NumBaseMaskElts);
       bool UseZero = isAnyZero(BaseMask);
       Res = extractSubVector(Res, SubIdx, DAG, DL, BaseMaskEltSizeInBits);
       Res = widenSubVector(Res, UseZero, Subtarget, DAG, DL, RootSizeInBits);
       return DAG.getBitcast(RootVT, Res);
     }
 
-    // TODO - handle AVX512 cases with X86ISD::SHUF128.
+    SmallVector<int, 4> Mask;
+    if (BaseMaskEltSizeInBits > 128) {
+      assert((BaseMaskEltSizeInBits % 128) == 0 && "Illegal mask size");
+      int MaskScale = BaseMaskEltSizeInBits / 128;
+      narrowShuffleMaskElts(MaskScale, BaseMask, Mask);
+    } else {
+      Mask.assign(BaseMask.begin(), BaseMask.end());
+    }
+
+    // Try to lower to vshuf64x2/vshuf32x4.
+    auto MatchSHUF128 = [](MVT ShuffleVT, const SDLoc &DL, ArrayRef<int> Mask,
+                           SDValue V1, SDValue V2, SelectionDAG &DAG) {
+      unsigned PermMask = 0;
+      // Insure elements came from the same Op.
+      SDValue Ops[2] = {DAG.getUNDEF(ShuffleVT), DAG.getUNDEF(ShuffleVT)};
+      for (int i = 0; i < 4; ++i) {
+        assert(Mask[i] >= -1 && "Illegal shuffle sentinel value");
+        if (Mask[i] < 0)
+          continue;
+
+        SDValue Op = Mask[i] >= 4 ? V2 : V1;
+        unsigned OpIndex = i / 2;
+        if (Ops[OpIndex].isUndef())
+          Ops[OpIndex] = Op;
+        else if (Ops[OpIndex] != Op)
+          return SDValue();
+
+        // Convert the 128-bit shuffle mask selection values into 128-bit
+        // selection bits defined by a vshuf64x2 instruction's immediate control
+        // byte.
+        PermMask |= (Mask[i] % 4) << (i * 2);
+      }
+
+      return DAG.getNode(X86ISD::SHUF128, DL, ShuffleVT,
+                         DAG.getBitcast(ShuffleVT, Ops[0]),
+                         DAG.getBitcast(ShuffleVT, Ops[1]),
+                         DAG.getTargetConstant(PermMask, DL, MVT::i8));
+    };
+
+    // FIXME: Is there a better way to do this? is256BitLaneRepeatedShuffleMask
+    // doesn't work because our mask is for 128 bits and we don't have an MVT
+    // to match that.
+    bool PreferPERMQ =
+        UnaryShuffle && isUndefOrInRange(Mask[0], 0, 2) &&
+        isUndefOrInRange(Mask[1], 0, 2) && isUndefOrInRange(Mask[2], 2, 4) &&
+        isUndefOrInRange(Mask[3], 2, 4) &&
+        (Mask[0] < 0 || Mask[2] < 0 || Mask[0] == (Mask[2] % 2)) &&
+        (Mask[1] < 0 || Mask[3] < 0 || Mask[1] == (Mask[3] % 2));
+
+    if (!isAnyZero(Mask) && !PreferPERMQ) {
+      if (SDValue V = MatchSHUF128(ShuffleVT, DL, Mask, V1, V2, DAG))
+        return DAG.getBitcast(RootVT, V);
+    }
   }
 
   // Handle 128-bit lane shuffles of 256-bit vectors.
   if (RootVT.is256BitVector() && NumBaseMaskElts == 2) {
     MVT ShuffleVT = (FloatDomain ? MVT::v4f64 : MVT::v4i64);
 
     // If the upper half is zeroable, then an extract+insert is more optimal
     // than using X86ISD::VPERM2X128. The insertion is free, even if it has to
     // zero the upper half.
     if (isUndefOrZero(BaseMask[1])) {
       if (Depth == 0 && Root.getOpcode() == ISD::INSERT_SUBVECTOR)
         return SDValue(); // Nothing to do!
       assert(isInRange(BaseMask[0], 0, 2) && "Unexpected lane shuffle");
       Res = DAG.getBitcast(ShuffleVT, V1);

 
       Res = DAG.getBitcast(ShuffleVT, V1);
       Res = DAG.getNode(X86ISD::VPERM2X128, DL, ShuffleVT, Res,
                         DAG.getUNDEF(ShuffleVT),
                         DAG.getTargetConstant(PermMask, DL, MVT::i8));
       return DAG.getBitcast(RootVT, Res);
     }
 
+    if (Depth == 0 && Root.getOpcode() == X86ISD::SHUF128)
+      return SDValue(); // Nothing to do!
+
     // TODO - handle AVX512VL cases with X86ISD::SHUF128.
-    if (!UnaryShuffle && !IsEVEXShuffle) {
+    if (!UnaryShuffle && !IsMaskedShuffle) {
       assert(llvm::all_of(BaseMask, [](int M) { return 0 <= M && M < 4; }) &&
              "Unexpected shuffle sentinel value");
       // Prefer blends to X86ISD::VPERM2X128.
       if (!((BaseMask[0] == 0 && BaseMask[1] == 3) ||
             (BaseMask[0] == 2 && BaseMask[1] == 1))) {
         unsigned PermMask = 0;
         PermMask |= ((BaseMask[0] & 3) << 0);
         PermMask |= ((BaseMask[1] & 3) << 4);

   // For masks that have been widened to 128-bit elements or more,
   // narrow back down to 64-bit elements.
   SmallVector<int, 64> Mask;
   if (BaseMaskEltSizeInBits > 64) {
     assert((BaseMaskEltSizeInBits % 64) == 0 && "Illegal mask size");
     int MaskScale = BaseMaskEltSizeInBits / 64;
     narrowShuffleMaskElts(MaskScale, BaseMask, Mask);
   } else {
-    Mask = SmallVector<int, 64>(BaseMask.begin(), BaseMask.end());
+    Mask.assign(BaseMask.begin(), BaseMask.end());
   }
 
   unsigned NumMaskElts = Mask.size();
   unsigned MaskEltSizeInBits = RootSizeInBits / NumMaskElts;
 
   // Determine the effective mask value type.
   FloatDomain &= (32 <= MaskEltSizeInBits);
   MVT MaskVT = FloatDomain ? MVT::getFloatingPointVT(MaskEltSizeInBits)

       if (isSequentialOrUndefInRange(Mask, 0, Scale, 0) &&
           isUndefOrZeroOrInRange(HiMask, Scale, NumMaskElts)) {
         return DAG.getBitcast(RootVT, V1);
       }
     }
 
     // Attempt to match against broadcast-from-vector.
     // Limit AVX1 to cases where we're loading+broadcasting a scalar element.
-    if ((Subtarget.hasAVX2() || (Subtarget.hasAVX() && 32 <= MaskEltSizeInBits))
-        && (!IsEVEXShuffle || NumRootElts == NumMaskElts)) {
+    if ((Subtarget.hasAVX2() ||
+         (Subtarget.hasAVX() && 32 <= MaskEltSizeInBits)) &&
+        (!IsMaskedShuffle || NumRootElts == NumMaskElts)) {
       SmallVector<int, 64> BroadcastMask(NumMaskElts, 0);
       if (isTargetShuffleEquivalent(Mask, BroadcastMask)) {
         if (V1.getValueType() == MaskVT &&
             V1.getOpcode() == ISD::SCALAR_TO_VECTOR &&
             MayFoldLoad(V1.getOperand(0))) {
           if (Depth == 0 && Root.getOpcode() == X86ISD::VBROADCAST)
             return SDValue(); // Nothing to do!
           Res = V1.getOperand(0);

         }
       }
     }
 
     SDValue NewV1 = V1; // Save operand in case early exit happens.
     if (matchUnaryShuffle(MaskVT, Mask, AllowFloatDomain, AllowIntDomain, NewV1,
                           DL, DAG, Subtarget, Shuffle, ShuffleSrcVT,
                           ShuffleVT) &&
-        (!IsEVEXShuffle || (NumRootElts == ShuffleVT.getVectorNumElements()))) {
+        (!IsMaskedShuffle ||
+         (NumRootElts == ShuffleVT.getVectorNumElements()))) {
       if (Depth == 0 && Root.getOpcode() == Shuffle)
         return SDValue(); // Nothing to do!
       Res = DAG.getBitcast(ShuffleSrcVT, NewV1);
       Res = DAG.getNode(Shuffle, DL, ShuffleVT, Res);
       return DAG.getBitcast(RootVT, Res);
     }
 
     if (matchUnaryPermuteShuffle(MaskVT, Mask, Zeroable, AllowFloatDomain,
                                  AllowIntDomain, Subtarget, Shuffle, ShuffleVT,
                                  PermuteImm) &&
-        (!IsEVEXShuffle || (NumRootElts == ShuffleVT.getVectorNumElements()))) {
+        (!IsMaskedShuffle ||
+         (NumRootElts == ShuffleVT.getVectorNumElements()))) {
       if (Depth == 0 && Root.getOpcode() == Shuffle)
         return SDValue(); // Nothing to do!
       Res = DAG.getBitcast(ShuffleVT, V1);
       Res = DAG.getNode(Shuffle, DL, ShuffleVT, Res,
                         DAG.getTargetConstant(PermuteImm, DL, MVT::i8));
       return DAG.getBitcast(RootVT, Res);
     }
   }
 
   SDValue NewV1 = V1; // Save operands in case early exit happens.
   SDValue NewV2 = V2;
   if (matchBinaryShuffle(MaskVT, Mask, AllowFloatDomain, AllowIntDomain, NewV1,
                          NewV2, DL, DAG, Subtarget, Shuffle, ShuffleSrcVT,
                          ShuffleVT, UnaryShuffle) &&
-      (!IsEVEXShuffle || (NumRootElts == ShuffleVT.getVectorNumElements()))) {
+      (!IsMaskedShuffle || (NumRootElts == ShuffleVT.getVectorNumElements()))) {
     if (Depth == 0 && Root.getOpcode() == Shuffle)
       return SDValue(); // Nothing to do!
     NewV1 = DAG.getBitcast(ShuffleSrcVT, NewV1);
     NewV2 = DAG.getBitcast(ShuffleSrcVT, NewV2);
     Res = DAG.getNode(Shuffle, DL, ShuffleVT, NewV1, NewV2);
     return DAG.getBitcast(RootVT, Res);
   }
 
   NewV1 = V1; // Save operands in case early exit happens.
   NewV2 = V2;
-  if (matchBinaryPermuteShuffle(
-          MaskVT, Mask, Zeroable, AllowFloatDomain, AllowIntDomain, NewV1,
-          NewV2, DL, DAG, Subtarget, Shuffle, ShuffleVT, PermuteImm) &&
-      (!IsEVEXShuffle || (NumRootElts == ShuffleVT.getVectorNumElements()))) {
+  if (matchBinaryPermuteShuffle(MaskVT, Mask, Zeroable, AllowFloatDomain,
+                                AllowIntDomain, NewV1, NewV2, DL, DAG,
+                                Subtarget, Shuffle, ShuffleVT, PermuteImm) &&
+      (!IsMaskedShuffle || (NumRootElts == ShuffleVT.getVectorNumElements()))) {
     if (Depth == 0 && Root.getOpcode() == Shuffle)
       return SDValue(); // Nothing to do!
     NewV1 = DAG.getBitcast(ShuffleVT, NewV1);
     NewV2 = DAG.getBitcast(ShuffleVT, NewV2);
     Res = DAG.getNode(Shuffle, DL, ShuffleVT, NewV1, NewV2,
                       DAG.getTargetConstant(PermuteImm, DL, MVT::i8));
     return DAG.getBitcast(RootVT, Res);
   }

llvm/test/CodeGen/X86/avx-intrinsics-x86-upgrade.ll

 ; X86-AVX-NEXT:    # xmm1 = xmm1[0],mem[0]
 ; X86-AVX-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0x58,0xc1]
 ; X86-AVX-NEXT:    vmovupd %xmm0, (%eax) # encoding: [0xc5,0xf9,0x11,0x00]
 ; X86-AVX-NEXT:    retl # encoding: [0xc3]
 ;
 ; X86-AVX512VL-LABEL: test_x86_sse2_storeu_pd:
 ; X86-AVX512VL:       # %bb.0:
 ; X86-AVX512VL-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-AVX512VL-NEXT:    vmovsd {{\.LCPI.*}}, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x0d,A,A,A,A]
+; X86-AVX512VL-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0x57,0xc9]
+; X86-AVX512VL-NEXT:    vmovhpd {{\.LCPI.*}}, %xmm1, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0x16,0x0d,A,A,A,A]
 ; X86-AVX512VL-NEXT:    # fixup A - offset: 4, value: {{\.LCPI.*}}, kind: FK_Data_4
-; X86-AVX512VL-NEXT:    # xmm1 = mem[0],zero
-; X86-AVX512VL-NEXT:    vpslldq $8, %xmm1, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0x73,0xf9,0x08]
-; X86-AVX512VL-NEXT:    # xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7]
+; X86-AVX512VL-NEXT:    # xmm1 = xmm1[0],mem[0]
 ; X86-AVX512VL-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x58,0xc1]
 ; X86-AVX512VL-NEXT:    vmovupd %xmm0, (%eax) # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x11,0x00]
 ; X86-AVX512VL-NEXT:    retl # encoding: [0xc3]
 ;
 ; X64-AVX-LABEL: test_x86_sse2_storeu_pd:
 ; X64-AVX:       # %bb.0:
 ; X64-AVX-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 # encoding: [0xc5,0xf1,0x57,0xc9]
 ; X64-AVX-NEXT:    vmovhpd {{.*}}(%rip), %xmm1, %xmm1 # encoding: [0xc5,0xf1,0x16,0x0d,A,A,A,A]
 ; X64-AVX-NEXT:    # fixup A - offset: 4, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
 ; X64-AVX-NEXT:    # xmm1 = xmm1[0],mem[0]
 ; X64-AVX-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0x58,0xc1]
 ; X64-AVX-NEXT:    vmovupd %xmm0, (%rdi) # encoding: [0xc5,0xf9,0x11,0x07]
 ; X64-AVX-NEXT:    retq # encoding: [0xc3]
 ;
 ; X64-AVX512VL-LABEL: test_x86_sse2_storeu_pd:
 ; X64-AVX512VL:       # %bb.0:
-; X64-AVX512VL-NEXT:    vmovsd {{.*}}(%rip), %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x0d,A,A,A,A]
+; X64-AVX512VL-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0x57,0xc9]
+; X64-AVX512VL-NEXT:    vmovhpd {{.*}}(%rip), %xmm1, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0x16,0x0d,A,A,A,A]
 ; X64-AVX512VL-NEXT:    # fixup A - offset: 4, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
-; X64-AVX512VL-NEXT:    # xmm1 = mem[0],zero
-; X64-AVX512VL-NEXT:    vpslldq $8, %xmm1, %xmm1 # EVEX TO VEX Compression encoding: [0xc5,0xf1,0x73,0xf9,0x08]
-; X64-AVX512VL-NEXT:    # xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7]
+; X64-AVX512VL-NEXT:    # xmm1 = xmm1[0],mem[0]
 ; X64-AVX512VL-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x58,0xc1]
 ; X64-AVX512VL-NEXT:    vmovupd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x11,0x07]
 ; X64-AVX512VL-NEXT:    retq # encoding: [0xc3]
   %a2 = fadd <2 x double> %a1, <double 0x0, double 0x4200000000000000>
   call void @llvm.x86.sse2.storeu.pd(i8* %a0, <2 x double> %a2)
   ret void
 }
 declare void @llvm.x86.sse2.storeu.pd(i8*, <2 x double>) nounwind

llvm/test/CodeGen/X86/avx512-cvt.ll

 ; DQNOVL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
 ; DQNOVL-NEXT:    vzeroupper
 ; DQNOVL-NEXT:    retq
   %b = sitofp <2 x i64> %a to <2 x double>
   ret <2 x double> %b
 }
 
 define <2 x float> @sltof2f32(<2 x i64> %a) {
-; NOVLDQ-LABEL: sltof2f32:
-; NOVLDQ:       # %bb.0:
-; NOVLDQ-NEXT:    vpextrq $1, %xmm0, %rax
-; NOVLDQ-NEXT:    vcvtsi2ss %rax, %xmm1, %xmm1
-; NOVLDQ-NEXT:    vmovq %xmm0, %rax
-; NOVLDQ-NEXT:    vcvtsi2ss %rax, %xmm2, %xmm0
-; NOVLDQ-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
-; NOVLDQ-NEXT:    retq
+; NODQ-LABEL: sltof2f32:
+; NODQ:       # %bb.0:
+; NODQ-NEXT:    vpextrq $1, %xmm0, %rax
+; NODQ-NEXT:    vcvtsi2ss %rax, %xmm1, %xmm1
+; NODQ-NEXT:    vmovq %xmm0, %rax
+; NODQ-NEXT:    vcvtsi2ss %rax, %xmm2, %xmm0
+; NODQ-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
+; NODQ-NEXT:    retq
 ;
 ; VLDQ-LABEL: sltof2f32:
 ; VLDQ:       # %bb.0:
 ; VLDQ-NEXT:    vcvtqq2ps %xmm0, %xmm0
 ; VLDQ-NEXT:    retq
 ;
-; VLNODQ-LABEL: sltof2f32:
-; VLNODQ:       # %bb.0:
-; VLNODQ-NEXT:    vpextrq $1, %xmm0, %rax
-; VLNODQ-NEXT:    vcvtsi2ss %rax, %xmm1, %xmm1
-; VLNODQ-NEXT:    vmovq %xmm0, %rax
-; VLNODQ-NEXT:    vcvtsi2ss %rax, %xmm2, %xmm0
-; VLNODQ-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; VLNODQ-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
-; VLNODQ-NEXT:    retq
-;
 ; DQNOVL-LABEL: sltof2f32:
 ; DQNOVL:       # %bb.0:
 ; DQNOVL-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; DQNOVL-NEXT:    vcvtqq2ps %zmm0, %ymm0
 ; DQNOVL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; DQNOVL-NEXT:    vzeroupper
 ; DQNOVL-NEXT:    retq
   %b = sitofp <2 x i64> %a to <2 x float>

llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll

; CHECK-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 2, i32 0, i32 3, i32 4>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
   ret <4 x i64> %res
 }
 define <2 x i64> @test_8xi64_to_2xi64_perm_mask0(<8 x i64> %vec) {
 ; CHECK-LABEL: test_8xi64_to_2xi64_perm_mask0:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpermpd {{.*#+}} zmm0 = zmm0[3,0,2,3,7,4,6,7]
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; CHECK-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
   %res = shufflevector <8 x i64> %vec, <8 x i64> undef, <2 x i32> <i32 3, i32 0>
   ret <2 x i64> %res
 }
 define <2 x i64> @test_masked_8xi64_to_2xi64_perm_mask0(<8 x i64> %vec, <2 x i64> %vec2, <2 x i64> %mask) {
 ; CHECK-LABEL: test_masked_8xi64_to_2xi64_perm_mask0:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[3,0,2,3,7,4,6,7]
+; CHECK-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; CHECK-NEXT:    vptestnmq %xmm2, %xmm2, %k1
-; CHECK-NEXT:    vpblendmq %xmm0, %xmm1, %xmm0 {%k1}
+; CHECK-NEXT:    valignq {{.*#+}} xmm1 {%k1} = xmm3[1],xmm0[0]
+; CHECK-NEXT:    vmovdqa %xmm1, %xmm0
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <2 x i32> <i32 3, i32 0>
   %cmp = icmp eq <2 x i64> %mask, zeroinitializer
   %res = select <2 x i1> %cmp, <2 x i64> %shuf, <2 x i64> %vec2
   ret <2 x i64> %res
 }
 
 define <2 x i64> @test_masked_z_8xi64_to_2xi64_perm_mask0(<8 x i64> %vec, <2 x i64> %mask) {
 ; CHECK-LABEL: test_masked_z_8xi64_to_2xi64_perm_mask0:
 ; CHECK:       # %bb.0:
+; CHECK-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; CHECK-NEXT:    vptestnmq %xmm1, %xmm1, %k1
-; CHECK-NEXT:    vpermq {{.*#+}} zmm0 {%k1} {z} = zmm0[3,0,2,3,7,4,6,7]
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; CHECK-NEXT:    valignq {{.*#+}} xmm0 {%k1} {z} = xmm2[1],xmm0[0]
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <2 x i32> <i32 3, i32 0>
   %cmp = icmp eq <2 x i64> %mask, zeroinitializer
   %res = select <2 x i1> %cmp, <2 x i64> %shuf, <2 x i64> zeroinitializer
   ret <2 x i64> %res
 }
 define <2 x i64> @test_masked_8xi64_to_2xi64_perm_mask1(<8 x i64> %vec, <2 x i64> %vec2, <2 x i64> %mask) {

llvm/test/CodeGen/X86/sse2-intrinsics-x86-upgrade.ll

 ; X86-AVX1-NEXT:    ## xmm1 = xmm1[0],mem[0]
 ; X86-AVX1-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x58,0xc1]
 ; X86-AVX1-NEXT:    vmovupd %xmm0, (%eax) ## encoding: [0xc5,0xf9,0x11,0x00]
 ; X86-AVX1-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X86-AVX512-LABEL: test_x86_sse2_storeu_pd:
 ; X86-AVX512:       ## %bb.0:
 ; X86-AVX512-NEXT:    movl {{[0-9]+}}(%esp), %eax ## encoding: [0x8b,0x44,0x24,0x04]
-; X86-AVX512-NEXT:    vmovsd LCPI11_0, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x0d,A,A,A,A]
+; X86-AVX512-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x57,0xc9]
+; X86-AVX512-NEXT:    vmovhpd LCPI11_0, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x16,0x0d,A,A,A,A]
 ; X86-AVX512-NEXT:    ## fixup A - offset: 4, value: LCPI11_0, kind: FK_Data_4
-; X86-AVX512-NEXT:    ## xmm1 = mem[0],zero
-; X86-AVX512-NEXT:    vpslldq $8, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x73,0xf9,0x08]
-; X86-AVX512-NEXT:    ## xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7]
+; X86-AVX512-NEXT:    ## xmm1 = xmm1[0],mem[0]
 ; X86-AVX512-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x58,0xc1]
 ; X86-AVX512-NEXT:    vmovupd %xmm0, (%eax) ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x11,0x00]
 ; X86-AVX512-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X64-SSE-LABEL: test_x86_sse2_storeu_pd:
 ; X64-SSE:       ## %bb.0:
 ; X64-SSE-NEXT:    xorpd %xmm1, %xmm1 ## encoding: [0x66,0x0f,0x57,0xc9]
 ; X64-SSE-NEXT:    movhpd {{.*}}(%rip), %xmm1 ## encoding: [0x66,0x0f,0x16,0x0d,A,A,A,A]

 ; X64-AVX1-NEXT:    ## fixup A - offset: 4, value: LCPI11_0-4, kind: reloc_riprel_4byte
 ; X64-AVX1-NEXT:    ## xmm1 = xmm1[0],mem[0]
 ; X64-AVX1-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0x58,0xc1]
 ; X64-AVX1-NEXT:    vmovupd %xmm0, (%rdi) ## encoding: [0xc5,0xf9,0x11,0x07]
 ; X64-AVX1-NEXT:    retq ## encoding: [0xc3]
 ;
 ; X64-AVX512-LABEL: test_x86_sse2_storeu_pd:
 ; X64-AVX512:       ## %bb.0:
-; X64-AVX512-NEXT:    vmovsd {{.*}}(%rip), %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x0d,A,A,A,A]
+; X64-AVX512-NEXT:    vxorpd %xmm1, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x57,0xc9]
+; X64-AVX512-NEXT:    vmovhpd {{.*}}(%rip), %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x16,0x0d,A,A,A,A]
 ; X64-AVX512-NEXT:    ## fixup A - offset: 4, value: LCPI11_0-4, kind: reloc_riprel_4byte
-; X64-AVX512-NEXT:    ## xmm1 = mem[0],zero
-; X64-AVX512-NEXT:    vpslldq $8, %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf1,0x73,0xf9,0x08]
-; X64-AVX512-NEXT:    ## xmm1 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7]
+; X64-AVX512-NEXT:    ## xmm1 = xmm1[0],mem[0]
 ; X64-AVX512-NEXT:    vaddpd %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x58,0xc1]
 ; X64-AVX512-NEXT:    vmovupd %xmm0, (%rdi) ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x11,0x07]
 ; X64-AVX512-NEXT:    retq ## encoding: [0xc3]
   %a2 = fadd <2 x double> %a1, <double 0x0, double 0x4200000000000000>
   call void @llvm.x86.sse2.storeu.pd(i8* %a0, <2 x double> %a2)
   ret void
 }
 declare void @llvm.x86.sse2.storeu.pd(i8*, <2 x double>) nounwind

llvm/test/CodeGen/X86/sse41.ll

 ; AVX1-LABEL: insertps_5:
 ; AVX1:       ## %bb.0:
 ; AVX1-NEXT:    vinsertps $92, %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0xc1,0x5c]
 ; AVX1-NEXT:    ## xmm0 = xmm0[0],xmm1[1],zero,zero
 ; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
 ;
 ; AVX512-LABEL: insertps_5:
 ; AVX512:       ## %bb.0:
-; AVX512-NEXT:    vpblendd $2, %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x02,0xc1,0x02]
-; AVX512-NEXT:    ## xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
-; AVX512-NEXT:    vmovq %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x7e,0xc0]
-; AVX512-NEXT:    ## xmm0 = xmm0[0],zero
+; AVX512-NEXT:    vinsertps $92, %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x79,0x21,0xc1,0x5c]
+; AVX512-NEXT:    ## xmm0 = xmm0[0],xmm1[1],zero,zero
 ; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %vecext = extractelement <4 x float> %A, i32 0
   %vecinit = insertelement <4 x float> undef, float %vecext, i32 0
   %vecext1 = extractelement <4 x float> %B, i32 1
   %vecinit2 = insertelement <4 x float> %vecinit, float %vecext1, i32 1
   %vecinit3 = insertelement <4 x float> %vecinit2, float 0.000000e+00, i32 2
   %vecinit4 = insertelement <4 x float> %vecinit3, float 0.000000e+00, i32 3
   ret <4 x float> %vecinit4

 ; AVX1-NEXT:    vinsertps $28, %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x21,0xc1,0x1c]
 ; AVX1-NEXT:    ## xmm0 = xmm0[0],xmm1[0],zero,zero
 ; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
 ;
 ; AVX512-LABEL: insertps_8:
 ; AVX512:       ## %bb.0:
 ; AVX512-NEXT:    vinsertps $28, %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe3,0x79,0x21,0xc1,0x1c]
 ; AVX512-NEXT:    ## xmm0 = xmm0[0],xmm1[0],zero,zero
-; AVX512-NEXT:    vmovq %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x7e,0xc0]
-; AVX512-NEXT:    ## xmm0 = xmm0[0],zero
 ; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %vecext = extractelement <4 x float> %A, i32 0
   %vecinit = insertelement <4 x float> undef, float %vecext, i32 0
   %vecext1 = extractelement <4 x float> %B, i32 0
   %vecinit2 = insertelement <4 x float> %vecinit, float %vecext1, i32 1
   %vecinit3 = insertelement <4 x float> %vecinit2, float 0.000000e+00, i32 2
   %vecinit4 = insertelement <4 x float> %vecinit3, float 0.000000e+00, i32 3
   ret <4 x float> %vecinit4

llvm/test/CodeGen/X86/vec-strict-inttofp-128.ll

 ; AVX512F-64-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512F-64-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-64-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-64-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512F-64-NEXT:    retq
 ;
 ; AVX512VL-32-LABEL: uitofp_v2i64_v2f64:
 ; AVX512VL-32:       # %bb.0:
-; AVX512VL-32-NEXT:    vpand {{\.LCPI.*}}, %xmm0, %xmm1
+; AVX512VL-32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-32-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
 ; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %xmm1, %xmm1
 ; AVX512VL-32-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vsubpd {{\.LCPI.*}}, %xmm0, %xmm0
 ; AVX512VL-32-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512VL-32-NEXT:    retl
 ;
 ; AVX512VL-64-LABEL: uitofp_v2i64_v2f64:
 ; AVX512VL-64:       # %bb.0:
-; AVX512VL-64-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
+; AVX512VL-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-64-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
 ; AVX512VL-64-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-64-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512VL-64-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-64-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-64-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512VL-64-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_v2i64_v2f64:

llvm/test/CodeGen/X86/vec-strict-inttofp-256.ll

 ; AVX512F-64-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX512F-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
 ; AVX512F-64-NEXT:    vsubpd %ymm2, %ymm0, %ymm0
 ; AVX512F-64-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512F-64-NEXT:    retq
 ;
 ; AVX512VL-32-LABEL: uitofp_v4i64_v4f64:
 ; AVX512VL-32:       # %bb.0:
-; AVX512VL-32-NEXT:    vpand {{\.LCPI.*}}, %ymm0, %ymm1
+; AVX512VL-32-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-32-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %ymm1, %ymm1
 ; AVX512VL-32-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX512VL-32-NEXT:    vpor {{\.LCPI.*}}, %ymm0, %ymm0
 ; AVX512VL-32-NEXT:    vsubpd {{\.LCPI.*}}{1to4}, %ymm0, %ymm0
 ; AVX512VL-32-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512VL-32-NEXT:    retl
 ;
 ; AVX512VL-64-LABEL: uitofp_v4i64_v4f64:
 ; AVX512VL-64:       # %bb.0:
-; AVX512VL-64-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm0, %ymm1
+; AVX512VL-64-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-64-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX512VL-64-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm1, %ymm1
 ; AVX512VL-64-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX512VL-64-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm0, %ymm0
 ; AVX512VL-64-NEXT:    vsubpd {{.*}}(%rip){1to4}, %ymm0, %ymm0
 ; AVX512VL-64-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512VL-64-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_v4i64_v4f64:

llvm/test/CodeGen/X86/vec_int_to_fp.ll

 ; AVX512F-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_2i64_to_2f64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
+; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
 ; AVX512VL-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_2i64_to_2f64:

 ; AVX512F-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
 ; AVX512F-NEXT:    vsubpd %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_4i64_to_4f64:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX512VL-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vsubpd {{.*}}(%rip){1to4}, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_4i64_to_4f64:

 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: sitofp_2i64_to_4f32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtsi2ss %rax, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovq %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtsi2ss %rax, %xmm2, %xmm0
-; AVX512VL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
+; AVX512VL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: sitofp_2i64_to_4f32:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vcvtqq2ps %zmm0, %ymm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512DQ-NEXT:    vzeroupper

 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: sitofp_2i64_to_4f32_zero:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vmovq %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtsi2ss %rax, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtsi2ss %rax, %xmm2, %xmm0
-; AVX512VL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; AVX512VL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
+; AVX512VL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: sitofp_2i64_to_4f32_zero:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vcvtqq2ps %zmm0, %ymm0
 ; AVX512DQ-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; AVX512DQ-NEXT:    vzeroupper

 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: sitofp_4i64_to_4f32_undef:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtsi2ss %rax, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovq %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtsi2ss %rax, %xmm2, %xmm0
-; AVX512VL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
+; AVX512VL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: sitofp_4i64_to_4f32_undef:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vcvtqq2ps %zmm0, %ymm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512DQ-NEXT:    vzeroupper

 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_2i64_to_4f32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtusi2ss %rax, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovq %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtusi2ss %rax, %xmm2, %xmm0
-; AVX512VL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
+; AVX512VL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_2i64_to_4f32:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vcvtuqq2ps %zmm0, %ymm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512DQ-NEXT:    vzeroupper

 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_2i64_to_2f32:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vmovq %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtusi2ss %rax, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtusi2ss %rax, %xmm2, %xmm0
-; AVX512VL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; AVX512VL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
+; AVX512VL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0],xmm0[0],zero,zero
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_2i64_to_2f32:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vcvtuqq2ps %zmm0, %ymm0
 ; AVX512DQ-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
 ; AVX512DQ-NEXT:    vzeroupper

 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_4i64_to_4f32_undef:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpextrq $1, %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtusi2ss %rax, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vmovq %xmm0, %rax
 ; AVX512VL-NEXT:    vcvtusi2ss %rax, %xmm2, %xmm0
-; AVX512VL-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
+; AVX512VL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],zero,zero
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_4i64_to_4f32_undef:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
 ; AVX512DQ-NEXT:    vcvtuqq2ps %zmm0, %ymm0
 ; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512DQ-NEXT:    vzeroupper

 ; AVX512F-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512F-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_load_2i64_to_2f64:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm1
+; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm0[0],xmm1[1],xmm0[2],xmm1[3]
 ; AVX512VL-NEXT:    vpor {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vpor {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    vsubpd {{.*}}(%rip), %xmm0, %xmm0
 ; AVX512VL-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_load_2i64_to_2f64:

 ; AVX512F-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25,1.9342813118337666E+25]
 ; AVX512F-NEXT:    vsubpd %ymm2, %ymm0, %ymm0
 ; AVX512F-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: uitofp_load_4i64_to_4f64:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vpandq {{.*}}(%rip){1to4}, %ymm0, %ymm1
+; AVX512VL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX512VL-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlq $32, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vporq {{.*}}(%rip){1to4}, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vsubpd {{.*}}(%rip){1to4}, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vaddpd %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: uitofp_load_4i64_to_4f64:

 ; AVX512F-NEXT:    vaddpd %xmm2, %xmm1, %xmm1
 ; AVX512F-NEXT:    vmovupd %xmm0, (%rdi)
 ; AVX512F-NEXT:    vmovupd %xmm1, 16(%rdi)
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: PR43609:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpaddq {{.*}}(%rip), %xmm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [4294967295,4294967295]
-; AVX512VL-NEXT:    vpand %xmm2, %xmm0, %xmm3
+; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm3 = xmm0[0],xmm2[1],xmm0[2],xmm2[3]
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm4 = [4841369599423283200,4841369599423283200]
 ; AVX512VL-NEXT:    vpor %xmm4, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm5 = [4985484787499139072,4985484787499139072]
 ; AVX512VL-NEXT:    vpor %xmm5, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovapd {{.*#+}} xmm6 = [1.9342813118337666E+25,1.9342813118337666E+25]
 ; AVX512VL-NEXT:    vsubpd %xmm6, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vaddpd %xmm0, %xmm3, %xmm0
-; AVX512VL-NEXT:    vpand %xmm2, %xmm1, %xmm2
+; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm2 = xmm1[0],xmm2[1],xmm1[2],xmm2[3]
 ; AVX512VL-NEXT:    vpor %xmm4, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsrlq $32, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpor %xmm5, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vsubpd %xmm6, %xmm1, %xmm1
 ; AVX512VL-NEXT:    vaddpd %xmm1, %xmm2, %xmm1
 ; AVX512VL-NEXT:    vmovapd {{.*#+}} xmm2 = [5.0E-1,5.0E-1]
 ; AVX512VL-NEXT:    vaddpd %xmm2, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vaddpd %xmm2, %xmm1, %xmm1

llvm/test/CodeGen/X86/vector-fshl-256.ll

 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm1[8],ymm3[8],ymm1[9],ymm3[9],ymm1[10],ymm3[10],ymm1[11],ymm3[11],ymm1[12],ymm3[12],ymm1[13],ymm3[13],ymm1[14],ymm3[14],ymm1[15],ymm3[15],ymm1[24],ymm3[24],ymm1[25],ymm3[25],ymm1[26],ymm3[26],ymm1[27],ymm3[27],ymm1[28],ymm3[28],ymm1[29],ymm3[29],ymm1[30],ymm3[30],ymm1[31],ymm3[31]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm4, %ymm4
+; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[1],ymm3[1],ymm1[2],ymm3[2],ymm1[3],ymm3[3],ymm1[4],ymm3[4],ymm1[5],ymm3[5],ymm1[6],ymm3[6],ymm1[7],ymm3[7],ymm1[16],ymm3[16],ymm1[17],ymm3[17],ymm1[18],ymm3[18],ymm1[19],ymm3[19],ymm1[20],ymm3[20],ymm1[21],ymm3[21],ymm1[22],ymm3[22],ymm1[23],ymm3[23]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero

llvm/test/CodeGen/X86/vector-fshl-512.ll

 ; AVX512VL-NEXT:    vpsllw $2, %ymm4, %ymm7
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm8 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm8, %ymm7
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm9
 ; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm7, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm7
 ; AVX512VL-NEXT:    vpaddb %ymm9, %ymm9, %ymm10
 ; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm7, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm7 = ymm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm7, %ymm7
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm11 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm7, %ymm11, %ymm7
-; AVX512VL-NEXT:    vpsrlw $8, %ymm7, %ymm7
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    vpxor %xmm7, %xmm7, %xmm7
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm11 = ymm2[8],ymm7[8],ymm2[9],ymm7[9],ymm2[10],ymm7[10],ymm2[11],ymm7[11],ymm2[12],ymm7[12],ymm2[13],ymm7[13],ymm2[14],ymm7[14],ymm2[15],ymm7[15],ymm2[24],ymm7[24],ymm2[25],ymm7[25],ymm2[26],ymm7[26],ymm2[27],ymm7[27],ymm2[28],ymm7[28],ymm2[29],ymm7[29],ymm2[30],ymm7[30],ymm2[31],ymm7[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
 ; AVX512VL-NEXT:    # ymm12 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm2, %ymm12, %ymm2
+; AVX512VL-NEXT:    vpmullw %ymm12, %ymm11, %ymm11
+; AVX512VL-NEXT:    vpsrlw $8, %ymm11, %ymm11
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm2[0],ymm7[0],ymm2[1],ymm7[1],ymm2[2],ymm7[2],ymm2[3],ymm7[3],ymm2[4],ymm7[4],ymm2[5],ymm7[5],ymm2[6],ymm7[6],ymm2[7],ymm7[7],ymm2[16],ymm7[16],ymm2[17],ymm7[17],ymm2[18],ymm7[18],ymm2[19],ymm7[19],ymm2[20],ymm7[20],ymm2[21],ymm7[21],ymm2[22],ymm7[22],ymm2[23],ymm7[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm13 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm2, %ymm13, %ymm2
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpackuswb %ymm7, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpackuswb %ymm11, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpor %ymm2, %ymm4, %ymm2
 ; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm3, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm8, %ymm5
 ; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm5, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm5
 ; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm5, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8],ymm7[8],ymm1[9],ymm7[9],ymm1[10],ymm7[10],ymm1[11],ymm7[11],ymm1[12],ymm7[12],ymm1[13],ymm7[13],ymm1[14],ymm7[14],ymm1[15],ymm7[15],ymm1[24],ymm7[24],ymm1[25],ymm7[25],ymm1[26],ymm7[26],ymm1[27],ymm7[27],ymm1[28],ymm7[28],ymm1[29],ymm7[29],ymm1[30],ymm7[30],ymm1[31],ymm7[31]
+; AVX512VL-NEXT:    vpmullw %ymm5, %ymm12, %ymm5
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpmullw %ymm5, %ymm11, %ymm5
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpmullw %ymm1, %ymm12, %ymm1
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm7[0],ymm1[1],ymm7[1],ymm1[2],ymm7[2],ymm1[3],ymm7[3],ymm1[4],ymm7[4],ymm1[5],ymm7[5],ymm1[6],ymm7[6],ymm1[7],ymm7[7],ymm1[16],ymm7[16],ymm1[17],ymm7[17],ymm1[18],ymm7[18],ymm1[19],ymm7[19],ymm1[20],ymm7[20],ymm1[21],ymm7[21],ymm1[22],ymm7[22],ymm1[23],ymm7[23]
+; AVX512VL-NEXT:    vpmullw %ymm1, %ymm13, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm3, %ymm1
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm1, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512BW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm3, %zmm4
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
+; AVX512BW-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512BW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512BW-NEXT:    kmovq %rax, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
 ; AVX512VBMI2-NEXT:    vpsllw $2, %zmm3, %zmm4
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VBMI2-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512VBMI2-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
 ; AVX512VBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512VBMI2-NEXT:    kmovq %rax, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VLBW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm3, %zmm4
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512VLBW-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLBW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512VLBW-NEXT:    kmovq %rax, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VLVBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpblendmb %zmm3, %zmm0, %zmm3 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm3, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm4, %zmm4
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm3, %zmm3, %zmm3 {%k1}
-; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm4, %zmm4
+; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
 ; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLVBMI2-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512VLVBMI2-NEXT:    vpackuswb %zmm4, %zmm1, %zmm1
 ; AVX512VLVBMI2-NEXT:    vporq %zmm1, %zmm3, %zmm1
 ; AVX512VLVBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512VLVBMI2-NEXT:    kmovq %rax, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
 ; AVX512VLVBMI2-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res

llvm/test/CodeGen/X86/vector-fshl-rot-256.ll

 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vpsllw $2, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0

llvm/test/CodeGen/X86/vector-fshl-rot-512.ll

 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm7
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm5, %ymm9, %ymm5
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm9 = ymm1[8],ymm5[8],ymm1[9],ymm5[9],ymm1[10],ymm5[10],ymm1[11],ymm5[11],ymm1[12],ymm5[12],ymm1[13],ymm5[13],ymm1[14],ymm5[14],ymm1[15],ymm5[15],ymm1[24],ymm5[24],ymm1[25],ymm5[25],ymm1[26],ymm5[26],ymm1[27],ymm5[27],ymm1[28],ymm5[28],ymm1[29],ymm5[29],ymm1[30],ymm5[30],ymm1[31],ymm5[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
 ; AVX512VL-NEXT:    # ymm10 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm1, %ymm10, %ymm1
+; AVX512VL-NEXT:    vpmullw %ymm10, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpsrlw $8, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm5[0],ymm1[1],ymm5[1],ymm1[2],ymm5[2],ymm1[3],ymm5[3],ymm1[4],ymm5[4],ymm1[5],ymm5[5],ymm1[6],ymm5[6],ymm1[7],ymm5[7],ymm1[16],ymm5[16],ymm1[17],ymm5[17],ymm1[18],ymm5[18],ymm1[19],ymm5[19],ymm1[20],ymm5[20],ymm1[21],ymm5[21],ymm1[22],ymm5[22],ymm1[23],ymm5[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm11 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm1, %ymm11, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm9, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm9, %ymm3
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm5[8],ymm0[9],ymm5[9],ymm0[10],ymm5[10],ymm0[11],ymm5[11],ymm0[12],ymm5[12],ymm0[13],ymm5[13],ymm0[14],ymm5[14],ymm0[15],ymm5[15],ymm0[24],ymm5[24],ymm0[25],ymm5[25],ymm0[26],ymm5[26],ymm0[27],ymm5[27],ymm0[28],ymm5[28],ymm0[29],ymm5[29],ymm0[30],ymm5[30],ymm0[31],ymm5[31]
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm10, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmullw %ymm0, %ymm10, %ymm0
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm5[0],ymm0[1],ymm5[1],ymm0[2],ymm5[2],ymm0[3],ymm5[3],ymm0[4],ymm5[4],ymm0[5],ymm5[5],ymm0[6],ymm5[6],ymm0[7],ymm5[7],ymm0[16],ymm5[16],ymm0[17],ymm5[17],ymm0[18],ymm5[18],ymm0[19],ymm5[19],ymm0[20],ymm5[20],ymm0[21],ymm5[21],ymm0[22],ymm5[22],ymm0[23],ymm5[23]
+; AVX512VL-NEXT:    vpmullw %ymm0, %ymm11, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512BW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256]
 ; AVX512VLBW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts

llvm/test/CodeGen/X86/vector-fshr-256.ll

 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm3 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15],ymm1[24],ymm2[24],ymm1[25],ymm2[25],ymm1[26],ymm2[26],ymm1[27],ymm2[27],ymm1[28],ymm2[28],ymm1[29],ymm2[29],ymm1[30],ymm2[30],ymm1[31],ymm2[31]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm3, %ymm2
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[16],ymm2[16],ymm1[17],ymm2[17],ymm1[18],ymm2[18],ymm1[19],ymm2[19],ymm1[20],ymm2[20],ymm1[21],ymm2[21],ymm1[22],ymm2[22],ymm1[23],ymm2[23]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpor %ymm2, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255]
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm2 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero

llvm/test/CodeGen/X86/vector-fshr-512.ll

 ; AVX512VL-NEXT:    vpsllw $2, %ymm3, %ymm4
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm4, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm6, %ymm6, %ymm8
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm4, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm3, %ymm3, %ymm4
 ; AVX512VL-NEXT:    vpaddb %ymm8, %ymm8, %ymm9
 ; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm4 = ymm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
-; AVX512VL-NEXT:    # ymm10 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm4, %ymm10, %ymm4
-; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm11 = ymm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm11, %ymm11
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm12 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm12, %ymm11, %ymm11
-; AVX512VL-NEXT:    vpsrlw $8, %ymm11, %ymm11
-; AVX512VL-NEXT:    vpackuswb %ymm4, %ymm11, %ymm4
-; AVX512VL-NEXT:    vpor %ymm4, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
-; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm3, %ymm2, %ymm2
+; AVX512VL-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm10 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm11 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm11, %ymm10, %ymm10
+; AVX512VL-NEXT:    vpsrlw $8, %ymm10, %ymm10
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm13 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    # ymm13 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm13, %ymm12, %ymm12
+; AVX512VL-NEXT:    vpsrlw $8, %ymm12, %ymm12
+; AVX512VL-NEXT:    vpackuswb %ymm10, %ymm12, %ymm10
+; AVX512VL-NEXT:    vpor %ymm3, %ymm10, %ymm3
+; AVX512VL-NEXT:    vpbroadcastq {{.*#+}} ymm10 = [18446744073709551360,18446744073709551360,18446744073709551360,18446744073709551360]
+; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm5, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm6, %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsllw $2, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm7, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpaddb %ymm0, %ymm0, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm9, %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm4[8],ymm1[9],ymm4[9],ymm1[10],ymm4[10],ymm1[11],ymm4[11],ymm1[12],ymm4[12],ymm1[13],ymm4[13],ymm1[14],ymm4[14],ymm1[15],ymm4[15],ymm1[24],ymm4[24],ymm1[25],ymm4[25],ymm1[26],ymm4[26],ymm1[27],ymm4[27],ymm1[28],ymm4[28],ymm1[29],ymm4[29],ymm1[30],ymm4[30],ymm1[31],ymm4[31]
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm11, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm10, %ymm3
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpmullw %ymm5, %ymm12, %ymm5
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm5, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm1[0],ymm4[0],ymm1[1],ymm4[1],ymm1[2],ymm4[2],ymm1[3],ymm4[3],ymm1[4],ymm4[4],ymm1[5],ymm4[5],ymm1[6],ymm4[6],ymm1[7],ymm4[7],ymm1[16],ymm4[16],ymm1[17],ymm4[17],ymm1[18],ymm4[18],ymm1[19],ymm4[19],ymm1[20],ymm4[20],ymm1[21],ymm4[21],ymm1[22],ymm4[22],ymm1[23],ymm4[23]
+; AVX512VL-NEXT:    vpmullw %ymm4, %ymm13, %ymm4
+; AVX512VL-NEXT:    vpsrlw $8, %ymm4, %ymm4
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm4, %ymm3
 ; AVX512VL-NEXT:    vpor %ymm3, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm0
+; AVX512VL-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512BW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
 ; AVX512BW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512BW-NEXT:    kmovq %rax, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VBMI2:       # %bb.0:
 ; AVX512VBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512VBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512VBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512VBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VBMI2-NEXT:    vpackuswb %zmm2, %zmm3, %zmm2
+; AVX512VBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512VBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512VBMI2-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
 ; AVX512VBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512VBMI2-NEXT:    kmovq %rax, %k1
 ; AVX512VBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VBMI2-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512VLBW-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VLBW-NEXT:    vpackuswb %zmm2, %zmm3, %zmm2
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512VLBW-NEXT:    kmovq %rax, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VLBW-NEXT:    retq
 ;
 ; AVX512VLVBMI2-LABEL: constant_funnnel_v64i8:
 ; AVX512VLVBMI2:       # %bb.0:
 ; AVX512VLVBMI2-NEXT:    vbroadcasti32x4 {{.*#+}} zmm2 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512VLVBMI2-NEXT:    # zmm2 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpsllw $4, %zmm0, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpsllw $2, %zmm0, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm2, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vpmovb2m %zmm2, %k1
 ; AVX512VLVBMI2-NEXT:    vpaddb %zmm0, %zmm0, %zmm0 {%k1}
-; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512VLVBMI2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLVBMI2-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
 ; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512VLVBMI2-NEXT:    vpackuswb %zmm2, %zmm3, %zmm2
+; AVX512VLVBMI2-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
+; AVX512VLVBMI2-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512VLVBMI2-NEXT:    vpackuswb %zmm3, %zmm2, %zmm2
 ; AVX512VLVBMI2-NEXT:    vporq %zmm2, %zmm0, %zmm0
 ; AVX512VLVBMI2-NEXT:    movabsq $72340172838076673, %rax # imm = 0x101010101010101
 ; AVX512VLVBMI2-NEXT:    kmovq %rax, %k1
 ; AVX512VLVBMI2-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
 ; AVX512VLVBMI2-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res
 }

llvm/test/CodeGen/X86/vector-fshr-rot-256.ll

 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vpsllw $2, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0

llvm/test/CodeGen/X86/vector-fshr-rot-512.ll

 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm7
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
-; AVX512VL-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm5, %ymm9, %ymm5
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm9 = ymm1[8],ymm5[8],ymm1[9],ymm5[9],ymm1[10],ymm5[10],ymm1[11],ymm5[11],ymm1[12],ymm5[12],ymm1[13],ymm5[13],ymm1[14],ymm5[14],ymm1[15],ymm5[15],ymm1[24],ymm5[24],ymm1[25],ymm5[25],ymm1[26],ymm5[26],ymm1[27],ymm5[27],ymm1[28],ymm5[28],ymm1[29],ymm5[29],ymm1[30],ymm5[30],ymm1[31],ymm5[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
 ; AVX512VL-NEXT:    # ymm10 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm1, %ymm10, %ymm1
+; AVX512VL-NEXT:    vpmullw %ymm10, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpsrlw $8, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm5[0],ymm1[1],ymm5[1],ymm1[2],ymm5[2],ymm1[3],ymm5[3],ymm1[4],ymm5[4],ymm1[5],ymm5[5],ymm1[6],ymm5[6],ymm1[7],ymm5[7],ymm1[16],ymm5[16],ymm1[17],ymm5[17],ymm1[18],ymm5[18],ymm1[19],ymm5[19],ymm1[20],ymm5[20],ymm1[21],ymm5[21],ymm1[22],ymm5[22],ymm1[23],ymm5[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
+; AVX512VL-NEXT:    # ymm11 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm1, %ymm11, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm9, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm9, %ymm3
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm5[8],ymm0[9],ymm5[9],ymm0[10],ymm5[10],ymm0[11],ymm5[11],ymm0[12],ymm5[12],ymm0[13],ymm5[13],ymm0[14],ymm5[14],ymm0[15],ymm5[15],ymm0[24],ymm5[24],ymm0[25],ymm5[25],ymm0[26],ymm5[26],ymm0[27],ymm5[27],ymm0[28],ymm5[28],ymm0[29],ymm5[29],ymm0[30],ymm5[30],ymm0[31],ymm5[31]
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm10, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmullw %ymm0, %ymm10, %ymm0
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm5[0],ymm0[1],ymm5[1],ymm0[2],ymm5[2],ymm0[3],ymm5[3],ymm0[4],ymm5[4],ymm0[5],ymm5[5],ymm0[6],ymm5[6],ymm0[7],ymm5[7],ymm0[16],ymm5[16],ymm0[17],ymm5[17],ymm0[18],ymm5[18],ymm0[19],ymm5[19],ymm0[20],ymm5[20],ymm0[21],ymm5[21],ymm0[22],ymm5[22],ymm0[23],ymm5[23]
+; AVX512VL-NEXT:    vpmullw %ymm0, %ymm11, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512BW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536,57344,41152,24704,8256,8192,24640,41088,57536]
 ; AVX512VLBW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
   ret <64 x i8> %res
 }
 
 ;
 ; Uniform Constant Shifts

llvm/test/CodeGen/X86/vector-idiv-udiv-512.ll

 ; AVX512F-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_divconstant_64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
-; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsubb %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm2[8],zmm0[9],zmm2[9],zmm0[10],zmm2[10],zmm0[11],zmm2[11],zmm0[12],zmm2[12],zmm0[13],zmm2[13],zmm0[14],zmm2[14],zmm0[15],zmm2[15],zmm0[24],zmm2[24],zmm0[25],zmm2[25],zmm0[26],zmm2[26],zmm0[27],zmm2[27],zmm0[28],zmm2[28],zmm0[29],zmm2[29],zmm0[30],zmm2[30],zmm0[31],zmm2[31],zmm0[40],zmm2[40],zmm0[41],zmm2[41],zmm0[42],zmm2[42],zmm0[43],zmm2[43],zmm0[44],zmm2[44],zmm0[45],zmm2[45],zmm0[46],zmm2[46],zmm0[47],zmm2[47],zmm0[56],zmm2[56],zmm0[57],zmm2[57],zmm0[58],zmm2[58],zmm0[59],zmm2[59],zmm0[60],zmm2[60],zmm0[61],zmm2[61],zmm0[62],zmm2[62],zmm0[63],zmm2[63]
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpsubb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
 ; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm2[0],zmm0[1],zmm2[1],zmm0[2],zmm2[2],zmm0[3],zmm2[3],zmm0[4],zmm2[4],zmm0[5],zmm2[5],zmm0[6],zmm2[6],zmm0[7],zmm2[7],zmm0[16],zmm2[16],zmm0[17],zmm2[17],zmm0[18],zmm2[18],zmm0[19],zmm2[19],zmm0[20],zmm2[20],zmm0[21],zmm2[21],zmm0[22],zmm2[22],zmm0[23],zmm2[23],zmm0[32],zmm2[32],zmm0[33],zmm2[33],zmm0[34],zmm2[34],zmm0[35],zmm2[35],zmm0[36],zmm2[36],zmm0[37],zmm2[37],zmm0[38],zmm2[38],zmm0[39],zmm2[39],zmm0[48],zmm2[48],zmm0[49],zmm2[49],zmm0[50],zmm2[50],zmm0[51],zmm2[51],zmm0[52],zmm2[52],zmm0[53],zmm2[53],zmm0[54],zmm2[54],zmm0[55],zmm2[55]
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
   %res = udiv <64 x i8> %a, <i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31, i8 32, i8 33, i8 34, i8 35, i8 36, i8 37, i8 38, i8 38, i8 37, i8 36, i8 35, i8 34, i8 33, i8 32, i8 31, i8 30, i8 29, i8 28, i8 27, i8 26, i8 25, i8 24, i8 23, i8 22, i8 21, i8 20, i8 19, i8 18, i8 17, i8 16, i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7>
   ret <64 x i8> %res
 }
 
 ;
 ; urem by 7
 ;

 ; AVX512F-NEXT:    vpand %ymm2, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpsubb %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_remconstant_64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
+; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
-; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsubb %zmm1, %zmm0, %zmm3
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm3[8],zmm2[8],zmm3[9],zmm2[9],zmm3[10],zmm2[10],zmm3[11],zmm2[11],zmm3[12],zmm2[12],zmm3[13],zmm2[13],zmm3[14],zmm2[14],zmm3[15],zmm2[15],zmm3[24],zmm2[24],zmm3[25],zmm2[25],zmm3[26],zmm2[26],zmm3[27],zmm2[27],zmm3[28],zmm2[28],zmm3[29],zmm2[29],zmm3[30],zmm2[30],zmm3[31],zmm2[31],zmm3[40],zmm2[40],zmm3[41],zmm2[41],zmm3[42],zmm2[42],zmm3[43],zmm2[43],zmm3[44],zmm2[44],zmm3[45],zmm2[45],zmm3[46],zmm2[46],zmm3[47],zmm2[47],zmm3[56],zmm2[56],zmm3[57],zmm2[57],zmm3[58],zmm2[58],zmm3[59],zmm2[59],zmm3[60],zmm2[60],zmm3[61],zmm2[61],zmm3[62],zmm2[62],zmm3[63],zmm2[63]
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpsubb %zmm2, %zmm0, %zmm3
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm4 = zmm3[8],zmm1[8],zmm3[9],zmm1[9],zmm3[10],zmm1[10],zmm3[11],zmm1[11],zmm3[12],zmm1[12],zmm3[13],zmm1[13],zmm3[14],zmm1[14],zmm3[15],zmm1[15],zmm3[24],zmm1[24],zmm3[25],zmm1[25],zmm3[26],zmm1[26],zmm3[27],zmm1[27],zmm3[28],zmm1[28],zmm3[29],zmm1[29],zmm3[30],zmm1[30],zmm3[31],zmm1[31],zmm3[40],zmm1[40],zmm3[41],zmm1[41],zmm3[42],zmm1[42],zmm3[43],zmm1[43],zmm3[44],zmm1[44],zmm3[45],zmm1[45],zmm3[46],zmm1[46],zmm3[47],zmm1[47],zmm3[56],zmm1[56],zmm3[57],zmm1[57],zmm3[58],zmm1[58],zmm3[59],zmm1[59],zmm3[60],zmm1[60],zmm3[61],zmm1[61],zmm3[62],zmm1[62],zmm3[63],zmm1[63]
 ; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm4, %zmm4
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm4, %zmm4
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm2 = zmm3[0],zmm2[0],zmm3[1],zmm2[1],zmm3[2],zmm2[2],zmm3[3],zmm2[3],zmm3[4],zmm2[4],zmm3[5],zmm2[5],zmm3[6],zmm2[6],zmm3[7],zmm2[7],zmm3[16],zmm2[16],zmm3[17],zmm2[17],zmm3[18],zmm2[18],zmm3[19],zmm2[19],zmm3[20],zmm2[20],zmm3[21],zmm2[21],zmm3[22],zmm2[22],zmm3[23],zmm2[23],zmm3[32],zmm2[32],zmm3[33],zmm2[33],zmm3[34],zmm2[34],zmm3[35],zmm2[35],zmm3[36],zmm2[36],zmm3[37],zmm2[37],zmm3[38],zmm2[38],zmm3[39],zmm2[39],zmm3[48],zmm2[48],zmm3[49],zmm2[49],zmm3[50],zmm2[50],zmm3[51],zmm2[51],zmm3[52],zmm2[52],zmm3[53],zmm2[53],zmm3[54],zmm2[54],zmm3[55],zmm2[55]
-; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpackuswb %zmm4, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpaddb %zmm1, %zmm2, %zmm1
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vpalignr {{.*#+}} zmm2 = zmm2[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,0,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,16,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,32,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,48]
-; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm2, %zmm2
-; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm2, %zmm2
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm3 = zmm3[0],zmm1[0],zmm3[1],zmm1[1],zmm3[2],zmm1[2],zmm3[3],zmm1[3],zmm3[4],zmm1[4],zmm3[5],zmm1[5],zmm3[6],zmm1[6],zmm3[7],zmm1[7],zmm3[16],zmm1[16],zmm3[17],zmm1[17],zmm3[18],zmm1[18],zmm3[19],zmm1[19],zmm3[20],zmm1[20],zmm3[21],zmm1[21],zmm3[22],zmm1[22],zmm3[23],zmm1[23],zmm3[32],zmm1[32],zmm3[33],zmm1[33],zmm3[34],zmm1[34],zmm3[35],zmm1[35],zmm3[36],zmm1[36],zmm3[37],zmm1[37],zmm3[38],zmm1[38],zmm3[39],zmm1[39],zmm3[48],zmm1[48],zmm3[49],zmm1[49],zmm3[50],zmm1[50],zmm3[51],zmm1[51],zmm3[52],zmm1[52],zmm3[53],zmm1[53],zmm3[54],zmm1[54],zmm3[55],zmm1[55]
+; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpackuswb %zmm4, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpaddb %zmm2, %zmm3, %zmm2
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm2[8],zmm1[8],zmm2[9],zmm1[9],zmm2[10],zmm1[10],zmm2[11],zmm1[11],zmm2[12],zmm1[12],zmm2[13],zmm1[13],zmm2[14],zmm1[14],zmm2[15],zmm1[15],zmm2[24],zmm1[24],zmm2[25],zmm1[25],zmm2[26],zmm1[26],zmm2[27],zmm1[27],zmm2[28],zmm1[28],zmm2[29],zmm1[29],zmm2[30],zmm1[30],zmm2[31],zmm1[31],zmm2[40],zmm1[40],zmm2[41],zmm1[41],zmm2[42],zmm1[42],zmm2[43],zmm1[43],zmm2[44],zmm1[44],zmm2[45],zmm1[45],zmm2[46],zmm1[46],zmm2[47],zmm1[47],zmm2[56],zmm1[56],zmm2[57],zmm1[57],zmm2[58],zmm1[58],zmm2[59],zmm1[59],zmm2[60],zmm1[60],zmm2[61],zmm1[61],zmm2[62],zmm1[62],zmm2[63],zmm1[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
+; AVX512BW-NEXT:    vpandq %zmm4, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm1 = zmm2[0],zmm1[0],zmm2[1],zmm1[1],zmm2[2],zmm1[2],zmm2[3],zmm1[3],zmm2[4],zmm1[4],zmm2[5],zmm1[5],zmm2[6],zmm1[6],zmm2[7],zmm1[7],zmm2[16],zmm1[16],zmm2[17],zmm1[17],zmm2[18],zmm1[18],zmm2[19],zmm1[19],zmm2[20],zmm1[20],zmm2[21],zmm1[21],zmm2[22],zmm1[22],zmm2[23],zmm1[23],zmm2[32],zmm1[32],zmm2[33],zmm1[33],zmm2[34],zmm1[34],zmm2[35],zmm1[35],zmm2[36],zmm1[36],zmm2[37],zmm1[37],zmm2[38],zmm1[38],zmm2[39],zmm1[39],zmm2[48],zmm1[48],zmm2[49],zmm1[49],zmm2[50],zmm1[50],zmm2[51],zmm1[51],zmm2[52],zmm1[52],zmm2[53],zmm1[53],zmm2[54],zmm1[54],zmm2[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpalignr {{.*#+}} zmm1 = zmm1[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,0,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,16,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,32,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,48]
+; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmullw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpandq %zmm3, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpandq %zmm4, %zmm1, %zmm1
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpsubb %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
   %res = urem <64 x i8> %a, <i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31, i8 32, i8 33, i8 34, i8 35, i8 36, i8 37, i8 38, i8 38, i8 37, i8 36, i8 35, i8 34, i8 33, i8 32, i8 31, i8 30, i8 29, i8 28, i8 27, i8 26, i8 25, i8 24, i8 23, i8 22, i8 21, i8 20, i8 19, i8 18, i8 17, i8 16, i8 15, i8 14, i8 13, i8 12, i8 11, i8 10, i8 9, i8 8, i8 7>
   ret <64 x i8> %res
 }

llvm/test/CodeGen/X86/vector-reduce-mul.ll

 ; AVX512BW-NEXT:    vmovq %xmm0, %rax
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: test_v2i64:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BWVL-NEXT:    vpsrlq $32, %xmm0, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm2, %xmm2
-; AVX512BWVL-NEXT:    vpsrlq $32, %xmm1, %xmm3
+; AVX512BWVL-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vmovq %xmm0, %rax
 ; AVX512BWVL-NEXT:    retq
 ;

 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BWVL-NEXT:    vpsrlq $32, %xmm0, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm2, %xmm2
-; AVX512BWVL-NEXT:    vpsrlq $32, %xmm1, %xmm3
+; AVX512BWVL-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vmovq %xmm0, %rax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq

 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BWVL-NEXT:    vpsrlq $32, %xmm0, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm2, %xmm2
-; AVX512BWVL-NEXT:    vpsrlq $32, %xmm1, %xmm3
+; AVX512BWVL-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vmovq %xmm0, %rax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq

 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
 ; AVX512BWVL-NEXT:    vpsrlq $32, %xmm0, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm2, %xmm2
-; AVX512BWVL-NEXT:    vpsrlq $32, %xmm1, %xmm3
+; AVX512BWVL-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm0[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX512BWVL-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm3, %xmm2
 ; AVX512BWVL-NEXT:    vpsllq $32, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vpmuludq %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpaddq %xmm2, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vmovq %xmm0, %rax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq

 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
-; AVX512BW-LABEL: test_v4i8:
-; AVX512BW:       # %bb.0:
-; AVX512BW-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
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovd %xmm0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: test_v4i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512VL-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
-; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovd %xmm0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_v4i8:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-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
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vmovd %xmm0, %eax
-; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512DQ-NEXT:    retq
+; AVX512-LABEL: test_v4i8:
+; AVX512:       # %bb.0:
+; AVX512-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
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
+; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v4i8(<4 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v8i8(<8 x i8> %a0) {
 ; SSE2-LABEL: test_v8i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,2,3,3]

 ; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
 ; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX-NEXT:    vmovd %xmm0, %eax
 ; AVX-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX-NEXT:    retq
 ;
-; AVX512BW-LABEL: test_v8i8:
-; AVX512BW:       # %bb.0:
-; AVX512BW-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
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovd %xmm0, %eax
-; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512BW-NEXT:    retq
-;
-; AVX512VL-LABEL: test_v8i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
-; AVX512VL-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
-; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpalignr {{.*#+}} xmm1 = xmm0[4,5,6,7,8,9,10,11,12,13,14,15,0,1,2,3]
-; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX512VL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512VL-NEXT:    vmovd %xmm0, %eax
-; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512VL-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_v8i8:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-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
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
-; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vpsrld $16, %xmm0, %xmm1
-; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX512DQ-NEXT:    vmovd %xmm0, %eax
-; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
-; AVX512DQ-NEXT:    retq
+; AVX512-LABEL: test_v8i8:
+; AVX512:       # %bb.0:
+; AVX512-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
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm1
+; AVX512-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX512-NEXT:    vmovd %xmm0, %eax
+; AVX512-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512-NEXT:    retq
   %1 = call i8 @llvm.experimental.vector.reduce.mul.v8i8(<8 x i8> %a0)
   ret i8 %1
 }
 
 define i8 @test_v16i8(<16 x i8> %a0) {
 ; SSE2-LABEL: test_v16i8:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    movdqa %xmm0, %xmm1

 ; AVX512DQVL-LABEL: test_v32i8:
 ; AVX512DQVL:       # %bb.0:
 ; AVX512DQVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
 ; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX512DQVL-NEXT:    vpmullw %xmm2, %xmm3, %xmm2
 ; AVX512DQVL-NEXT:    vmovdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
 ; AVX512DQVL-NEXT:    vpand %xmm3, %xmm2, %xmm2
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512DQVL-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
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm3, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm3, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm3, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpackuswb %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper

 ; AVX512BWVL-NEXT:    vpmullw %zmm1, %zmm0, %zmm0
 ; AVX512BWVL-NEXT:    vpand %ymm3, %ymm0, %ymm1
 ; AVX512BWVL-NEXT:    vpackuswb %ymm4, %ymm1, %ymm1
 ; AVX512BWVL-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
 ; AVX512BWVL-NEXT:    vpmullw %xmm3, %xmm2, %xmm2
 ; AVX512BWVL-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
 ; AVX512BWVL-NEXT:    vpmovwb %ymm2, %xmm2
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm2[8],xmm0[8],xmm2[9],xmm0[9],xmm2[10],xmm0[10],xmm2[11],xmm0[11],xmm2[12],xmm0[12],xmm2[13],xmm0[13],xmm2[14],xmm0[14],xmm2[15],xmm0[15]
 ; AVX512BWVL-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq

 ; AVX512DQVL-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX512DQVL-NEXT:    vpmullw %xmm0, %xmm3, %xmm0
 ; AVX512DQVL-NEXT:    vpmullw %xmm0, %xmm2, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512DQVL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm1, %xmm0, %xmm2
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm2, %xmm2
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm2, %xmm2
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper

 ; AVX512BWVL-NEXT:    vpmullw %zmm4, %zmm2, %zmm2
 ; AVX512BWVL-NEXT:    vpand %ymm1, %ymm2, %ymm1
 ; AVX512BWVL-NEXT:    vpackuswb %ymm3, %ymm1, %ymm1
 ; AVX512BWVL-NEXT:    vextracti128 $1, %ymm1, %xmm1
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15]
 ; AVX512BWVL-NEXT:    vpmullw %xmm3, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BWVL-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX512BWVL-NEXT:    vpmullw %xmm0, %xmm1, %xmm0
 ; AVX512BWVL-NEXT:    vpmullw %xmm0, %xmm2, %xmm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm0
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq

 ; AVX512DQVL-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vextracti128 $1, %ymm0, %xmm3
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm3, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm2, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm2, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} 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]
+; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpand %xmm2, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpackuswb %xmm3, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm0, %xmm1
 ; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vmovd %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper

llvm/test/CodeGen/X86/vector-rotate-256.ll

 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm1
 ; AVX512VL-NEXT:    vpsllw $2, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpaddb %ymm1, %ymm1, %ymm3
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm2, %ymm3, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
-; AVX512VL-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm2[8],ymm0[9],ymm2[9],ymm0[10],ymm2[10],ymm0[11],ymm2[11],ymm0[12],ymm2[12],ymm0[13],ymm2[13],ymm0[14],ymm2[14],ymm0[15],ymm2[15],ymm0[24],ymm2[24],ymm0[25],ymm2[25],ymm0[26],ymm2[26],ymm0[27],ymm2[27],ymm0[28],ymm2[28],ymm0[29],ymm2[29],ymm0[30],ymm2[30],ymm0[31],ymm2[31]
+; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm3, %ymm3
+; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm2[0],ymm0[1],ymm2[1],ymm0[2],ymm2[2],ymm0[3],ymm2[3],ymm0[4],ymm2[4],ymm0[5],ymm2[5],ymm0[6],ymm2[6],ymm0[7],ymm2[7],ymm0[16],ymm2[16],ymm0[17],ymm2[17],ymm0[18],ymm2[18],ymm0[19],ymm2[19],ymm0[20],ymm2[20],ymm0[21],ymm2[21],ymm0[22],ymm2[22],ymm0[23],ymm2[23]
 ; AVX512VL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v32i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm1
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0

llvm/test/CodeGen/X86/vector-rotate-512.ll

 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm5, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm4, %ymm4, %ymm7
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm5
 ; AVX512VL-NEXT:    vpaddb %ymm7, %ymm7, %ymm8
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm9 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
-; AVX512VL-NEXT:    # ymm9 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm5, %ymm9, %ymm5
-; AVX512VL-NEXT:    vpsrlw $8, %ymm5, %ymm5
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    vpxor %xmm5, %xmm5, %xmm5
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm9 = ymm1[8],ymm5[8],ymm1[9],ymm5[9],ymm1[10],ymm5[10],ymm1[11],ymm5[11],ymm1[12],ymm5[12],ymm1[13],ymm5[13],ymm1[14],ymm5[14],ymm1[15],ymm5[15],ymm1[24],ymm5[24],ymm1[25],ymm5[25],ymm1[26],ymm5[26],ymm1[27],ymm5[27],ymm1[28],ymm5[28],ymm1[29],ymm5[29],ymm1[30],ymm5[30],ymm1[31],ymm5[31]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [256,128,64,32,16,8,4,2,256,128,64,32,16,8,4,2]
 ; AVX512VL-NEXT:    # ymm10 = mem[0,1,0,1]
-; AVX512VL-NEXT:    vpmullw %ymm1, %ymm10, %ymm1
+; AVX512VL-NEXT:    vpmullw %ymm10, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpsrlw $8, %ymm9, %ymm9
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0],ymm5[0],ymm1[1],ymm5[1],ymm1[2],ymm5[2],ymm1[3],ymm5[3],ymm1[4],ymm5[4],ymm1[5],ymm5[5],ymm1[6],ymm5[6],ymm1[7],ymm5[7],ymm1[16],ymm5[16],ymm1[17],ymm5[17],ymm1[18],ymm5[18],ymm1[19],ymm5[19],ymm1[20],ymm5[20],ymm1[21],ymm5[21],ymm1[22],ymm5[22],ymm1[23],ymm5[23]
+; AVX512VL-NEXT:    vbroadcasti128 {{.*#+}} ymm11 = [256,2,4,8,16,32,64,128,256,2,4,8,16,32,64,128]
+; AVX512VL-NEXT:    # ymm11 = mem[0,1,0,1]
+; AVX512VL-NEXT:    vpmullw %ymm1, %ymm11, %ymm1
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512VL-NEXT:    vpackuswb %ymm5, %ymm1, %ymm1
+; AVX512VL-NEXT:    vpackuswb %ymm9, %ymm1, %ymm1
 ; AVX512VL-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512VL-NEXT:    vpsllw $4, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpand %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm2
 ; AVX512VL-NEXT:    vpsllw $2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpand %ymm6, %ymm3, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm7, %ymm3, %ymm2, %ymm2
 ; AVX512VL-NEXT:    vpaddb %ymm2, %ymm2, %ymm3
 ; AVX512VL-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
-; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpmullw %ymm3, %ymm9, %ymm3
+; AVX512VL-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8],ymm5[8],ymm0[9],ymm5[9],ymm0[10],ymm5[10],ymm0[11],ymm5[11],ymm0[12],ymm5[12],ymm0[13],ymm5[13],ymm0[14],ymm5[14],ymm0[15],ymm5[15],ymm0[24],ymm5[24],ymm0[25],ymm5[25],ymm0[26],ymm5[26],ymm0[27],ymm5[27],ymm0[28],ymm5[28],ymm0[29],ymm5[29],ymm0[30],ymm5[30],ymm0[31],ymm5[31]
+; AVX512VL-NEXT:    vpmullw %ymm3, %ymm10, %ymm3
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm3, %ymm3
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmullw %ymm0, %ymm10, %ymm0
+; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm5[0],ymm0[1],ymm5[1],ymm0[2],ymm5[2],ymm0[3],ymm5[3],ymm0[4],ymm5[4],ymm0[5],ymm5[5],ymm0[6],ymm5[6],ymm0[7],ymm5[7],ymm0[16],ymm5[16],ymm0[17],ymm5[17],ymm0[18],ymm5[18],ymm0[19],ymm5[19],ymm0[20],ymm5[20],ymm0[21],ymm5[21],ymm0[22],ymm5[22],ymm0[23],ymm5[23]
+; AVX512VL-NEXT:    vpmullw %ymm0, %ymm11, %ymm0
 ; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512VL-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512VL-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_rotate_v64i8:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
 ; AVX512BW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512BW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512BW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512BW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512BW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512BW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512VLBW-LABEL: constant_rotate_v64i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    vbroadcasti32x4 {{.*#+}} zmm1 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256]
 ; AVX512VLBW-NEXT:    # zmm1 = mem[0,1,2,3,0,1,2,3,0,1,2,3,0,1,2,3]
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpsllw $4, %zmm0, %zmm2
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm2, %zmm2
 ; AVX512VLBW-NEXT:    vpblendmb %zmm2, %zmm0, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpsllw $2, %zmm2, %zmm3
 ; AVX512VLBW-NEXT:    vpandq {{.*}}(%rip), %zmm3, %zmm3
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
 ; AVX512VLBW-NEXT:    vpaddb %zmm1, %zmm1, %zmm1
 ; AVX512VLBW-NEXT:    vpmovb2m %zmm1, %k1
 ; AVX512VLBW-NEXT:    vpaddb %zmm2, %zmm2, %zmm2 {%k1}
-; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512VLBW-NEXT:    vpunpckhbw {{.*#+}} zmm3 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpsrlw $8, %zmm3, %zmm3
+; AVX512VLBW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512VLBW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512VLBW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512VLBW-NEXT:    vpackuswb %zmm3, %zmm0, %zmm0
 ; AVX512VLBW-NEXT:    vporq %zmm0, %zmm2, %zmm0
 ; AVX512VLBW-NEXT:    retq
   %shl = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>
   %lshr = lshr <64 x i8> %a, <i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>
   %or = or <64 x i8> %shl, %lshr
   ret <64 x i8> %or
 }
 

llvm/test/CodeGen/X86/vector-shift-lshr-256.ll

 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BW-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQVL-LABEL: constant_shift_v32i8:
 ; AVX512DQVL:       # %bb.0:
-; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; AVX512DQVL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; AVX512DQVL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512DQVL-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31]
+; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm2, %ymm2
+; AVX512DQVL-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; AVX512DQVL-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23]
 ; AVX512DQVL-NEXT:    vpmullw {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpackuswb %ymm1, %ymm0, %ymm0
+; AVX512DQVL-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; AVX512DQVL-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: constant_shift_v32i8:
 ; AVX512BWVL:       # %bb.0:
 ; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero
 ; AVX512BWVL-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BWVL-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BWVL-NEXT:    retq

llvm/test/CodeGen/X86/vector-shift-lshr-512.ll

 ; AVX512DQ-NEXT:    vpmullw %ymm5, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vpackuswb %ymm3, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512BW-LABEL: constant_shift_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm1, %zmm1
-; AVX512BW-NEXT:    vpsrlw $8, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
-; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpunpckhbw {{.*#+}} zmm2 = zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[12],zmm1[12],zmm0[13],zmm1[13],zmm0[14],zmm1[14],zmm0[15],zmm1[15],zmm0[24],zmm1[24],zmm0[25],zmm1[25],zmm0[26],zmm1[26],zmm0[27],zmm1[27],zmm0[28],zmm1[28],zmm0[29],zmm1[29],zmm0[30],zmm1[30],zmm0[31],zmm1[31],zmm0[40],zmm1[40],zmm0[41],zmm1[41],zmm0[42],zmm1[42],zmm0[43],zmm1[43],zmm0[44],zmm1[44],zmm0[45],zmm1[45],zmm0[46],zmm1[46],zmm0[47],zmm1[47],zmm0[56],zmm1[56],zmm0[57],zmm1[57],zmm0[58],zmm1[58],zmm0[59],zmm1[59],zmm0[60],zmm1[60],zmm0[61],zmm1[61],zmm0[62],zmm1[62],zmm0[63],zmm1[63]
+; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm2, %zmm2
+; AVX512BW-NEXT:    vpsrlw $8, %zmm2, %zmm2
+; AVX512BW-NEXT:    vpunpcklbw {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[16],zmm1[16],zmm0[17],zmm1[17],zmm0[18],zmm1[18],zmm0[19],zmm1[19],zmm0[20],zmm1[20],zmm0[21],zmm1[21],zmm0[22],zmm1[22],zmm0[23],zmm1[23],zmm0[32],zmm1[32],zmm0[33],zmm1[33],zmm0[34],zmm1[34],zmm0[35],zmm1[35],zmm0[36],zmm1[36],zmm0[37],zmm1[37],zmm0[38],zmm1[38],zmm0[39],zmm1[39],zmm0[48],zmm1[48],zmm0[49],zmm1[49],zmm0[50],zmm1[50],zmm0[51],zmm1[51],zmm0[52],zmm1[52],zmm0[53],zmm1[53],zmm0[54],zmm1[54],zmm0[55],zmm1[55]
 ; AVX512BW-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
-; AVX512BW-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpackuswb %zmm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
   %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
   ret <64 x i8> %shift
 }
 
 ;
 ; Uniform Constant Shifts
 ;

llvm/test/CodeGen/X86/vector-shuffle-128-v8.ll

 ; AVX2-FAST-LABEL: shuffle_v8i16_01100110:
 ; AVX2-FAST:       # %bb.0:
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,2,3,0,1,0,1,2,3,2,3,0,1]
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i16_01100110:
 ; AVX512VL-SLOW:       # %bb.0:
 ; AVX512VL-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,1,0,4,5,6,7]
-; AVX512VL-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
+; AVX512VL-SLOW-NEXT:    vpbroadcastq %xmm0, %xmm0
 ; AVX512VL-SLOW-NEXT:    retq
 ;
 ; AVX512VL-FAST-LABEL: shuffle_v8i16_01100110:
 ; AVX512VL-FAST:       # %bb.0:
 ; AVX512VL-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,2,3,0,1,0,1,2,3,2,3,0,1]
 ; AVX512VL-FAST-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v8i16_01100110:

 ; AVX2-FAST-LABEL: shuffle_v8i16_01u0u110:
 ; AVX2-FAST:       # %bb.0:
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,2,3,0,1,0,1,2,3,2,3,0,1]
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i16_01u0u110:
 ; AVX512VL-SLOW:       # %bb.0:
 ; AVX512VL-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,1,0,4,5,6,7]
-; AVX512VL-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
+; AVX512VL-SLOW-NEXT:    vpbroadcastq %xmm0, %xmm0
 ; AVX512VL-SLOW-NEXT:    retq
 ;
 ; AVX512VL-FAST-LABEL: shuffle_v8i16_01u0u110:
 ; AVX512VL-FAST:       # %bb.0:
 ; AVX512VL-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,2,3,0,1,0,1,2,3,2,3,0,1]
 ; AVX512VL-FAST-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v8i16_01u0u110:

llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll

; CHECK-NEXT:    ret{{[l|q]}}
   %3 = bitcast <32 x i16> %1 to <64 x i8>
   %4 = bitcast <32 x i16> %2 to <64 x i8>
   %5 = tail call <64 x i8> @llvm.x86.avx512.mask.pshuf.b.512(<64 x i8> %3, <64 x i8> <i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <64 x i8> undef, i64 -1)
   %6 = tail call <64 x i8> @llvm.x86.avx512.mask.pshuf.b.512(<64 x i8> %4, <64 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 0, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14>, <64 x i8> undef, i64 -1)
   %7 = or <64 x i8> %5, %6
   ret <64 x i8> %7
 }
 
+define <32 x i16> @combine_vpermi2var_32i16_as_rotate(<32 x i16> %a0) {
+; CHECK-LABEL: combine_vpermi2var_32i16_as_rotate:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vprold $16, %zmm0, %zmm0
+; CHECK-NEXT:    ret{{[l|q]}}
+  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %a0, <32 x i16> <i16 1, i16 0, i16 3, i16 2, i16 4, i16 5, i16 6, i16 7, i16 9, i16 8, i16 11, i16 10, i16 12, i16 13, i16 14, i16 15, i16 17, i16 16, i16 19, i16 18, i16 20, i16 21, i16 22, i16 23, i16 25, i16 24, i16 27, i16 26, i16 28, i16 29, i16 30, i16 31>)
+  %2 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %1, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 5, i16 4, i16 7, i16 6, i16 8, i16 9, i16 10, i16 11, i16 13, i16 12, i16 15, i16 14, i16 16, i16 17, i16 18, i16 19, i16 21, i16 20, i16 23, i16 22, i16 24, i16 25, i16 26, i16 27, i16 29, i16 28, i16 31, i16 30>)
+  ret <32 x i16> %2
+}
+
 define <32 x i16> @combine_vpermi2var_32i16_as_pshufb(<32 x i16> %a0) {
 ; CHECK-LABEL: combine_vpermi2var_32i16_as_pshufb:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13,18,19,16,17,22,23,20,21,26,27,24,25,30,31,28,29,34,35,32,33,38,39,36,37,42,43,40,41,46,47,44,45,50,51,48,49,54,55,52,53,58,59,56,57,62,63,60,61]
+; CHECK-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[2,3,0,1,4,5,6,7,10,11,8,9,14,15,12,13,18,19,16,17,20,21,22,23,26,27,24,25,30,31,28,29,34,35,32,33,36,37,38,39,42,43,40,41,46,47,44,45,50,51,48,49,52,53,54,55,58,59,56,57,62,63,60,61]
 ; CHECK-NEXT:    ret{{[l|q]}}
-  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %a0, <32 x i16> <i16 1, i16 0, i16 3, i16 2, i16 4, i16 5, i16 6, i16 7, i16 9, i16 8, i16 11, i16 10, i16 12, i16 13, i16 14, i16 15, i16 17, i16 16, i16 19, i16 18, i16 20, i16 21, i16 22, i16 23, i16 25, i16 24, i16 27, i16 26, i16 28, i16 29, i16 30, i16 31>)
+  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %a0, <32 x i16> <i16 1, i16 0, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 9, i16 8, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 16, i16 18, i16 19, i16 20, i16 21, i16 22, i16 23, i16 25, i16 24, i16 26, i16 27, i16 28, i16 29, i16 30, i16 31>)
   %2 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %1, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 5, i16 4, i16 7, i16 6, i16 8, i16 9, i16 10, i16 11, i16 13, i16 12, i16 15, i16 14, i16 16, i16 17, i16 18, i16 19, i16 21, i16 20, i16 23, i16 22, i16 24, i16 25, i16 26, i16 27, i16 29, i16 28, i16 31, i16 30>)
   ret <32 x i16> %2
 }
 
 define <32 x i16> @combine_vpermi2var_32i16_identity(<32 x i16> %x0, <32 x i16> %x1) {
 ; CHECK-LABEL: combine_vpermi2var_32i16_identity:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    ret{{[l|q]}}

llvm/test/CodeGen/X86/vector-shuffle-combining-avx512f.ll

; X64-AVX512BW-NEXT:    retq
   %res0 = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>, <16 x float> %x0, <16 x float> %x1, i16 %m)
   %res1 = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> <i32 15, i32 30, i32 13, i32 28, i32 11, i32 26, i32 9, i32 24, i32 7, i32 22, i32 5, i32 20, i32 3, i32 18, i32 1, i32 16>, <16 x float> %res0, <16 x float> %res0, i16 %m)
   ret <16 x float> %res1
 }
 
 define <16 x float> @combine_vpermt2var_16f32_vmovddup(<16 x float> %x0, <16 x float> %x1) {
 ; CHECK-LABEL: combine_vpermt2var_16f32_vmovddup:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} zmm1 = [0,1,0,1,4,5,4,5,8,9,8,9,12,13,12,13]
-; CHECK-NEXT:    vpermt2ps %zmm0, %zmm1, %zmm0
+; CHECK-NEXT:    vmovddup {{.*#+}} zmm0 = zmm0[0,0,2,2,4,4,6,6]
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res0 = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5, i32 8, i32 9, i32 8, i32 9, i32 12, i32 13, i32 12, i32 13>, <16 x float> %x0, <16 x float> %x1, i16 -1)
   ret <16 x float> %res0
 }
 define <16 x float> @combine_vpermt2var_16f32_vmovddup_load(<16 x float> *%p0, <16 x float> %x1) {
 ; X86-LABEL: combine_vpermt2var_16f32_vmovddup_load:
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    vmovaps (%eax), %zmm1
-; X86-NEXT:    vmovaps {{.*#+}} zmm0 = [0,1,0,1,4,5,4,5,8,9,8,9,12,13,12,13]
-; X86-NEXT:    vpermi2ps %zmm0, %zmm1, %zmm0
+; X86-NEXT:    vmovddup {{.*#+}} zmm0 = mem[0,0,2,2,4,4,6,6]
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_vpermt2var_16f32_vmovddup_load:
 ; X64:       # %bb.0:
-; X64-NEXT:    vmovaps (%rdi), %zmm1
-; X64-NEXT:    vmovaps {{.*#+}} zmm0 = [0,1,0,1,4,5,4,5,8,9,8,9,12,13,12,13]
-; X64-NEXT:    vpermi2ps %zmm0, %zmm1, %zmm0
+; X64-NEXT:    vmovddup {{.*#+}} zmm0 = mem[0,0,2,2,4,4,6,6]
 ; X64-NEXT:    retq
   %x0 = load <16 x float>, <16 x float> *%p0
   %res0 = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5, i32 8, i32 9, i32 8, i32 9, i32 12, i32 13, i32 12, i32 13>, <16 x float> %x0, <16 x float> %x1, i16 -1)
   ret <16 x float> %res0
 }
 define <16 x float> @combine_vpermt2var_16f32_vmovddup_mask(<16 x float> %x0, <16 x float> %x1, i16 %m) {
 ; X86-LABEL: combine_vpermt2var_16f32_vmovddup_mask:
 ; X86:       # %bb.0:
; CHECK-NEXT:    ret{{[l|q]}}
   %res0 = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8>, <16 x i32> %x0, <16 x i32> %x1, i16 -1)
   %res1 = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> <i32 0, i32 16, i32 2, i32 18, i32 4, i32 20, i32 6, i32 22, i32 8, i32 24, i32 10, i32 26, i32 12, i32 28, i32 14, i32 30>, <16 x i32> %res0, <16 x i32> %res0, i16 -1)
   ret <16 x i32> %res1
 }
 
 define <16 x i32> @combine_vpermt2var_16i32_as_vpsrlq(<16 x i32> %x0) {
 ; CHECK-LABEL: combine_vpermt2var_16i32_as_vpsrlq:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [1,16,3,16,5,16,7,16,9,16,11,16,13,16,15,16]
-; CHECK-NEXT:    vpermt2d %zmm1, %zmm2, %zmm0
+; CHECK-NEXT:    vpsrlq $32, %zmm0, %zmm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res0 = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> <i32 1, i32 16, i32 3, i32 16, i32 5, i32 16, i32 7, i32 16, i32 9, i32 16, i32 11, i32 16, i32 13, i32 16, i32 15, i32 16>, <16 x i32> %x0, <16 x i32> zeroinitializer, i16 -1)
   ret <16 x i32> %res0
 }
 
 define <16 x i32> @combine_vpermt2var_16i32_as_vpsllq(<16 x i32> %x0) {
 ; CHECK-LABEL: combine_vpermt2var_16i32_as_vpsllq:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; CHECK-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [16,0,16,2,16,4,16,6,16,8,16,10,16,12,16,14]
-; CHECK-NEXT:    vpermt2d %zmm1, %zmm2, %zmm0
+; CHECK-NEXT:    vpsllq $32, %zmm0, %zmm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res0 = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> <i32 16, i32 0, i32 16, i32 2, i32 16, i32 4, i32 16, i32 6, i32 16, i32 8, i32 16, i32 10, i32 16, i32 12, i32 16, i32 14>, <16 x i32> %x0, <16 x i32> zeroinitializer, i16 -1)
   ret <16 x i32> %res0
 }
 
 define <8 x double> @combine_vpermi2var_vpermt2var_8f64_as_vperm2(<8 x double> %x0, <8 x double> %x1) {
 ; X86-LABEL: combine_vpermi2var_vpermt2var_8f64_as_vperm2:
 ; X86:       # %bb.0:

llvm/test/CodeGen/X86/vector-shuffle-v1.ll

 ; AVX512VL-LABEL: shuf2i1_1_2:
 ; AVX512VL:       # %bb.0:
 ; AVX512VL-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vptestmq %xmm0, %xmm0, %k1
 ; AVX512VL-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} {z}
 ; AVX512VL-NEXT:    movq $-1, %rax
 ; AVX512VL-NEXT:    vmovq %rax, %xmm2
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,2]
-; AVX512VL-NEXT:    vpermi2q %xmm2, %xmm1, %xmm3
-; AVX512VL-NEXT:    vptestmq %xmm3, %xmm3, %k1
+; AVX512VL-NEXT:    vpalignr {{.*#+}} xmm1 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vptestmq %xmm1, %xmm1, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512VL-NEXT:    retq
 ;
 ; VL_BW_DQ-LABEL: shuf2i1_1_2:
 ; VL_BW_DQ:       # %bb.0:
 ; VL_BW_DQ-NEXT:    vpsllq $63, %xmm0, %xmm0
 ; VL_BW_DQ-NEXT:    vpmovq2m %xmm0, %k0
 ; VL_BW_DQ-NEXT:    movq $-1, %rax
 ; VL_BW_DQ-NEXT:    vmovq %rax, %xmm0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm1
-; VL_BW_DQ-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,2]
-; VL_BW_DQ-NEXT:    vpermi2q %xmm0, %xmm1, %xmm2
-; VL_BW_DQ-NEXT:    vpmovq2m %xmm2, %k0
+; VL_BW_DQ-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
+; VL_BW_DQ-NEXT:    vpmovq2m %xmm0, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm0
 ; VL_BW_DQ-NEXT:    retq
   %b = shufflevector <2 x i1> %a, <2 x i1> <i1 1, i1 0>, <2 x i32> <i32 1, i32 2>
   ret <2 x i1> %b
 }
 
 
 define <4 x i1> @shuf4i1_3_2_10(<4 x i1> %a) {