[X86 isel] Fix permute mask calculation in lowerShuffleAsUNPCKAndPermute

[zhuhan0]

Summary

This fixes issue 62242

This code block can potentially swap the order of V1 and V2 in Ops and therefore
also in the unpck instruction generated.

SDValue &Op = Ops[Elt & 1];
if (M < NumElts && (Op.isUndef() || Op == V1))
    Op = V1;
else if (NumElts <= M && (Op.isUndef() || Op == V2)) {
    Op = V2;
    NormM -= NumElts;
} else
    return SDValue();

But the permute mask is calculated assuming the first operand being V1 and
second V2, therefore causing a mis-compile.

First check if the input operands are swapped, and then calculate the permute
mask based on that.

[RKSimon]

Chaging the permute mask calculation based on whether V1 and V2 are swapped could be troublesome

I'm not sure I follow why you think this could be a problem?

llvm/lib/Target/X86/X86ISelLowering.cpp

13341

We're going to see regressions from this, because by forcing the V1/V2 unpack order we lose any chance of making use of a better shuffles by V2/V1

[zhuhan0]

In D148843#4286930, @RKSimon wrote:

Chaging the permute mask calculation based on whether V1 and V2 are swapped could be troublesome

I'm not sure I follow why you think this could be a problem?

Just thought the code would be ugly. Nothing major. I'll look into this since the current fix is introducing a regression.

llvm/lib/Target/X86/X86ISelLowering.cpp

13341

I see. Curious how does the order of the operands affect the quality of generated shuffles? Is it with optimization purpose to set the ops like this SDValue &Op = Ops[Elt & 1];?

[zhuhan0]

Modify permute mask instead.

[zhuhan0]

Looks like ef38880 also introduced some miscompiles in the vector-interleaved* tests. With this change, the tests have been reverted to pre-ef38880 version.

[zhuhan0]

Use isInRange.

[RKSimon]

llvm/lib/Target/X86/X86ISelLowering.cpp

13338

Is the OpsSwapped logic necessary?

for (int Elt = 0; Elt != NumElts; ++Elt) {
  int M = Mask[Elt];
  if (M < 0)
    continue;
  int NormM = M;
  if (NumElts <= M)
    NormM -= NumElts;
  bool IsFirstOp = M < NumElts;
  int BaseMaskElt =
      NumLaneElts * (NormM / NumLaneElts) + (2 * (NormM % NumHalfLaneElts));
  if ((IsFirstOp && V1 == Ops[0]) || (!IsFirstOp && V2 == Ops[0]))
    PermuteMask[Elt] = BaseMaskElt;
  else if if ((IsFirstOp && V1 == Ops[1]) || (!IsFirstOp && V2 == Ops[1]))
    PermuteMask[Elt] = BaseMaskElt + 1;
}

I think there probably needs to be an assertion there as well someplace.

[zhuhan0]

Address comment.

[RKSimon]

LGTM - cheers

llvm/lib/Target/X86/X86ISelLowering.cpp

static SDValue lowerShuffleAsUNPCKAndPermute(const SDLoc &DL, MVT VT,
                                              ArrayRef<int> Mask,
                                              SelectionDAG &DAG) {
   int NumElts = Mask.size();
   int NumLanes = VT.getSizeInBits() / 128;
   int NumLaneElts = NumElts / NumLanes;
   int NumHalfLaneElts = NumLaneElts / 2;
 
   bool MatchLo = true, MatchHi = true;
-  SDValue Ops[2] = {DAG.getUNDEF(VT), DAG.getUNDEF(VT)};
+  SDValue PrevOp = DAG.getUNDEF(VT);
 
   // Determine UNPCKL/UNPCKH type and operand order.
   for (int Elt = 0; Elt != NumElts; ++Elt) {
     int M = Mask[Elt];
     if (M < 0)
       continue;
 
+    // Check that Mask is interleaving elements from V1 and V2.
     // Normalize the mask value depending on whether it's V1 or V2.
     int NormM = M;
-    SDValue &Op = Ops[Elt & 1];
-    if (M < NumElts && (Op.isUndef() || Op == V1))
-      Op = V1;
-    else if (NumElts <= M && (Op.isUndef() || Op == V2)) {
-      Op = V2;
+    SDValue CurrOp = DAG.getUNDEF(VT);
+    if (M < NumElts)
+      CurrOp = V1;
+    else if (NumElts <= M) {
+      CurrOp = V2;
       NormM -= NumElts;
-    } else
+    }
+    if (!PrevOp.isUndef() && CurrOp == PrevOp)
       return SDValue();
+    PrevOp = CurrOp;
 
     bool MatchLoAnyLane = false, MatchHiAnyLane = false;
     for (int Lane = 0; Lane != NumElts; Lane += NumLaneElts) {
       int Lo = Lane, Mid = Lane + NumHalfLaneElts, Hi = Lane + NumLaneElts;
       MatchLoAnyLane |= isUndefOrInRange(NormM, Lo, Mid);
       MatchHiAnyLane |= isUndefOrInRange(NormM, Mid, Hi);
       if (MatchLoAnyLane || MatchHiAnyLane) {
         assert((MatchLoAnyLane ^ MatchHiAnyLane) &&
static SDValue lowerShuffleAsUNPCKAndPermute(const SDLoc &DL, MVT VT,
   for (int Elt = 0; Elt != NumElts; ++Elt) {
     int M = Mask[Elt];
     if (NumElts <= M)
       PermuteMask[Elt] = NumLaneElts * ((M - NumElts) / NumLaneElts) +
                          (2 * (M % NumHalfLaneElts)) + 1;
     else if (0 <= M)
       PermuteMask[Elt] =
           NumLaneElts * (M / NumLaneElts) + (2 * (M % NumHalfLaneElts));
   }
 
   unsigned UnpckOp = MatchLo ? X86ISD::UNPCKL : X86ISD::UNPCKH;
-  SDValue Unpck = DAG.getNode(UnpckOp, DL, VT, Ops);
+  SDValue Unpck = DAG.getNode(UnpckOp, DL, VT, {V1, V2});
   return DAG.getVectorShuffle(VT, DL, Unpck, DAG.getUNDEF(VT), PermuteMask);
 }
 
 /// Try to lower a shuffle as a permute of the inputs followed by an
 /// UNPCK instruction.
 ///
 /// This specifically targets cases where we end up with alternating between
 /// the two inputs, and so can permute them into something that feeds a single

llvm/test/CodeGen/X86/pr62242.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+
+define <4 x i32> @unpck_permute_canonicalized_mask(<4 x i32> %a, <4 x i32> %b) {
+; AVX2-LABEL: unpck_permute_canonicalized_mask:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX2-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,2,2,0]
+; AVX2-NEXT:    retq
+  ; This mask will get canonicalized to vector_shuffle<6, 3, -1, 2> %b, %a.
+  ; Make sure the generated permute masks are still correct.
+  %shuffle = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 2, i32 7, i32 undef, i32 6>
+  ret <4 x i32> %shuffle
+}

llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-7.ll

 ; AVX1-ONLY-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; AVX1-ONLY-NEXT:    vmovaps %xmm0, 208(%rax)
 ; AVX1-ONLY-NEXT:    addq $40, %rsp
 ; AVX1-ONLY-NEXT:    vzeroupper
 ; AVX1-ONLY-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: store_i16_stride7_vf16:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    pushq %rax
+; AVX2-SLOW-NEXT:    subq $72, %rsp
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm6
-; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm5
-; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %ymm4
-; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm3
-; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm2
-; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm7
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm14
+; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %ymm15
+; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm4
+; AVX2-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm5
+; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm3
+; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <3,u,u,u,4,u,u,4>
-; AVX2-SLOW-NEXT:    vpermd %ymm7, %ymm0, %ymm0
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm6[0,3,2,3,4,7,6,7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm8[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpermd %ymm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[0,3,2,3,4,7,6,7]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm1[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,3,u,u,u,4,u,u>
-; AVX2-SLOW-NEXT:    vpermd %ymm5, %ymm8, %ymm8
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm4[0,0,0,0,4,5,6,7,8,8,8,8,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,3,u,u,u,4,u,u>
+; AVX2-SLOW-NEXT:    vpermd %ymm14, %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm15[0,0,0,0,4,5,6,7,8,8,8,8,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm9 = ymm9[0,1,2,3,7,7,7,7,8,9,10,11,15,15,15,15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm1, %ymm9, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,3,u,u,u,4,u>
-; AVX2-SLOW-NEXT:    vpermd %ymm3, %ymm8, %ymm8
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm2[0,1,0,3,4,5,4,7]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,3,u,u,u,4,u>
+; AVX2-SLOW-NEXT:    vpermd %ymm4, %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm5[0,1,0,3,4,5,4,7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm5, %ymm11
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm9 = ymm9[0,1,2,3,4,4,7,7,8,9,10,11,12,12,15,15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm1, %ymm9, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <3,u,u,3,u,u,u,4>
-; AVX2-SLOW-NEXT:    vpermd %ymm1, %ymm9, %ymm9
+; AVX2-SLOW-NEXT:    vpermd %ymm3, %ymm9, %ymm9
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm1, %ymm9, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %xmm10
-; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm11
-; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm5
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm5[4],xmm10[4],xmm5[5],xmm10[5],xmm5[6],xmm10[6],xmm5[7],xmm10[7]
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,2,3,3,4,5,6,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm13
-; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm3
-; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm3[4],xmm13[4],xmm3[5],xmm13[5],xmm3[6],xmm13[6],xmm3[7],xmm13[7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm8[2,1,2,3,4,5,6,7]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,5,4]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,1,3]
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm4
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm13[4],xmm4[5],xmm13[5],xmm4[6],xmm13[6],xmm4[7],xmm13[7]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,4]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm8, %ymm0, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa (%r9), %xmm9
-; AVX2-SLOW-NEXT:    vmovdqa (%r8), %xmm12
-; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm12[4],xmm9[4],xmm12[5],xmm9[5],xmm12[6],xmm9[6],xmm12[7],xmm9[7]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm1, %ymm0, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqa (%r9), %xmm1
+; AVX2-SLOW-NEXT:    vmovdqa (%r8), %xmm9
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm12 = xmm9[4],xmm1[4],xmm9[5],xmm1[5],xmm9[6],xmm1[6],xmm9[7],xmm1[7]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,1,1]
 ; AVX2-SLOW-NEXT:    vpbroadcastd 8(%rax), %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm8, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm12, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm10[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm11[1,1,2,2]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3],xmm0[4],xmm1[5,6],xmm0[7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm3[3,3,3,3,4,5,6,7]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm13[1,1,2,3]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm15[0,1],xmm1[2],xmm15[3,4],xmm1[5],xmm15[6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[1,1,2,2]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3],xmm0[4],xmm2[5,6],xmm0[7]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm4[3,3,3,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm8 = xmm13[1,1,2,3]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} xmm2 = xmm8[0,1],xmm2[2],xmm8[3,4],xmm2[5],xmm8[6,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,1]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm0, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm12[0],xmm9[0],xmm12[1],xmm9[1],xmm12[2],xmm9[2],xmm12[3],xmm9[3]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm9 = xmm0[0,1,2,3,4,5,7,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[0,2,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-SLOW-NEXT:    vpbroadcastd 4(%rax), %ymm12
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm9, %ymm12, %ymm9
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm1, %ymm9, %ymm9
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm6[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm6, %ymm8
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm7[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm12[0,1,2],ymm1[3],ymm12[4,5],ymm1[6],ymm12[7,8,9,10],ymm1[11],ymm12[12,13],ymm1[14],ymm12[15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm12 = ymm4[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm5[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm15[0],ymm12[1],ymm15[2,3],ymm12[4],ymm15[5,6,7,8],ymm12[9],ymm15[10,11],ymm12[12],ymm15[13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm1, %ymm12, %ymm1
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm12 = ymm2[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm6
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[2,2,2,3,6,6,6,7]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm2[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0,1],ymm15[2],ymm12[3,4],ymm15[5],ymm12[6,7,8,9],ymm15[10],ymm12[11,12],ymm15[13],ymm12[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm1, %ymm12, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm14[2,3,3,3,6,7,7,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm1, %ymm12, %ymm12
-; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm13[0],xmm3[0],xmm13[1],xmm3[1],xmm13[2],xmm3[2],xmm13[3],xmm3[3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
-; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm10 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm10 = xmm10[0,1,3,2,4,5,6,7]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm10 = xmm10[0,0,1,1]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm1, %ymm10, %ymm1
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm0, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm1[0],xmm9[1],xmm1[1],xmm9[2],xmm1[2],xmm9[3],xmm1[3]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,7,6]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,3,3]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
+; AVX2-SLOW-NEXT:    vpbroadcastd 4(%rax), %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm1, %ymm8, %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm2, %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm6[4],ymm7[4],ymm6[5],ymm7[5],ymm6[6],ymm7[6],ymm6[7],ymm7[7],ymm6[12],ymm7[12],ymm6[13],ymm7[13],ymm6[14],ymm7[14],ymm6[15],ymm7[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,3,3,4,6,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm8 = ymm14[4],ymm15[4],ymm14[5],ymm15[5],ymm14[6],ymm15[6],ymm14[7],ymm15[7],ymm14[12],ymm15[12],ymm14[13],ymm15[13],ymm14[14],ymm15[14],ymm14[15],ymm15[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm2, %ymm8, %ymm2
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm8 = ymm11[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm11, %ymm9
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[2,2,2,3,6,6,6,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm3[3,3,3,3,7,7,7,7]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm12[2],ymm8[3,4],ymm12[5],ymm8[6,7,8,9],ymm12[10],ymm8[11,12],ymm12[13],ymm8[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm2, %ymm8, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm11[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm2, %ymm8, %ymm12
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm13[0],xmm4[0],xmm13[1],xmm4[1],xmm13[2],xmm4[2],xmm13[3],xmm4[3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm5[0],xmm10[0],xmm5[1],xmm10[1],xmm5[2],xmm10[2],xmm5[3],xmm10[3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm8[0,1,3,2,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm2, %ymm8, %ymm2
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
-; AVX2-SLOW-NEXT:    vpbroadcastd (%rax), %ymm10
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm0, %ymm10, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm1, %ymm0, %ymm10
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm5[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm5, %ymm3
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
+; AVX2-SLOW-NEXT:    vpbroadcastd (%rax), %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm0, %ymm8, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm2, %ymm0, %ymm10
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm14[2,2,2,2,6,6,6,6]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0,1],ymm0[2],ymm2[3,4],ymm0[5],ymm2[6,7,8,9],ymm0[10],ymm2[11,12],ymm0[13],ymm2[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm8[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0],ymm11[1],ymm1[2,3],ymm11[4],ymm1[5,6,7,8],ymm11[9],ymm1[10,11],ymm11[12],ymm1[13,14,15]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm1[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm6[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm11[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm11[2],ymm1[3,4],ymm11[5],ymm1[6,7,8,9],ymm11[10],ymm1[11,12],ymm11[13],ymm1[14,15]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[1,2,2,3,5,6,6,7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm3, %ymm4
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm9[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm14[0,1,2,2,4,5,6,6]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm1, %ymm11, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,1,1,1,5,5,5,5]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm5 = ymm8[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[0,0,2,1,4,4,6,5]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7,8,9],ymm1[10],ymm5[11,12],ymm1[13],ymm5[14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm11[0,1,2,2,4,5,6,6]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm1, %ymm2, %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[1,1,1,1,5,5,5,5]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm7[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,2,1,4,4,6,5]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,1,1,3,4,5,5,7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm4 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[0,0,0,0,4,4,4,4]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7,8,9],ymm4[10],ymm3[11,12],ymm4[13],ymm3[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[0,0,2,1,4,4,6,5]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm6[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,0,0,4,4,4,4]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7,8,9,10],ymm3[11],ymm2[12,13],ymm3[14],ymm2[15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm14[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm15[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,0,0,0,4,4,4,4]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7,8,9],ymm3[10],ymm2[11,12],ymm3[13],ymm2[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[0,0,2,1,4,4,6,5]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm9[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,0,0,0,4,4,4,4]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7,8,9,10],ymm2[11],ymm3[12,13],ymm2[14],ymm3[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,3,3]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm14[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm11[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm2, 96(%rax)
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 128(%rax)
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 160(%rax)
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm10, (%rax)
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm12, 192(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm9, 32(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovaps %ymm0, 32(%rax)
+; AVX2-SLOW-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 64(%rax)
-; AVX2-SLOW-NEXT:    popq %rax
+; AVX2-SLOW-NEXT:    addq $72, %rsp
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: store_i16_stride7_vf16:
 ; AVX2-FAST:       # %bb.0:
 ; AVX2-FAST-NEXT:    pushq %rax
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm5
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm7
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm8
+; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm9
 ; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm4
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm6
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm15
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm5
+; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm6
 ; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm3
 ; AVX2-FAST-NEXT:    vmovdqa (%rax), %ymm1
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm4[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7,8,9],ymm0[10],ymm8[11,12],ymm0[13],ymm8[14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm7[u,u,8,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,u,u,u,u,26,27,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm5[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm8 = ymm9[0],ymm8[1],ymm9[2,3],ymm8[4],ymm9[5,6,7,8],ymm8[9],ymm9[10,11],ymm8[12],ymm9[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [4,5,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm8, %ymm8
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm3[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm15[1,2,2,3,5,6,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm9 = ymm10[0,1],ymm9[2],ymm10[3,4],ymm9[5],ymm10[6,7,8,9],ymm9[10],ymm10[11,12],ymm9[13],ymm10[14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm9, %ymm8, %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15]
+; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [0,1,4,5,4,5,5,7]
+; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm7, %ymm0
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm4[2,2,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm10, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [4,5,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm10, %ymm10
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm3[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm12 = ymm6[1,2,2,3,5,6,6,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0,1],ymm11[2],ymm12[3,4],ymm11[5],ymm12[6,7,8,9],ymm11[10],ymm12[11,12],ymm11[13],ymm12[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm11, %ymm10, %ymm10
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm0, %ymm10, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm5[1,1,1,1,5,5,5,5]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,u,u,20,21,24,25,u,u,22,23,22,23]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7,8,9],ymm0[10],ymm8[11,12],ymm0[13],ymm8[14,15]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm8[1,1,1,1,5,5,5,5]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,u,u,20,21,24,25,u,u,22,23,22,23]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm10[0,1],ymm0[2],ymm10[3,4],ymm0[5],ymm10[6,7,8,9],ymm0[10],ymm10[11,12],ymm0[13],ymm10[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm8 = ymm6[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm4[0,1,1,3,4,5,5,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm8 = ymm9[0,1],ymm8[2],ymm9[3,4],ymm8[5],ymm9[6,7,8,9],ymm8[10],ymm9[11,12],ymm8[13],ymm9[14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [0,1,4,5,4,5,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm8, %ymm8
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm5[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm4[0,1,1,3,4,5,5,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm0, %ymm10, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm15[0,0,2,1,4,4,6,5]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,u,u,18,19,20,21,u,u,20,21]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm9 = ymm10[0,1,2],ymm9[3],ymm10[4,5],ymm9[6],ymm10[7,8,9,10],ymm9[11],ymm10[12,13],ymm9[14],ymm10[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,3,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm9, %ymm8, %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm7, %ymm7
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm6[0,0,2,1,4,4,6,5]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,u,u,18,19,20,21,u,u,20,21]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7,8,9,10],ymm10[11],ymm11[12,13],ymm10[14],ymm11[15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,3,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm10, %ymm7, %ymm7
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm7, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <3,u,u,u,4,u,u,4>
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm7[0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
+; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm9[0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,3,u,u,u,4,u,u>
 ; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm8, %ymm8
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm5[0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,3,u,u,u,4,u>
-; AVX2-FAST-NEXT:    vpermd %ymm15, %ymm8, %ymm8
+; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm8, %ymm8
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm3[0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <3,u,u,3,u,u,u,4>
 ; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm9, %ymm9
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm10
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm11
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm9
+; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm10
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,3]
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm13
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm14
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm14[4],xmm13[4],xmm14[5],xmm13[5],xmm14[6],xmm13[6],xmm14[7],xmm13[7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm8[0,2,3,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm0, %ymm8, %ymm12
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %xmm8
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm12
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm13
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm11[0,2,3,3,4,5,6,7]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm0, %ymm11, %ymm11
+; AVX2-FAST-NEXT:    vmovdqa (%r9), %xmm15
 ; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm2
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm2[4],xmm8[4],xmm2[5],xmm8[5],xmm2[6],xmm8[6],xmm2[7],xmm8[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm2[4],xmm15[4],xmm2[5],xmm15[5],xmm2[6],xmm15[6],xmm2[7],xmm15[7]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm14 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-FAST-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpbroadcastd 8(%rax), %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm0, %ymm12, %ymm12
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm13[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm14[1,1,2,2]
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm0, %ymm11, %ymm11
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm12[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm13[1,1,2,2]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3],xmm0[4],xmm1[5,6],xmm0[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm11[u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm9 = xmm10[1,1,2,3]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm9[0,1],xmm1[2],xmm9[3,4],xmm1[5],xmm9[6,7]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm10[u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm7 = xmm9[1,1,2,3]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm7[0,1],xmm1[2],xmm7[3,4],xmm1[5],xmm7[6,7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm2[0],xmm8[0],xmm2[1],xmm8[1],xmm2[2],xmm8[2],xmm2[3],xmm8[3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm2[0],xmm15[0],xmm2[1],xmm15[1],xmm2[2],xmm15[2],xmm2[3],xmm15[3]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
-; AVX2-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm2, %ymm8, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm1, %ymm2, %ymm8
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,14,15,u,u,u,u,u,u,u,u,28,29,u,u,u,u,30,31,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7,8,9,10],ymm1[11],ymm2[12,13],ymm1[14],ymm2[15]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[u,u,12,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,u,u,u,u,30,31,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1],ymm4[2,3],ymm2[4],ymm4[5,6,7,8],ymm2[9],ymm4[10,11],ymm2[12],ymm4[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm7
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm15, %ymm2, %ymm7, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm1, %ymm2, %ymm15
+; AVX2-FAST-NEXT:    vpshufhw $180, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,6,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [2,3,6,7,6,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm4, %ymm2
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm15[3,3,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[3,3,3,3,7,7,7,7]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7,8,9],ymm2[10],ymm3[11,12],ymm2[13],ymm3[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,7,3,3,7,7,6,7]
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3]
+; AVX2-FAST-NEXT:    vpshufb %xmm14, %xmm2, %xmm2
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,1]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm14[0],xmm13[0],xmm14[1],xmm13[1],xmm14[2],xmm13[2],xmm14[3],xmm13[3]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX2-FAST-NEXT:    vpbroadcastd (%rax), %ymm3
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm0, %ymm3, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, 96(%rax)
+; AVX2-FAST-NEXT:    vmovdqa %ymm8, 96(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm2, 128(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm2, 160(%rax)
 ; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%rax)
 ; AVX2-FAST-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm8, 32(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm12, 64(%rax)
+; AVX2-FAST-NEXT:    vmovdqa %ymm15, 32(%rax)
+; AVX2-FAST-NEXT:    vmovdqa %ymm11, 64(%rax)
 ; AVX2-FAST-NEXT:    popq %rax
 ; AVX2-FAST-NEXT:    vzeroupper
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX2-FAST-PERLANE-LABEL: store_i16_stride7_vf16:
 ; AVX2-FAST-PERLANE:       # %bb.0:
 ; AVX2-FAST-PERLANE-NEXT:    subq $40, %rsp
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <3,u,u,u,4,u,u,4>
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm7, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm6[0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,3,u,u,u,4,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm4, %ymm8, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm5[0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,3,u,u,u,4,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm12, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm15[0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm9, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,3,u,u,u,4,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm8, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm2[0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,3,u,u,u,4,u>
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm4, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm5[0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm9, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <3,u,u,3,u,u,u,4>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm1, %ymm9, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm9, %ymm9
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm9, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm10
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm11
 ; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm12
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm2
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm2[4],xmm12[4],xmm2[5],xmm12[5],xmm2[6],xmm12[6],xmm2[7],xmm12[7]
-; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm8[0,2,3,3,4,5,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm13
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm13[4],xmm6[4],xmm13[5],xmm6[5],xmm13[6],xmm6[6],xmm13[7],xmm6[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,2,3,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %xmm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm14
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm14[4],xmm9[4],xmm14[5],xmm9[5],xmm14[6],xmm9[6],xmm14[7],xmm9[7]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,1,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm9
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm9[4],xmm1[4],xmm9[5],xmm1[5],xmm9[6],xmm1[6],xmm9[7],xmm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm14 = xmm14[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,1,1]
 ; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd 8(%rax), %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm8, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm14, %ymm0, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm0, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm12[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[1,1,2,2]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3],xmm0[4],xmm1[5,6],xmm0[7]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm11[u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm15 = xmm10[1,1,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm15[0,1],xmm1[2],xmm15[3,4],xmm1[5],xmm15[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm6[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm2 = xmm13[1,1,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3],xmm0[4],xmm2[5,6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm2 = xmm11[u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm8 = xmm10[1,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm8[0,1],xmm2[2],xmm8[3,4],xmm2[5],xmm8[6,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,1]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm0, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm14[0],xmm9[0],xmm14[1],xmm9[1],xmm14[2],xmm9[2],xmm14[3],xmm9[3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm9 = xmm0[0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd 4(%rax), %ymm14
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm9, %ymm14, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm1, %ymm9, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm0, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm1[0],xmm9[1],xmm1[1],xmm9[2],xmm1[2],xmm9[3],xmm1[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd 4(%rax), %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm1, %ymm8, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm2, %ymm1, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,14,15,u,u,u,u,u,u,u,u,28,29,u,u,u,u,30,31,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm7[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm7, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm14[0,1,2],ymm1[3],ymm14[4,5],ymm1[6],ymm14[7,8,9,10],ymm1[11],ymm14[12,13],ymm1[14],ymm14[15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm14 = ymm5[u,u,12,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,u,u,u,u,30,31,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm4, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6,7,8],ymm14[9],ymm15[10,11],ymm14[12],ymm15[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm1, %ymm14, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm3[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm15 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm13, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7,8,9],ymm14[10],ymm15[11,12],ymm14[13],ymm15[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm1, %ymm14, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm13[2,3,3,3,6,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm1, %ymm14, %ymm14
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm10 = xmm2[0],xmm12[0],xmm2[1],xmm12[1],xmm2[2],xmm12[2],xmm2[3],xmm12[3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,1,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm1, %ymm10, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm7[4],ymm9[4],ymm7[5],ymm9[5],ymm7[6],ymm9[6],ymm7[7],ymm9[7],ymm7[12],ymm9[12],ymm7[13],ymm9[13],ymm7[14],ymm9[14],ymm7[15],ymm9[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm1[u,u,u,u,u,u,u,u,14,15,12,13,14,15,12,13,16,17,18,19,24,25,26,27,30,31,28,29,30,31,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm8 = ymm12[4],ymm15[4],ymm12[5],ymm15[5],ymm12[6],ymm15[6],ymm12[7],ymm15[7],ymm12[12],ymm15[12],ymm12[13],ymm15[13],ymm12[14],ymm15[14],ymm12[15],ymm15[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm8[8,9,10,11,14,15,12,13,u,u,u,u,u,u,u,u,24,25,26,27,30,31,28,29,30,31,28,29,30,31,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm2, %ymm8, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm4, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm4[3,3,3,3,7,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm14 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm8 = ymm14[0,1],ymm8[2],ymm14[3,4],ymm8[5],ymm14[6,7,8,9],ymm8[10],ymm14[11,12],ymm8[13],ymm14[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u,255,255,255,255,255,255,255,255,0,0,0,0,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm2, %ymm8, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm3[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm2, %ymm8, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,1]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm13[0],xmm6[0],xmm13[1],xmm6[1],xmm13[2],xmm6[2],xmm13[3],xmm6[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm2, %ymm8, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd (%rax), %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm0, %ymm10, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm0, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm4[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm5, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd (%rax), %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm2, %ymm0, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm12[2,2,2,2,6,6,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0,1],ymm0[2],ymm2[3,4],ymm0[5],ymm2[6,7,8,9],ymm0[10],ymm2[11,12],ymm0[13],ymm2[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,8,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,u,u,u,u,26,27,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm9[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm11[0],ymm1[1],ymm11[2,3],ymm1[4],ymm11[5,6,7,8],ymm1[9],ymm11[10,11],ymm1[12],ymm11[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,2,3,6,7,4,5,u,u,u,u,u,u,u,u,16,17,18,19,22,23,20,21,22,23,20,21,28,29,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm8[1,2,2,3,5,6,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm11[0,1],ymm1[2],ymm11[3,4],ymm1[5],ymm11[6,7,8,9],ymm1[10],ymm11[11,12],ymm1[13],ymm11[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm13[0,1,2,2,4,5,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm1, %ymm11, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[1,1,1,1,5,5,5,5]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,u,u,20,21,24,25,u,u,22,23,22,23]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0,1],ymm1[2],ymm5[3,4],ymm1[5],ymm5[6,7,8,9],ymm1[10],ymm5[11,12],ymm1[13],ymm5[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,1,1,1,5,5,5,5]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,u,u,20,21,24,25,u,u,22,23,22,23]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm4 = ymm3[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7,8,9],ymm4[10],ymm3[11,12],ymm4[13],ymm3[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm8[0,0,2,1,4,4,6,5]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,u,u,18,19,20,21,u,u,20,21]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7,8,9,10],ymm3[11],ymm2[12,13],ymm3[14],ymm2[15]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm15[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm12[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7,8,9],ymm2[10],ymm3[11,12],ymm2[13],ymm3[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,0,2,1,4,4,6,5]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,u,u,18,19,20,21,u,u,20,21]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7,8,9,10],ymm2[11],ymm3[12,13],ymm2[14],ymm3[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm13[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm11[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-PERLANE-NEXT:    vmovups (%rsp), %ymm2 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, 96(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 128(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 160(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, (%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm14, 192(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    addq $40, %rsp
 ; AVX2-FAST-PERLANE-NEXT:    vzeroupper
 ; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-SLOW-LABEL: store_i16_stride7_vf16:
 ; AVX512F-SLOW:       # %bb.0:
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm5
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm6
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm6
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm7
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm3
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm4
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm2
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r10), %ymm1
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm4[u,u,u,u,u,u],zero,zero,ymm4[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm4[16,17,u,u,u,u,u,u,u,u]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm4[u,u,u,u,u,u],zero,zero,ymm4[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm4[16,17,u,u,u,u,u,u,u,u]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm3[u,u,u,u,u,u,14,15],zero,zero,ymm3[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm3[u,u,u,u,u,u,u,u]
-; AVX512F-SLOW-NEXT:    vpor %ymm7, %ymm8, %ymm7
+; AVX512F-SLOW-NEXT:    vpor %ymm5, %ymm8, %ymm5
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm8
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm9
 ; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm10 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm10 = xmm10[0,2,3,3,4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm10, %zmm7
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,zero,ymm6[14,15,u,u,u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm6[16,17,u,u,u,u],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm5[12,13,14,15],zero,zero,ymm5[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm5[u,u,u,u,u,u,u,u,16,17,18,19]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm10, %zmm5
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,zero,ymm7[14,15,u,u,u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm7[16,17,u,u,u,u],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm6[12,13,14,15],zero,zero,ymm6[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm6[u,u,u,u,u,u,u,u,16,17,18,19]
 ; AVX512F-SLOW-NEXT:    vpor %ymm10, %ymm11, %ymm12
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm10
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm11
 ; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm13 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm13 = xmm13[2,1,2,3,4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,4,5,5,4]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,0,1,3]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm13, %zmm15
-; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm15
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm2[u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[16,17,u,u,u,u]
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm15
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm2[u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[16,17,u,u,u,u]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm0[u,u,u,u,u,u,u,u,u,u,14,15],zero,zero,ymm0[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm0[u,u,u,u]
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm17
-; AVX512F-SLOW-NEXT:    vpor %ymm7, %ymm12, %ymm7
+; AVX512F-SLOW-NEXT:    vpor %ymm5, %ymm12, %ymm5
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm12
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm13
 ; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
 ; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm14 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
 ; AVX512F-SLOW-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm14, %xmm18
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpbroadcastd 8(%r10), %ymm7
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpbroadcastd 8(%r10), %ymm5
 ; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm16 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpandnq %ymm7, %ymm16, %ymm7
+; AVX512F-SLOW-NEXT:    vpandnq %ymm5, %ymm16, %ymm5
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm14 = ymm1[12,13,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[14,15,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[16,17,u,u]
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm16
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm7, %zmm7
-; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm7
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm15, %zmm7
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm5, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm15, %zmm5
 ; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
 ; AVX512F-SLOW-NEXT:    vprold $16, %xmm11, %xmm14
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm10[1,1,2,3]
 ; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm14 = xmm15[0,1],xmm14[2],xmm15[3,4],xmm14[5],xmm15[6,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,2,1]

 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm8, %zmm8
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm8
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm3[0,1,1,3,4,5,5,7]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
 ; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7,8,9],ymm9[10],ymm0[11,12],ymm9[13],ymm0[14,15]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm5[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm6[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0],ymm10[1],ymm9[2,3],ymm10[4],ymm9[5,6,7,8],ymm10[9],ymm9[10,11],ymm10[12],ymm9[13,14,15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm3[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm10[0,1],ymm9[2],ymm10[3,4],ymm9[5],ymm10[6,7,8,9],ymm9[10],ymm10[11,12],ymm9[13],ymm10[14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm5[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm6[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm9 = ymm6[4],ymm7[4],ymm6[5],ymm7[5],ymm6[6],ymm7[6],ymm6[7],ymm7[7],ymm6[12],ymm7[12],ymm6[13],ymm7[13],ymm6[14],ymm7[14],ymm6[15],ymm7[15]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm9[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm3[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
 ; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm10, %zmm9
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[2,2,2,3,6,6,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm9
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm7[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7,8,9],ymm6[10],ymm7[11,12],ymm6[13],ymm7[14,15]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm6, %zmm6
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm6 = zmm6[2,2,2,3,6,6,6,7]
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm6
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm17[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm2[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7,8,9,10],ymm0[11],ymm10[12,13],ymm0[14],ymm10[15]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm2[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm7[0,1,2],ymm0[3],ymm7[4,5],ymm0[6],ymm7[7,8,9,10],ymm0[11],ymm7[12,13],ymm0[14],ymm7[15]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
-; AVX512F-SLOW-NEXT:    vprold $16, %ymm2, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm17[1,2,2,3,5,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm16[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm16[0,1,2,2,4,5,6,6]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm10, %zmm10
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm10
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm10
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm5[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm5 = ymm6[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm5[3],ymm0[4,5],ymm5[6],ymm0[7,8,9,10],ymm5[11],ymm0[12,13],ymm5[14],ymm0[15]
+; AVX512F-SLOW-NEXT:    vprold $16, %ymm2, %ymm7
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm17[1,2,2,3,5,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm10[0,1],ymm7[2],ymm10[3,4],ymm7[5],ymm10[6,7,8,9],ymm7[10],ymm10[11,12],ymm7[13],ymm10[14,15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm16[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm16[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm7, %zmm7
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm7
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm7
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm9[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,3,3,4,6,7,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2,3],ymm4[4],ymm3[5,6,7,8],ymm4[9],ymm3[10,11],ymm4[12],ymm3[13,14,15]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm3[4],ymm4[4],ymm3[5],ymm4[5],ymm3[6],ymm4[6],ymm3[7],ymm4[7],ymm3[12],ymm4[12],ymm3[13],ymm4[13],ymm3[14],ymm4[14],ymm3[15],ymm4[15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,3,3,3,6,7,7,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm3
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm17[3,3,3,3,7,7,7,7]
 ; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm2[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,3,6,6,6,7]
 ; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm16[2,3,3,3,6,7,7,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm10, 128(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm7, 128(%rax)
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm8, (%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm7, 64(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm5, 64(%rax)
 ; AVX512F-SLOW-NEXT:    vzeroupper
 ; AVX512F-SLOW-NEXT:    retq
 ;
 ; AVX512F-FAST-LABEL: store_i16_stride7_vf16:
 ; AVX512F-FAST:       # %bb.0:
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm4
 ; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %ymm5
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %ymm2
-; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm4
+; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm3
 ; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm1
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,ymm5[14,15,u,u,u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm5[16,17,u,u,u,u],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm3[12,13,14,15],zero,zero,ymm3[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm3[u,u,u,u,u,u,u,u,16,17,18,19]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm4[12,13,14,15],zero,zero,ymm4[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm4[u,u,u,u,u,u,u,u,16,17,18,19]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm4, %ymm17
 ; AVX512F-FAST-NEXT:    vpor %ymm6, %ymm7, %ymm6
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %xmm8
 ; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm9
 ; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,3]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm7, %zmm6
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm4[u,u,u,u,u,u],zero,zero,ymm4[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm4[16,17,u,u,u,u,u,u,u,u]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm3[u,u,u,u,u,u],zero,zero,ymm3[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[16,17,u,u,u,u,u,u,u,u]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm2[u,u,u,u,u,u,14,15],zero,zero,ymm2[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm2[u,u,u,u,u,u,u,u]
 ; AVX512F-FAST-NEXT:    vpor %ymm7, %ymm10, %ymm7
 ; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %xmm10
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %xmm11
 ; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm12 = xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
 ; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm12 = xmm12[0,2,3,3,4,5,6,7]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,2,1]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm12, %zmm15
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm15
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm1[u,u,u,u,u,u,u,u,u,u],zero,zero,ymm1[14,15,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm1[16,17,u,u,u,u]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm0[u,u,u,u,u,u,u,u,u,u,14,15],zero,zero,ymm0[u,u,u,u,u,u,u,u,u,u,16,17],zero,zero,ymm0[u,u,u,u]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm17
 ; AVX512F-FAST-NEXT:    vpor %ymm6, %ymm7, %ymm6
 ; AVX512F-FAST-NEXT:    vmovdqa (%r9), %xmm12
 ; AVX512F-FAST-NEXT:    vmovdqa (%r8), %xmm13
 ; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm7, %xmm7
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm0, %xmm18
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm7, %xmm7
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm4, %xmm18
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,1]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm7, %zmm16
 ; AVX512F-FAST-NEXT:    vpbroadcastd 8(%r10), %ymm6
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
 ; AVX512F-FAST-NEXT:    vpandn %ymm6, %ymm7, %ymm6
 ; AVX512F-FAST-NEXT:    vbroadcasti64x4 {{.*#+}} zmm7 = mem[0,1,2,3,0,1,2,3]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm14 = ymm7[12,13,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm7[14,15,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm7[16,17,u,u]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm6, %zmm6
 ; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm16, %zmm6
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm15, %zmm6
 ; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm14 = xmm14[0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,1,3]
 ; AVX512F-FAST-NEXT:    vprold $16, %xmm9, %xmm15
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm8[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm15[2],xmm0[3,4],xmm15[5],xmm0[6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm14, %zmm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm8[1,1,2,3]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm15[2],xmm4[3,4],xmm15[5],xmm4[6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm14, %zmm4
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
 ; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm11 = xmm11[1,1,2,2]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm10 = xmm11[0],xmm10[1],xmm11[2,3],xmm10[4],xmm11[5,6],xmm10[7]
 ; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3]
 ; AVX512F-FAST-NEXT:    vmovdqa64 %xmm18, %xmm9
 ; AVX512F-FAST-NEXT:    vpshufb %xmm9, %xmm8, %xmm8
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm8, %zmm8
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,0,1,1,4,4,5,5]
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm9 = xmm0[0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm8
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm9 = xmm4[0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,0,1]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm9, %zmm0
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,1,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm9, %zmm4
 ; AVX512F-FAST-NEXT:    vpbroadcastd (%r10), %ymm9
 ; AVX512F-FAST-NEXT:    vpbroadcastd 4(%r10), %ymm10
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm9, %zmm9
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm9
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm9
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm9
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm4[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,1,1,3,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7,8,9],ymm0[10],ymm8[11,12],ymm0[13],ymm8[14,15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,2]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,u,u,u,u,26,27,u,u,u,u,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm3[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm8 = ymm10[0],ymm8[1],ymm10[2,3],ymm8[4],ymm10[5,6,7,8],ymm8[9],ymm10[10,11],ymm8[12],ymm10[13,14,15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm8 = ymm10[0,1],ymm8[2],ymm10[3,4],ymm8[5],ymm10[6,7,8,9],ymm8[10],ymm10[11,12],ymm8[13],ymm10[14,15]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm3[1,1,1,1,5,5,5,5]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,u,u,20,21,24,25,u,u,22,23,22,23]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm12
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm8 = ymm12[4],ymm5[4],ymm12[5],ymm5[5],ymm12[6],ymm5[6],ymm12[7],ymm5[7],ymm12[12],ymm5[12],ymm12[13],ymm5[13],ymm12[14],ymm5[14],ymm12[15],ymm5[15]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm4 = ymm8[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,4,u,u,5,5,u>
+; AVX512F-FAST-NEXT:    vpermd %ymm4, %ymm10, %ymm4
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm3[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm2[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm10, %zmm4
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm2[2,2,2,2,6,6,6,6]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm10, %zmm8
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[2,2,2,3,6,6,6,7]
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm17[0,0,2,1,4,4,6,5]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm17[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,u,u,20,21,24,25,u,u,22,23,22,23]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0,1],ymm11[2],ymm5[3,4],ymm11[5],ymm5[6,7,8,9],ymm11[10],ymm5[11,12],ymm11[13],ymm5[14,15]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm5, %zmm5
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm5 = zmm5[2,2,2,3,6,6,6,7]
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm5
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm4 = ymm0[0,0,2,1,4,4,6,5]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,u,u,18,19,20,21,u,u,20,21]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3],ymm10[4,5],ymm0[6],ymm10[7,8,9,10],ymm0[11],ymm10[12,13],ymm0[14],ymm10[15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm4 = ymm10[0,1,2],ymm4[3],ymm10[4,5],ymm4[6],ymm10[7,8,9,10],ymm4[11],ymm10[12,13],ymm4[14],ymm10[15]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,3,3]
 ; AVX512F-FAST-NEXT:    vprold $16, %ymm1, %ymm10
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm17[1,2,2,3,5,6,6,7]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm0[1,2,2,3,5,6,6,7]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vbroadcasti64x4 {{.*#+}} zmm10 = [0,13,4,0,0,14,5,0,0,13,4,0,0,14,5,0]
-; AVX512F-FAST-NEXT:    # zmm10 = mem[0,1,2,3,0,1,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm4, %zmm4
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm10 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
 ; AVX512F-FAST-NEXT:    vpermd %zmm7, %zmm10, %zmm10
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm10
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm10
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,14,15,u,u,u,u,u,u,u,u,28,29,u,u,u,u,30,31,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm3[0,1,2],ymm0[3],ymm3[4,5],ymm0[6],ymm3[7,8,9,10],ymm0[11],ymm3[12,13],ymm0[14],ymm3[15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,u,u,u,u,30,31,u,u,u,u,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6,7,8],ymm3[9],ymm2[10,11],ymm3[12],ymm2[13,14,15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <6,u,u,u,7,u,u,7>
-; AVX512F-FAST-NEXT:    vpermd %ymm7, %ymm0, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm17[3,3,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm10
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm10
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm8[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,6,u,u,7,7,7,u>
+; AVX512F-FAST-NEXT:    vpermd %ymm4, %ymm5, %ymm4
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[12],ymm3[12],ymm2[13],ymm3[13],ymm2[14],ymm3[14],ymm2[15],ymm3[15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,6,u,u,7,7,u>
+; AVX512F-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm4, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <6,u,u,u,7,u,u,7>
+; AVX512F-FAST-NEXT:    vpermd %ymm7, %ymm3, %ymm3
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,3,7,7,7,7]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7,8,9],ymm3[10],ymm1[11,12],ymm3[13],ymm1[14,15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa %ymm1, 192(%rax)
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm3, %ymm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa %ymm0, 192(%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm10, 128(%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm9, (%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm6, 64(%rax)
 ; AVX512F-FAST-NEXT:    vzeroupper
 ; AVX512F-FAST-NEXT:    retq
 ;
 ; AVX512BW-LABEL: store_i16_stride7_vf16:
 ; AVX512BW:       # %bb.0:

 ; AVX1-ONLY-NEXT:    vzeroupper
 ; AVX1-ONLY-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: store_i16_stride7_vf32:
 ; AVX2-SLOW:       # %bb.0:
 ; AVX2-SLOW-NEXT:    subq $600, %rsp # imm = 0x258
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm5
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm10
-; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm13
-; AVX2-SLOW-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm12
-; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm6
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm6
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm3
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm13
+; AVX2-SLOW-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm9
 ; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %ymm14
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm11
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm10
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%r8), %ymm7
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%r9), %ymm8
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <3,u,u,u,4,u,u,4>
-; AVX2-SLOW-NEXT:    vpermd %ymm10, %ymm0, %ymm1
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm12[0,3,2,3,4,7,6,7]
+; AVX2-SLOW-NEXT:    vpermd %ymm2, %ymm0, %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm11
+; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[0,3,2,3,4,7,6,7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm3, %ymm12
+; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,3,u,u,u,4,u,u>
 ; AVX2-SLOW-NEXT:    vpermd %ymm9, %ymm2, %ymm4
 ; AVX2-SLOW-NEXT:    vpermd %ymm5, %ymm0, %ymm0
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm13[0,3,2,3,4,7,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm6[0,3,2,3,4,7,6,7]
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm5 = ymm5[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm5, %ymm0
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm11[0,0,0,0,4,5,6,7,8,8,8,8,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm10[0,0,0,0,4,5,6,7,8,8,8,8,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,7,7,7,7,8,9,10,11,15,15,15,15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm4, %ymm3, %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-SLOW-NEXT:    vpermd %ymm6, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpermd %ymm13, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm14[0,0,0,0,4,5,6,7,8,8,8,8,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,7,7,7,7,8,9,10,11,15,15,15,15]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm2, %ymm3, %ymm2
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,3,u,u,u,4,u>
 ; AVX2-SLOW-NEXT:    vpermd %ymm7, %ymm3, %ymm5
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[0,1,0,3,4,5,4,7]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,4,7,7,8,9,10,11,12,12,15,15]

 ; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpermd %ymm5, %ymm4, %ymm4
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm3, %ymm4, %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm3, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-SLOW-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[2,2,2,2,6,6,6,6]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-SLOW-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm12[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm10[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,2,2,3,5,6,6,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm11[4],ymm12[4],ymm11[5],ymm12[5],ymm11[6],ymm12[6],ymm11[7],ymm12[7],ymm11[12],ymm12[12],ymm11[13],ymm12[13],ymm11[14],ymm12[14],ymm11[15],ymm12[15]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm1[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[1,2,2,3,5,6,6,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm8[2,1,2,3,6,5,6,7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm1[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[2,1,2,3,6,5,6,7]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm13[0,1,2,2,4,5,6,6]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm13[0,1,2,2,4,5,6,6]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm12[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm10[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm11[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm9[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,3,3,4,6,7,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm9[4],ymm10[4],ymm9[5],ymm10[5],ymm9[6],ymm10[6],ymm9[7],ymm10[7],ymm9[12],ymm10[12],ymm9[13],ymm10[13],ymm9[14],ymm10[14],ymm9[15],ymm10[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,3,3,6,7,7,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vpbroadcastd 60(%r8), %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm8[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[3,3,3,3]

 ; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
 ; AVX2-SLOW-NEXT:    vpshufb %xmm5, %xmm3, %xmm3
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,1,1]
 ; AVX2-SLOW-NEXT:    vpbroadcastd 8(%rax), %ymm5
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm3, %ymm5, %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm3, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm10[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,2,6,6,6,6]
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm7[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7,8,9,10],ymm2[11],ymm3[12,13],ymm2[14],ymm3[15]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm9[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6,7,8],ymm3[9],ymm4[10,11],ymm3[12],ymm4[13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm12[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,2,2,3,6,6,6,7]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm7[4],ymm12[4],ymm7[5],ymm12[5],ymm7[6],ymm12[6],ymm7[7],ymm12[7],ymm7[12],ymm12[12],ymm7[13],ymm12[13],ymm7[14],ymm12[14],ymm7[15],ymm12[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm3[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,3,3,4,6,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm2, %ymm4, %ymm2
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm11[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7,8,9],ymm4[10],ymm3[11,12],ymm4[13],ymm3[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm11[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,2,3,6,6,6,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7,8,9],ymm5[10],ymm4[11,12],ymm5[13],ymm4[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm13[2,3,3,3,6,7,7,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm8[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7,8,9],ymm4[10],ymm3[11,12],ymm4[13],ymm3[14,15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm10[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm7[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6,7,8],ymm4[9],ymm5[10,11],ymm4[12],ymm5[13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm13[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm2, %ymm4, %ymm2
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm8[2,2,2,2,6,6,6,6]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7,8,9],ymm5[10],ymm4[11,12],ymm5[13],ymm4[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm3[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm4 = ymm12[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm4, %ymm3, %ymm3
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm4 = ymm11[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm11[1,2,2,3,5,6,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm10[1,2,2,3,5,6,6,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm4 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7,8,9],ymm4[10],ymm5[11,12],ymm4[13],ymm5[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,2]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm13[0,1,2,2,4,5,6,6]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3

 ; AVX2-SLOW-NEXT:    # ymm6 = mem[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7,8,9],ymm6[10],ymm5[11,12],ymm6[13],ymm5[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,2]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm7[1,1,1,1,5,5,5,5]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm10[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm12[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm5 = ymm7[0,1],ymm5[2],ymm7[3,4],ymm5[5],ymm7[6,7,8,9],ymm5[10],ymm7[11,12],ymm5[13],ymm7[14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm8[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm9[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7,8,9],ymm8[10],ymm7[11,12],ymm8[13],ymm7[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,2]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm5, %ymm7, %ymm5
 ; AVX2-SLOW-NEXT:    vpshufd $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm6 = mem[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm7 = mem[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1,2],ymm6[3],ymm7[4,5],ymm6[6],ymm7[7,8,9,10],ymm6[11],ymm7[12,13],ymm6[14],ymm7[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,2,3,3]
 ; AVX2-SLOW-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm7 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm6, %ymm7, %ymm6
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm11[0,0,2,1,4,4,6,5]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm12[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm10[0,0,2,1,4,4,6,5]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm11[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm9[0,1,2],ymm7[3],ymm9[4,5],ymm7[6],ymm9[7,8,9,10],ymm7[11],ymm9[12,13],ymm7[14],ymm9[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,3,3]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm13[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm7, %ymm9, %ymm7
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm4, %ymm6, %ymm4

 ; AVX2-SLOW-NEXT:    addq $600, %rsp # imm = 0x258
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: store_i16_stride7_vf32:
 ; AVX2-FAST:       # %bb.0:
 ; AVX2-FAST-NEXT:    subq $328, %rsp # imm = 0x148
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm10
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm14
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm6
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm1
+; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm2
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm13
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm4
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm8
-; AVX2-FAST-NEXT:    vmovdqa (%rax), %ymm11
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm10[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vmovdqa %ymm6, %ymm7
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [6,7,3,3,7,7,6,7]
-; AVX2-FAST-NEXT:    vpermd %ymm11, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7,8,9],ymm2[10],ymm3[11,12],ymm2[13],ymm3[14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm2, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [4,5,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpermd %ymm11, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
-; AVX2-FAST-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[1,2,2,3,5,6,6,7]
+; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm8
+; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm9
+; AVX2-FAST-NEXT:    vmovdqa (%rax), %ymm7
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
+; AVX2-FAST-NEXT:    vmovdqa %ymm2, %ymm11
+; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm14
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,6,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm12[4],ymm13[4],ymm12[5],ymm13[5],ymm12[6],ymm13[6],ymm12[7],ymm13[7],ymm12[12],ymm13[12],ymm12[13],ymm13[13],ymm12[14],ymm13[14],ymm12[15],ymm13[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [2,3,6,7,6,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,7,3,3,7,7,6,7]
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm8[3,3,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0,1],ymm3[2],ymm4[3,4],ymm3[5],ymm4[6,7,8,9],ymm3[10],ymm4[11,12],ymm3[13],ymm4[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm3, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,5,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm9[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
+; AVX2-FAST-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[1,2,2,3,5,6,6,7]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm3, %ymm2
+; AVX2-FAST-NEXT:    vmovdqu %ymm7, (%rsp) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm14[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX2-FAST-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm10[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6,7,8],ymm2[9],ymm3[10,11],ymm2[12],ymm3[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm12[2,2,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7,8,9],ymm3[10],ymm2[11,12],ymm3[13],ymm2[14,15]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,4,5,5,7]
+; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm3, %ymm0
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm6
-; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %ymm9
+; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm9, %ymm1
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %ymm3
 ; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %ymm2
-; AVX2-FAST-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm15[0,1],ymm2[2],ymm15[3,4],ymm2[5],ymm15[6,7,8,9],ymm2[10],ymm15[11,12],ymm2[13],ymm15[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm15, %ymm1, %ymm2, %ymm12
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm14, %ymm0
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm10[1,1,1,1,5,5,5,5]
+; AVX2-FAST-NEXT:    vpblendvb %ymm15, %ymm1, %ymm2, %ymm10
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm11, %ymm0
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm14[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm13[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[0,1,1,3,4,5,5,7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm12[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm15, %ymm0, %ymm1, %ymm15
 ; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %ymm5
-; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %ymm7
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm7, %ymm3
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm5[0,0,2,1,4,4,6,5]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm10[3],ymm3[4,5],ymm10[6],ymm3[7,8,9,10],ymm10[11],ymm3[12,13],ymm10[14],ymm3[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm3[2,2,3,3]
-; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %ymm3
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,4,5,4,5,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm10, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm8, %ymm4
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm8[0,0,2,1,4,4,6,5]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm4 = ymm4[0,1,2],ymm10[3],ymm4[4,5],ymm10[6],ymm4[7,8,9,10],ymm10[11],ymm4[12,13],ymm10[14],ymm4[15]
-; AVX2-FAST-NEXT:    vpermd %ymm11, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,3,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm4, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm12, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm5[0,0,2,1,4,4,6,5]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm14[3],ymm1[4,5],ymm14[6],ymm1[7,8,9,10],ymm14[11],ymm1[12,13],ymm14[14],ymm1[15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm1[2,2,3,3]
+; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %ymm4
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,1,4,5,4,5,5,7]
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm11, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm14, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm9, %ymm3
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm8[0,0,2,1,4,4,6,5]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm14[3],ymm3[4,5],ymm14[6],ymm3[7,8,9,10],ymm14[11],ymm3[12,13],ymm14[14],ymm3[15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm11, %ymm14
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm3, %ymm14, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm10, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm15, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm15, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <3,u,u,u,4,u,u,4>
 ; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm0, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
-; AVX2-FAST-NEXT:    vpshufb %ymm2, %ymm9, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm4, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm7, %ymm10
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm10, %ymm11
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm2, %ymm14, %ymm2
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,3,u,u,u,4,u,u>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpermd %ymm15, %ymm2, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
-; AVX2-FAST-NEXT:    vmovdqu (%rsp), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm6, %ymm12
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm4, %ymm12, %ymm4
-; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm13, %ymm10
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm2, %ymm10, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,3,u,u,u,4,u>
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm2, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm7, %ymm12
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm3
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm0, %ymm3, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,3,u,u,u,4,u,u>
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm3, %ymm10
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm14, %ymm9, %ymm15
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm10, %ymm15, %ymm10
+; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm3, %ymm3
+; AVX2-FAST-NEXT:    vpshufb %ymm14, %ymm13, %ymm12
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm3, %ymm12, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm11, %ymm10, %ymm10
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,3,u,u,u,4,u>
+; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm1, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
+; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm2, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm4, %ymm12, %ymm4
-; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm3, %ymm12, %ymm3
+; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm8, %ymm10
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm2, %ymm10, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <3,u,u,3,u,u,u,4>
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm10, %ymm12
+; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm8, %ymm11
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm1, %ymm11, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <3,u,u,3,u,u,u,4>
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm11, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm4, %ymm12, %ymm4
-; AVX2-FAST-NEXT:    vpermd %ymm11, %ymm10, %ymm10
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm2, %ymm10, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm15[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm9[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [5,6,2,3,6,7,5,6]
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,5,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm3, %ymm12, %ymm3
+; AVX2-FAST-NEXT:    vpermd (%rsp), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm1, %ymm11, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm10, %ymm3, %ymm3
+; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm15[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm0[4],ymm7[4],ymm0[5],ymm7[5],ymm0[6],ymm7[6],ymm0[7],ymm7[7],ymm0[12],ymm7[12],ymm0[13],ymm7[13],ymm0[14],ymm7[14],ymm0[15],ymm7[15]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,4,5,4,5,5,7]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm6[2,2,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm8[2],ymm3[3,4],ymm8[5],ymm3[6,7,8,9],ymm8[10],ymm3[11,12],ymm8[13],ymm3[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm3, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [5,6,2,3,6,7,5,6]
+; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm3, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,5,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm3, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,6,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm6[4],ymm9[4],ymm6[5],ymm9[5],ymm6[6],ymm9[6],ymm6[7],ymm9[7],ymm6[12],ymm9[12],ymm6[13],ymm9[13],ymm6[14],ymm9[14],ymm6[15],ymm9[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [2,3,6,7,6,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vpbroadcastd 60(%r8), %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm7[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm2[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[3,3,3,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [6,7,3,3,7,7,6,7]
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm7
+; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm3
+; AVX2-FAST-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %xmm2
-; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %xmm0
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa %xmm0, (%rsp) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm5 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX2-FAST-NEXT:    vpshufb %xmm5, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa %xmm5, %xmm6
+; AVX2-FAST-NEXT:    vmovdqa %xmm5, %xmm7
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,1]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm7[0],xmm2[1],xmm7[1],xmm2[2],xmm7[2],xmm2[3],xmm7[3]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
+; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
 ; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,1,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm13
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm0, %ymm1, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm11
 ; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm10
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3]
 ; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm0, %xmm3
 ; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm15
 ; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm9
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm9[0],xmm15[0],xmm9[1],xmm15[1],xmm9[2],xmm15[2],xmm9[3],xmm15[3]
-; AVX2-FAST-NEXT:    vpshufb %xmm6, %xmm5, %xmm5
+; AVX2-FAST-NEXT:    vpshufb %xmm7, %xmm5, %xmm5
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm5, %ymm3, %ymm2
+; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm5, %ymm3, %ymm7
 ; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = [0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
 ; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm6, %xmm3
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,0,1]
 ; AVX2-FAST-NEXT:    vpbroadcastd 32(%rax), %ymm5
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm3, %ymm5, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %xmm0
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm12
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm12[0],xmm0[0],xmm12[1],xmm0[1],xmm12[2],xmm0[2],xmm12[3],xmm0[3]
+; AVX2-FAST-NEXT:    vmovdqa (%r9), %xmm14
+; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm5
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm14[0],xmm5[1],xmm14[1],xmm5[2],xmm14[2],xmm5[3],xmm14[3]
 ; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm0, %xmm8
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,0,1]
-; AVX2-FAST-NEXT:    vpbroadcastd (%rax), %ymm14
-; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm8, %ymm14, %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm13, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, (%rsp) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm2, %ymm8, %ymm1
+; AVX2-FAST-NEXT:    vpbroadcastd (%rax), %ymm13
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm8, %ymm13, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = [255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm12, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
-; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm7, %xmm8
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm13 = xmm5[1,1,2,2]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm8 = xmm13[0],xmm8[1],xmm13[2,3],xmm8[4],xmm13[5,6],xmm8[7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm13 = <u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm7, %ymm8, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm7 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %xmm13, %xmm3, %xmm14
+; AVX2-FAST-NEXT:    vpshufb %xmm7, %xmm3, %xmm8
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm12 = xmm2[1,1,2,2]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm8 = xmm12[0],xmm8[1],xmm12[2,3],xmm8[4],xmm12[5,6],xmm8[7]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm12 = <u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u>
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm2[1,1,2,3]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm14[2],xmm4[3,4],xmm14[5],xmm4[6,7]
+; AVX2-FAST-NEXT:    vpshufb %xmm12, %xmm2, %xmm13
+; AVX2-FAST-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm1[1,1,2,3]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm13[2],xmm4[3,4],xmm13[5],xmm4[6,7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm8, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm11, %xmm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm8, %ymm4, %ymm4
+; AVX2-FAST-NEXT:    vpshufb %xmm7, %xmm11, %xmm7
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm8 = xmm10[1,1,2,2]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm8[0],xmm1[1],xmm8[2,3],xmm1[4],xmm8[5,6],xmm1[7]
-; AVX2-FAST-NEXT:    vpshufb %xmm13, %xmm15, %xmm8
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm13 = xmm9[1,1,2,3]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm8 = xmm13[0,1],xmm8[2],xmm13[3,4],xmm8[5],xmm13[6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm7 = xmm8[0],xmm7[1],xmm8[2,3],xmm7[4],xmm8[5,6],xmm7[7]
+; AVX2-FAST-NEXT:    vpshufb %xmm12, %xmm15, %xmm8
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm12 = xmm9[1,1,2,3]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm8 = xmm12[0,1],xmm8[2],xmm12[3,4],xmm8[5],xmm12[6,7]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm8, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm7, %ymm8, %ymm7
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = [0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
 ; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm6, %xmm6
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX2-FAST-NEXT:    vpbroadcastd 36(%rax), %ymm13
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm6, %ymm13, %ymm6
+; AVX2-FAST-NEXT:    vpbroadcastd 36(%rax), %ymm12
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm6, %ymm12, %ymm6
 ; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
 ; AVX2-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm8
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm0, %ymm8, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm4, %ymm6, %ymm4
-; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm5[4],xmm7[4],xmm5[5],xmm7[5],xmm5[6],xmm7[6],xmm5[7],xmm7[7]
+; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm7, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
-; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm1, %xmm1
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm6[0,2,3,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm1, %ymm6, %ymm1
+; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm6, %xmm6
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,1,3]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm7[0,2,3,3,4,5,6,7]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm6, %ymm7, %ymm6
 ; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm15[4],xmm9[4],xmm15[5],xmm9[5],xmm15[6],xmm9[6],xmm15[7],xmm9[7]
 ; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm10[4],xmm11[4],xmm10[5],xmm11[5],xmm10[6],xmm11[6],xmm10[7],xmm11[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm7 = xmm10[4],xmm11[4],xmm10[5],xmm11[5],xmm10[6],xmm11[6],xmm10[7],xmm11[7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm6[0,2,3,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm2, %ymm6, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm6 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm6 = xmm3[4],mem[4],xmm3[5],mem[5],xmm3[6],mem[6],xmm3[7],mem[7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm7 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX2-FAST-NEXT:    vpshufb %xmm7, %xmm6, %xmm6
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm7[0,2,3,3,4,5,6,7]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm2, %ymm7, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm7 # 16-byte Folded Reload
+; AVX2-FAST-NEXT:    # xmm7 = xmm1[4],mem[4],xmm1[5],mem[5],xmm1[6],mem[6],xmm1[7],mem[7]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm7, %xmm7
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpbroadcastd 40(%rax), %ymm8
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm12, %xmm3 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm3 = xmm12[4],mem[4],xmm12[5],mem[5],xmm12[6],mem[6],xmm12[7],mem[7]
-; AVX2-FAST-NEXT:    vpshufb %xmm7, %xmm3, %xmm3
+; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm7
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm14[4],xmm5[5],xmm14[5],xmm5[6],xmm14[6],xmm5[7],xmm14[7]
+; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm3, %xmm3
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpbroadcastd 8(%rax), %ymm5
 ; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm3, %ymm5, %ymm3
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm6, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm7, %ymm6, %ymm6
 ; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm3, %ymm2, %ymm2
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm3, 96(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm3, 320(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm3, 128(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm3, 352(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm3, 160(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm3, 192(%rax)
 ; AVX2-FAST-NEXT:    vmovdqa %ymm2, 64(%rax)
 ; AVX2-FAST-NEXT:    vmovdqa %ymm0, 32(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, (%rax)
-; AVX2-FAST-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 224(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, 288(%rax)
+; AVX2-FAST-NEXT:    vmovdqa %ymm6, 288(%rax)
 ; AVX2-FAST-NEXT:    vmovdqa %ymm4, 256(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 416(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 384(%rax)
 ; AVX2-FAST-NEXT:    addq $328, %rsp # imm = 0x148
 ; AVX2-FAST-NEXT:    vzeroupper
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX2-FAST-PERLANE-LABEL: store_i16_stride7_vf32:
 ; AVX2-FAST-PERLANE:       # %bb.0:
 ; AVX2-FAST-PERLANE-NEXT:    subq $616, %rsp # imm = 0x268
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm14
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm13
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdx), %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdx), %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm15
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rcx), %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %ymm8
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <3,u,u,u,4,u,u,4>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm8, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm10, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm4, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,3,u,u,u,4,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm9, %ymm3, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm10, %ymm3, %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = [0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm12, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm14, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm7, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm9, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm11, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm5, %ymm2, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm13, %ymm3, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm14, %ymm3, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm15, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,3,u,u,u,4,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm7, %ymm3, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm8, %ymm3, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm0, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm12, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm7, %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm5, %ymm3, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm5, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm3, %ymm2, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rax), %ymm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <3,u,u,3,u,u,u,4>
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm6, %ymm3, %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm6, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm6, %ymm9
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm5, %ymm3, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm2, %ymm3, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm1, %ymm4, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm10[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm10[2,2,2,2,6,6,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,2,2,3,5,6,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm13[0,1,2,2,4,5,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm12[4],ymm13[4],ymm12[5],ymm13[5],ymm12[6],ymm13[6],ymm12[7],ymm13[7],ymm12[12],ymm13[12],ymm12[13],ymm13[13],ymm12[14],ymm13[14],ymm12[15],ymm13[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm1[0,1,2,3,6,7,4,5,u,u,u,u,u,u,u,u,16,17,18,19,22,23,20,21,22,23,20,21,28,29,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm9[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm11[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm9[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,14,15,12,13,14,15,12,13,16,17,18,19,24,25,26,27,30,31,28,29,30,31,28,29]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm10[4],ymm11[4],ymm10[5],ymm11[5],ymm10[6],ymm11[6],ymm10[7],ymm11[7],ymm10[12],ymm11[12],ymm10[13],ymm11[13],ymm10[14],ymm11[14],ymm10[15],ymm11[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[8,9,10,11,14,15,12,13,u,u,u,u,u,u,u,u,24,25,26,27,30,31,28,29,30,31,28,29,30,31,28,29]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd 60(%r8), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm12[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
+; AVX2-FAST-PERLANE-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm7[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[3,3,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm13[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[2,3,3,3,6,7,7,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %xmm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]

 ; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm6[4],xmm8[4],xmm6[5],xmm8[5],xmm6[6],xmm8[6],xmm6[7],xmm8[7]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm3, %xmm3
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,1,1]
 ; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd 8(%rax), %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm3, %ymm5, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255,255,255,0,0,0,0,0,0,0,0,255,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm2, %ymm0, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm3, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm8[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1,2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7,8,9,10],ymm2[11],ymm3[12,13],ymm2[14],ymm3[15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm11[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm4 = ymm9[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6,7,8],ymm3[9],ymm4[10,11],ymm3[12],ymm4[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[8,9,10,11,14,15,12,13,u,u,u,u,u,u,u,u,24,25,26,27,30,31,28,29,30,31,28,29,30,31,28,29]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm12[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0,1],ymm3[2],ymm4[3,4],ymm3[5],ymm4[6,7,8,9],ymm3[10],ymm4[11,12],ymm3[13],ymm4[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[2,3,3,3,6,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm4 = ymm9[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7,8,9],ymm4[10],ymm3[11,12],ymm4[13],ymm3[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm10[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm12[4],ymm14[4],ymm12[5],ymm14[5],ymm12[6],ymm14[6],ymm12[7],ymm14[7],ymm12[12],ymm14[12],ymm12[13],ymm14[13],ymm12[14],ymm14[14],ymm12[15],ymm14[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm3[u,u,u,u,u,u,u,u,14,15,12,13,14,15,12,13,16,17,18,19,24,25,26,27,30,31,28,29,30,31,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm4, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm4 = ymm10[3,3,3,3,7,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,u,u,28,29,26,27,u,u,30,31,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7,8,9],ymm4[10],ymm5[11,12],ymm4[13],ymm5[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm13[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm2, %ymm4, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm8[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6,7,8],ymm4[9],ymm5[10,11],ymm4[12],ymm5[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7,8,9],ymm5[10],ymm4[11,12],ymm5[13],ymm4[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[0,1,2,3,6,7,4,5,u,u,u,u,u,u,u,u,16,17,18,19,22,23,20,21,22,23,20,21,28,29,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm13[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm12[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm4, %ymm3, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm11[u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm10[1,2,2,3,5,6,6,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm5[0,1],ymm4[2],ymm5[3,4],ymm4[5],ymm5[6,7,8,9],ymm4[10],ymm5[11,12],ymm4[13],ymm5[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm14[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm13[0,1,2,2,4,5,6,6]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm4, %ymm5, %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm6 = mem[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7,8,9],ymm6[10],ymm5[11,12],ymm6[13],ymm5[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm6[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm7 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7,8,9],ymm6[10],ymm7[11,12],ymm6[13],ymm7[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,2]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm4, %ymm10, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm6 = ymm8[1,1,1,1,5,5,5,5]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm4, %ymm14, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm6 = ymm12[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm4[0,1],ymm6[2],ymm4[3,4],ymm6[5],ymm4[6,7,8,9],ymm6[10],ymm4[11,12],ymm6[13],ymm4[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm11[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm9[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm9[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm8[0,1],ymm6[2],ymm8[3,4],ymm6[5],ymm8[6,7,8,9],ymm6[10],ymm8[11,12],ymm6[13],ymm8[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,2]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm4, %ymm6, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm7, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm8 = mem[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm7[0,1,2],ymm8[3],ymm7[4,5],ymm8[6],ymm7[7,8,9,10],ymm8[11],ymm7[12,13],ymm8[14],ymm7[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm8 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm13, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm12[0,0,2,1,4,4,6,5]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm11, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm10[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm6[0,1,2],ymm8[3],ymm6[4,5],ymm8[6],ymm6[7,8,9,10],ymm8[11],ymm6[12,13],ymm8[14],ymm6[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,2,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm14[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm13[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm6, %ymm8, %ymm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm5, %ymm7, %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm4, %ymm6, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm6, 96(%rax)

 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 384(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    addq $616, %rsp # imm = 0x268
 ; AVX2-FAST-PERLANE-NEXT:    vzeroupper
 ; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-SLOW-LABEL: store_i16_stride7_vf32:
 ; AVX512F-SLOW:       # %bb.0:
-; AVX512F-SLOW-NEXT:    subq $792, %rsp # imm = 0x318
+; AVX512F-SLOW-NEXT:    subq $552, %rsp # imm = 0x228
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm4, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm20
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm5, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm26
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm2, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm26
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm1, %ymm3
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm24
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm3, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm2, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm21
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm3, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm3, %ymm22
-; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm1, %ymm7
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm7, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm14
 ; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm7, %ymm14, %ymm1
+; AVX512F-SLOW-NEXT:    vpshufb %ymm7, %ymm14, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm15
 ; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm15, %ymm2
-; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm11
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm10, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm12
-; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm12, %ymm1
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm15, %ymm9
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm9, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm4
+; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm2
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm1
+; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm1, %ymm9
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm1, %ymm3
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm16
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm3, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm5
-; AVX512F-SLOW-NEXT:    vpshufb %ymm4, %ymm5, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm4
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm4, %ymm2
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm1
+; AVX512F-SLOW-NEXT:    vpshufb %ymm5, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm1, %ymm3
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm5
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm5, %ymm6
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm6, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %ymm13
-; AVX512F-SLOW-NEXT:    vpshufb %ymm7, %ymm13, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %ymm11
+; AVX512F-SLOW-NEXT:    vpshufb %ymm7, %ymm11, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %ymm7
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm7, %ymm2
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm7, %ymm8
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm8, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm11[0],xmm3[0],xmm11[1],xmm3[1],xmm11[2],xmm3[2],xmm11[3],xmm3[3]
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm3, %xmm6
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm0[0,1,2,3,4,5,7,6]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm2[0],xmm4[0],xmm2[1],xmm4[1],xmm2[2],xmm4[2],xmm2[3],xmm4[3]
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm2, %xmm6
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm8 = xmm0[0,1,2,3,4,5,7,6]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <0,1,0,1,0,1,1,3,16,18,19,19,19,19,u,u>
-; AVX512F-SLOW-NEXT:    vpermi2d %zmm2, %zmm0, %zmm1
+; AVX512F-SLOW-NEXT:    vpermi2d %zmm8, %zmm0, %zmm1
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm15[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm14[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,3,6,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm0[2],ymm2[3,4],ymm0[5],ymm2[6,7,8,9],ymm0[10],ymm2[11,12],ymm0[13],ymm2[14,15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm8 = ymm14[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[2,2,2,3,6,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7,8,9],ymm0[10],ymm8[11,12],ymm0[13],ymm8[14,15]
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %xmm8
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [4,5,4,5,4,5,6,7,16,17,16,17,16,17,17,19]
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %xmm0
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm0[0],xmm8[0],xmm0[1],xmm8[1],xmm0[2],xmm8[2],xmm0[3],xmm8[3]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm9[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermi2d %zmm3, %zmm2, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm8[4],xmm0[5],xmm8[5],xmm0[6],xmm8[6],xmm0[7],xmm8[7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm9[0,1,2,3,4,5,7,6]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm1, %xmm17
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <16,18,19,19,19,19,u,u,0,1,0,1,2,3,2,3>
-; AVX512F-SLOW-NEXT:    vpermi2d %zmm2, %zmm0, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [4,5,4,5,4,5,6,7,16,17,16,17,16,17,17,19]
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %xmm13
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm13[0],xmm8[0],xmm13[1],xmm8[1],xmm13[2],xmm8[2],xmm13[3],xmm8[3]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm12 = xmm1[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermi2d %zmm12, %zmm0, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm13[4],xmm8[4],xmm13[5],xmm8[5],xmm13[6],xmm8[6],xmm13[7],xmm8[7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,7,6]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-SLOW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm2, %xmm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = <16,18,19,19,19,19,u,u,0,1,0,1,2,3,2,3>
+; AVX512F-SLOW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-SLOW-NEXT:    vprold $16, %ymm14, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm15[1,2,2,3,5,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm0[2],ymm2[3,4],ymm0[5],ymm2[6,7,8,9],ymm0[10],ymm2[11,12],ymm0[13],ymm2[14,15]
-; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm15[1,2,2,3,5,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm15[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm14[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1,2],ymm0[3],ymm3[4,5],ymm0[6],ymm3[7,8,9,10],ymm0[11],ymm3[12,13],ymm0[14],ymm3[15]
-; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [2,2,3,3,10,9,11,10]
-; AVX512F-SLOW-NEXT:    vpermi2q %zmm2, %zmm3, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm0
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[0,2,3,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm14[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0,1,2],ymm0[3],ymm12[4,5],ymm0[6],ymm12[7,8,9,10],ymm0[11],ymm12[12,13],ymm0[14],ymm12[15]
+; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm28 = [2,2,3,3,10,9,11,10]
+; AVX512F-SLOW-NEXT:    vpermi2q %zmm1, %zmm12, %zmm28
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm13
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm13[4],xmm0[4],xmm13[5],xmm0[5],xmm13[6],xmm0[6],xmm13[7],xmm0[7]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,2,3,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm25 = ymm1[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm14
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm12 = xmm14[4],xmm1[4],xmm14[5],xmm1[5],xmm14[6],xmm1[6],xmm14[7],xmm1[7]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm12 = xmm12[2,1,2,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm12 = xmm12[0,1,2,3,4,5,5,4]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm27 = ymm12[0,0,1,3]
+; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512F-SLOW-NEXT:    vpbroadcastd 8(%rax), %ymm12
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpandn %ymm12, %ymm15, %ymm12
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rax), %ymm15
-; AVX512F-SLOW-NEXT:    vpbroadcastd 8(%rax), %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpandn %ymm2, %ymm3, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm15, %ymm9
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm2, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm11[4],xmm6[4],xmm11[5],xmm6[5],xmm11[6],xmm6[6],xmm11[7],xmm6[7]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm10[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm12[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm9[0,1],ymm11[2],ymm9[3,4],ymm11[5],ymm9[6,7,8,9],ymm11[10],ymm9[11,12],ymm11[13],ymm9[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm8, %ymm31
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm5[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm4[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm11[0,1],ymm9[2],ymm11[3,4],ymm9[5],ymm11[6,7,8,9],ymm9[10],ymm11[11,12],ymm9[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm8, %ymm29
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %ymm11
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm11[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpandn %ymm9, %ymm14, %ymm9
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm11, %ymm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm3, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm13[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm7[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm3[0,1,2],ymm9[3],ymm3[4,5],ymm9[6],ymm3[7,8,9,10],ymm9[11],ymm3[12,13],ymm9[14],ymm3[15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm8, %ymm30
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm4[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0,1],ymm14[2],ymm3[3,4],ymm14[5],ymm3[6,7,8,9],ymm14[10],ymm3[11,12],ymm14[13],ymm3[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm18
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm5[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0],ymm4[1],ymm3[2,3],ymm4[4],ymm3[5,6,7,8],ymm4[9],ymm3[10,11],ymm4[12],ymm3[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm10[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm12[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6,7,8],ymm4[9],ymm5[10,11],ymm4[12],ymm5[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm19
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm12[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm10[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0,1,2],ymm10[3],ymm4[4,5],ymm10[6],ymm4[7,8,9,10],ymm10[11],ymm4[12,13],ymm10[14],ymm4[15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm16
-; AVX512F-SLOW-NEXT:    vprold $16, %ymm13, %ymm4
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm7[1,2,2,3,5,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm12[0,1],ymm4[2],ymm12[3,4],ymm4[5],ymm12[6,7,8,9],ymm4[10],ymm12[11,12],ymm4[13],ymm12[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm27
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm7[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm7 = ymm13[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm15, %ymm10
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm12, %zmm10
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm12 = xmm6[4],xmm4[4],xmm6[5],xmm4[5],xmm6[6],xmm4[6],xmm6[7],xmm4[7]
+; AVX512F-SLOW-NEXT:    vpshufb %xmm8, %xmm12, %xmm12
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm8, %xmm20
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm31 = ymm12[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm9, %ymm6
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm9[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm16, %ymm8
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm16[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm4 = ymm12[0,1],ymm4[2],ymm12[3,4],ymm4[5],ymm12[6,7,8,9],ymm4[10],ymm12[11,12],ymm4[13],ymm12[14,15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm18 = ymm4[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm3[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm5[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1],ymm12[2],ymm9[3,4],ymm12[5],ymm9[6,7,8,9],ymm12[10],ymm9[11,12],ymm12[13],ymm9[14,15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm9[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %ymm12
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm29 = ymm12[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm29 = ymm29[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm30 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpandnq %ymm29, %ymm30, %ymm29
+; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm12, %ymm2
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm29, %zmm2, %zmm29
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm11, %ymm4
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm11[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm7[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm11[3],ymm2[4,5],ymm11[6],ymm2[7,8,9,10],ymm11[11],ymm2[12,13],ymm11[14],ymm2[15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm2[2,2,3,3]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm3, %ymm2
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm5[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm10[0,1],ymm3[2],ymm10[3,4],ymm3[5],ymm10[6,7,8,9],ymm3[10],ymm10[11,12],ymm3[13],ymm10[14,15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm5 = ymm5[4],ymm2[4],ymm5[5],ymm2[5],ymm5[6],ymm2[6],ymm5[7],ymm2[7],ymm5[12],ymm2[12],ymm5[13],ymm2[13],ymm5[14],ymm2[14],ymm5[15],ymm2[15]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm10 = ymm8[4],ymm6[4],ymm8[5],ymm6[5],ymm8[6],ymm6[6],ymm8[7],ymm6[7],ymm8[12],ymm6[12],ymm8[13],ymm6[13],ymm8[14],ymm6[14],ymm8[15],ymm6[15]
+; AVX512F-SLOW-NEXT:    vprold $16, %ymm4, %ymm3
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm7[1,2,2,3,5,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm3 = ymm6[0,1],ymm3[2],ymm6[3,4],ymm3[5],ymm6[6,7,8,9],ymm3[10],ymm6[11,12],ymm3[13],ymm6[14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm7[3,3,3,3,7,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm7 = ymm4[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[2,2,2,3,6,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm5 = ymm7[0,1],ymm4[2],ymm7[3,4],ymm4[5],ymm7[6,7,8,9],ymm4[10],ymm7[11,12],ymm4[13],ymm7[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm4
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm7
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm7[4],xmm4[4],xmm7[5],xmm4[5],xmm7[6],xmm4[6],xmm7[7],xmm4[7]
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm4[0],xmm7[0],xmm4[1],xmm7[1],xmm4[2],xmm7[2],xmm4[3],xmm7[3]
-; AVX512F-SLOW-NEXT:    vprold $16, %xmm7, %xmm7
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[1,1,2,3]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm7[2],xmm4[3,4],xmm7[5],xmm4[6,7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm4, %xmm1, %xmm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm20, %ymm9
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm9[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm26[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm7[0],ymm0[1],ymm7[2,3],ymm0[4],ymm7[5,6,7,8],ymm0[9],ymm7[10,11],ymm0[12],ymm7[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm21, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm10[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm22[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm12[0,1,2],ymm0[3],ymm12[4,5],ymm0[6],ymm12[7,8,9,10],ymm0[11],ymm12[12,13],ymm0[14],ymm12[15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm17, %xmm0
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm6, %xmm6
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm2, %xmm12
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm17, %xmm7
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm1
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7,8,9],ymm6[10],ymm7[11,12],ymm6[13],ymm7[14,15]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm1[0],xmm14[0],xmm1[1],xmm14[1],xmm1[2],xmm14[2],xmm1[3],xmm14[3]
+; AVX512F-SLOW-NEXT:    vprold $16, %xmm14, %xmm14
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm14[2],xmm1[3,4],xmm14[5],xmm1[6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm17
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm5 = ymm5[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm23 = ymm5[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm10[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm22 = ymm14[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm13[0],xmm0[0],xmm13[1],xmm0[1],xmm13[2],xmm0[2],xmm13[3],xmm0[3]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm13 = xmm13[1,1,2,2]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm8 = xmm13[0],xmm0[1],xmm13[2,3],xmm0[4],xmm13[5,6],xmm0[7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm21 = ymm10[0,2,3,3,4,6,7,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm30 = ymm3[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm20, %xmm1
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm7, %xmm4
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm13
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm7, %xmm2, %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm23
-; AVX512F-SLOW-NEXT:    vprold $16, %xmm0, %xmm2
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm14 = xmm1[1,1,2,3]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm2 = xmm14[0,1],xmm2[2],xmm14[3,4],xmm2[5],xmm14[6,7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm24
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm1
-; AVX512F-SLOW-NEXT:    vpshufb %xmm4, %xmm1, %xmm4
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm13[0],xmm0[0],xmm13[1],xmm0[1],xmm13[2],xmm0[2],xmm13[3],xmm0[3]
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm19
+; AVX512F-SLOW-NEXT:    vprold $16, %xmm0, %xmm1
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm14 = xmm13[1,1,2,3]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm14[0,1],xmm1[2],xmm14[3,4],xmm1[5],xmm14[6,7]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm0[4],xmm13[4],xmm0[5],xmm13[5],xmm0[6],xmm13[6],xmm0[7],xmm13[7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm20 = ymm6[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm13
+; AVX512F-SLOW-NEXT:    vpshufb %xmm2, %xmm13, %xmm2
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm14
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm13 = xmm14[1,1,2,2]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm13[0],xmm4[1],xmm13[2,3],xmm4[4],xmm13[5,6],xmm4[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm25
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm14[0],xmm1[0],xmm14[1],xmm1[1],xmm14[2],xmm1[2],xmm14[3],xmm1[3]
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm14[4],xmm1[4],xmm14[5],xmm1[5],xmm14[6],xmm1[6],xmm14[7],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm21
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm20, %ymm0
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm0[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm9 = xmm14[1,1,2,2]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm2 = xmm9[0],xmm2[1],xmm9[2,3],xmm2[4],xmm9[5,6],xmm2[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm14[0],xmm13[0],xmm14[1],xmm13[1],xmm14[2],xmm13[2],xmm14[3],xmm13[3]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm13 = xmm14[4],xmm13[4],xmm14[5],xmm13[5],xmm14[6],xmm13[6],xmm14[7],xmm13[7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm26, %ymm6
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm6[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm26[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm9[0,1],ymm14[2],ymm9[3,4],ymm14[5],ymm9[6,7,8,9],ymm14[10],ymm9[11,12],ymm14[13],ymm9[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm28
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm14 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm26[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm14[0,1],ymm7[2],ymm14[3,4],ymm7[5],ymm14[6,7,8,9],ymm7[10],ymm14[11,12],ymm7[13],ymm14[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm20
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm10[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm10, %ymm7
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm22[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm14[0,1],ymm10[2],ymm14[3,4],ymm10[5],ymm14[6,7,8,9],ymm10[10],ymm14[11,12],ymm10[13],ymm14[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm17
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm24, %ymm7
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm24[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm10[0,1],ymm14[2],ymm10[3,4],ymm14[5],ymm10[6,7,8,9],ymm14[10],ymm10[11,12],ymm14[13],ymm10[14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm12[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm24 = ymm12[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm26 = ymm4[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, (%rsp), %zmm25, %zmm25 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm27, %zmm27 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm25, %zmm27
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm25 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm25 = zmm31[0,1,2,3],zmm25[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm25
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm27, %zmm25
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm18, %zmm0, %zmm18
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm0, %zmm16
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm18, %zmm16
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm0, %zmm11
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm11 = zmm0[0,1,2,3],zmm11[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm29, %zmm11
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm16, %zmm11
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm17[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm5[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm8[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm7[4],ymm6[4],ymm7[5],ymm6[5],ymm7[6],ymm6[6],ymm7[7],ymm6[7],ymm7[12],ymm6[12],ymm7[13],ymm6[13],ymm7[14],ymm6[14],ymm7[15],ymm6[15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm9 = xmm9[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm9 = xmm9[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm7[2],ymm0[3,4],ymm7[5],ymm0[6,7,8,9],ymm7[10],ymm0[11,12],ymm7[13],ymm0[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm8[4],ymm12[4],ymm8[5],ymm12[5],ymm8[6],ymm12[6],ymm8[7],ymm12[7],ymm8[12],ymm12[12],ymm8[13],ymm12[13],ymm8[14],ymm12[14],ymm8[15],ymm12[15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm12[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm12[0,1],ymm8[2],ymm12[3,4],ymm8[5],ymm12[6,7,8,9],ymm8[10],ymm12[11,12],ymm8[13],ymm12[14,15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm12 = ymm7[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,2,3,3,4,6,7,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm18 = ymm19[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[2,1,2,3,4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,5,4]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,1,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm31[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm29 = ymm29[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm30 = ymm30[2,2,3,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm31 = ymm18[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm2 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm19[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm16[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm22[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm9 = ymm7[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0],ymm9[1],ymm0[2,3],ymm9[4],ymm0[5,6,7,8],ymm9[9],ymm0[10,11],ymm9[12],ymm0[13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm11[0,1,2,2,4,5,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[2,3,3,3,6,7,7,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm16 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm3 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm22 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm16, %zmm22, %zmm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm16 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm6 = zmm6[0,1,2,3],zmm16[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm6
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm6
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm3, %zmm3
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm29, %zmm7, %zmm14
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm14
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm30, %zmm0, %zmm3
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm3 = zmm7[0,1,2,3],zmm3[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm3
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm14, %zmm3
-; AVX512F-SLOW-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm14 = mem[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm8[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm16 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm16 = mem[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm18 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm18 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm13 = xmm13[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm13 = xmm13[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm19 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm19 = mem[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm24 = ymm24[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpshuflw $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm10 = mem[2,1,2,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,4,5,5,4]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,1,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm26 = ymm25[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm21, %xmm7
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm7[0,2,3,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm28 = ymm28[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm29 = ymm20[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm30 = ymm17[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm31, %zmm2
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm4, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm2, %zmm4, %zmm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm15[2,3,3,3,6,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm31 = ymm15[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm13 = xmm13[0,2,3,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm7[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm17 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm23, %zmm17, %zmm23
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm22 = ymm22[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm21 = ymm21[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm21, %zmm22, %zmm21
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm22 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm23, %zmm22, %zmm21
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm23 = ymm15[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm27 = ymm15[0,1,1,3,4,5,5,7]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[0,1,2,2,4,5,6,6]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm27, %zmm5
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm5
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm9[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm11[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm1, %zmm1
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm1
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm12, %zmm5
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,1,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm8, %zmm8
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm8
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm13[0,1,1,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm18, %zmm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm23, %zmm19, %zmm9
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm5, %zmm4, %zmm9
-; AVX512F-SLOW-NEXT:    vpbroadcastd (%rax), %ymm4
-; AVX512F-SLOW-NEXT:    vpbroadcastd 4(%rax), %ymm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm4, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm4
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm4
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpbroadcastd 32(%rax), %ymm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm2, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm2
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm2
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm24, %zmm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm26, %zmm7
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm5, %zmm22, %zmm7
-; AVX512F-SLOW-NEXT:    vpbroadcastd 36(%rax), %ymm5
-; AVX512F-SLOW-NEXT:    vpbroadcastd 40(%rax), %ymm8
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm5, %zmm5
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm5
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm29, %zmm28, %zmm7
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm30, %zmm0
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm0
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm31[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm15[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm7, %zmm7
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm7
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm7
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm20, %zmm30, %zmm20
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm21, %zmm20
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm17 = ymm24[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm17, %zmm14, %zmm14
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm20, %zmm14
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm26, %zmm16
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm16, %zmm4
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm6[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm9[0,1,1,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm6, %zmm5, %zmm5
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm12[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm18, %zmm6, %zmm6
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm5, %zmm22, %zmm6
+; AVX512F-SLOW-NEXT:    vpbroadcastd (%rax), %ymm5
+; AVX512F-SLOW-NEXT:    vpbroadcastd 4(%rax), %ymm9
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm5, %zmm5
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm5
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm23[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vpbroadcastd 32(%rax), %ymm9
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm4, %zmm4
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm4
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm4
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm2, %zmm2
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm2
+; AVX512F-SLOW-NEXT:    vpbroadcastd 36(%rax), %ymm1
+; AVX512F-SLOW-NEXT:    vpbroadcastd 40(%rax), %ymm3
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm1
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm1
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm10, %zmm0
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm7[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm8, %zmm2
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm2
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm27[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm15[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm28, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm7, 128(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm3, 320(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm5, 256(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm2, 192(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm6, 64(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm4, (%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm1, 384(%rax)
-; AVX512F-SLOW-NEXT:    addq $792, %rsp # imm = 0x318
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm0, 128(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm11, 320(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm1, 256(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm4, 192(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm25, 64(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm5, (%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm14, 384(%rax)
+; AVX512F-SLOW-NEXT:    addq $552, %rsp # imm = 0x228
 ; AVX512F-SLOW-NEXT:    vzeroupper
 ; AVX512F-SLOW-NEXT:    retq
 ;
 ; AVX512F-FAST-LABEL: store_i16_stride7_vf32:
 ; AVX512F-FAST:       # %bb.0:
 ; AVX512F-FAST-NEXT:    subq $712, %rsp # imm = 0x2C8
 ; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
 ; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %ymm2
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm15, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm2, %ymm2
 ; AVX512F-FAST-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %ymm6
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
 ; AVX512F-FAST-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm3, %ymm3
-; AVX512F-FAST-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm2
+; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
+; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm9
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128]
+; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm8
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm8, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm9
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u>
 ; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm9, %ymm3
-; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm10
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm10, %ymm4
-; AVX512F-FAST-NEXT:    vporq %ymm3, %ymm4, %ymm19
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %xmm3
+; AVX512F-FAST-NEXT:    vporq %ymm2, %ymm3, %ymm20
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %xmm6
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %xmm7
-; AVX512F-FAST-NEXT:    vpshufb %ymm12, %ymm6, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %ymm12
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm12, %ymm2
-; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %ymm11
+; AVX512F-FAST-NEXT:    vpshufb %ymm12, %ymm11, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %ymm4
+; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vpor %ymm2, %ymm3, %ymm2
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %ymm13
 ; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm13, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %ymm14
-; AVX512F-FAST-NEXT:    vpshufb %ymm15, %ymm14, %ymm2
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm14, %ymm2
 ; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %ymm15
-; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm15, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %ymm0
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm0, %ymm2
-; AVX512F-FAST-NEXT:    vporq %ymm1, %ymm2, %ymm17
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm3[4],xmm7[5],xmm3[5],xmm7[6],xmm3[6],xmm7[7],xmm3[7]
-; AVX512F-FAST-NEXT:    vpshufb %xmm8, %xmm1, %xmm1
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3]
+; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm15, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %ymm5
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
+; AVX512F-FAST-NEXT:    vporq %ymm2, %ymm1, %ymm18
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm4[4],ymm11[4],ymm4[5],ymm11[5],ymm4[6],ymm11[6],ymm4[7],ymm11[7],ymm4[12],ymm11[12],ymm4[13],ymm11[13],ymm4[14],ymm11[14],ymm4[15],ymm11[15]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm1[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm6[4],xmm7[5],xmm6[5],xmm7[6],xmm6[6],xmm7[7],xmm6[7]
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm6[0],xmm7[1],xmm6[1],xmm7[2],xmm6[2],xmm7[3],xmm6[3]
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,1,1,3,8,8,9,9]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm6 = [0,1,1,3,8,8,9,9]
 ; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm7, %xmm2
-; AVX512F-FAST-NEXT:    vpermi2q %zmm1, %zmm2, %zmm4
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpermi2q %zmm1, %zmm2, %zmm6
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa (%r9), %xmm1
 ; AVX512F-FAST-NEXT:    vmovdqa (%r8), %xmm2
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm4, %xmm5
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm6, %xmm10
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm4, %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm11 = [0,0,0,1,8,9,9,11]
-; AVX512F-FAST-NEXT:    vpermi2q %zmm5, %zmm4, %zmm11
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm6, %xmm6
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm12 = [0,0,0,1,8,9,9,11]
+; AVX512F-FAST-NEXT:    vpermi2q %zmm10, %zmm6, %zmm12
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm19, %zmm0, %zmm2
-; AVX512F-FAST-NEXT:    vpshufb %xmm8, %xmm1, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = <0,0,1,1,12,13,u,15>
-; AVX512F-FAST-NEXT:    vpermi2q %zmm2, %zmm1, %zmm4
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31>
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm0, %zmm2
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = <0,0,1,1,12,13,u,15>
+; AVX512F-FAST-NEXT:    vpermi2q %zmm2, %zmm1, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31>
 ; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm7, %xmm1
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm9, %ymm2
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm10[3,3,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm8, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa %ymm0, %ymm7
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[3,3,3,3,7,7,7,7]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7,8,9],ymm3[10],ymm2[11,12],ymm3[13],ymm2[14,15]
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [2,2,2,3,8,8,8,9]
-; AVX512F-FAST-NEXT:    vpermi2q %zmm1, %zmm2, %zmm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vprold $16, %ymm9, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm10[1,2,2,3,5,6,6,7]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [2,2,2,3,8,8,8,9]
+; AVX512F-FAST-NEXT:    vpermi2q %zmm1, %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vprold $16, %ymm8, %ymm1
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm9[1,2,2,3,5,6,6,7]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
-; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm9, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm10[0,0,2,1,4,4,6,5]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
+; AVX512F-FAST-NEXT:    vpshufb %ymm6, %ymm8, %ymm1
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[0,0,2,1,4,4,6,5]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm1[0,1,2],ymm3[3],ymm1[4,5],ymm3[6],ymm1[7,8,9,10],ymm3[11],ymm1[12,13],ymm3[14],ymm1[15]
-; AVX512F-FAST-NEXT:    vmovdqa (%rax), %ymm11
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [2,2,3,3,10,9,11,10]
-; AVX512F-FAST-NEXT:    vpermi2q %zmm2, %zmm3, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm11, %ymm3
-; AVX512F-FAST-NEXT:    vpbroadcastd 8(%rax), %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpandn %ymm4, %ymm7, %ymm4
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm28
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm4 = ymm12[1,1,1,1,5,5,5,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7,8,9],ymm4[10],ymm3[11,12],ymm4[13],ymm3[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa (%rax), %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm13[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm7 = ymm14[0,1,1,3,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm7[0,1],ymm3[2],ymm7[3,4],ymm3[5],ymm7[6,7,8,9],ymm3[10],ymm7[11,12],ymm3[13],ymm7[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm30
-; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm15, %ymm5
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm7 = ymm0[0,0,2,1,4,4,6,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0,1,2],ymm7[3],ymm5[4,5],ymm7[6],ymm5[7,8,9,10],ymm7[11],ymm5[12,13],ymm7[14],ymm5[15]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm25 = <0,1,u,3,10,10,11,11>
-; AVX512F-FAST-NEXT:    vpermi2q %zmm5, %zmm17, %zmm25
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,4,u,u,u,5,u>
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %ymm9
-; AVX512F-FAST-NEXT:    vpermd %ymm9, %ymm5, %ymm5
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpandn %ymm5, %ymm7, %ymm5
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm9, %ymm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm2, %zmm17
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm7 = ymm12[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm7[0],ymm2[1],ymm7[2,3],ymm2[4],ymm7[5,6,7,8],ymm2[9],ymm7[10,11],ymm2[12],ymm7[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm31
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm12[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm6[0,1,2],ymm2[3],ymm6[4,5],ymm2[6],ymm6[7,8,9,10],ymm2[11],ymm6[12,13],ymm2[14],ymm6[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm30 = [2,2,3,3,10,9,11,10]
+; AVX512F-FAST-NEXT:    vpermi2q %zmm2, %zmm3, %zmm30
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; AVX512F-FAST-NEXT:    vpbroadcastd 8(%rax), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpandn %ymm3, %ymm9, %ymm3
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm31
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm4[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm11[2],ymm2[3,4],ymm11[5],ymm2[6,7,8,9],ymm11[10],ymm2[11,12],ymm11[13],ymm2[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, (%rsp) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm12 = ymm14[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm12[2],ymm2[3,4],ymm12[5],ymm2[6,7,8,9],ymm12[10],ymm2[11,12],ymm12[13],ymm2[14,15]
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm14[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm11[0,1],ymm2[2],ymm11[3,4],ymm2[5],ymm11[6,7,8,9],ymm2[10],ymm11[11,12],ymm2[13],ymm11[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm13, %ymm2
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm13 = ymm14[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm13[0],ymm2[1],ymm13[2,3],ymm2[4],ymm13[5,6,7,8],ymm2[9],ymm13[10,11],ymm2[12],ymm13[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm24
-; AVX512F-FAST-NEXT:    vprold $16, %ymm15, %ymm2
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm0[1,2,2,3,5,6,6,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm14[0,1],ymm2[2],ymm14[3,4],ymm2[5],ymm14[6,7,8,9],ymm2[10],ymm14[11,12],ymm2[13],ymm14[14,15]
+; AVX512F-FAST-NEXT:    vpshufb %ymm6, %ymm15, %ymm6
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm5[0,0,2,1,4,4,6,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm6[0,1,2],ymm11[3],ymm6[4,5],ymm11[6],ymm6[7,8,9,10],ymm11[11],ymm6[12,13],ymm11[14],ymm6[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm25 = <0,1,u,3,10,10,11,11>
+; AVX512F-FAST-NEXT:    vpermi2q %zmm6, %zmm18, %zmm25
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %ymm6
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,4,5,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm6, %ymm1, %ymm11
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm26 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpandnq %ymm11, %ymm26, %ymm11
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm0, %zmm12
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm14[4],ymm13[4],ymm14[5],ymm13[5],ymm14[6],ymm13[6],ymm14[7],ymm13[7],ymm14[12],ymm13[12],ymm14[13],ymm13[13],ymm14[14],ymm13[14],ymm14[15],ymm13[15]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,2,2,2,6,6,6,6]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm13[0,1],ymm14[2],ymm13[3,4],ymm14[5],ymm13[6,7,8,9],ymm14[10],ymm13[11,12],ymm14[13],ymm13[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm20
+; AVX512F-FAST-NEXT:    vprold $16, %ymm15, %ymm14
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm5[1,2,2,3,5,6,6,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm10[0,1],ymm14[2],ymm10[3,4],ymm14[5],ymm10[6,7,8,9],ymm14[10],ymm10[11,12],ymm14[13],ymm10[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm15, %ymm2
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0,1],ymm0[2],ymm2[3,4],ymm0[5],ymm2[6,7,8,9],ymm0[10],ymm2[11,12],ymm0[13],ymm2[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm22
-; AVX512F-FAST-NEXT:    vmovdqa64 (%rax), %zmm27
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm29 = <u,5,u,u,u,6,u,u,30,u,u,u,31,u,u,31>
-; AVX512F-FAST-NEXT:    vpermi2d %zmm27, %zmm9, %zmm29
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %xmm9
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %xmm10
-; AVX512F-FAST-NEXT:    vprold $16, %xmm10, %xmm2
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm8 = xmm9[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm8[0,1],xmm2[2],xmm8[3,4],xmm2[5],xmm8[6,7]
+; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm15, %ymm10
+; AVX512F-FAST-NEXT:    vmovdqa64 (%rax), %zmm26
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[3,3,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm10[0,1],ymm5[2],ymm10[3,4],ymm5[5],ymm10[6,7,8,9],ymm5[10],ymm10[11,12],ymm5[13],ymm10[14,15]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %xmm10
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm15 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm11 = <u,5,u,u,u,6,u,u,30,u,u,u,31,u,u,31>
+; AVX512F-FAST-NEXT:    vpermi2d %zmm26, %zmm6, %zmm11
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm27 = [2,3,6,7,6,7,7,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm15, %ymm27, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm0
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm2, %xmm6
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm5, %xmm2
-; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %xmm5
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %xmm2
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm5, %xmm3
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm8 = xmm2[1,1,2,2]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm8[0],xmm3[1],xmm8[2,3],xmm3[4],xmm8[5,6],xmm3[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm3, %ymm19
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vprold $16, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3,4],xmm0[5],xmm1[6,7]
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %xmm7
+; AVX512F-FAST-NEXT:    vprold $16, %xmm7, %xmm6
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm15 = xmm10[1,1,2,3]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm15[0,1],xmm6[2],xmm15[3,4],xmm6[5],xmm15[6,7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm22
+; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %xmm8
+; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm9
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm6, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm18
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm10[4],xmm7[5],xmm10[5],xmm7[6],xmm10[6],xmm7[7],xmm10[7]
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm6, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm23
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
+; AVX512F-FAST-NEXT:    vpshufb %xmm6, %xmm2, %xmm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm14 = xmm1[1,1,2,2]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm14[0],xmm0[1],xmm14[2,3],xmm0[4],xmm14[5,6],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm24
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3]
+; AVX512F-FAST-NEXT:    vprold $16, %xmm9, %xmm9
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[1,1,2,3]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm8[0,1],xmm9[2],xmm8[3,4],xmm9[5],xmm8[6,7]
 ; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm21
-; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %xmm1
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm1, %xmm4
+; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %xmm9
+; AVX512F-FAST-NEXT:    vpshufb %xmm6, %xmm9, %xmm6
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %xmm0
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm0[4],xmm9[4],xmm0[5],xmm9[5],xmm0[6],xmm9[6],xmm0[7],xmm9[7]
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm0[0],xmm9[0],xmm0[1],xmm9[1],xmm0[2],xmm9[2],xmm0[3],xmm9[3]
 ; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2,3],xmm4[4],xmm0[5,6],xmm4[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm23
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm6[1],xmm0[2,3],xmm6[4],xmm0[5,6],xmm6[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm19
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm6 = ymm5[4],ymm3[4],ymm5[5],ymm3[5],ymm5[6],ymm3[6],ymm5[7],ymm3[7],ymm5[12],ymm3[12],ymm5[13],ymm3[13],ymm5[14],ymm3[14],ymm5[15],ymm3[15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vpermd %ymm6, %ymm27, %ymm29
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm15 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm9, %xmm9
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm17
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,6,u,u,7,7,7,u>
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm6[4],ymm4[4],ymm6[5],ymm4[5],ymm6[6],ymm4[6],ymm6[7],ymm4[7],ymm6[12],ymm4[12],ymm6[13],ymm4[13],ymm6[14],ymm4[14],ymm6[15],ymm4[15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm13 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vpermd %ymm13, %ymm2, %ymm27
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm10[0],xmm7[0],xmm10[1],xmm7[1],xmm10[2],xmm7[2],xmm10[3],xmm7[3]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm14, %xmm7
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm16
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm13 = ymm3[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm5[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm13 = ymm14[0,1],ymm13[2],ymm14[3,4],ymm13[5],ymm14[6,7,8,9],ymm13[10],ymm14[11,12],ymm13[13],ymm14[14,15]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm14 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[2,2,2,2,6,6,6,6]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm14[0,1],ymm2[2],ymm14[3,4],ymm2[5],ymm14[6,7,8,9],ymm2[10],ymm14[11,12],ymm2[13],ymm14[14,15]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,4,5,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm28
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm14 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm6[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm14[0,1],ymm0[2],ymm14[3,4],ymm0[5],ymm14[6,7,8,9],ymm0[10],ymm14[11,12],ymm0[13],ymm14[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <6,u,u,u,7,u,u,7>
 ; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm0, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm7[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm4 = ymm14[0],ymm4[1],ymm14[2,3],ymm4[4],ymm14[5,6,7,8],ymm4[9],ymm14[10,11],ymm4[12],ymm14[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm4, %ymm26
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm2[4],xmm5[4],xmm2[5],xmm5[5],xmm2[6],xmm5[6],xmm2[7],xmm5[7]
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1],xmm2[2],xmm5[2],xmm2[3],xmm5[3]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm5 = [0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
-; AVX512F-FAST-NEXT:    vpshufb %xmm5, %xmm1, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm18
-; AVX512F-FAST-NEXT:    vpshufb %xmm5, %xmm2, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm20
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm2[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm4 = ymm15[0,1,2],ymm5[3],ymm15[4,5],ymm5[6],ymm15[7,8,9,10],ymm5[11],ymm15[12,13],ymm5[14],ymm15[15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm4, %ymm16
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm9 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm8, %xmm8
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm9, %xmm9
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm10 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm7[0,1,1,3,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm15[0,1],ymm10[2],ymm15[3,4],ymm10[5],ymm15[6,7,8,9],ymm10[10],ymm15[11,12],ymm10[13],ymm15[14,15]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm13 = ymm7[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3,4],ymm13[5],ymm15[6,7,8,9],ymm13[10],ymm15[11,12],ymm13[13],ymm15[14,15]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm7 = ymm2[1,1,1,1,5,5,5,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm7 = ymm15[0,1],ymm7[2],ymm15[3,4],ymm7[5],ymm15[6,7,8,9],ymm7[10],ymm15[11,12],ymm7[13],ymm15[14,15]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm1[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm12 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0],ymm15[1],ymm12[2,3],ymm15[4],ymm12[5,6,7,8],ymm15[9],ymm12[10,11],ymm15[12],ymm12[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <6,u,u,u,7,u,u,7>
-; AVX512F-FAST-NEXT:    vpermd %ymm11, %ymm15, %ymm15
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm27, %zmm11
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[0,2,3,3,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,1,3]
-; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm27 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm30 = ymm30[2,1,3,2]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm31 = ymm31[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $246, (%rsp), %ymm2 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm2 = mem[2,1,3,3]
-; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm1 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm24[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm5 = mem[2,1,3,2]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm22[2,2,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm3 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm6, %zmm6 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm3, %zmm24, %zmm6
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm28
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm28
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm27, %zmm3, %zmm3
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm6 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm30, %zmm6, %zmm6
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm6
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm25, %zmm17
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm17
-; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm3 = mem[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq $208, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm6 = mem[0,0,1,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm19 = ymm19[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm14 = xmm14[0,2,3,3,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm21 = ymm21[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm22 = ymm18[0,1,1,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm25 = ymm26[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm20[0,1,1,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm30 = ymm16[2,1,3,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,2,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm31, %zmm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm2, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm2
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm2
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm29
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm3, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm19, %zmm2
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm24, %zmm2
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm14, %ymm14
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm26, %zmm3
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm8 = xmm8[0,2,3,3,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm18[0,0,1,3]
+; AVX512F-FAST-NEXT:    vpermq $234, (%rsp), %ymm26 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm26 = mem[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm0 = mem[2,1,3,2]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm8, %zmm8 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm10, %zmm10 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm18 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm8, %zmm18, %zmm10
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm31
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm10, %zmm31
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm26, %zmm4, %zmm8
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm4, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm25, %zmm12
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm20[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm10 = mem[2,1,3,2]
+; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm20 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm20 = mem[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm22 = ymm22[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[0,0,1,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm25 = ymm24[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm15[0,2,3,3,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm26 = ymm21[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,1,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm19[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm17[0,1,1,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm16[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,1,3,2]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm12
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = <u,u,4,u,u,5,5,u,u,14,u,u,15,15,u,u>
+; AVX512F-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm8, %zmm8 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm24, %zmm8
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm10, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm11
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm23, %zmm22, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm25, %zmm6
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm18, %zmm6
 ; AVX512F-FAST-NEXT:    vpbroadcastd 36(%rax), %ymm0
-; AVX512F-FAST-NEXT:    vpbroadcastd 40(%rax), %ymm3
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm0
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm21, %zmm8, %zmm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm23, %zmm22, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm3
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm27, %zmm25, %zmm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm30, %zmm4
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm2, %zmm1, %zmm4
-; AVX512F-FAST-NEXT:    vpbroadcastd (%rax), %ymm1
-; AVX512F-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpbroadcastd 40(%rax), %ymm8
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm26, %zmm7, %zmm6
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm15, %zmm9, %zmm7
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm7
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm29, %zmm6
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm27, %zmm5
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm6, %zmm24, %zmm5
+; AVX512F-FAST-NEXT:    vpbroadcastd (%rax), %ymm6
+; AVX512F-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm8
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm6, %zmm6
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm6
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm6
+; AVX512F-FAST-NEXT:    vpbroadcastd 32(%rax), %ymm7
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm14, %zmm7
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm7
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm7
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm13, %zmm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm28, %zmm1, %zmm1
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm1
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm1
-; AVX512F-FAST-NEXT:    vpbroadcastd 32(%rax), %ymm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm15, %zmm2
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm2
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm13, %zmm10, %zmm3
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm7, %zmm4
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
-; AVX512F-FAST-NEXT:    vpermd %zmm11, %zmm3, %zmm3
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
+; AVX512F-FAST-NEXT:    vpermd %zmm3, %zmm2, %zmm2
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm30, %zmm2
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm2
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm17, 320(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm3, 128(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm2, 192(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm28, 64(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm1, (%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm12, 320(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm2, 128(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm7, 192(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm31, 64(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm6, (%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm0, 256(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm29, 384(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm11, 384(%rax)
 ; AVX512F-FAST-NEXT:    addq $712, %rsp # imm = 0x2C8
 ; AVX512F-FAST-NEXT:    vzeroupper
 ; AVX512F-FAST-NEXT:    retq
 ;
 ; AVX512BW-LABEL: store_i16_stride7_vf32:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512BW-NEXT:    movq {{[0-9]+}}(%rsp), %r10

 ; AVX2-SLOW-NEXT:    subq $1720, %rsp # imm = 0x6B8
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-SLOW-NEXT:    vmovdqa 64(%r8), %ymm4
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%r8), %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 64(%r9), %ymm5
-; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 64(%r9), %ymm6
+; AVX2-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%r9), %ymm7
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm9
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rax), %ymm8
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 64(%rax), %ymm6
-; AVX2-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 64(%rax), %ymm5
+; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,3,u,u,u,4,u>
 ; AVX2-SLOW-NEXT:    vpermd %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm10
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,3,4,5,4,7]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm1[0,1,2,3,4,4,7,7,8,9,10,11,12,12,15,15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <3,u,u,3,u,u,u,4>
 ; AVX2-SLOW-NEXT:    vpermd %ymm3, %ymm10, %ymm2
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm7[0,1,0,3,4,5,4,7]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,4,7,7,8,9,10,11,12,12,15,15]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm2, %ymm3, %ymm2
 ; AVX2-SLOW-NEXT:    vpermd %ymm4, %ymm10, %ymm3
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm5[0,1,0,3,4,5,4,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm6[0,1,0,3,4,5,4,7]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,4,4,7,7,8,9,10,11,12,12,15,15]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm3, %ymm4, %ymm1
 ; AVX2-SLOW-NEXT:    vpermd %ymm9, %ymm11, %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm3, %ymm0
 ; AVX2-SLOW-NEXT:    vpermd %ymm8, %ymm11, %ymm3
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
-; AVX2-SLOW-NEXT:    vpermd %ymm6, %ymm11, %ymm3
+; AVX2-SLOW-NEXT:    vpermd %ymm5, %ymm11, %ymm3
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,3,2,3,4,7,6,7]
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm3[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm4
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <3,u,u,u,4,u,u,4>

 ; AVX2-SLOW-NEXT:    vpermd %ymm3, %ymm9, %ymm3
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm3, %ymm2, %ymm2
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 96(%rsi), %ymm1
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm1[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm1, %ymm4
+; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vmovdqa 96(%rdi), %ymm2
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[1,1,1,1,5,5,5,5]
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm6
+; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
 ; AVX2-SLOW-NEXT:    vmovdqa 96(%rcx), %ymm2
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm2[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm5
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vmovdqa 96(%rdx), %ymm3
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]

 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa 96(%rax), %ymm2
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm9
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[2,2,2,2,6,6,6,6]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm5, %ymm10
+; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa %ymm3, %ymm5
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[2,2,2,2,6,6,6,6]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm4[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,2,2,3,5,6,6,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm6[4],ymm4[4],ymm6[5],ymm4[5],ymm6[6],ymm4[6],ymm6[7],ymm4[7],ymm6[12],ymm4[12],ymm6[13],ymm4[13],ymm6[14],ymm4[14],ymm6[15],ymm4[15]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm1[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[1,2,2,3,5,6,6,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm8[2,1,2,3,6,5,6,7]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm1[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm8[2,1,2,3,6,5,6,7]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[0,1,2,2,4,5,6,6]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm9[0,1,2,2,4,5,6,6]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm4[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm5[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,3,3,4,6,7,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm5[4],ymm10[4],ymm5[5],ymm10[5],ymm5[6],ymm10[6],ymm5[7],ymm10[7],ymm5[12],ymm10[12],ymm5[13],ymm10[13],ymm5[14],ymm10[14],ymm5[15],ymm10[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,3,3,6,7,7,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vpbroadcastd 124(%r8), %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm8[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[3,3,3,3]

 ; AVX2-SLOW-NEXT:    vpshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm6 = mem[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm7 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7,8,9],ymm6[10],ymm7[11,12],ymm6[13],ymm7[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm5, %ymm6, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm5, %ymm6, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshuflw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm6 = mem[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm7 = mem[1,1,1,1,5,5,5,5]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm6 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7,8,9],ymm7[10],ymm6[11,12],ymm7[13],ymm6[14,15]
 ; AVX2-SLOW-NEXT:    vpshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm7 = mem[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,0,0,0,4,4,4,4]
 ; AVX2-SLOW-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm8 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm8[0,1],ymm7[2],ymm8[3,4],ymm7[5],ymm8[6,7,8,9],ymm7[10],ymm8[11,12],ymm7[13],ymm8[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,2,2,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm6, %ymm7, %ymm0
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm6, %ymm7, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshuflw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm7 = mem[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm8 = mem[1,1,1,1,5,5,5,5]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7,8,9],ymm8[10],ymm7[11,12],ymm8[13],ymm7[14,15]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm0[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm1[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[0,0,0,0,4,4,4,4]
-; AVX2-SLOW-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm9 = mem[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm0[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm9[0,1],ymm8[2],ymm9[3,4],ymm8[5],ymm9[6,7,8,9],ymm8[10],ymm9[11,12],ymm8[13],ymm9[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,2,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm7, %ymm8, %ymm7
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm5[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm7, %ymm8, %ymm14
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm15[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[0,0,0,0,4,4,4,4]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm13[0,0,2,1,4,4,6,5]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm3[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1,2],ymm9[3],ymm8[4,5],ymm9[6],ymm8[7,8,9,10],ymm9[11],ymm8[12,13],ymm9[14],ymm8[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,3,3]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm14[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm5[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm8, %ymm9, %ymm8
+; AVX2-SLOW-NEXT:    vpshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm9 = mem[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm2[0,0,2,1,4,4,6,5]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm4[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1,2],ymm11[3],ymm9[4,5],ymm11[6],ymm9[7,8,9,10],ymm11[11],ymm9[12,13],ymm11[14],ymm9[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,3,3]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm15[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm6[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm9, %ymm11, %ymm9
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm1[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm9, %ymm11, %ymm9
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm2[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[0,0,0,0,4,4,4,4]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm6[0,0,2,1,4,4,6,5]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm10[0,0,2,1,4,4,6,5]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3],ymm11[4,5],ymm12[6],ymm11[7,8,9,10],ymm12[11],ymm11[12,13],ymm12[14],ymm11[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,2,3,3]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm3[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm7[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,2,2,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm11, %ymm12, %ymm10
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm12
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm11, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm11, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm8 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm7, %ymm10, %ymm7
-; AVX2-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm14, %ymm12, %ymm8
+; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm11, %ymm7, %ymm8
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb %ymm11, %ymm8, %ymm8
 ; AVX2-SLOW-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm9 = mem[2,2,2,2,6,6,6,6]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7,8,9],ymm9[10],ymm8[11,12],ymm9[13],ymm8[14,15]
-; AVX2-SLOW-NEXT:    vpshufhw $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm9 = mem[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm7[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm10[0],ymm9[1],ymm10[2,3],ymm9[4],ymm10[5,6,7,8],ymm9[9],ymm10[10,11],ymm9[12],ymm10[13,14,15]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm9 = ymm9[4],mem[4],ymm9[5],mem[5],ymm9[6],mem[6],ymm9[7],mem[7],ymm9[12],mem[12],ymm9[13],mem[13],ymm9[14],mem[14],ymm9[15],mem[15]
+; AVX2-SLOW-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm9[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,1,1,3,4,5,5,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm8, %ymm9, %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm8, %ymm9, %ymm8
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vpshufb %ymm11, %ymm8, %ymm9
-; AVX2-SLOW-NEXT:    vmovdqa %ymm11, %ymm8
-; AVX2-SLOW-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm10 = mem[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1],ymm10[2],ymm9[3,4],ymm10[5],ymm9[6,7,8,9],ymm10[10],ymm9[11,12],ymm10[13],ymm9[14,15]
-; AVX2-SLOW-NEXT:    vpshufhw $225, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm10 = mem[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[2,2,2,2,6,6,6,6]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm11, %ymm13
 ; AVX2-SLOW-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
 ; AVX2-SLOW-NEXT:    # ymm11 = mem[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6,7,8],ymm10[9],ymm11[10,11],ymm10[12],ymm11[13,14,15]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1],ymm11[2],ymm9[3,4],ymm11[5],ymm9[6,7,8,9],ymm11[10],ymm9[11,12],ymm11[13],ymm9[14,15]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm11 = ymm11[4],mem[4],ymm11[5],mem[5],ymm11[6],mem[6],ymm11[7],mem[7],ymm11[12],mem[12],ymm11[13],mem[13],ymm11[14],mem[14],ymm11[15],mem[15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm9, %ymm10, %ymm9
-; AVX2-SLOW-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm0, %ymm10
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm9[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm10[0,1],ymm11[2],ymm10[3,4],ymm11[5],ymm10[6,7,8,9],ymm11[10],ymm10[11,12],ymm11[13],ymm10[14,15]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm8[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[2,2,2,2,6,6,6,6]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm11[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm9, %ymm12, %ymm9
+; AVX2-SLOW-NEXT:    vpshufb %ymm13, %ymm1, %ymm12
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm0[2,2,2,2,6,6,6,6]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0,1],ymm13[2],ymm12[3,4],ymm13[5],ymm12[6,7,8,9],ymm13[10],ymm12[11,12],ymm13[13],ymm12[14,15]
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm0[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0],ymm11[1],ymm12[2,3],ymm11[4],ymm12[5,6,7,8],ymm11[9],ymm12[10,11],ymm11[12],ymm12[13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm5[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm11[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm13[1,2,2,3,5,6,6,7]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm13, %ymm5
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0,1],ymm11[2],ymm12[3,4],ymm11[5],ymm12[6,7,8,9],ymm11[10],ymm12[11,12],ymm11[13],ymm12[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm14[0,1,2,2,4,5,6,6]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm4[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm12 = ymm12[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm4
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7,8,9],ymm12[10],ymm14[11,12],ymm12[13],ymm14[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm15[0,1,2,2,4,5,6,6]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm12, %ymm14, %ymm12
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm1[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm13 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm13 = ymm0[4],mem[4],ymm0[5],mem[5],ymm0[6],mem[6],ymm0[7],mem[7],ymm0[12],mem[12],ymm0[13],mem[13],ymm0[14],mem[14],ymm0[15],mem[15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,2,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm13[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm0, %ymm12, %ymm14, %ymm12
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm15[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm14 = ymm14[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm6[1,2,2,3,5,6,6,7]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm6, %ymm2
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[1,2,2,3,5,6,6,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7,8,9],ymm14[10],ymm15[11,12],ymm14[13],ymm15[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[0,1,2,2,4,5,6,6]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm5[0,1,2,2,4,5,6,6]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm14, %ymm15, %ymm13
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm14, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm1 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpblendvb %ymm14, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12, %ymm6 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm10, %ymm13, %ymm10
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm7[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $235, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm12 = mem[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm11 = ymm11[0,1,2],ymm12[3],ymm11[4,5],ymm12[6],ymm11[7,8,9,10],ymm12[11],ymm11[12,13],ymm12[14],ymm11[15]
-; AVX2-SLOW-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm12 = mem[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $235, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm13 = mem[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0],ymm13[1],ymm12[2,3],ymm13[4],ymm12[5,6,7,8],ymm13[9],ymm12[10,11],ymm13[12],ymm12[13,14,15]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm14, %ymm15, %ymm14
+; AVX2-SLOW-NEXT:    vpshuflw $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm15 = mem[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm15 = ymm15[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[1,2,2,3,5,6,6,7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm4, %ymm5
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm15[2],ymm1[3,4],ymm15[5],ymm1[6,7,8,9],ymm15[10],ymm1[11,12],ymm15[13],ymm1[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm6[0,1,2,2,4,5,6,6]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm1, %ymm15, %ymm1
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm2[3,3,3,3,4,5,6,7,11,11,11,11,12,13,14,15]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, %ymm6
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm15 = ymm15[0,1,2,3,4,4,4,4,8,9,10,11,12,12,12,12]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm10[1,2,2,3,5,6,6,7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm10, %ymm4
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm15[2],ymm2[3,4],ymm15[5],ymm2[6,7,8,9],ymm15[10],ymm2[11,12],ymm15[13],ymm2[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm7[0,1,2,2,4,5,6,6]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm2, %ymm15, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14, %ymm3 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm9, %ymm1, %ymm9
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm12, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm1 = ymm1[4],mem[4],ymm1[5],mem[5],ymm1[6],mem[6],ymm1[7],mem[7],ymm1[12],mem[12],ymm1[13],mem[13],ymm1[14],mem[14],ymm1[15],mem[15]
+; AVX2-SLOW-NEXT:    vpshufhw $180, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm2 = mem[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,3,3,4,6,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm2, %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm2[4],ymm8[4],ymm2[5],ymm8[5],ymm2[6],ymm8[6],ymm2[7],ymm8[7],ymm2[12],ymm8[12],ymm2[13],ymm8[13],ymm2[14],ymm8[14],ymm2[15],ymm8[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm11[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,2,3,3,4,6,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm10, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm10 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm10 = ymm8[4],mem[4],ymm8[5],mem[5],ymm8[6],mem[6],ymm8[7],mem[7],ymm8[12],mem[12],ymm8[13],mem[13],ymm8[14],mem[14],ymm8[15],mem[15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm13[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[0,2,3,3,4,6,7,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,2,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm11, %ymm12, %ymm11
-; AVX2-SLOW-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm12 = mem[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $235, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm13 = mem[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0,1,2],ymm13[3],ymm12[4,5],ymm13[6],ymm12[7,8,9,10],ymm13[11],ymm12[12,13],ymm13[14],ymm12[15]
-; AVX2-SLOW-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm13 = mem[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $235, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm14 = mem[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm13 = ymm13[0],ymm14[1],ymm13[2,3],ymm14[4],ymm13[5,6,7,8],ymm14[9],ymm13[10,11],ymm14[12],ymm13[13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,2,2,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm12, %ymm13, %ymm12
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm0[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm14 = ymm8[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm13 = ymm13[0,1,2],ymm14[3],ymm13[4,5],ymm14[6],ymm13[7,8,9,10],ymm14[11],ymm13[12,13],ymm14[14],ymm13[15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm9[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $235, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm15 = mem[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,2,2,2,6,6,6,6]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm14 = ymm14[0],ymm15[1],ymm14[2,3],ymm15[4],ymm14[5,6,7,8],ymm15[9],ymm14[10,11],ymm15[12],ymm14[13,14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,1,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm13, %ymm14, %ymm13
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm5[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm15 = mem[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm11, %ymm10, %ymm10
+; AVX2-SLOW-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm11 = mem[3,3,3,3,7,7,7,7]
+; AVX2-SLOW-NEXT:    vpshufhw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm12 = mem[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[2,2,2,3,6,6,6,7]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0,1],ymm11[2],ymm12[3,4],ymm11[5],ymm12[6,7,8,9],ymm11[10],ymm12[11,12],ymm11[13],ymm12[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,2,2,3]
+; AVX2-SLOW-NEXT:    vpshufd $254, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm12 = mem[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm5[3,3,3,3,7,7,7,7]
+; AVX2-SLOW-NEXT:    vpshufhw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm14 = mem[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[2,2,2,3,6,6,6,7]
+; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm12 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7,8,9],ymm12[10],ymm14[11,12],ymm12[13],ymm14[14,15]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,2,2,3]
+; AVX2-SLOW-NEXT:    vpshufd $254, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm14 = mem[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,2]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm12, %ymm14, %ymm12
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm4[3,3,3,3,7,7,7,7]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm15 = ymm6[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,2,2,3,6,6,6,7]
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7,8,9],ymm14[10],ymm15[11,12],ymm14[13],ymm15[14,15]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpshufd $254, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm15 = mem[2,3,3,3,6,7,7,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm7[2,3,3,3,6,7,7,7]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm0, %ymm14, %ymm15, %ymm14
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm1 = mem[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,3,6,6,6,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm15[2],ymm1[3,4],ymm15[5],ymm1[6,7,8,9],ymm15[10],ymm1[11,12],ymm15[13],ymm1[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpshufd $254, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm15 = mem[2,3,3,3,6,7,7,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm0, %ymm1, %ymm15, %ymm1
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm2[3,3,3,3,7,7,7,7]
-; AVX2-SLOW-NEXT:    vpshufhw $249, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm2 = mem[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,3,6,6,6,7]
-; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm15[2],ymm2[3,4],ymm15[5],ymm2[6,7,8,9],ymm15[10],ymm2[11,12],ymm15[13],ymm2[14,15]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[2,3,3,3,6,7,7,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,2]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm0, %ymm2, %ymm15, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm11, %ymm14, %ymm11
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm12, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm13, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm14, %ymm15, %ymm13
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm1, %ymm11, %ymm1
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm2, %ymm12, %ymm2
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm10, %ymm13, %ymm10
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm2, 544(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm2, 320(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm2, 96(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 640(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm10, 608(%rax)
+; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovaps %ymm3, 544(%rax)
+; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovaps %ymm3, 320(%rax)
+; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovaps %ymm3, 96(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm10, 640(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 608(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 576(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 416(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm6, 384(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, 416(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm9, 384(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 352(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm11, 192(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 192(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 160(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 128(%rax)
 ; AVX2-SLOW-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 768(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 736(%rax)

 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-NEXT:    vmovdqa 96(%rax), %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <3,u,u,3,u,u,u,4>
 ; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm2
-; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm3
+; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm13
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,2,1,4,4,6,5]
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm13, %ymm1
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5],ymm2[6],ymm1[7,8,9,10],ymm2[11],ymm1[12,13],ymm2[14],ymm1[15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[2,2,3,3]
 ; AVX2-FAST-NEXT:    vmovdqa (%rax), %ymm3
 ; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,4,5,4,5,5,7]
 ; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm1, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm4
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm5
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm2, %ymm3, %ymm8
-; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %ymm5
-; AVX2-FAST-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %ymm6
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm2
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm5[0,0,2,1,4,4,6,5]
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm2, %ymm3, %ymm9
+; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %ymm6
+; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %ymm2
+; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7,8,9,10],ymm3[11],ymm2[12,13],ymm3[14],ymm2[15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,3,3]
-; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %ymm13
-; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm4, %ymm3
-; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm2, %ymm3, %ymm7
-; AVX2-FAST-NEXT:    vmovdqa 64(%r9), %ymm9
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm9, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm9, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %ymm7
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm5, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa %ymm5, %ymm8
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm2, %ymm3, %ymm5
+; AVX2-FAST-NEXT:    vmovdqa 64(%r9), %ymm11
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm11, %ymm0
+; AVX2-FAST-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 64(%r8), %ymm2
 ; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3],ymm0[4,5],ymm2[6],ymm0[7,8,9,10],ymm2[11],ymm0[12,13],ymm2[14],ymm0[15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3]
-; AVX2-FAST-NEXT:    vmovdqa 64(%rax), %ymm2
-; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm4, %ymm2
-; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm11
+; AVX2-FAST-NEXT:    vmovdqa 64(%rax), %ymm12
+; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm8, %ymm2
+; AVX2-FAST-NEXT:    vmovdqu %ymm12, (%rsp) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm10
 ; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm2
 ; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
+; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm12
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm15
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm15[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm12[0,1,1,3,4,5,5,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm3
+; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm14
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm14[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-NEXT:    vmovdqu %ymm14, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa %ymm2, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm9, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm1[1,1,1,1,5,5,5,5]
+; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm2[2],ymm0[3,4],ymm2[5],ymm0[6,7,8,9],ymm2[10],ymm0[11,12],ymm2[13],ymm0[14,15]
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %ymm4
+; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %ymm2
 ; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm1[0,1,1,3,4,5,5,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm8[0,1],ymm2[2],ymm8[3,4],ymm2[5],ymm8[6,7,8,9],ymm2[10],ymm8[11,12],ymm2[13],ymm8[14,15]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm4[0,1,1,3,4,5,5,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm9[0,1],ymm2[2],ymm9[3,4],ymm2[5],ymm9[6,7,8,9],ymm2[10],ymm9[11,12],ymm2[13],ymm9[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
-; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm7, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm4
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm5, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 64(%rdi), %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 64(%rsi), %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm0, %ymm7
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm1[1,1,1,1,5,5,5,5]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7,8,9],ymm8[10],ymm7[11,12],ymm8[13],ymm7[14,15]
+; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm1[1,1,1,1,5,5,5,5]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm0[0,1],ymm2[2],ymm0[3,4],ymm2[5],ymm0[6,7,8,9],ymm2[10],ymm0[11,12],ymm2[13],ymm0[14,15]
 ; AVX2-FAST-NEXT:    vmovdqa 64(%rdx), %ymm2
 ; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 64(%rcx), %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm14[0,1],ymm0[2],ymm14[3,4],ymm0[5],ymm14[6,7,8,9],ymm0[10],ymm14[11,12],ymm0[13],ymm14[14,15]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm2[0,1,1,3,4,5,5,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3,4],ymm0[5],ymm15[6,7,8,9],ymm0[10],ymm15[11,12],ymm0[13],ymm15[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,2]
-; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm11, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm10, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u>
+; AVX2-FAST-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm13, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[1,2,2,3,5,6,6,7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm2[1,2,2,3,5,6,6,7]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm3[0,1],ymm1[2],ymm3[3,4],ymm1[5],ymm3[6,7,8,9],ymm1[10],ymm3[11,12],ymm1[13],ymm3[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [4,5,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [4,5,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm8, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm10, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm5, %ymm3
 ; AVX2-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm3
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm7[1,2,2,3,5,6,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm14[0,1],ymm3[2],ymm14[3,4],ymm3[5],ymm14[6,7,8,9],ymm3[10],ymm14[11,12],ymm3[13],ymm14[14,15]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm6[1,2,2,3,5,6,6,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm15[0,1],ymm3[2],ymm15[3,4],ymm3[5],ymm15[6,7,8,9],ymm3[10],ymm15[11,12],ymm3[13],ymm15[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqa %ymm13, %ymm8
-; AVX2-FAST-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm10, %ymm14
-; AVX2-FAST-NEXT:    vmovdqa %ymm10, %ymm13
-; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm3, %ymm14, %ymm3
-; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm9, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm10[1,2,2,3,5,6,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm14[0,1],ymm0[2],ymm14[3,4],ymm0[5],ymm14[6,7,8,9],ymm0[10],ymm14[11,12],ymm0[13],ymm14[14,15]
+; AVX2-FAST-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm8, %ymm15
+; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm3, %ymm15, %ymm3
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm11, %ymm0
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm11[1,2,2,3,5,6,6,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3,4],ymm0[5],ymm15[6,7,8,9],ymm0[10],ymm15[11,12],ymm0[13],ymm15[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,2]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpermd %ymm9, %ymm13, %ymm14
-; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm0, %ymm14, %ymm0
+; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm8, %ymm15
+; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm15, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
-; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm15, %ymm11
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm12[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm11 = ymm11[0,1],ymm14[2],ymm11[3,4],ymm14[5],ymm11[6,7,8,9],ymm14[10],ymm11[11,12],ymm14[13],ymm11[14,15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm0, %ymm14
-; AVX2-FAST-NEXT:    vmovdqa %ymm12, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
+; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm14, %ymm10
 ; AVX2-FAST-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    # ymm15 = mem[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6,7,8],ymm14[9],ymm15[10,11],ymm14[12],ymm15[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm11, %ymm14, %ymm11
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm10[0,1],ymm15[2],ymm10[3,4],ymm15[5],ymm10[6,7,8,9],ymm15[10],ymm10[11,12],ymm15[13],ymm10[14,15]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm15 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm15 = ymm0[4],mem[4],ymm0[5],mem[5],ymm0[6],mem[6],ymm0[7],mem[7],ymm0[12],mem[12],ymm0[13],mem[13],ymm0[14],mem[14],ymm0[15],mem[15]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm15[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [0,1,4,5,4,5,5,7]
+; AVX2-FAST-NEXT:    vpermd %ymm14, %ymm0, %ymm14
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm10, %ymm14, %ymm10
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm11, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm10, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm12[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm11[2],ymm1[3,4],ymm11[5],ymm1[6,7,8,9],ymm11[10],ymm1[11,12],ymm11[13],ymm1[14,15]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm11, %ymm11
-; AVX2-FAST-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    # ymm15 = mem[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm11 = ymm15[0],ymm11[1],ymm15[2,3],ymm11[4],ymm15[5,6,7,8],ymm11[9],ymm15[10,11],ymm11[12],ymm15[13,14,15]
+; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm10 = mem[2,2,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm10[2],ymm1[3,4],ymm10[5],ymm1[6,7,8,9],ymm10[10],ymm1[11,12],ymm10[13],ymm1[14,15]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm10 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm10 = ymm10[4],mem[4],ymm10[5],mem[5],ymm10[6],mem[6],ymm10[7],mem[7],ymm10[12],mem[12],ymm10[13],mem[13],ymm10[14],mem[14],ymm10[15],mem[15]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm10[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm0, %ymm12
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa %ymm3, %ymm15
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm11, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm1, %ymm12, %ymm1
 ; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    # ymm3 = mem[2,2,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm9, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm8[2,2,2,2,6,6,6,6]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7,8,9],ymm3[10],ymm1[11,12],ymm3[13],ymm1[14,15]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm3, %ymm3
-; AVX2-FAST-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    # ymm11 = mem[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm11[0],ymm3[1],ymm11[2,3],ymm3[4],ymm11[5,6,7,8],ymm3[9],ymm11[10,11],ymm3[12],ymm11[13,14,15]
+; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm3 = ymm3[4],mem[4],ymm3[5],mem[5],ymm3[6],mem[6],ymm3[7],mem[7],ymm3[12],mem[12],ymm3[13],mem[13],ymm3[14],mem[14],ymm3[15],mem[15]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm3[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm0, %ymm12
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm15, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm1, %ymm12, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31>
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7,8,9],ymm3[10],ymm1[11,12],ymm3[13],ymm1[14,15]
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm13, %ymm1
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm12 = ymm2[3,3,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm12[2],ymm1[3,4],ymm12[5],ymm1[6,7,8,9],ymm12[10],ymm1[11,12],ymm12[13],ymm1[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,7,3,3,7,7,6,7]
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm3
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm7[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0,1],ymm14[2],ymm3[3,4],ymm14[5],ymm3[6,7,8,9],ymm14[10],ymm3[11,12],ymm14[13],ymm3[14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm2, %ymm14
-; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm3, %ymm14, %ymm3
-; AVX2-FAST-NEXT:    vmovdqu (%rsp), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm2, %ymm12
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm12, %ymm1
+; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm5, %ymm12
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm13 = ymm6[3,3,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0,1],ymm13[2],ymm12[3,4],ymm13[5],ymm12[6,7,8,9],ymm13[10],ymm12[11,12],ymm13[13],ymm12[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,2,2,3]
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm2, %ymm13
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm4, %ymm0
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm10[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm14[2],ymm0[3,4],ymm14[5],ymm0[6,7,8,9],ymm14[10],ymm0[11,12],ymm14[13],ymm0[14,15]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm13 = ymm11[3,3,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm13[2],ymm0[3,4],ymm13[5],ymm0[6,7,8,9],ymm13[10],ymm0[11,12],ymm13[13],ymm0[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermd %ymm9, %ymm2, %ymm14
-; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm0, %ymm14, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm6, %ymm10, %ymm11
+; AVX2-FAST-NEXT:    vpermd (%rsp), %ymm2, %ymm13 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm0, %ymm13, %ymm0
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm13 = ymm15[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = [4,6,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm6, %ymm13
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm2[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm11 = ymm14[0,1,2],ymm11[3],ymm14[4,5],ymm11[6],ymm14[7,8,9,10],ymm11[11],ymm14[12,13],ymm11[14],ymm14[15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm5, %ymm14
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm13[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0],ymm14[1],ymm15[2,3],ymm14[4],ymm15[5,6,7,8],ymm14[9],ymm15[10,11],ymm14[12],ymm15[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm11, %ymm14, %ymm11
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm11, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm14 = ymm4[4],ymm2[4],ymm4[5],ymm2[5],ymm4[6],ymm2[6],ymm4[7],ymm2[7],ymm4[12],ymm2[12],ymm4[13],ymm2[13],ymm4[14],ymm2[14],ymm4[15],ymm2[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [2,3,6,7,6,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm14, %ymm7, %ymm14
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm13, %ymm14, %ymm13
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm13, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm10[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm6, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm6, %ymm5, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm14[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm11[0,1,2],ymm1[3],ymm11[4,5],ymm1[6],ymm11[7,8,9,10],ymm1[11],ymm11[12,13],ymm1[14],ymm11[15]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm4, %ymm8, %ymm11
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm12[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm11 = ymm15[0],ymm11[1],ymm15[2,3],ymm11[4],ymm15[5,6,7,8],ymm11[9],ymm15[10,11],ymm11[12],ymm15[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm1, %ymm11, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa %ymm7, %ymm12
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm6, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3],ymm3[4,5],ymm1[6],ymm3[7,8,9,10],ymm1[11],ymm3[12,13],ymm1[14],ymm3[15]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm7[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm11 = ymm15[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm11[0],ymm3[1],ymm11[2,3],ymm3[4],ymm11[5,6,7,8],ymm3[9],ymm11[10,11],ymm3[12],ymm11[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm10 = ymm5[4],ymm15[4],ymm5[5],ymm15[5],ymm5[6],ymm15[6],ymm5[7],ymm15[7],ymm5[12],ymm15[12],ymm5[13],ymm15[13],ymm5[14],ymm15[14],ymm5[15],ymm15[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm10[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vpermd %ymm10, %ymm7, %ymm10
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm1, %ymm10, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm12, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm6, %ymm1
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm3[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm7, %ymm3
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <3,u,u,u,4,u,u,4>
-; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm10, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpermd %ymm14, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa %ymm3, %ymm2
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm5, %ymm3
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm4, %ymm2
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm3, %ymm2, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,3,u,u,u,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm6, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm4 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm8, %ymm5
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT:    vpermd %ymm15, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm7, %ymm6
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm5, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm0, %ymm3, %ymm0
-; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm1, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm2, %ymm5, %ymm2
+; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm0 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vmovdqa %ymm3, %ymm10
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm3, %ymm3
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm6, %ymm9
+; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm3, %ymm9, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,3,u,u,u,4,u,u>
+; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm9, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
+; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm2, %ymm7
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm8, %ymm7, %ymm7
+; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm9, %ymm6
+; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm15, %ymm5
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm6, %ymm5, %ymm5
+; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm4 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm4, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm0, %ymm7, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm5, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm3, %ymm2, %ymm2
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,3,u,u,u,4,u>
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm3 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm4, %ymm4
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm4 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
 ; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm4, %ymm6, %ymm4
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm7, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vmovdqu (%rsp), %ymm7 # 32-byte Reload
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm7, %ymm7
 ; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm6, %ymm7, %ymm5
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <3,u,u,3,u,u,u,4>
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm6 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm3, %ymm6, %ymm3
 ; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm6 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm4, %ymm6, %ymm4
-; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm6 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vpermd (%rsp), %ymm8, %ymm6 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm0, %ymm3, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm1, %ymm4, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm6, %ymm2, %ymm5, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm6[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,1,1,3,4,5,5,7]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm5[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,4,2,1,6,5,6,5]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm1, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm8[1,1,2,2,4,5,6,7,9,9,10,10,12,13,14,15]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,4,5,4,5,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm9, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,1,4,5,4,5,5,7]
+; AVX2-FAST-NEXT:    vpermd %ymm9, %ymm10, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[2,2,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [5,6,2,3,6,7,5,6]
-; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,5,2,2,6,6,6,6]
-; AVX2-FAST-NEXT:    vpermd %ymm9, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm3[4],ymm6[4],ymm3[5],ymm6[5],ymm3[6],ymm6[6],ymm3[7],ymm6[7],ymm3[12],ymm6[12],ymm3[13],ymm6[13],ymm3[14],ymm6[14],ymm3[15],ymm6[15]
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm10, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[2,2,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7,8,9],ymm3[10],ymm2[11,12],ymm3[13],ymm2[14,15]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm2, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [5,6,2,3,6,7,5,6]
+; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,5,2,2,6,6,6,6]
+; AVX2-FAST-NEXT:    vpermd %ymm9, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,6,3,3,7,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm4[4],ymm5[4],ymm4[5],ymm5[5],ymm4[6],ymm5[6],ymm4[7],ymm5[7],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15]
+; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,3,6,7,6,7,7,7]
+; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[3,3,3,3,7,7,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vpbroadcastd 124(%r8), %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm8[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[3,3,3,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [6,7,3,3,7,7,6,7]

 ; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa %xmm1, %xmm4
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[0,0,1,1]
 ; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm3
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
 ; AVX2-FAST-NEXT:    vmovdqa %xmm3, %xmm15
 ; AVX2-FAST-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, %xmm11
+; AVX2-FAST-NEXT:    vmovdqa %xmm0, %xmm14
 ; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
 ; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm2, %xmm2
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %xmm2
 ; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
 ; AVX2-FAST-NEXT:    vpshufb %xmm4, %xmm1, %xmm1
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
-; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm14
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %xmm6
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm6[0],xmm14[0],xmm6[1],xmm14[1],xmm6[2],xmm14[2],xmm6[3],xmm14[3]
-; AVX2-FAST-NEXT:    vmovdqa %xmm6, %xmm10
-; AVX2-FAST-NEXT:    vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm11
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %xmm10
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3]
+; AVX2-FAST-NEXT:    vmovdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm2, %xmm2
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,1,3]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 64(%rsi), %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 64(%rdi), %xmm2
 ; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill

 ; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm1, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm1, {{[-0-9]+}}(%r{{[sb]}}p), %ymm2, %ymm2 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm7, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
-; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm11, %xmm1
+; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm14, %xmm1
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm15[1,1,2,2]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0],xmm1[1],xmm2[2,3],xmm1[4],xmm2[5,6],xmm1[7]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,u,u,6,7,u,u,u,u,8,9,u,u,u,u>
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm3, %xmm3
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm15[1,1,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm14[1,1,2,3]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1],xmm3[2],xmm4[3,4],xmm3[5],xmm4[6,7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm14, %xmm3
+; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm11, %xmm3
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm6 = xmm10[1,1,2,2]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm6[0],xmm3[1],xmm6[2,3],xmm3[4],xmm6[5,6],xmm3[7]
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm14, %xmm6
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm15, %xmm6
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm7 = xmm11[1,1,2,3]
 ; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm6 = xmm7[0,1],xmm6[2],xmm7[3,4],xmm6[5],xmm7[6,7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,1,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm3, %ymm6, %ymm3
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm5, %xmm6

 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm1, %ymm4, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm3, %ymm6, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm5, %ymm12, %ymm13
 ; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm0, %ymm2, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm15[4],xmm0[5],xmm15[5],xmm0[6],xmm15[6],xmm0[7],xmm15[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm14[4],xmm0[5],xmm14[5],xmm0[6],xmm14[6],xmm0[7],xmm14[7]
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
 ; AVX2-FAST-NEXT:    # xmm1 = xmm1[4],mem[4],xmm1[5],mem[5],xmm1[6],mem[6],xmm1[7],mem[7]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
 ; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,3]
 ; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,2,3,3,4,5,6,7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm14[4],xmm11[4],xmm14[5],xmm11[5],xmm14[6],xmm11[6],xmm14[7],xmm11[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm15[4],xmm11[4],xmm15[5],xmm11[5],xmm15[6],xmm11[6],xmm15[7],xmm11[7]
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm4, %xmm4 # 16-byte Folded Reload
 ; AVX2-FAST-NEXT:    # xmm4 = xmm4[4],mem[4],xmm4[5],mem[5],xmm4[6],mem[6],xmm4[7],mem[7]
 ; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,3]
 ; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm4[0,2,3,3,4,5,6,7]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm1, %ymm4, %ymm1

 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 800(%rax)
 ; AVX2-FAST-NEXT:    addq $1240, %rsp # imm = 0x4D8
 ; AVX2-FAST-NEXT:    vzeroupper
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX2-FAST-PERLANE-LABEL: store_i16_stride7_vf64:
 ; AVX2-FAST-PERLANE:       # %bb.0:
-; AVX2-FAST-PERLANE-NEXT:    subq $1544, %rsp # imm = 0x608
+; AVX2-FAST-PERLANE-NEXT:    subq $1528, %rsp # imm = 0x5F8
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rdi), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rsi), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, (%rsp) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rdx), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rcx), %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%r8), %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rdx), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rcx), %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%r8), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <3,u,u,u,4,u,u,4>
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm12, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = [0,1,0,1,14,15,14,15,8,9,10,11,12,13,14,15,16,17,16,17,30,31,30,31,24,25,26,27,28,29,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm1, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm1, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm5, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,3,u,u,u,4,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm8, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm8, %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,1,0,1,0,1,0,1,14,15,14,15,14,15,14,15,16,17,16,17,16,17,16,17,30,31,30,31,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm5, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm15, %ymm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm4, %ymm5, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm0, %ymm4, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm5, %ymm6, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm0, %ymm5, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,3,u,u,u,4,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm6, %ymm1, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm0, %ymm4, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%r9), %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm4, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm4, %ymm1, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm0, %ymm5, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%r9), %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm6 = ymm5[0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm5, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm6, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rax), %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rax), %ymm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <3,u,u,3,u,u,u,4>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm5, %ymm9, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm6, %ymm9, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = [0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm12, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm1, %ymm12, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm0, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm1, %ymm7, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm1, %ymm7, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rdi), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm12, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rsi), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm0, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm7, %ymm12, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm7, %ymm12, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm8, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm1, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm7, %ymm12, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdx), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rcx), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm8, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm1, %ymm14
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm12, %ymm14, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rdx), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm8, %ymm8
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rcx), %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm0, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm8, %ymm10, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm6, %ymm7, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm3, %ymm12, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm2, %ymm8, %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,3,u,u,u,4,u>
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm12, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm6, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm4, %ymm7, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm2, %ymm12, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm3, %ymm8, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,3,u,u,u,4,u>
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm11, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = [0,1,2,3,4,5,6,7,0,1,0,1,14,15,14,15,16,17,18,19,20,21,22,23,16,17,16,17,30,31,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm4, %ymm8
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u,u,u,u,u,u,u,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm12, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm12, %ymm13
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm6, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm11, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm4, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm8, %ymm12, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%r8), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm13, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%r9), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm3, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%r8), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm11, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%r9), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm2, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm12, %ymm11, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm9, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm9, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm7, %ymm11, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rax), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm9, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rax), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm9, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm8, %ymm11, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rax), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm3, %ymm9, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rax), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm9, %ymm9
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm10, %ymm9, %ymm9
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm0, %ymm7, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm8, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm2, %ymm9, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm3, %ymm9, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu (%rsp), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[1,1,1,1,5,5,5,5]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu (%rsp), %ymm8 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm8[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[0,1,1,3,4,5,5,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm15[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm3[0,1],ymm1[2],ymm3[3,4],ymm1[5],ymm3[6,7,8,9],ymm1[10],ymm3[11,12],ymm1[13],ymm3[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm4[1,1,2,2,4,5,6,7,9,9,10,10,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm5[1,1,2,2,4,5,6,7,9,9,10,10,12,13,14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[2,2,2,2,6,6,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[2,2,2,2,6,6,6,6]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[12],ymm8[12],ymm7[13],ymm8[13],ymm7[14],ymm8[14],ymm7[15],ymm8[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm1[0,1,2,3,6,7,4,5,u,u,u,u,u,u,u,u,16,17,18,19,22,23,20,21,22,23,20,21,28,29,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm9[1,2,2,3,5,6,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[0,1,2,2,4,5,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm0, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm0, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm6[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm0, %ymm3, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8,9,10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm4[4],ymm15[4],ymm4[5],ymm15[5],ymm4[6],ymm15[6],ymm4[7],ymm15[7],ymm4[12],ymm15[12],ymm4[13],ymm15[13],ymm4[14],ymm15[14],ymm4[15],ymm15[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,14,15,12,13,14,15,12,13,16,17,18,19,24,25,26,27,30,31,28,29,30,31,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[8,9,10,11,14,15,12,13,u,u,u,u,u,u,u,u,24,25,26,27,30,31,28,29,30,31,28,29,30,31,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd 124(%r8), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u,255,255,255,255,255,255,255,255,0,0,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm4[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
+; AVX2-FAST-PERLANE-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm5[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[3,3,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u,255,255,255,255,255,255,255,255,255,255,0,0,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[2,3,3,3,6,7,7,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, (%rsp) # 16-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill

 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm2, %ymm0, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm5, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm7, %ymm4, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm6, %ymm3, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm0, %ymm4
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm5 = mem[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7,8,9],ymm5[10],ymm4[11,12],ymm5[13],ymm4[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm5 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm6 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm5 = ymm6[0,1],ymm5[2],ymm6[3,4],ymm5[5],ymm6[6,7,8,9],ymm5[10],ymm6[11,12],ymm5[13],ymm6[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm4, %ymm5, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = <255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm1, %ymm4, %ymm5, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm0, %ymm5
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $85, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm6 = mem[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7,8,9],ymm6[10],ymm5[11,12],ymm6[13],ymm5[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm7 = mem[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm14[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm7 = ymm12[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1],ymm6[2],ymm7[3,4],ymm6[5],ymm7[6,7,8,9],ymm6[10],ymm7[11,12],ymm6[13],ymm7[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm5, %ymm6, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm1, %ymm5, %ymm6, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm12, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm0, %ymm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm7 = ymm13[1,1,1,1,5,5,5,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm6[0,1],ymm7[2],ymm6[3,4],ymm7[5],ymm6[6,7,8,9],ymm7[10],ymm6[11,12],ymm7[13],ymm6[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm7 = ymm1[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm0[0,1,1,3,4,5,5,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm7 = ymm15[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm8 = ymm2[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm8[0,1],ymm7[2],ymm8[3,4],ymm7[5],ymm8[6,7,8,9],ymm7[10],ymm8[11,12],ymm7[13],ymm8[14,15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm6, %ymm7, %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm1, %ymm6, %ymm7, %ymm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm14, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm1, %ymm8
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm9 = ymm4[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1,2],ymm9[3],ymm8[4,5],ymm9[6],ymm8[7,8,9,10],ymm9[11],ymm8[12,13],ymm9[14],ymm8[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm9 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm5, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm6[0,0,2,1,4,4,6,5]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm3[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1,2],ymm11[3],ymm9[4,5],ymm11[6],ymm9[7,8,9,10],ymm11[11],ymm9[12,13],ymm11[14],ymm9[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm11 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm9, %ymm11, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm3, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm2[0,0,2,1,4,4,6,5]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm0[0,0,2,1,4,4,6,5]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm7[0,1,2],ymm11[3],ymm7[4,5],ymm11[6],ymm7[7,8,9,10],ymm11[11],ymm7[12,13],ymm11[14],ymm7[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,3,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm11 = mem[0,1,1,3,4,5,5,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm7, %ymm11, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm8 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm15, %ymm7, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm6, %ymm7, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm7, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm7, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm8 = mem[2,2,2,2,6,6,6,6]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm7[0,1],ymm8[2],ymm7[3,4],ymm8[5],ymm7[6,7,8,9],ymm8[10],ymm7[11,12],ymm8[13],ymm7[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm8, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm9 = mem[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm8 = ymm9[0],ymm8[1],ymm9[2,3],ymm8[4],ymm9[5,6,7,8],ymm8[9],ymm9[10,11],ymm8[12],ymm9[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm8 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm8 = ymm8[4],mem[4],ymm8[5],mem[5],ymm8[6],mem[6],ymm8[7],mem[7],ymm8[12],mem[12],ymm8[13],mem[13],ymm8[14],mem[14],ymm8[15],mem[15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm7, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm9 = mem[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7,8,9],ymm9[10],ymm8[11,12],ymm9[13],ymm8[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm9, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm10 = mem[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm9 = ymm10[0],ymm9[1],ymm10[2,3],ymm9[4],ymm10[5,6,7,8],ymm9[9],ymm10[10,11],ymm9[12],ymm10[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <0,1,2,3,6,7,4,5,u,u,u,u,u,u,u,u,16,17,18,19,22,23,20,21,22,23,20,21,28,29,30,31>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm8, %ymm9
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm8, %ymm9, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm1, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm10 = ymm0[2,2,2,2,6,6,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u,u,u,u,u,0,0,0,0,255,255,255,255,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm7, %ymm9, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm14, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm6, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm10 = ymm12[2,2,2,2,6,6,6,6]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1],ymm10[2],ymm9[3,4],ymm10[5],ymm9[6,7,8,9],ymm10[10],ymm9[11,12],ymm10[13],ymm9[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm12, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm11 = ymm13[2,2,2,2,6,6,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0],ymm10[1],ymm11[2,3],ymm10[4],ymm11[5,6,7,8],ymm10[9],ymm11[10,11],ymm10[12],ymm11[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm6, %ymm10 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm10 = ymm6[4],mem[4],ymm6[5],mem[5],ymm6[6],mem[6],ymm6[7],mem[7],ymm6[12],mem[12],ymm6[13],mem[13],ymm6[14],mem[14],ymm6[15],mem[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm9, %ymm10, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm14, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm14, %ymm15
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm12 = ymm4[1,2,2,3,5,6,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0,1],ymm11[2],ymm12[3,4],ymm11[5],ymm12[6,7,8,9],ymm11[10],ymm12[11,12],ymm11[13],ymm12[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm12 = ymm7[0,1,2,2,4,5,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm5, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm6[1,2,2,3,5,6,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm12 = ymm14[0,1],ymm12[2],ymm14[3,4],ymm12[5],ymm14[6,7,8,9],ymm12[10],ymm14[11,12],ymm12[13],ymm14[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm4[0,1,2,2,4,5,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm12, %ymm14, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm3, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm2[1,2,2,3,5,6,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm10 = ymm14[0,1],ymm10[2],ymm14[3,4],ymm10[5],ymm14[6,7,8,9],ymm10[10],ymm14[11,12],ymm10[13],ymm14[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm11, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm10, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm9, %ymm11, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm15, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm12 = ymm2[2,2,2,2,6,6,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm11 = ymm11[0,1],ymm12[2],ymm11[3,4],ymm12[5],ymm11[6,7,8,9],ymm12[10],ymm11[11,12],ymm12[13],ymm11[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm13, %ymm12 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm12 = ymm13[4],mem[4],ymm13[5],mem[5],ymm13[6],mem[6],ymm13[7],mem[7],ymm13[12],mem[12],ymm13[13],mem[13],ymm13[14],mem[14],ymm13[15],mem[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm6, %ymm12, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm11, %ymm13, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,8,9,u,u,u,u,u,u,u,u,22,23,u,u,u,u,24,25,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm1, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm4[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm14 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7,8,9],ymm14[10],ymm15[11,12],ymm14[13],ymm15[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm6[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm14, %ymm15, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm5, %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm15[2],ymm1[3,4],ymm15[5],ymm1[6,7,8,9],ymm15[10],ymm1[11,12],ymm15[13],ymm1[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm5[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm15, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm14 = ymm2[0,1,2,2,4,5,6,6]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm10, %ymm14, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm2, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm0[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm13 = ymm15[0,1],ymm13[2],ymm15[3,4],ymm13[5],ymm15[6,7,8,9],ymm13[10],ymm15[11,12],ymm13[13],ymm15[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[0,1,2,2,4,5,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm13, %ymm15, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = [255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm0 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm8, %ymm12, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm9, %ymm10, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29>
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm0, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm11 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1,2],ymm10[3],ymm11[4,5],ymm10[6],ymm11[7,8,9,10],ymm10[11],ymm11[12,13],ymm10[14],ymm11[15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm11, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm12 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0],ymm11[1],ymm12[2,3],ymm11[4],ymm12[5,6,7,8],ymm11[9],ymm12[10,11],ymm11[12],ymm12[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm7, %ymm14, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm9, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm11, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm9 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm9 = ymm4[4],mem[4],ymm4[5],mem[5],ymm4[6],mem[6],ymm4[7],mem[7],ymm4[12],mem[12],ymm4[13],mem[13],ymm4[14],mem[14],ymm4[15],mem[15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,14,15,12,13,14,15,12,13,16,17,18,19,24,25,26,27,30,31,28,29,30,31,28,29>
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm4, %ymm8, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <8,9,10,11,14,15,12,13,u,u,u,u,u,u,u,u,24,25,26,27,30,31,28,29,30,31,28,29,30,31,28,29>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm9, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm9 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm9 = ymm9[4],mem[4],ymm9[5],mem[5],ymm9[6],mem[6],ymm9[7],mem[7],ymm9[12],mem[12],ymm9[13],mem[13],ymm9[14],mem[14],ymm9[15],mem[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm4, %ymm10, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u,255,255,255,255,0,0,0,0,u,u,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm1, %ymm10, %ymm11, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm9, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm11, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm10, %ymm9, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm4, %ymm12, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{[-0-9]+}}(%r{{[sb]}}p), %ymm4, %ymm11 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm11 = ymm4[4],mem[4],ymm4[5],mem[5],ymm4[6],mem[6],ymm4[7],mem[7],ymm4[12],mem[12],ymm4[13],mem[13],ymm4[14],mem[14],ymm4[15],mem[15]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm11, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm13, %ymm14
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm12 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm11 = ymm12[0,1,2],ymm11[3],ymm12[4,5],ymm11[6],ymm12[7,8,9,10],ymm11[11],ymm12[12,13],ymm11[14],ymm12[15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm12, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm13 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm12 = ymm13[0],ymm12[1],ymm13[2,3],ymm12[4],ymm13[5,6,7,8],ymm12[9],ymm13[10,11],ymm12[12],ymm13[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm1, %ymm11, %ymm12, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm14, %ymm12, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm10, %ymm11, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31>
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm4, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm13 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm12 = ymm13[0,1,2],ymm12[3],ymm13[4,5],ymm12[6],ymm13[7,8,9,10],ymm12[11],ymm13[12,13],ymm12[14],ymm13[15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm13, %ymm13
-; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # ymm14 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm13 = ymm14[0],ymm13[1],ymm14[2,3],ymm13[4],ymm14[5,6,7,8],ymm13[9],ymm14[10,11],ymm13[12],ymm14[13,14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,1,3,3]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm1, %ymm12, %ymm13, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm15, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm12 = ymm12[0,1],ymm13[2],ymm12[3,4],ymm13[5],ymm12[6,7,8,9],ymm13[10],ymm12[11,12],ymm13[13],ymm12[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm13 = ymm6[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm4, %ymm13
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm15 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm14 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7,8,9],ymm15[10],ymm14[11,12],ymm15[13],ymm14[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm7[2,3,3,3,6,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0,u,u,u,u,u,u,u,u,255,255,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm14, %ymm15, %ymm14
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm5, %ymm15
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm15[0,1],ymm1[2],ymm15[3,4],ymm1[5],ymm15[6,7,8,9],ymm1[10],ymm15[11,12],ymm1[13],ymm15[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm4[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm13 = ymm13[0,1],ymm15[2],ymm13[3,4],ymm15[5],ymm13[6,7,8,9],ymm15[10],ymm13[11,12],ymm15[13],ymm13[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm5[2,3,3,3,6,7,7,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm1, %ymm15, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm3, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm13, %ymm15, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm2, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vpshufd $255, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
 ; AVX2-FAST-PERLANE-NEXT:    # ymm15 = mem[3,3,3,3,7,7,7,7]
-; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm13 = ymm13[0,1],ymm15[2],ymm13[3,4],ymm15[5],ymm13[6,7,8,9],ymm15[10],ymm13[11,12],ymm15[13],ymm13[14,15]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,2,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm2[2,3,3,3,6,7,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm11 = ymm11[0,1],ymm15[2],ymm11[3,4],ymm15[5],ymm11[6,7,8,9],ymm15[10],ymm11[11,12],ymm15[13],ymm11[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[2,3,3,3,6,7,7,7]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,1,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm13, %ymm15, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm10, %ymm14, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm11, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm12, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm11, %ymm15, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = [0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0,255,255,255,255,255,255,255,255,0,0,0,0,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm8, %ymm12, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm9, %ymm13, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm10, %ymm11, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, 544(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, 320(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, 96(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 640(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm9, 608(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, 640(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 608(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 576(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 416(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm8, 384(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm9, 416(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 384(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 352(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, 192(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 160(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm8, 192(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm7, 160(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 128(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 768(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 736(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 704(%rax)

 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 864(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 832(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 800(%rax)
-; AVX2-FAST-PERLANE-NEXT:    addq $1544, %rsp # imm = 0x608
+; AVX2-FAST-PERLANE-NEXT:    addq $1528, %rsp # imm = 0x5F8
 ; AVX2-FAST-PERLANE-NEXT:    vzeroupper
 ; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-SLOW-LABEL: store_i16_stride7_vf64:
 ; AVX512F-SLOW:       # %bb.0:
-; AVX512F-SLOW-NEXT:    subq $2440, %rsp # imm = 0x988
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rcx), %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdx), %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdi), %ymm7
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rsi), %ymm15
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm20
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm2, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm24
-; AVX512F-SLOW-NEXT:    vporq %ymm0, %ymm1, %ymm19
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm15, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm7, %ymm1
+; AVX512F-SLOW-NEXT:    subq $2504, %rsp # imm = 0x9C8
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rcx), %ymm14
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdx), %ymm9
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rsi), %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdi), %ymm4
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm14, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm9, %ymm1
 ; AVX512F-SLOW-NEXT:    vporq %ymm0, %ymm1, %ymm16
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%r9), %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm0, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm2, %ymm8
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm31
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%r8), %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm10, %ymm0, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm18
-; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rcx), %ymm6
-; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm6, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdx), %ymm5
-; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm5, %ymm2
-; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rsi), %ymm4
-; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm4, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdi), %ymm3
-; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm3, %ymm0
-; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm17
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm4, %ymm2
+; AVX512F-SLOW-NEXT:    vporq %ymm1, %ymm2, %ymm19
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%r9), %ymm2
+; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm3, %ymm5
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm22
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%r8), %ymm3
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm3, %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm3, %ymm18
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm10, %ymm1, %ymm1
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rcx), %ymm10
+; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm10, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdx), %ymm7
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm7, %ymm2
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm1, %ymm1
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rsi), %ymm2
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm24
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdi), %ymm11
+; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm11, %ymm2
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm5, %ymm1, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm5, %ymm8
+; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm2, %ymm2
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm2, %ymm2
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm2, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm21
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm2, %ymm2
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm5
+; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm5, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm12
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm12, %ymm3
+; AVX512F-SLOW-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm6
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm6, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm3, %ymm1
 ; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm2, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm14
-; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm14, %ymm9
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm9, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm9
-; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm9, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm12
-; AVX512F-SLOW-NEXT:    vpshufb %ymm13, %ymm12, %ymm13
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm13, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %ymm13
-; AVX512F-SLOW-NEXT:    vpshufb %ymm10, %ymm13, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %ymm1
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm1, %ymm11
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm8, %ymm22
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm11, %ymm0
+; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm13, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %ymm2
+; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm2, %ymm15
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm8, %ymm20
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm15, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vprold $16, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm13[1,2,2,3,5,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm11[0,1],ymm0[2],ymm11[3,4],ymm0[5],ymm11[6,7,8,9],ymm0[10],ymm11[11,12],ymm0[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm1[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[2,2,2,3,6,6,6,7]
+; AVX512F-SLOW-NEXT:    vprold $16, %ymm2, %ymm0
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm13[1,2,2,3,5,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3,4],ymm0[5],ymm15[6,7,8,9],ymm0[10],ymm15[11,12],ymm0[13],ymm15[14,15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm15 = ymm2[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[2,2,2,3,6,6,6,7]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm13[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm11[0,1],ymm8[2],ymm11[3,4],ymm8[5],ymm11[6,7,8,9],ymm8[10],ymm11[11,12],ymm8[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [2,1,3,2,10,10,10,11]
-; AVX512F-SLOW-NEXT:    vpermi2q %zmm8, %zmm0, %zmm10
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm20, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm10, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm8, %ymm17
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm24[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm15[0,1],ymm8[2],ymm15[3,4],ymm8[5],ymm15[6,7,8,9],ymm8[10],ymm15[11,12],ymm8[13],ymm15[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [2,1,3,2,10,10,10,11]
+; AVX512F-SLOW-NEXT:    vpermi2q %zmm8, %zmm0, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm14, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm21
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm9[2,2,2,2,6,6,6,6]
 ; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm8[2],ymm0[3,4],ymm8[5],ymm0[6,7,8,9],ymm8[10],ymm0[11,12],ymm8[13],ymm0[14,15]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm10[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm24[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1],ymm8[2,3],ymm0[4],ymm8[5,6,7,8],ymm0[9],ymm8[10,11],ymm0[12],ymm8[13,14,15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm9[4],ymm14[4],ymm9[5],ymm14[5],ymm9[6],ymm14[6],ymm9[7],ymm14[7],ymm9[12],ymm14[12],ymm9[13],ymm14[13],ymm9[14],ymm14[14],ymm9[15],ymm14[15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,3,3,6,7,7,7]
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm15[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm7[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1],ymm8[2,3],ymm0[4],ymm8[5,6,7,8],ymm0[9],ymm8[10,11],ymm0[12],ymm8[13,14,15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm15[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm7[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1,2],ymm0[3],ymm8[4,5],ymm0[6],ymm8[7,8,9,10],ymm0[11],ymm8[12,13],ymm0[14],ymm8[15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm17, %ymm15
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm4[4],ymm15[4],ymm4[5],ymm15[5],ymm4[6],ymm15[6],ymm4[7],ymm15[7],ymm4[12],ymm15[12],ymm4[13],ymm15[13],ymm4[14],ymm15[14],ymm4[15],ymm15[15]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm8[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,3,3,4,6,7,7]
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-SLOW-NEXT:    vmovdqa 96(%r9), %ymm0
 ; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm8 = ymm0[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[3,3,3,3]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm8[3,3,3,3]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm0[2,1,2,3,6,5,6,7]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm8 = ymm8[0,0,3,3,4,5,6,7,8,8,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,2]
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm24[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm10[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm11[2],ymm8[3,4],ymm11[5],ymm8[6,7,8,9],ymm11[10],ymm8[11,12],ymm11[13],ymm8[14,15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm8[2,2,2,2]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm9[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm14[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7,8,9],ymm9[10],ymm8[11,12],ymm9[13],ymm8[14,15]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm15[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm11[0,1],ymm7[2],ymm11[3,4],ymm7[5],ymm11[6,7,8,9],ymm7[10],ymm11[11,12],ymm7[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vprold $16, %ymm0, %ymm11
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm22, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb %ymm10, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm19, %zmm8
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm16, %zmm7
-; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm7
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm4[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm15[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm9 = ymm15[0,1],ymm9[2],ymm15[3,4],ymm9[5],ymm15[6,7,8,9],ymm9[10],ymm15[11,12],ymm9[13],ymm15[14,15]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vprold $16, %ymm0, %ymm15
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm20, %ymm1
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm16, %zmm8
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm19, %zmm9
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm9
 ; AVX512F-SLOW-NEXT:    vmovdqa 96(%r8), %ymm8
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm8[1,2,2,3,5,6,6,7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm8[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm8[1,2,2,3,5,6,6,7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm16 = ymm8[0,0,2,1,4,4,6,5]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm8[u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm8[14,15],zero,zero,ymm8[u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm8[16,17],zero,zero,ymm8[u,u],zero,zero
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm19 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm19, %ymm7, %ymm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm17 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm17, %ymm9, %ymm8
 ; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm8, %ymm0
-; AVX512F-SLOW-NEXT:    vextracti64x4 $1, %zmm7, %ymm7
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm15[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm7, %ymm8
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm11[2,2,2,2]
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm8, %ymm7
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm7
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm7[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vextracti64x4 $1, %zmm9, %ymm8
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm16[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm8, %ymm9
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm15[2,2,2,2]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm9, %ymm8
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm8
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm8[4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512F-SLOW-NEXT:    vmovdqa 96(%rax), %ymm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm0[0,1,2,2,4,5,6,6]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm0[2,3,3,3,6,7,7,7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm0[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpandnq %ymm7, %ymm19, %ymm7
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm0[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm0[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm0[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpandnq %ymm8, %ymm17, %ymm8
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
 ; AVX512F-SLOW-NEXT:    vpbroadcastd 72(%rax), %ymm0
-; AVX512F-SLOW-NEXT:    vpandn %ymm0, %ymm10, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rax), %ymm7
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm7, %ymm11
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%r9), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%r8), %xmm15
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm15[4],xmm0[4],xmm15[5],xmm0[5],xmm15[6],xmm0[6],xmm15[7],xmm0[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm16
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm11, %xmm11
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpbroadcastd 8(%rax), %ymm11
-; AVX512F-SLOW-NEXT:    vpandnq %ymm11, %ymm10, %ymm20
-; AVX512F-SLOW-NEXT:    vmovdqa (%rax), %ymm11
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm11, %ymm10
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm11, %ymm30
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm20, %zmm10
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm2[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm14[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm11[0,1],ymm10[2],ymm11[3,4],ymm10[5],ymm11[6,7,8,9],ymm10[10],ymm11[11,12],ymm10[13],ymm11[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm9[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm12[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm10 = ymm10[0,1],ymm11[2],ymm10[3,4],ymm11[5],ymm10[6,7,8,9],ymm11[10],ymm10[11,12],ymm11[13],ymm10[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm10, %ymm26
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %ymm11
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm11[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpandnq %ymm10, %ymm19, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb %ymm8, %ymm11, %ymm8
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm11, %ymm27
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm8, %zmm8
+; AVX512F-SLOW-NEXT:    vpandn %ymm0, %ymm14, %ymm8
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rax), %ymm4
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm4, %ymm9
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm8, %zmm8
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm13[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm1[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm8[3],ymm1[4,5],ymm8[6],ymm1[7,8,9,10],ymm8[11],ymm1[12,13],ymm8[14],ymm1[15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm25
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm17, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb %ymm10, %ymm2, %ymm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm14[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm8[2],ymm1[3,4],ymm8[5],ymm1[6,7,8,9],ymm8[10],ymm1[11,12],ymm8[13],ymm1[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm14[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6,7,8],ymm2[9],ymm1[10,11],ymm2[12],ymm1[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm9[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%r9), %xmm8
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%r8), %xmm9
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm15 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm15, %xmm15
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm20
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpbroadcastd 8(%rax), %ymm15
+; AVX512F-SLOW-NEXT:    vpandnq %ymm15, %ymm14, %ymm16
+; AVX512F-SLOW-NEXT:    vmovdqa (%rax), %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm0, %ymm14
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm16, %zmm14
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm5[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm12[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm15[0,1],ymm14[2],ymm15[3,4],ymm14[5],ymm15[6,7,8,9],ymm14[10],ymm15[11,12],ymm14[13],ymm15[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm6[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm3[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm15 = ymm14[0,1],ymm15[2],ymm14[3,4],ymm15[5],ymm14[6,7,8,9],ymm15[10],ymm14[11,12],ymm15[13],ymm14[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm15, %ymm23
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %ymm0
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm0[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpandnq %ymm14, %ymm17, %ymm14
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm0, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm26
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm1, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm13[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm15 = ymm2[0,1,2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7,8,9,10],ymm1[11],ymm2[12,13],ymm1[14],ymm2[15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm15, %ymm19
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm21, %ymm14
+; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm5, %ymm1
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm12[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm9[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5],ymm2[6],ymm1[7,8,9,10],ymm2[11],ymm1[12,13],ymm2[14],ymm1[15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm4[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm15 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm12 = ymm12[4],ymm5[4],ymm12[5],ymm5[5],ymm12[6],ymm5[6],ymm12[7],ymm5[7],ymm12[12],ymm5[12],ymm12[13],ymm5[13],ymm12[14],ymm5[14],ymm12[15],ymm5[15]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm3[4],ymm6[4],ymm3[5],ymm6[5],ymm3[6],ymm6[6],ymm3[7],ymm6[7],ymm3[12],ymm6[12],ymm3[13],ymm6[13],ymm3[14],ymm6[14],ymm3[15],ymm6[15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm24, %ymm3
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm3[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm4[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm3[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm6[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm11[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm11[4],ymm3[4],ymm11[5],ymm3[5],ymm11[6],ymm3[6],ymm11[7],ymm3[7],ymm11[12],ymm3[12],ymm11[13],ymm3[13],ymm11[14],ymm3[14],ymm11[15],ymm3[15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm10[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm14 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm14, %ymm28
-; AVX512F-SLOW-NEXT:    vpshufb %ymm10, %ymm6, %ymm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm31, %ymm8
-; AVX512F-SLOW-NEXT:    vprold $16, %ymm31, %ymm1
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm10, %ymm1
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm22, %ymm0
+; AVX512F-SLOW-NEXT:    vprold $16, %ymm22, %ymm1
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm18[1,2,2,3,5,6,6,7]
 ; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm8[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm31, %ymm9
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm0[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm18[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm8[3],ymm2[4,5],ymm8[6],ymm2[7,8,9,10],ymm8[11],ymm2[12,13],ymm8[14],ymm2[15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm12 = [2,2,3,3,10,9,11,10]
-; AVX512F-SLOW-NEXT:    vpermt2q %zmm1, %zmm12, %zmm2
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm18[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1,2],ymm3[3],ymm2[4,5],ymm3[6],ymm2[7,8,9,10],ymm3[11],ymm2[12,13],ymm3[14],ymm2[15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [2,2,3,3,10,9,11,10]
+; AVX512F-SLOW-NEXT:    vpermt2q %zmm1, %zmm24, %zmm2
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[0,1,1,3,4,5,5,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm7[0,1,2,2,4,5,6,6]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm1, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm11 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm2, %zmm11, %zmm1
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm1
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm6[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6,7,8],ymm2[9],ymm1[10,11],ymm2[12],ymm1[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm4[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm2[3],ymm1[4,5],ymm2[6],ymm1[7,8,9,10],ymm2[11],ymm1[12,13],ymm2[14],ymm1[15]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm7[4],ymm10[4],ymm7[5],ymm10[5],ymm7[6],ymm10[6],ymm7[7],ymm10[7],ymm7[12],ymm10[12],ymm7[13],ymm10[13],ymm7[14],ymm10[14],ymm7[15],ymm10[15]
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm18[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm9[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm0[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,2,3,6,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm13 = [4,5,4,5,4,5,6,7,16,17,16,17,16,17,17,19]
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%r9), %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%r8), %xmm6
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm8[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermt2d %zmm4, %zmm13, %zmm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm7[2,3,3,3,6,7,7,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpbroadcastd 96(%rax), %ymm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm4, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm1, %zmm5, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rsi), %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdi), %xmm7
-; AVX512F-SLOW-NEXT:    vprold $16, %xmm1, %xmm9
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm10 = xmm7[1,1,2,3]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm4 = xmm10[0,1],xmm9[2],xmm10[3,4],xmm9[5],xmm10[6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm5 = ymm2[0,1],ymm1[2],ymm2[3,4],ymm1[5],ymm2[6,7,8,9],ymm1[10],ymm2[11,12],ymm1[13],ymm2[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [4,5,4,5,4,5,6,7,16,17,16,17,16,17,17,19]
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%r9), %xmm2
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%r8), %xmm3
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm10 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm10[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermt2d %zmm1, %zmm0, %zmm5
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm4[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vpbroadcastd 96(%rax), %ymm1
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rsi), %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdi), %xmm5
+; AVX512F-SLOW-NEXT:    vprold $16, %xmm0, %xmm6
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm7 = xmm5[1,1,2,3]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm4 = xmm7[0,1],xmm6[2],xmm7[3,4],xmm6[5],xmm7[6,7]
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm7[0],xmm1[0],xmm7[1],xmm1[1],xmm7[2],xmm1[2],xmm7[3],xmm1[3]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm4, %xmm23
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm7[4],xmm1[5],xmm7[5],xmm1[6],xmm7[6],xmm1[7],xmm7[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
 ; AVX512F-SLOW-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rcx), %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdx), %xmm7
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm2, %xmm1, %xmm9
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm10 = xmm7[1,1,2,2]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm4 = xmm10[0],xmm9[1],xmm10[2,3],xmm9[4],xmm10[5,6],xmm9[7]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rcx), %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa 96(%rdx), %xmm5
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm4, %xmm0, %xmm6
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm4, %xmm11
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm7 = xmm5[1,1,2,2]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm4 = xmm7[0],xmm6[1],xmm7[2,3],xmm6[4],xmm7[5,6],xmm6[7]
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm7[0],xmm1[0],xmm7[1],xmm1[1],xmm7[2],xmm1[2],xmm7[3],xmm1[3]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
 ; AVX512F-SLOW-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm1[4],xmm7[5],xmm1[5],xmm7[6],xmm1[6],xmm7[7],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm6[4],xmm3[4],xmm6[5],xmm3[5],xmm6[6],xmm3[6],xmm6[7],xmm3[7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm3 = xmm8[0,1,2,3,4,5,7,6]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm10[0,1,2,3,4,5,7,6]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm20, %xmm1
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm0, %xmm3
 ; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm0 = <16,18,19,19,19,19,u,u,0,1,0,1,2,3,2,3>
-; AVX512F-SLOW-NEXT:    vpermt2d %zmm3, %zmm0, %zmm1
-; AVX512F-SLOW-NEXT:    vpbroadcastd 100(%rax), %ymm3
-; AVX512F-SLOW-NEXT:    vpbroadcastd 104(%rax), %ymm6
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm6, %zmm3, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm31 = [65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm1, %zmm31, %zmm0
+; AVX512F-SLOW-NEXT:    vpermt2d %zmm2, %zmm0, %zmm3
+; AVX512F-SLOW-NEXT:    vpbroadcastd 100(%rax), %ymm2
+; AVX512F-SLOW-NEXT:    vpbroadcastd 104(%rax), %ymm4
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdi), %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rsi), %xmm3
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm3[4],xmm1[4],xmm3[5],xmm1[5],xmm3[6],xmm1[6],xmm3[7],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm8, %xmm19
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; AVX512F-SLOW-NEXT:    vprold $16, %xmm3, %xmm3
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1],xmm3[2],xmm1[3,4],xmm3[5],xmm1[6,7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rcx), %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdx), %xmm6
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm6[4],xmm1[4],xmm6[5],xmm1[5],xmm6[6],xmm1[6],xmm6[7],xmm1[7]
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm6[0],xmm1[0],xmm6[1],xmm1[1],xmm6[2],xmm1[2],xmm6[3],xmm1[3]
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[1,1,2,2]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm6[0],xmm1[1],xmm6[2,3],xmm1[4],xmm6[5,6],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm16, %xmm0
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm15[0],xmm0[0],xmm15[1],xmm0[1],xmm15[2],xmm0[2],xmm15[3],xmm0[3]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,7,6]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm6 = <0,1,0,1,0,1,1,3,16,18,19,19,19,19,u,u>
-; AVX512F-SLOW-NEXT:    vpermt2d %zmm1, %zmm6, %zmm0
-; AVX512F-SLOW-NEXT:    vpbroadcastd 64(%rax), %ymm1
-; AVX512F-SLOW-NEXT:    vpbroadcastd 68(%rax), %ymm7
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm1, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm0, %zmm1, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm7
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm7[4],xmm0[4],xmm7[5],xmm0[5],xmm7[6],xmm0[6],xmm7[7],xmm0[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm14, %xmm22
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3]
-; AVX512F-SLOW-NEXT:    vprold $16, %xmm7, %xmm7
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm7[2],xmm0[3,4],xmm7[5],xmm0[6,7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm7
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm7[4],xmm0[4],xmm7[5],xmm0[5],xmm7[6],xmm0[6],xmm7[7],xmm0[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm4, %xmm20
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm7[0],xmm0[0],xmm7[1],xmm0[1],xmm7[2],xmm0[2],xmm7[3],xmm0[3]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm4, %xmm29
-; AVX512F-SLOW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[1,1,2,2]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm7[0],xmm0[1],xmm7[2,3],xmm0[4],xmm7[5,6],xmm0[7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm24
-; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm7
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm15 = xmm7[4],xmm0[4],xmm7[5],xmm0[5],xmm7[6],xmm0[6],xmm7[7],xmm0[7]
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm7[0],xmm0[0],xmm7[1],xmm0[1],xmm7[2],xmm0[2],xmm7[3],xmm0[3]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm7 = xmm0[0,1,2,3,4,5,7,6]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermt2d %zmm7, %zmm6, %zmm0
-; AVX512F-SLOW-NEXT:    vpbroadcastd (%rax), %ymm6
-; AVX512F-SLOW-NEXT:    vpbroadcastd 4(%rax), %ymm7
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm6, %zmm2
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm0, %zmm1, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm9[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm21, %ymm4
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm21[1,1,1,1,5,5,5,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm21
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm9[0,1,2,3,5,4,6,7,8,9,10,11,13,12,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm4, %ymm10
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7,8],ymm0[9],ymm1[10,11],ymm0[12],ymm1[13,14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdi), %xmm3
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rsi), %xmm4
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm16
+; AVX512F-SLOW-NEXT:    vprold $16, %xmm4, %xmm4
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,2,3]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm3[0,1],xmm4[2],xmm3[3,4],xmm4[5],xmm3[6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm30
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rcx), %xmm4
+; AVX512F-SLOW-NEXT:    vmovdqa 64(%rdx), %xmm5
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm29
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm11, %xmm7
+; AVX512F-SLOW-NEXT:    vpshufb %xmm11, %xmm4, %xmm4
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[1,1,2,2]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0],xmm4[1],xmm5[2,3],xmm4[4],xmm5[5,6],xmm4[7]
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm17
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm5 = xmm4[0,1,2,3,4,5,7,6]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm4[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm9 = <0,1,0,1,0,1,1,3,16,18,19,19,19,19,u,u>
+; AVX512F-SLOW-NEXT:    vpermt2d %zmm5, %zmm9, %zmm6
+; AVX512F-SLOW-NEXT:    vpbroadcastd 64(%rax), %ymm4
+; AVX512F-SLOW-NEXT:    vpbroadcastd 68(%rax), %ymm5
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm4, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm6, %zmm10, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm5
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm6
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm8 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm31
+; AVX512F-SLOW-NEXT:    vprold $16, %xmm6, %xmm6
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[1,1,2,3]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1],xmm6[2],xmm5[3,4],xmm6[5],xmm5[6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm28
+; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm5
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm11
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm11[4],xmm5[4],xmm11[5],xmm5[5],xmm11[6],xmm5[6],xmm11[7],xmm5[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm11[0],xmm5[0],xmm11[1],xmm5[1],xmm11[2],xmm5[2],xmm11[3],xmm5[3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm27
+; AVX512F-SLOW-NEXT:    vpshufb %xmm7, %xmm5, %xmm5
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm11 = xmm11[1,1,2,2]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm11[0],xmm5[1],xmm11[2,3],xmm5[4],xmm11[5,6],xmm5[7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm22
+; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm11
+; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm13
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm13 = xmm11[0,1,2,3,4,5,7,6]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm11[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermt2d %zmm13, %zmm9, %zmm11
+; AVX512F-SLOW-NEXT:    vpbroadcastd (%rax), %ymm9
+; AVX512F-SLOW-NEXT:    vpbroadcastd 4(%rax), %ymm13
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm9, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm11, %zmm10, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm0[1,2,3,3,4,5,6,7,9,10,11,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,2,1,4,4,6,5]
 ; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vprold $16, %ymm4, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm7[1,2,2,3,5,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm7[0,0,2,1,4,4,6,5]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm6[3],ymm1[4,5],ymm6[6],ymm1[7,8,9,10],ymm6[11],ymm1[12,13],ymm6[14],ymm1[15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm4[1,1,1,1,5,5,5,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm9[0,1],ymm10[2],ymm9[3,4],ymm10[5],ymm9[6,7,8,9],ymm10[10],ymm9[11,12],ymm10[13],ymm9[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm7, %ymm18
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm4[4],ymm0[4],ymm4[5],ymm0[5],ymm4[6],ymm0[6],ymm4[7],ymm0[7],ymm4[12],ymm0[12],ymm4[13],ymm0[13],ymm4[14],ymm0[14],ymm4[15],ymm0[15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm25
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm0[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm4[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm10[0,1],ymm9[2],ymm10[3,4],ymm9[5],ymm10[6,7,8,9],ymm9[10],ymm10[11,12],ymm9[13],ymm10[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm7, %ymm21
+; AVX512F-SLOW-NEXT:    vpshufb %ymm14, %ymm0, %ymm9
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm0, %ymm7
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm4[2,2,2,2,6,6,6,6]
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm4, %ymm13
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm9[0,1],ymm10[2],ymm9[3,4],ymm10[5],ymm9[6,7,8,9],ymm10[10],ymm9[11,12],ymm10[13],ymm9[14,15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm20
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vprold $16, %ymm0, %ymm9
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm4[1,2,2,3,5,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm11 = ymm10[0,1],ymm9[2],ymm10[3,4],ymm9[5],ymm10[6,7,8,9],ymm9[10],ymm10[11,12],ymm9[13],ymm10[14,15]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm0[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,0,0,0,4,4,4,4]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm4[0,0,2,1,4,4,6,5]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm14 = ymm9[0,1,2],ymm10[3],ymm9[4,5],ymm10[6],ymm9[7,8,9,10],ymm10[11],ymm9[12,13],ymm10[14],ymm9[15]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[0,2,3,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm3[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[2,1,2,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,3]
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpermt2q %zmm0, %zmm12, %zmm1
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm30[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm30[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm1, %xmm9
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm5, %xmm2
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm6[0,2,3,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm8[2,1,2,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm2[0,0,1,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpermpd $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm1 = mem[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm23[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm19[2,2,3,3]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm15[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm12[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpermt2q %zmm11, %zmm24, %zmm14
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm5[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm5, %ymm8
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm1, %zmm11, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm3[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,0,0,0,4,4,4,4]
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm18
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm16
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm6[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm3[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6,7,8],ymm1[9],ymm0[10,11],ymm1[12],ymm0[13,14,15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm10[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm9[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,2,6,6,6,6]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm1[3],ymm0[4,5],ymm1[6],ymm0[7,8,9,10],ymm1[11],ymm0[12,13],ymm1[14],ymm0[15]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm7[3,3,3,3,7,7,7,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm4[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,2,2,3,6,6,6,7]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm7 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm24
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm14, %zmm24
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm13[4],ymm7[4],ymm13[5],ymm7[5],ymm13[6],ymm7[6],ymm13[7],ymm7[7],ymm13[12],ymm7[12],ymm13[13],ymm7[13],ymm13[14],ymm7[14],ymm13[15],ymm7[15]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[3,3,3,3,7,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm0[0,1,2,3,5,6,7,7,8,9,10,11,13,14,15,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,2,2,3,6,6,6,7]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} ymm6 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7,8,9],ymm2[10],ymm3[11,12],ymm2[13],ymm3[14,15]
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %xmm1
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm6[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermt2d %zmm11, %zmm13, %zmm7
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm30[2,3,3,3,6,7,7,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpbroadcastd 32(%rax), %ymm11
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm2, %zmm30
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm7, %zmm5, %zmm30
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm5
-; AVX512F-SLOW-NEXT:    vprold $16, %xmm5, %xmm7
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm13 = xmm2[1,1,2,3]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm13[0,1],xmm7[2],xmm13[3,4],xmm7[5],xmm13[6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %xmm5
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm3[0],xmm5[1],xmm3[1],xmm5[2],xmm3[2],xmm5[3],xmm3[3]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm4[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [4,5,4,5,4,5,6,7,16,17,16,17,16,17,17,19]
+; AVX512F-SLOW-NEXT:    vpermt2d %zmm11, %zmm0, %zmm6
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm8[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vpbroadcastd 32(%rax), %ymm14
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm11, %zmm23
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm23
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm6
+; AVX512F-SLOW-NEXT:    vprold $16, %xmm6, %xmm11
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm1[1,1,2,3]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm15[0,1],xmm11[2],xmm15[3,4],xmm11[5],xmm15[6,7]
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm2[0],xmm5[0],xmm2[1],xmm5[1],xmm2[2],xmm5[2],xmm2[3],xmm5[3]
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm5[4],xmm2[4],xmm5[5],xmm2[5],xmm5[6],xmm2[6],xmm5[7],xmm2[7]
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3]
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm1[4],xmm6[5],xmm1[5],xmm6[6],xmm1[6],xmm6[7],xmm1[7]
 ; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm15, %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm23, %xmm3
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm3, %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm8, %xmm3
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm14, %xmm4
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm13, %xmm5
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm19, %xmm8
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm13 = xmm8[2,1,2,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,4,5,5,4]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm19 = ymm13[0,0,1,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm2[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm20, %xmm0
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm0[0,2,3,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm23 = ymm5[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm22, %xmm14
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm14[2,1,2,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,5,5,4]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm20 = ymm7[0,0,1,3]
-; AVX512F-SLOW-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm22 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm22 = mem[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm26 = ymm26[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm25 = ymm25[2,2,3,3]
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm12[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm1[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm12[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm1[0,2,3,3,4,6,7,7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm5[4],xmm3[4],xmm5[5],xmm3[5],xmm5[6],xmm3[6],xmm5[7],xmm3[7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm19 = ymm26[0,1,2,2,4,5,6,6]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm26[2,3,3,3,6,7,7,7]
 ; AVX512F-SLOW-NEXT:    vpermpd $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    # ymm0 = mem[2,2,2,3]
 ; AVX512F-SLOW-NEXT:    vmovups %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpermpd $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm0 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpermpd $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm0 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vmovups %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpermpd $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm0 = mem[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,7,6]
+; AVX512F-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; AVX512F-SLOW-NEXT:    vpshufb %xmm9, %xmm0, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm16, %xmm0
+; AVX512F-SLOW-NEXT:    vpshufb %xmm9, %xmm0, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm31, %xmm2
+; AVX512F-SLOW-NEXT:    vpshufb %xmm9, %xmm2, %xmm2
+; AVX512F-SLOW-NEXT:    vpshufb %xmm9, %xmm8, %xmm5
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %xmm9, %xmm1, %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm13 = ymm12[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm31 = ymm13[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermpd $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm3 = mem[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vmovups %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm26 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm26 = mem[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm0[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm30[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm29, %xmm0
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm0[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm15[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm17[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm17 = ymm2[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm28[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm27, %xmm0
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm0[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm11 = xmm11[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm22[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm18 = ymm18[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,7,6]
 ; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm0 = <16,18,19,19,19,19,u,u,0,1,0,1,2,3,2,3>
-; AVX512F-SLOW-NEXT:    vpermt2d %zmm6, %zmm0, %zmm1
-; AVX512F-SLOW-NEXT:    vpbroadcastd 36(%rax), %ymm6
-; AVX512F-SLOW-NEXT:    vpbroadcastd 40(%rax), %ymm13
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm6, %zmm13
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm1, %zmm31, %zmm13
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm6
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm6[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
+; AVX512F-SLOW-NEXT:    vpermt2d %zmm4, %zmm0, %zmm1
+; AVX512F-SLOW-NEXT:    vpbroadcastd 36(%rax), %ymm4
+; AVX512F-SLOW-NEXT:    vpbroadcastd 40(%rax), %ymm22
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm22, %zmm4, %zmm22
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm22
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm1
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm4 = xmm1[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm0
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm0[1,1,2,2]
-; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm15[0],xmm1[1],xmm15[2,3],xmm1[4],xmm15[5,6],xmm1[7]
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3]
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm6[4],xmm0[5],xmm6[5],xmm0[6],xmm6[6],xmm0[7],xmm6[7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm9 = xmm0[1,1,2,2]
+; AVX512F-SLOW-NEXT:    vpblendw {{.*#+}} xmm2 = xmm9[0],xmm4[1],xmm9[2,3],xmm4[4],xmm9[5,6],xmm4[7]
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm27[0,1,2,2,4,5,6,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm27[2,3,3,3,6,7,7,7]
-; AVX512F-SLOW-NEXT:    vpermpd $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm0 = mem[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vmovups %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm31 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm31 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm28 = ymm28[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm27 = mem[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm3[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm7 = mem[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpshuflw $180, {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm14 = mem[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm14 = xmm14[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm5 = mem[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm4[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm4 = mem[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm29, %xmm0
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm0[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm24[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm21 = ymm21[2,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm17 = ymm17[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm18 = ymm18[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm25, %ymm6
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm6[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm0[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm21[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm20[2,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm30 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm30 = mem[0,2,2,3]
 ; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm1 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm1, %zmm24, %zmm0
-; AVX512F-SLOW-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm1 = mem[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm1 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm30, %zmm0, %zmm30
+; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm0 = mem[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm29 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm29 = mem[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm29, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm29 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm30, %zmm29, %zmm0
+; AVX512F-SLOW-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm30 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm30 = mem[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm30
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm20 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm30, %ymm20
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm30 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm30 = zmm1[0,1,2,3],zmm30[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $234, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm30 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm28 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm1 = zmm2[0,1,2,3],zmm1[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $234, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm1 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm29 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm10 = zmm10[0,1,2,3],zmm29[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $234, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm10 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm28 = zmm2[0,1,2,3],zmm28[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpternlogq $234, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm28 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm19, %zmm19 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm29 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm2, %zmm29, %zmm19
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm1, %zmm2, %zmm19
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm23, %zmm1 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm20, %zmm20 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm1, %zmm29, %zmm20
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm10, %zmm2, %zmm20
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm25 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm1, %zmm25, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm30, %zmm3, %zmm2
 ; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm22, %zmm1, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm26, %zmm2, %zmm2
-; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm23 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $248, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm23 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm25, %zmm0, %zmm10
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm22 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm10 = zmm22[0,1,2,3],zmm10[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $234, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm10 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogd $226, 124(%r8){1to8}, %ymm1, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm3[0,1,2,3],zmm0[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm1 = mem[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm22 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm22 = mem[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm22, %zmm1, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm22 = [65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm0, %zmm22, %zmm1
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm12[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm15[2,1,3,2]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpternlogq $184, {{[-0-9]+}}(%r{{[sb]}}p), %zmm22, %zmm0 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm8, %zmm7
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm14[0,1,1,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm8, %zmm5
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm7, %zmm8, %zmm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm11, %zmm4
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm6, %zmm3
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm4, %zmm8, %zmm3
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm25 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm5, %zmm4, %zmm25
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm26 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm3, %zmm4, %zmm26
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm31, %zmm3, %zmm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm27, %zmm28, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm3, %zmm5, %zmm4
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm17, %zmm21, %zmm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm18, %zmm6
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm3, %zmm5, %zmm6
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm22 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm4, %zmm3, %zmm22
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm27 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm6, %zmm3, %zmm27
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm30 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm30, %zmm25, %zmm1
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm28, %zmm3, %zmm1
 ; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm3 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm3, %zmm24, %zmm4
-; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm3 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshuflw $180, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm5 = mem[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,1,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm3, %zmm3
-; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm5 = mem[2,1,3,3]
-; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm6 = mem[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm6, %zmm5, %zmm5
-; AVX512F-SLOW-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm21 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm21 = mem[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpshuflw $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm7 = mem[2,1,2,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,5,5,4]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm27 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm27, %zmm28 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm28
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm21 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $248, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm21 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm30 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm27 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm30 = zmm27[0,1,2,3],zmm30[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpternlogq $234, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3, %zmm30 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogd $226, 124(%r8){1to8}, %ymm3, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm3
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm20[0,1,2,3],zmm0[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm3 = mem[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm27 = mem[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm27, %zmm3, %zmm3
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm27 = [65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0]
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm0, %zmm27, %zmm3
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm19[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm19 = ymm14[2,1,3,2]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm19, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $184, {{[-0-9]+}}(%r{{[sb]}}p), %zmm27, %zmm0 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm16, %zmm13
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,1,1,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm15, %zmm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm15 = [65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0]
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm13, %zmm15, %zmm8
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm17, %zmm10
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,1,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm11, %zmm7
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm10, %zmm15, %zmm7
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm20 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm8, %zmm10, %zmm20
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm27 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm7, %zmm10, %zmm27
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm31[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm8, %zmm7
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm26, %zmm8, %zmm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535]
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm7, %zmm10, %zmm8
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm18, %zmm4
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm9, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm4, %zmm10, %zmm5
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm19 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm8, %zmm4, %zmm19
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm5, %zmm4, %zmm24
+; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm4 = mem[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
+; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm5 = mem[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm7 = mem[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm5, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm4, %zmm29, %zmm5
+; AVX512F-SLOW-NEXT:    vpshufhw $180, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm4 = mem[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshuflw $180, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm7 = mem[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,1,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm4, %zmm4
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm7 = ymm12[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,2,3,3,4,6,7,7]
 ; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    # ymm8 = mem[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm9 = mem[0,2,3,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm12 = mem[0,2,2,3]
-; AVX512F-SLOW-NEXT:    vpshuflw $180, {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm14 = mem[0,1,3,2,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm14 = xmm14[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm15 = mem[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm7, %zmm7
+; AVX512F-SLOW-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm8 = mem[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vpshuflw $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm9 = mem[2,1,2,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,4,5,5,4]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,1,3]
+; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm10 = mem[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm11 = mem[0,2,3,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vpshufhw $180, {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm13 = mem[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[2,3,3,3,6,7,7,7]
+; AVX512F-SLOW-NEXT:    vpshuflw $180, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm15 = mem[0,1,3,2,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm15[0,0,1,1]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm6 = ymm6[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[0,2,3,3,4,6,7,7]
 ; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm16 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    # ymm16 = mem[0,0,1,1]
 ; AVX512F-SLOW-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm17 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    # ymm17 = mem[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpshuflw $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm11 = mem[2,1,2,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,4,5,5,4]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,1,3]
+; AVX512F-SLOW-NEXT:    vpshuflw $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm12 = mem[2,1,2,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm12 = xmm12[0,1,2,3,4,5,5,4]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,1,3]
 ; AVX512F-SLOW-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm18 # 32-byte Folded Reload
 ; AVX512F-SLOW-NEXT:    # ymm18 = mem[0,0,1,1]
-; AVX512F-SLOW-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm6 = mem[0,2,3,3,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,2,1]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm3, %zmm24, %zmm5
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm14[0,1,1,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm12, %zmm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm12
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm3, %zmm24, %zmm12
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm14 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm5, %zmm3, %zmm14
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm12, %zmm3, %zmm30
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm21, %zmm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm8, %zmm5
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm3, %zmm29, %zmm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm17, %zmm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm6, %zmm18, %zmm6
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm3, %zmm29, %zmm6
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0]
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm5, %zmm3, %zmm7
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm6, %zmm3, %zmm13
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm10
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm0
+; AVX512F-SLOW-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm14 = mem[0,2,3,3,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,2,1]
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm4, %zmm29, %zmm7
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm13[0,2,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm15[0,1,1,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm4, %zmm4
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,1,3,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm6, %zmm6
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm4, %zmm29, %zmm6
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm13 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm7, %zmm4, %zmm13
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm6, %zmm4, %zmm23
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm8, %zmm4
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm10, %zmm6
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm4, %zmm25, %zmm6
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm17, %zmm4
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm18, %zmm7
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm4, %zmm25, %zmm7
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0]
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm6, %zmm4, %zmm8
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm7, %zmm4, %zmm22
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm28, %zmm30
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm0
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm10, 320(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm13, 256(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm30, 192(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm27, 128(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm20, 64(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm26, (%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm25, 448(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm7, 704(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm14, 640(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm22, 576(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm19, 512(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm30, 320(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm22, 256(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm23, 192(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm24, 128(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm1, 64(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm27, (%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm20, 448(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm8, 704(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm13, 640(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm19, 576(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm2, 512(%rax)
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm0, 384(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm23, 768(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm1, 832(%rax)
-; AVX512F-SLOW-NEXT:    addq $2440, %rsp # imm = 0x988
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm21, 768(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm3, 832(%rax)
+; AVX512F-SLOW-NEXT:    addq $2504, %rsp # imm = 0x9C8
 ; AVX512F-SLOW-NEXT:    vzeroupper
 ; AVX512F-SLOW-NEXT:    retq
 ;
 ; AVX512F-FAST-LABEL: store_i16_stride7_vf64:
 ; AVX512F-FAST:       # %bb.0:
-; AVX512F-FAST-NEXT:    subq $2200, %rsp # imm = 0x898
+; AVX512F-FAST-NEXT:    subq $2168, %rsp # imm = 0x878
 ; AVX512F-FAST-NEXT:    vmovdqa 96(%rsi), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rdi), %ymm6
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rcx), %ymm3
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rdx), %ymm15
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm3, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm3, %ymm25
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm15, %ymm3
-; AVX512F-FAST-NEXT:    vporq %ymm0, %ymm3, %ymm17
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm20
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm5
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm6, %ymm23
-; AVX512F-FAST-NEXT:    vporq %ymm4, %ymm5, %ymm16
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rdi), %ymm2
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15]
+; AVX512F-FAST-NEXT:    vmovdqa %ymm2, %ymm5
+; AVX512F-FAST-NEXT:    vmovdqa %ymm1, %ymm4
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm0[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rcx), %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rdx), %ymm9
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm23
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm9, %ymm3
+; AVX512F-FAST-NEXT:    vporq %ymm1, %ymm3, %ymm17
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128,128,128,128,128,128,128,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm4, %ymm19
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u,u,u,u,u,16,17,18,19>
+; AVX512F-FAST-NEXT:    vpshufb %ymm14, %ymm5, %ymm4
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm5, %ymm18
+; AVX512F-FAST-NEXT:    vporq %ymm3, %ymm4, %ymm25
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128,128,128]
-; AVX512F-FAST-NEXT:    vmovdqa 64(%r9), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa %ymm5, %ymm6
-; AVX512F-FAST-NEXT:    vmovdqa 64(%r8), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 64(%r9), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm3, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa %ymm5, %ymm10
+; AVX512F-FAST-NEXT:    vmovdqa 64(%r8), %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,12,13,14,15,128,128,u,u,u,u,u,u,u,u,u,u,16,17,128,128,u,u,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm1, %ymm5
-; AVX512F-FAST-NEXT:    vporq %ymm4, %ymm5, %ymm24
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rcx), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rdx), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm1, %ymm5
-; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm5, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rsi), %ymm1
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm30
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rdi), %ymm14
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm14, %ymm5
-; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm5, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm6, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa %ymm6, %ymm10
-; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm1, %ymm5
-; AVX512F-FAST-NEXT:    vporq %ymm4, %ymm5, %ymm31
-; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm1, %ymm5
-; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm5, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm5
-; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm5, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %ymm13
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm13, %ymm2
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %ymm12
-; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm12, %ymm4
-; AVX512F-FAST-NEXT:    vpor %ymm2, %ymm4, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %ymm11
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm11, %ymm2
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %ymm7
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm7, %ymm0
-; AVX512F-FAST-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm4, %ymm4
+; AVX512F-FAST-NEXT:    vporq %ymm3, %ymm4, %ymm16
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rcx), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm3, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rdx), %ymm13
+; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm13, %ymm4
+; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rsi), %ymm4
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm4, %ymm26
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rdi), %ymm5
+; AVX512F-FAST-NEXT:    vpshufb %ymm14, %ymm5, %ymm4
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm5, %ymm27
+; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm3, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm4, %ymm4
+; AVX512F-FAST-NEXT:    vporq %ymm3, %ymm4, %ymm20
+; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm3, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm4, %ymm4
+; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm3, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm4, (%rsp) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm14, %ymm4, %ymm4
+; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %ymm15
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm15, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %ymm11
+; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm11, %ymm2
+; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %ymm12
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm12, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %ymm6
+; AVX512F-FAST-NEXT:    vpshufb %ymm14, %ymm6, %ymm1
+; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %ymm6
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm6, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %ymm7
+; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm7, %ymm2
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %ymm5
-; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm5, %ymm9
-; AVX512F-FAST-NEXT:    vporq %ymm9, %ymm0, %ymm22
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm6[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm9[2],ymm0[3,4],ymm9[5],ymm0[6,7,8,9],ymm9[10],ymm0[11,12],ymm9[13],ymm0[14,15]
-; AVX512F-FAST-NEXT:    vprold $16, %ymm5, %ymm9
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm6[1,2,2,3,5,6,6,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7,8,9],ymm9[10],ymm8[11,12],ymm9[13],ymm8[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [2,1,3,2,10,10,10,11]
-; AVX512F-FAST-NEXT:    vpermi2q %zmm0, %zmm8, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u>
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm20, %ymm1
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm1, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm29
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm23[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1],ymm8[2,3],ymm0[4],ymm8[5,6,7,8],ymm0[9],ymm8[10,11],ymm0[12],ymm8[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29>
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm1, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm20
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm23[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1,2],ymm0[3],ymm8[4,5],ymm0[6],ymm8[7,8,9,10],ymm0[11],ymm8[12,13],ymm0[14],ymm8[15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm25, %ymm4
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm4, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm21
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm15[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm8[2],ymm0[3,4],ymm8[5],ymm0[6,7,8,9],ymm8[10],ymm0[11,12],ymm8[13],ymm0[14,15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm4, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm25
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm15[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0],ymm0[1],ymm8[2,3],ymm0[4],ymm8[5,6,7,8],ymm0[9],ymm8[10,11],ymm0[12],ymm8[13,14,15]
+; AVX512F-FAST-NEXT:    vmovdqa %ymm10, %ymm4
+; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm5, %ymm14
+; AVX512F-FAST-NEXT:    vporq %ymm14, %ymm2, %ymm22
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm6[4],ymm12[4],ymm6[5],ymm12[5],ymm6[6],ymm12[6],ymm6[7],ymm12[7],ymm6[12],ymm12[12],ymm6[13],ymm12[13],ymm6[14],ymm12[14],ymm6[15],ymm12[15]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm2[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm14, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm7[3,3,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0,1],ymm14[2],ymm2[3,4],ymm14[5],ymm2[6,7,8,9],ymm14[10],ymm2[11,12],ymm14[13],ymm2[14,15]
+; AVX512F-FAST-NEXT:    vprold $16, %ymm5, %ymm14
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm7[1,2,2,3,5,6,6,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm10 = ymm10[0,1],ymm14[2],ymm10[3,4],ymm14[5],ymm10[6,7,8,9],ymm14[10],ymm10[11,12],ymm14[13],ymm10[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [2,1,3,2,10,10,10,11]
+; AVX512F-FAST-NEXT:    vpermi2q %zmm2, %zmm10, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27>
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm23, %ymm0
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm21
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm9[2,2,2,2,6,6,6,6]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm10[2],ymm2[3,4],ymm10[5],ymm2[6,7,8,9],ymm10[10],ymm2[11,12],ymm10[13],ymm2[14,15]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm9[4],ymm0[4],ymm9[5],ymm0[5],ymm9[6],ymm0[6],ymm9[7],ymm0[7],ymm9[12],ymm0[12],ymm9[13],ymm0[13],ymm9[14],ymm0[14],ymm9[15],ymm0[15]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm0[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm9[0,1],ymm2[2],ymm9[3,4],ymm2[5],ymm9[6,7,8,9],ymm2[10],ymm9[11,12],ymm2[13],ymm9[14,15]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,3,2]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm19, %ymm1
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm9
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm19
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm10 = ymm18[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm9 = ymm9[0,1],ymm10[2],ymm9[3,4],ymm10[5],ymm9[6,7,8,9],ymm10[10],ymm9[11,12],ymm10[13],ymm9[14,15]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm17, %zmm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm25, %zmm9
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm9
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <5,u,u,u,6,u,u,6>
+; AVX512F-FAST-NEXT:    vmovdqa 96(%r8), %ymm10
+; AVX512F-FAST-NEXT:    vpermd %ymm10, %ymm2, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm4[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm15[0,1,1,3,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7,8,9],ymm0[10],ymm8[11,12],ymm0[13],ymm8[14,15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,3,2]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23>
-; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm1, %ymm8
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm18
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm9 = ymm23[1,1,1,1,5,5,5,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm8 = ymm8[0,1],ymm9[2],ymm8[3,4],ymm9[5],ymm8[6,7,8,9],ymm9[10],ymm8[11,12],ymm9[13],ymm8[14,15]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm17, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm16, %zmm8
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <5,u,u,u,6,u,u,6>
-; AVX512F-FAST-NEXT:    vmovdqa 96(%r8), %ymm9
-; AVX512F-FAST-NEXT:    vpermd %ymm9, %ymm0, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,4,u,u,u,5,u,u>
+; AVX512F-FAST-NEXT:    vpermd %ymm10, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm10[u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm10[14,15],zero,zero,ymm10[u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm10[16,17],zero,zero,ymm10[u,u],zero,zero
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm17 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm17, %ymm9, %ymm10
+; AVX512F-FAST-NEXT:    vmovdqa 96(%r9), %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,4,u,u,u,5,u,u>
-; AVX512F-FAST-NEXT:    vpermd %ymm9, %ymm0, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm9[u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm9[14,15],zero,zero,ymm9[u,u],zero,zero,zero,zero,zero,zero,zero,zero,ymm9[16,17],zero,zero,ymm9[u,u],zero,zero
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm16 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm16, %ymm8, %ymm9
-; AVX512F-FAST-NEXT:    vmovdqa 96(%r9), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm1, %ymm15
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm9, %ymm15
-; AVX512F-FAST-NEXT:    vextracti64x4 $1, %zmm8, %ymm8
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm8, %ymm0
-; AVX512F-FAST-NEXT:    vprold $16, %ymm1, %ymm8
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,2]
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm8
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm15[0,1,2,3],zmm0[4,5,6,7]
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm0, %ymm14
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm10, %ymm14
+; AVX512F-FAST-NEXT:    vextracti64x4 $1, %zmm9, %ymm9
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm9, %ymm2
+; AVX512F-FAST-NEXT:    vprold $16, %ymm0, %ymm9
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,2,2]
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm9
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm2
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm14[0,1,2,3],zmm2[4,5,6,7]
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-FAST-NEXT:    vmovdqa64 64(%rax), %zmm23
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <u,5,u,u,u,6,u,u,30,u,u,u,31,u,u,31>
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rax), %ymm0
-; AVX512F-FAST-NEXT:    vpermi2d %zmm23, %zmm0, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm17 = [0,1,4,5,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm17, %ymm8
-; AVX512F-FAST-NEXT:    vpandnq %ymm8, %ymm16, %ymm8
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm0, %ymm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 64(%rax), %zmm18
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = <u,5,u,u,u,6,u,u,30,u,u,u,31,u,u,31>
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rax), %ymm2
+; AVX512F-FAST-NEXT:    vpermi2d %zmm18, %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm29 = [0,1,4,5,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm2, %ymm29, %ymm9
+; AVX512F-FAST-NEXT:    vpandnq %ymm9, %ymm17, %ymm9
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [12,13,128,128,128,128,128,128,128,128,128,128,128,128,14,15,128,128,128,128,128,128,128,128,128,128,128,128,16,17,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm2, %zmm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpbroadcastd 72(%rax), %ymm0
-; AVX512F-FAST-NEXT:    vpandn %ymm0, %ymm1, %ymm8
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rax), %ymm10
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm10, %ymm9
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm8, %zmm27
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm24, %zmm0, %zmm15
-; AVX512F-FAST-NEXT:    vmovdqa 64(%r9), %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm31 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpbroadcastd 72(%rax), %ymm2
+; AVX512F-FAST-NEXT:    vpandnq %ymm2, %ymm31, %ymm9
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rax), %ymm8
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm8, %ymm10
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm9, %zmm30
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm16, %zmm0, %zmm14
+; AVX512F-FAST-NEXT:    vmovdqa 64(%r9), %xmm10
 ; AVX512F-FAST-NEXT:    vmovdqa 64(%r8), %xmm9
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm9[4],xmm2[4],xmm9[5],xmm2[5],xmm9[6],xmm2[6],xmm9[7],xmm2[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm2, %xmm28
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [0,0,1,1,12,13,14,15]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm15, %zmm24, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm26 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $248, %zmm26, %zmm0, %zmm27
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm27, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm16 = [0,0,1,1,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm14, %zmm16, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm23 = [65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $248, %zmm23, %zmm0, %zmm30
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm30, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vpbroadcastd 8(%rax), %ymm0
-; AVX512F-FAST-NEXT:    vpandn %ymm0, %ymm1, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa (%rax), %ymm1
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm15
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm19
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm15, %zmm0, %zmm1
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm31, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa (%r9), %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r8), %xmm8
-; AVX512F-FAST-NEXT:    vmovdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm15 = xmm8[4],xmm4[4],xmm8[5],xmm4[5],xmm8[6],xmm4[6],xmm8[7],xmm4[7]
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm15, %xmm15
-; AVX512F-FAST-NEXT:    vpermt2q %zmm0, %zmm24, %zmm15
-; AVX512F-FAST-NEXT:    vpternlogq $248, %zmm26, %zmm15, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm13[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm12[0,1,1,3,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm15[0,1],ymm0[2],ymm15[3,4],ymm0[5],ymm15[6,7,8,9],ymm0[10],ymm15[11,12],ymm0[13],ymm15[14,15]
+; AVX512F-FAST-NEXT:    vpandnq %ymm0, %ymm31, %ymm30
+; AVX512F-FAST-NEXT:    vmovdqa (%rax), %ymm0
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm0, %ymm14
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm28
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm30, %zmm30
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm0, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqa (%r9), %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa (%r8), %xmm0
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm14 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm0, %xmm24
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm1, %xmm25
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm14, %xmm14
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm20
+; AVX512F-FAST-NEXT:    vpermt2q %zmm2, %zmm16, %zmm14
+; AVX512F-FAST-NEXT:    vpternlogq $248, %zmm23, %zmm14, %zmm30
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm30, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm15[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm14 = ymm11[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm14[0,1],ymm0[2],ymm14[3,4],ymm0[5],ymm14[6,7,8,9],ymm0[10],ymm14[11,12],ymm0[13],ymm14[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm18, %ymm4
-; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm11, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm15 = ymm7[1,1,1,1,5,5,5,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm15[2],ymm0[3,4],ymm15[5],ymm0[6,7,8,9],ymm15[10],ymm0[11,12],ymm15[13],ymm0[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm19, %ymm14
+; AVX512F-FAST-NEXT:    vpshufb %ymm14, %ymm12, %ymm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm6[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm6[2],ymm0[3,4],ymm6[5],ymm0[6,7,8,9],ymm6[10],ymm0[11,12],ymm6[13],ymm0[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm5, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[0,0,2,1,4,4,6,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm2[3],ymm0[4,5],ymm2[6],ymm0[7,8,9,10],ymm2[11],ymm0[12,13],ymm2[14],ymm0[15]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <0,1,u,3,10,10,11,11>
-; AVX512F-FAST-NEXT:    vpermi2q %zmm0, %zmm22, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21>
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm5, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa %ymm1, %ymm5
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm7[0,0,2,1,4,4,6,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2],ymm3[3],ymm0[4,5],ymm3[6],ymm0[7,8,9,10],ymm3[11],ymm0[12,13],ymm3[14],ymm0[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = <0,1,u,3,10,10,11,11>
+; AVX512F-FAST-NEXT:    vpermi2q %zmm0, %zmm22, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %ymm0
-; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm17, %ymm2
-; AVX512F-FAST-NEXT:    vpandnq %ymm2, %ymm16, %ymm2
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm0, %ymm1
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm29, %ymm3
+; AVX512F-FAST-NEXT:    vpandnq %ymm3, %ymm17, %ymm3
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm0, %ymm1
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm1
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm29, %ymm5
-; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm11, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm20, %ymm6
-; AVX512F-FAST-NEXT:    vpshufb %ymm6, %ymm11, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1,2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7,8,9,10],ymm1[11],ymm2[12,13],ymm1[14],ymm2[15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm21, %ymm3
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm13, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm12[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7,8,9],ymm2[10],ymm1[11,12],ymm2[13],ymm1[14,15]
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm11[4],ymm15[4],ymm11[5],ymm15[5],ymm11[6],ymm15[6],ymm11[7],ymm15[7],ymm11[12],ymm15[12],ymm11[13],ymm15[13],ymm11[14],ymm15[14],ymm11[15],ymm15[15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm25, %ymm15
-; AVX512F-FAST-NEXT:    vpshufb %ymm15, %ymm13, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm12[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6,7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13,14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm21, %ymm4
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm15, %ymm1
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm11[2,2,2,2,6,6,6,6]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm3[2],ymm1[3,4],ymm3[5],ymm1[6,7,8,9],ymm3[10],ymm1[11,12],ymm3[13],ymm1[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 (%rax), %zmm21
+; AVX512F-FAST-NEXT:    vmovdqa64 (%rax), %zmm15
 ; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <u,21,u,u,u,22,u,u,14,u,u,u,15,u,u,15>
-; AVX512F-FAST-NEXT:    vpermi2d %zmm0, %zmm21, %zmm1
+; AVX512F-FAST-NEXT:    vpermi2d %zmm0, %zmm15, %zmm1
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm30, %ymm2
-; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm2, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm14[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm26, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm27, %ymm1
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm26
+; AVX512F-FAST-NEXT:    vpshufb %ymm14, %ymm0, %ymm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm27[1,1,1,1,5,5,5,5]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm2, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm30, %ymm4
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm14[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7,8],ymm0[9],ymm1[10,11],ymm0[12],ymm1[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm5[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
 ; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm13[0,1,1,3,4,5,5,7]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm5, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa %ymm5, %ymm11
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm2[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vmovdqa %ymm2, %ymm7
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm2, %ymm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm13[2,2,2,2,6,6,6,6]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vprold $16, %ymm2, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[1,2,2,3,5,6,6,7]
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vprold $16, %ymm3, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm4[1,2,2,3,5,6,6,7]
 ; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm2, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa %ymm2, %ymm3
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,0,2,1,4,4,6,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3],ymm1[4,5],ymm2[6],ymm1[7,8,9,10],ymm2[11],ymm1[12,13],ymm2[14],ymm1[15]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm12 = [2,2,3,3,10,9,11,10]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm0, %zmm12, %zmm2
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm23, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
-; AVX512F-FAST-NEXT:    vpermd %zmm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm2, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm15, %ymm11, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm7[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2,3],ymm0[4],ymm2[5,6,7,8],ymm0[9],ymm2[10,11],ymm0[12],ymm2[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm6, %ymm4, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm14[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3],ymm2[4,5],ymm0[6],ymm2[7,8,9,10],ymm0[11],ymm2[12,13],ymm0[14],ymm2[15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm22 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31>
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm22, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm2[2],ymm0[3,4],ymm2[5],ymm0[6,7,8,9],ymm2[10],ymm0[11,12],ymm2[13],ymm0[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [2,2,2,3,8,8,8,9]
-; AVX512F-FAST-NEXT:    vmovdqa 96(%r9), %xmm5
-; AVX512F-FAST-NEXT:    vmovdqa 96(%r8), %xmm6
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm7, %xmm2
-; AVX512F-FAST-NEXT:    vmovdqa %xmm3, %xmm14
-; AVX512F-FAST-NEXT:    vpermt2q %zmm2, %zmm4, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,7,3,3,7,7,6,7]
-; AVX512F-FAST-NEXT:    vpermd %ymm10, %ymm2, %ymm3
-; AVX512F-FAST-NEXT:    vpbroadcastd 96(%rax), %ymm4
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rsi), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rdi), %xmm4
-; AVX512F-FAST-NEXT:    vprold $16, %xmm0, %xmm10
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm11 = xmm4[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm11[0,1],xmm10[2],xmm11[3,4],xmm10[5],xmm11[6,7]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm0[0],xmm4[1],xmm0[1],xmm4[2],xmm0[2],xmm4[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm30
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm0, %xmm23
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rcx), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa 96(%rdx), %xmm10
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm15 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
-; AVX512F-FAST-NEXT:    vpshufb %xmm15, %xmm0, %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa %xmm15, %xmm8
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm11 = xmm10[1,1,2,2]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm11[0],xmm4[1],xmm11[2,3],xmm4[4],xmm11[5,6],xmm4[7]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm10[0],xmm0[0],xmm10[1],xmm0[1],xmm10[2],xmm0[2],xmm10[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm10[4],xmm0[4],xmm10[5],xmm0[5],xmm10[6],xmm0[6],xmm10[7],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm15 = [0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
-; AVX512F-FAST-NEXT:    vpshufb %xmm15, %xmm7, %xmm5
-; AVX512F-FAST-NEXT:    vmovdqa %xmm15, %xmm7
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [0,1,1,3,8,8,9,9]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm0, %zmm3, %zmm5
-; AVX512F-FAST-NEXT:    vpbroadcastd 100(%rax), %ymm0
-; AVX512F-FAST-NEXT:    vpbroadcastd 104(%rax), %ymm6
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm0
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm3, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa %ymm3, %ymm5
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm4[0,0,2,1,4,4,6,5]
+; AVX512F-FAST-NEXT:    vmovdqa %ymm4, %ymm6
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2],ymm3[3],ymm1[4,5],ymm3[6],ymm1[7,8,9,10],ymm3[11],ymm1[12,13],ymm3[14],ymm1[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [2,2,3,3,10,9,11,10]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm0, %zmm3, %zmm1
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm18, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
+; AVX512F-FAST-NEXT:    vpermd %zmm0, %zmm3, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm13[4],ymm2[4],ymm13[5],ymm2[5],ymm13[6],ymm2[6],ymm13[7],ymm2[7],ymm13[12],ymm2[12],ymm13[13],ymm2[13],ymm13[14],ymm2[14],ymm13[15],ymm2[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm4, %ymm16
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm6[3,3,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7,8,9],ymm1[10],ymm0[11,12],ymm1[13],ymm0[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [2,2,2,3,8,8,8,9]
+; AVX512F-FAST-NEXT:    vmovdqa 96(%r9), %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa 96(%r8), %xmm5
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm3[0],xmm5[1],xmm3[1],xmm5[2],xmm3[2],xmm5[3],xmm3[3]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = [0,1,2,3,6,7,4,5,6,7,4,5,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm0, %xmm6
+; AVX512F-FAST-NEXT:    vpermt2q %zmm6, %zmm2, %zmm1
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = [6,7,3,3,7,7,6,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm8, %ymm6, %ymm2
+; AVX512F-FAST-NEXT:    vpbroadcastd 96(%rax), %ymm6
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm2, %zmm2
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rsi), %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rdi), %xmm2
+; AVX512F-FAST-NEXT:    vprold $16, %xmm1, %xmm6
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm7 = xmm2[1,1,2,3]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm6 = xmm7[0,1],xmm6[2],xmm7[3,4],xmm6[5],xmm7[6,7]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm6, %xmm30
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; AVX512F-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rcx), %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa 96(%rdx), %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm7 = <u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9>
+; AVX512F-FAST-NEXT:    vpshufb %xmm7, %xmm1, %xmm6
+; AVX512F-FAST-NEXT:    vmovdqa %xmm7, %xmm12
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm7 = xmm2[1,1,2,2]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm6 = xmm7[0],xmm6[1],xmm7[2,3],xmm6[4],xmm7[5,6],xmm6[7]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm14 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
+; AVX512F-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm5[4],xmm3[4],xmm5[5],xmm3[5],xmm5[6],xmm3[6],xmm5[7],xmm3[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm20, %xmm2
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm13 = [0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
+; AVX512F-FAST-NEXT:    vpshufb %xmm13, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa %xmm13, %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [0,1,1,3,8,8,9,9]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm1, %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vpbroadcastd 100(%rax), %ymm1
+; AVX512F-FAST-NEXT:    vpbroadcastd 104(%rax), %ymm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa 64(%rcx), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rdx), %xmm5
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm31
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vmovdqa %xmm8, %xmm2
-; AVX512F-FAST-NEXT:    vpshufb %xmm8, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[1,1,2,2]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0],xmm0[1],xmm5[2,3],xmm0[4],xmm5[5,6],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm18
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rdx), %xmm1
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm13 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm13, %xmm23
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm2, %xmm20
+; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,2]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1],xmm1[2,3],xmm0[4],xmm1[5,6],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa 64(%rdi), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa 64(%rsi), %xmm5
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm15 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm15, %xmm17
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm15 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3]
-; AVX512F-FAST-NEXT:    vprold $16, %xmm5, %xmm5
+; AVX512F-FAST-NEXT:    vmovdqa 64(%rsi), %xmm1
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm11 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm11, %xmm22
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm11 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-FAST-NEXT:    vprold $16, %xmm1, %xmm1
 ; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm5[2],xmm0[3,4],xmm5[5],xmm0[6,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2],xmm0[3,4],xmm1[5],xmm0[6,7]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm28, %xmm0
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm0[0],xmm9[1],xmm0[1],xmm9[2],xmm0[2],xmm9[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vmovdqa %xmm7, %xmm13
-; AVX512F-FAST-NEXT:    vpshufb %xmm7, %xmm0, %xmm5
-; AVX512F-FAST-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm6 = [0,0,0,1,8,9,9,11]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm5, %zmm6, %zmm0
-; AVX512F-FAST-NEXT:    vpbroadcastd 64(%rax), %ymm5
-; AVX512F-FAST-NEXT:    vpbroadcastd 68(%rax), %ymm7
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm5, %zmm3
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm0, %zmm8, %zmm3
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm0, %xmm1
+; AVX512F-FAST-NEXT:    vmovdqa %xmm3, %xmm5
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [0,0,0,1,8,9,9,11]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm1, %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vpbroadcastd 64(%rax), %ymm1
+; AVX512F-FAST-NEXT:    vpbroadcastd 68(%rax), %ymm3
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm0, %zmm1, %zmm3
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %xmm5
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm16
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm7 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[1,1,2,2]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0],xmm0[1],xmm5[2,3],xmm0[4],xmm5[5,6],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %xmm3
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm6, %xmm21
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm10 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,2,2]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm3[0],xmm0[1],xmm3[2,3],xmm0[4],xmm3[5,6],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm31
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm9
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm9[4],xmm0[4],xmm9[5],xmm0[5],xmm9[6],xmm0[6],xmm9[7],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm2, %xmm28
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm0[0],xmm9[0],xmm0[1],xmm9[1],xmm0[2],xmm9[2],xmm0[3],xmm9[3]
-; AVX512F-FAST-NEXT:    vprold $16, %xmm9, %xmm9
+; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm3
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm3[4],xmm0[4],xmm3[5],xmm0[5],xmm3[6],xmm0[6],xmm3[7],xmm0[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm6, %xmm19
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm13 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3]
+; AVX512F-FAST-NEXT:    vprold $16, %xmm3, %xmm3
 ; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm9[2],xmm0[3,4],xmm9[5],xmm0[6,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2],xmm0[3,4],xmm3[5],xmm0[6,7]
 ; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm27
-; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX512F-FAST-NEXT:    vpunpcklwd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; AVX512F-FAST-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm13, %xmm0, %xmm9
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm13, %xmm29
-; AVX512F-FAST-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa %xmm14, %xmm2
-; AVX512F-FAST-NEXT:    vpermt2q %zmm9, %zmm6, %zmm0
-; AVX512F-FAST-NEXT:    vpbroadcastd (%rax), %ymm6
-; AVX512F-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm9
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm6, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm0, %zmm8, %zmm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm3[1,1,1,1,5,5,5,5]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm6[2],ymm0[3,4],ymm6[5],ymm0[6,7,8,9],ymm6[10],ymm0[11,12],ymm6[13],ymm0[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm26
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[10,11,8,9,u,u,u,u,u,u,u,u,u,u,u,u,26,27,24,25,u,u,u,u,26,27,24,25,u,u,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm3[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vmovdqa %ymm3, %ymm14
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm6[0],ymm0[1],ymm6[2,3],ymm0[4],ymm6[5,6,7,8],ymm0[9],ymm6[10,11],ymm0[12],ymm6[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm25
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm25, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm24, %xmm3
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm3[0],xmm0[0],xmm3[1],xmm0[1],xmm3[2],xmm0[2],xmm3[3],xmm0[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm5, %xmm0, %xmm3
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa %xmm4, %xmm12
+; AVX512F-FAST-NEXT:    vpermt2q %zmm3, %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vpbroadcastd (%rax), %ymm2
+; AVX512F-FAST-NEXT:    vpbroadcastd 4(%rax), %ymm3
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm2
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm0, %zmm1, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
 ; AVX512F-FAST-NEXT:    vprold $16, %ymm3, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm9[1,2,2,3,5,6,6,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm6[0,1],ymm0[2],ymm6[3,4],ymm0[5],ymm6[6,7,8,9],ymm0[10],ymm6[11,12],ymm0[13],ymm6[14,15]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm9[0,0,2,1,4,4,6,5]
-; AVX512F-FAST-NEXT:    vmovdqa %ymm9, %ymm10
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm6[0,1,2],ymm8[3],ymm6[4,5],ymm8[6],ymm6[7,8,9,10],ymm8[11],ymm6[12,13],ymm8[14],ymm6[15]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm0, %zmm12, %zmm6
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm9[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm12[0,1,1,3,4,5,5,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm8[0,1],ymm0[2],ymm8[3,4],ymm0[5],ymm8[6,7,8,9],ymm0[10],ymm8[11,12],ymm0[13],ymm8[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm24
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm8 = ymm12[2,2,2,2,6,6,6,6]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm8[2],ymm0[3,4],ymm8[5],ymm0[6,7,8,9],ymm8[10],ymm0[11,12],ymm8[13],ymm0[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm20
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm19, %zmm21, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm8 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
-; AVX512F-FAST-NEXT:    vpermd %zmm0, %zmm8, %zmm0
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm6, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm9[14,15,12,13,u,u,u,u,u,u,u,u,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29,u,u,u,u]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5,6,7,8],ymm0[9],ymm1[10,11],ymm0[12],ymm1[13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm4, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm11, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm21
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm7, %xmm12
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %xmm9
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %xmm6
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm6[0],xmm9[0],xmm6[1],xmm9[1],xmm6[2],xmm9[2],xmm6[3],xmm9[3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm4, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm5[1,2,2,3,5,6,6,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7,8,9],ymm0[10],ymm1[11,12],ymm0[13],ymm1[14,15]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,18,19,20,21,18,19,20,21]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm5[0,0,2,1,4,4,6,5]
+; AVX512F-FAST-NEXT:    vmovdqa %ymm5, %ymm6
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm1[0,1,2],ymm2[3],ymm1[4,5],ymm2[6],ymm1[7,8,9,10],ymm2[11],ymm1[12,13],ymm2[14],ymm1[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm26, %ymm9
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm9[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpermd %ymm1, %ymm29, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [2,2,3,3,10,9,11,10]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm0, %zmm1, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqu (%rsp), %ymm5 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm7 = ymm5[4],ymm1[4],ymm5[5],ymm1[5],ymm5[6],ymm1[6],ymm5[7],ymm1[7],ymm5[12],ymm1[12],ymm5[13],ymm1[13],ymm5[14],ymm1[14],ymm5[15],ymm1[15]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm7[0,1,3,2,4,5,6,7,8,9,11,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm29, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,14,15,12,13,u,u,u,u,30,31,28,29,u,u,u,u,30,31,28,29]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm4[0,1,2],ymm1[3],ymm4[4,5],ymm1[6],ymm4[7,8,9,10],ymm1[11],ymm4[12,13],ymm1[14],ymm4[15]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,18,19,20,21,24,25,26,27,22,23,22,23]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[1,1,1,1,5,5,5,5]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm5[2],ymm0[3,4],ymm5[5],ymm0[6,7,8,9],ymm5[10],ymm0[11,12],ymm5[13],ymm0[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm26
+; AVX512F-FAST-NEXT:    vpshufhw $180, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm0 = mem[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [2,3,6,7,6,7,7,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm5, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm22, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm3, %ymm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm4 = ymm10[3,3,3,3,7,7,7,7]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1],ymm4[2],ymm1[3,4],ymm4[5],ymm1[6,7,8,9],ymm4[10],ymm1[11,12],ymm4[13],ymm1[14,15]
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %xmm7
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %xmm8
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm8[0],xmm7[0],xmm8[1],xmm7[1],xmm8[2],xmm7[2],xmm8[3],xmm7[3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm4, %xmm11
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [2,2,2,3,8,8,8,9]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm11, %zmm0, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [6,7,3,3,7,7,6,7]
-; AVX512F-FAST-NEXT:    vpermd %ymm19, %ymm0, %ymm2
-; AVX512F-FAST-NEXT:    vpbroadcastd 32(%rax), %ymm11
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm2, %zmm22
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm22
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm8[4],xmm7[4],xmm8[5],xmm7[5],xmm8[6],xmm7[6],xmm8[7],xmm7[7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm30, %xmm0
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vpshufhw $180, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm0 = mem[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm16, %ymm4
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm4[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm28, %zmm15, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = <u,u,4,u,u,u,5,u,u,13,u,u,u,14,u,u>
+; AVX512F-FAST-NEXT:    vpermd %zmm0, %zmm1, %zmm25
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm25
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm1[0,0,1,1,2,2,3,3,8,8,9,9,10,10,11,11]
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[0,1,1,3,4,5,5,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm2[0,1],ymm0[2],ymm2[3,4],ymm0[5],ymm2[6,7,8,9],ymm0[10],ymm2[11,12],ymm0[13],ymm2[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm24
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,u,u,24,25,26,27,u,u,26,27,26,27]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm4[2,2,2,2,6,6,6,6]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm2[2],ymm0[3,4],ymm2[5],ymm0[6,7,8,9],ymm2[10],ymm0[11,12],ymm2[13],ymm0[14,15]
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm15, %xmm8
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm5, %xmm11
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %xmm7
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm5 = xmm7[0],xmm0[0],xmm7[1],xmm0[1],xmm7[2],xmm0[2],xmm7[3],xmm0[3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm5, %xmm3
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm0 = ymm4[4],ymm1[4],ymm4[5],ymm1[5],ymm4[6],ymm1[6],ymm4[7],ymm1[7],ymm4[12],ymm1[12],ymm4[13],ymm1[13],ymm4[14],ymm1[14],ymm4[15],ymm1[15]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm5, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,28,29,26,27,28,29,26,27,28,29,30,31,30,31]
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} ymm2 = ymm6[3,3,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm2[2],ymm0[3,4],ymm2[5],ymm0[6,7,8,9],ymm2[10],ymm0[11,12],ymm2[13],ymm0[14,15]
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %xmm8
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %xmm2
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm2[0],xmm8[0],xmm2[1],xmm8[1],xmm2[2],xmm8[2],xmm2[3],xmm8[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm6, %xmm5
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [2,2,2,3,8,8,8,9]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm5, %zmm1, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = [0,1,2,3,0,1,2,3,6,7,4,5,6,7,4,5]
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm14, %xmm3
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm29, %xmm2
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm4, %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [0,1,1,3,8,8,9,9]
-; AVX512F-FAST-NEXT:    vpermt2q %zmm1, %zmm2, %zmm4
-; AVX512F-FAST-NEXT:    vpbroadcastd 36(%rax), %ymm1
-; AVX512F-FAST-NEXT:    vpbroadcastd 40(%rax), %ymm5
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm1, %zmm10
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm10
-; AVX512F-FAST-NEXT:    vprold $16, %xmm0, %xmm1
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm7[1,1,2,3]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm4[0,1],xmm1[2],xmm4[3,4],xmm1[5],xmm4[6,7]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm20, %xmm3
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm3, %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm3, %ymm17
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm10, %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm3, %ymm18
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %xmm4
+; AVX512F-FAST-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm12 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm12, %xmm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm17, %xmm15
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm15, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm28, %xmm2
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm23, %xmm3
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm3, %xmm3
-; AVX512F-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm9[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
-; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm6[1,1,2,2]
-; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm4[0],xmm0[1],xmm4[2,3],xmm0[4],xmm4[5,6],xmm0[7]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm23 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm23 = mem[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm4 = mem[2,1,3,3]
-; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm3 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm0 = mem[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm6[4],xmm9[4],xmm6[5],xmm9[5],xmm6[6],xmm9[6],xmm6[7],xmm9[7]
-; AVX512F-FAST-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm9 = ymm5[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[3,3,3,3]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm9[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [4,6,3,3,7,7,7,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm1, %ymm12, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm7[0,1,2,3,4,5,7,6,8,9,10,11,12,13,15,14]
+; AVX512F-FAST-NEXT:    vpermd %ymm1, %ymm12, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm20 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm20 = mem[2,2,2,3]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [6,7,3,3,7,7,6,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm28, %ymm1, %ymm1
+; AVX512F-FAST-NEXT:    vpbroadcastd 32(%rax), %ymm7
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm1, %zmm16
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm16
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm8[4],xmm2[5],xmm8[5],xmm2[6],xmm8[6],xmm2[7],xmm8[7]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm30, %xmm1
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm11, %xmm4
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm13, %xmm5
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %xmm1
+; AVX512F-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm8 = xmm3[0],xmm1[0],xmm3[1],xmm1[1],xmm3[2],xmm1[2],xmm3[3],xmm1[3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm8, %xmm7
+; AVX512F-FAST-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm2, %xmm8
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm0[0,1,2,3,5,5,7,6,8,9,10,11,13,13,15,14]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[3,3,3,3]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,u,u,u,u,u,u,u,u]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,2,2,2]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm31, %xmm5
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm14 = xmm5[0,2,3,3,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,3]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm16, %xmm5
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm13 = xmm5[0,2,3,3,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,3]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm23, %xmm13
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm12 = xmm13[0,2,3,3,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm6[0,1,2,3,8,9,10,11,14,15,12,13,14,15,12,13]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm7 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm22, %xmm11
+; AVX512F-FAST-NEXT:    vpshufb %xmm7, %xmm11, %xmm14
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,1,3]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm21, %xmm6
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm6[0,2,3,3,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm19, %xmm6
+; AVX512F-FAST-NEXT:    vpshufb %xmm7, %xmm6, %xmm10
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,1,3]
 ; AVX512F-FAST-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Folded Reload
 ; AVX512F-FAST-NEXT:    # ymm6 = mem[2,1,3,2]
-; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm7 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermpd $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm5 = mem[0,2,2,3]
-; AVX512F-FAST-NEXT:    vmovups %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpermpd $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm5 = mem[2,1,3,3]
-; AVX512F-FAST-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpermpd $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm5 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vmovups %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpermpd $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm5 = mem[0,2,2,3]
-; AVX512F-FAST-NEXT:    vmovups %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm31 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm31 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm30 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm30 = mem[0,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm9 = mem[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm30 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm30 = mem[2,2,2,3]
 ; AVX512F-FAST-NEXT:    vpermq $182, {{[-0-9]+}}(%r{{[sb]}}p), %ymm29 # 32-byte Folded Reload
 ; AVX512F-FAST-NEXT:    # ymm29 = mem[2,1,3,2]
 ; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm28 # 32-byte Folded Reload
 ; AVX512F-FAST-NEXT:    # ymm28 = mem[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm19 = ymm21[0,1,1,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm18 = ymm18[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm17 = ymm8[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm8 = mem[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm21 = ymm12[0,1,1,3]
-; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm12 = mem[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm16 = ymm11[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm27[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm26[2,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm26 = ymm25[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm25 = ymm24[2,1,3,2]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm24 = ymm20[2,2,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm23, %zmm4
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm4, %zmm3, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %ymm20 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm20
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm20, %ymm15
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm4 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm23 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm4, %zmm23, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $226, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm1 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm13, %zmm13 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm13, %zmm23, %zmm2
-; AVX512F-FAST-NEXT:    vpternlogq $226, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm2 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm4, %zmm4
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm23 = ymm17[0,1,1,3]
+; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm19 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm19 = mem[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm21 = ymm4[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm4 = mem[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm22 = ymm18[0,1,1,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm17 = ymm31[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm18 = ymm5[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm27[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm26 = ymm26[2,2,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm24[2,1,3,2]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [0,1,1,3,8,8,9,9]
+; AVX512F-FAST-NEXT:    vpermt2q %zmm8, %zmm7, %zmm0
+; AVX512F-FAST-NEXT:    vpbroadcastd 36(%rax), %ymm8
+; AVX512F-FAST-NEXT:    vpbroadcastd 40(%rax), %ymm31
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm31, %zmm8, %zmm8
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
+; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm31 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm31 = mem[2,2,2,3]
+; AVX512F-FAST-NEXT:    vprold $16, %xmm1, %xmm0
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm13 = xmm3[1,1,2,3]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm13[0,1],xmm0[2],xmm13[3,4],xmm0[5],xmm13[6,7]
+; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm20, %zmm13 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [0,1,4,5,4,5,5,7,12,14,11,11,15,15,15,15]
+; AVX512F-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm20 = [65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm20, %zmm13
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %ymm24 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm13, %ymm24
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm24, %ymm15
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm12, %zmm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm14, %zmm12 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm24, %zmm12
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $226, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm12 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm11 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm10, %zmm10 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm11, %zmm24, %zmm10
+; AVX512F-FAST-NEXT:    vpternlogq $226, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm10 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm6 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm6, %zmm6
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm6
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm20 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm20 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm5 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogd $226, 124(%r8){1to8}, %ymm4, %ymm0
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm9
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm6, %zmm6
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm6
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm14 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm14 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm11 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm11 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vextracti64x4 $1, %zmm13, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogd $226, 124(%r8){1to8}, %ymm7, %ymm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm15[0,1,2,3],zmm0[4,5,6,7]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0]
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm0, %zmm4, %zmm9
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0,65535,65535,65535,65535,65535,65535,0]
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm0, %zmm2, %zmm7
 ; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm13 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm13 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm18, %zmm19, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm17, %zmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm7, %zmm4
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm21, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm16, %zmm8
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm7, %zmm8
+; AVX512F-FAST-NEXT:    vpternlogq $184, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm13 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm19, %zmm23, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm21, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm4, %zmm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm17, %zmm22, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm18, %zmm5
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm4, %zmm5
 ; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535]
 ; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm21 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm4, %zmm0, %zmm21
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm17 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm8, %zmm0, %zmm17
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm30, %zmm31, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm28, %zmm29, %zmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm7, %zmm4
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm26, %zmm27, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm24, %zmm25, %zmm8
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm7, %zmm8
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm2, %zmm0, %zmm21
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm22 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm5, %zmm0, %zmm22
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm30, %zmm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm28, %zmm29, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535,65535,65535,0,0,65535,65535,65535]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm4, %zmm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm26, %zmm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm31, %zmm27, %zmm5
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm4, %zmm5
 ; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0]
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm26 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm4, %zmm0, %zmm26
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm29 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm8, %zmm0, %zmm29
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm3, %zmm4
-; AVX512F-FAST-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm0 = mem[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm7 = mem[0,1,1,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm7 = mem[2,1,3,3]
-; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm8 = mem[0,0,1,1]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm7, %zmm7
-; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm8 = mem[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq $208, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm27 = mem[0,0,1,3]
-; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm24 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm24 = mem[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
-; AVX512F-FAST-NEXT:    # xmm11 = mem[0,2,3,3,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq $232, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm12 = mem[0,2,2,3]
-; AVX512F-FAST-NEXT:    vpermq $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm25 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm25 = mem[0,1,1,3]
-; AVX512F-FAST-NEXT:    vpermq $246, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm15 = mem[2,1,3,3]
-; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm16 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm16 = mem[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm28 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm28 = mem[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpermq $208, {{[-0-9]+}}(%r{{[sb]}}p), %ymm18 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm18 = mem[0,0,1,3]
-; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm19 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm19 = mem[0,0,1,1]
-; AVX512F-FAST-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Folded Reload
-; AVX512F-FAST-NEXT:    # xmm14 = mem[0,2,3,3,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,2,1]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm3, %zmm7
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm25, %zmm12, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm12
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm3, %zmm12
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535]
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm23 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm2, %zmm0, %zmm23
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm5, %zmm0, %zmm25
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [4,5,2,3,4,5,6,7,8,9,10,11,10,11,8,9]
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm4[u,u,4,5,u,u,u,u,6,7,u,u,u,u,8,9]
+; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; AVX512F-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm5[1,1,2,2]
+; AVX512F-FAST-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0],xmm2[1],xmm3[2,3],xmm2[4],xmm3[5,6],xmm2[7]
+; AVX512F-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,1,4,5,4,5,5,7,12,14,11,11,15,15,15,15]
+; AVX512F-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %zmm4, %zmm4 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    vpermq $234, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm5 = mem[2,2,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, (%rsp), %zmm5, %zmm5 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm4, %zmm20, %zmm5
+; AVX512F-FAST-NEXT:    vpermq $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm4 = mem[0,1,1,3]
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm9, %zmm4
+; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm9 = mem[0,0,1,1]
 ; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm15 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm7, %zmm0, %zmm15
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm12, %zmm0, %zmm22
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm27, %zmm8, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm24, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm23, %zmm3
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm18, %zmm28, %zmm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm19, %zmm7
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm23, %zmm7
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm15, %zmm9
+; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm26 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm26 = mem[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,1,3]
+; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm27 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm27 = mem[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpshuflw $248, {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
+; AVX512F-FAST-NEXT:    # xmm15 = mem[0,2,3,3,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq $212, {{[-0-9]+}}(%r{{[sb]}}p), %ymm17 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm17 = mem[0,1,1,3]
+; AVX512F-FAST-NEXT:    vpermq $80, {{[-0-9]+}}(%r{{[sb]}}p), %ymm18 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm18 = mem[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpermq $96, {{[-0-9]+}}(%r{{[sb]}}p), %ymm19 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm19 = mem[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,1,1]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[0,2,3,3,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,0,2,1]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm4, %zmm20, %zmm9
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm17, %zmm4, %zmm4
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm17 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm18, %zmm17, %zmm17
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm4, %zmm20, %zmm17
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535]
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm18 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm9, %zmm4, %zmm18
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm17, %zmm4, %zmm16
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm26, %zmm1
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm15, %zmm27, %zmm4
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm1, %zmm24, %zmm4
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm19, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm1
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm24, %zmm1
 ; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0,0,65535,65535,65535,65535,0,0]
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm3, %zmm0, %zmm8
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm7, %zmm0, %zmm10
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm5
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm13
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm4, %zmm0, %zmm2
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm1, %zmm0, %zmm8
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm11
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm13
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm5, 320(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm10, 256(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm22, 192(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm29, 128(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm2, 64(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm17, (%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm11, 320(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm8, 256(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm16, 192(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm25, 128(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm10, 64(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm22, (%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm21, 448(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm8, 704(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm15, 640(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm26, 576(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm1, 512(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm2, 704(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm18, 640(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm23, 576(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm12, 512(%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm13, 384(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm20, 768(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm9, 832(%rax)
-; AVX512F-FAST-NEXT:    addq $2200, %rsp # imm = 0x898
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm14, 768(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm7, 832(%rax)
+; AVX512F-FAST-NEXT:    addq $2168, %rsp # imm = 0x878
 ; AVX512F-FAST-NEXT:    vzeroupper
 ; AVX512F-FAST-NEXT:    retq
 ;
 ; AVX512BW-LABEL: store_i16_stride7_vf64:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    subq $136, %rsp
 ; AVX512BW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm2

llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-5.ll

 ;
 ; AVX1-ONLY-LABEL: store_i32_stride5_vf4:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    vmovaps (%rdi), %xmm0
 ; AVX1-ONLY-NEXT:    vmovaps (%rsi), %xmm1
 ; AVX1-ONLY-NEXT:    vmovaps (%rdx), %xmm2
 ; AVX1-ONLY-NEXT:    vmovaps (%rcx), %xmm3
 ; AVX1-ONLY-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm4
-; AVX1-ONLY-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm5
-; AVX1-ONLY-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm6
-; AVX1-ONLY-NEXT:    vunpcklps {{.*#+}} ymm5 = ymm5[0],ymm6[0],ymm5[1],ymm6[1],ymm5[4],ymm6[4],ymm5[5],ymm6[5]
-; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm5 = ymm5[u,u,1,0,u,u,u,6]
 ; AVX1-ONLY-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm1
-; AVX1-ONLY-NEXT:    vunpcklps {{.*#+}} ymm7 = ymm4[0],ymm1[0],ymm4[1],ymm1[1],ymm4[4],ymm1[4],ymm4[5],ymm1[5]
-; AVX1-ONLY-NEXT:    vpermilpd {{.*#+}} ymm7 = ymm7[0,0,3,3]
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1],ymm5[2,3],ymm7[4,5,6],ymm5[7]
+; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm5 = ymm1[u,u,u,2,u,u,u,7]
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0,1],ymm4[2],ymm5[3,4,5,6,7]
+; AVX1-ONLY-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm6
+; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm7 = ymm6[1,u,u,u,6,u,u,u]
+; AVX1-ONLY-NEXT:    vbroadcastss 8(%rcx), %ymm8
+; AVX1-ONLY-NEXT:    vunpcklps {{.*#+}} ymm7 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[4],ymm8[4],ymm7[5],ymm8[5]
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm5 = ymm7[0,1],ymm5[2,3],ymm7[4,5],ymm5[6,7]
 ; AVX1-ONLY-NEXT:    vbroadcastf128 {{.*#+}} ymm7 = mem[0,1,0,1]
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm5 = ymm5[0],ymm7[1],ymm5[2,3,4,5],ymm7[6],ymm5[7]
+; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[u,0,u,u,u,5,u,u]
+; AVX1-ONLY-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[1,1],ymm4[1,1],ymm1[5,5],ymm4[5,5]
+; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm4 = ymm6[u,u,u,0,u,u,u,5]
+; AVX1-ONLY-NEXT:    vbroadcastss (%rdx), %ymm6
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm6[2],ymm4[3,4,5,6,7]
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3],ymm1[4,5,6],ymm4[7]
 ; AVX1-ONLY-NEXT:    vinsertf128 $1, %xmm7, %ymm0, %ymm0
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm5[1,2,3],ymm0[4],ymm5[5,6,7]
-; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[u,u,u,2,u,u,u,7]
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0,1],ymm4[2],ymm1[3,4,5,6,7]
-; AVX1-ONLY-NEXT:    vpermilps {{.*#+}} ymm4 = ymm6[1,u,u,u,6,u,u,u]
-; AVX1-ONLY-NEXT:    vbroadcastss 8(%rcx), %ymm5
-; AVX1-ONLY-NEXT:    vunpcklps {{.*#+}} ymm4 = ymm4[0],ymm5[0],ymm4[1],ymm5[1],ymm4[4],ymm5[4],ymm4[5],ymm5[5]
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm1 = ymm4[0,1],ymm1[2,3],ymm4[4,5],ymm1[6,7]
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm1 = ymm1[0],ymm7[1],ymm1[2,3,4,5],ymm7[6],ymm1[7]
-; AVX1-ONLY-NEXT:    vshufps {{.*#+}} xmm2 = xmm2[3,3],xmm3[3,3]
-; AVX1-ONLY-NEXT:    vbroadcastss 12(%rsi), %xmm3
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} xmm2 = xmm3[0],xmm2[1,2,3]
-; AVX1-ONLY-NEXT:    vblendps {{.*#+}} xmm2 = xmm2[0,1,2],xmm7[3]
-; AVX1-ONLY-NEXT:    vmovaps %xmm2, 64(%r9)
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX1-ONLY-NEXT:    vshufps {{.*#+}} xmm1 = xmm2[3,3],xmm3[3,3]
+; AVX1-ONLY-NEXT:    vbroadcastss 12(%rsi), %xmm2
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} xmm1 = xmm2[0],xmm1[1,2,3]
+; AVX1-ONLY-NEXT:    vblendps {{.*#+}} xmm1 = xmm1[0,1,2],xmm7[3]
+; AVX1-ONLY-NEXT:    vmovaps %xmm1, 64(%r9)
 ; AVX1-ONLY-NEXT:    vmovaps %ymm0, (%r9)
-; AVX1-ONLY-NEXT:    vmovaps %ymm1, 32(%r9)
+; AVX1-ONLY-NEXT:    vmovaps %ymm5, 32(%r9)
 ; AVX1-ONLY-NEXT:    vzeroupper
 ; AVX1-ONLY-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: store_i32_stride5_vf4:
 ; AVX2-SLOW:       # %bb.0:
 ; AVX2-SLOW-NEXT:    vmovaps (%rdi), %xmm0
 ; AVX2-SLOW-NEXT:    vmovaps (%rsi), %xmm1
 ; AVX2-SLOW-NEXT:    vmovaps (%rdx), %xmm2

llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-3.ll

; AVX512BW-NEXT:    retq
   %interleaved.vec = shufflevector <24 x i8> %3, <24 x i8> poison, <24 x i32> <i32 0, i32 8, i32 16, i32 1, i32 9, i32 17, i32 2, i32 10, i32 18, i32 3, i32 11, i32 19, i32 4, i32 12, i32 20, i32 5, i32 13, i32 21, i32 6, i32 14, i32 22, i32 7, i32 15, i32 23>
   store <24 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride3_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride3_vf16:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa (%rdi), %xmm2
+; SSE-NEXT:    movdqa (%rdi), %xmm3
 ; SSE-NEXT:    movdqa (%rsi), %xmm4
 ; SSE-NEXT:    movdqa (%rdx), %xmm1
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm0[0,1,2,3,4,4,6,5]
 ; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,255]
 ; SSE-NEXT:    movdqa %xmm0, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm4[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm3[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pandn %xmm2, %xmm5
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm2[0,1,2,3,4,5,5,6]
 ; SSE-NEXT:    pand %xmm0, %xmm6
 ; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255]
-; SSE-NEXT:    pand %xmm5, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm5, %xmm3
-; SSE-NEXT:    pandn %xmm7, %xmm3
-; SSE-NEXT:    por %xmm6, %xmm3
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm4[0],xmm6[1],xmm4[1],xmm6[2],xmm4[2],xmm6[3],xmm4[3],xmm6[4],xmm4[4],xmm6[5],xmm4[5],xmm6[6],xmm4[6],xmm6[7],xmm4[7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm5, %xmm7
-; SSE-NEXT:    pandn %xmm6, %xmm5
-; SSE-NEXT:    por %xmm7, %xmm5
-; SSE-NEXT:    pand %xmm0, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    movdqa %xmm0, %xmm7
-; SSE-NEXT:    pandn %xmm6, %xmm7
-; SSE-NEXT:    por %xmm5, %xmm7
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    pandn %xmm4, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
-; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm2, %xmm6
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm1[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm5, %xmm2
+; SSE-NEXT:    por %xmm6, %xmm2
+; SSE-NEXT:    movdqa %xmm3, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3],xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
+; SSE-NEXT:    pxor %xmm6, %xmm6
+; SSE-NEXT:    movdqa %xmm5, %xmm7
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm6[8],xmm7[9],xmm6[9],xmm7[10],xmm6[10],xmm7[11],xmm6[11],xmm7[12],xmm6[12],xmm7[13],xmm6[13],xmm7[14],xmm6[14],xmm7[15],xmm6[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3],xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm5[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[2,3],xmm7[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm8[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm5, %xmm7
+; SSE-NEXT:    pand %xmm0, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm1[0,1,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,5,6,6]
+; SSE-NEXT:    pandn %xmm5, %xmm0
+; SSE-NEXT:    por %xmm7, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm6[8],xmm4[9],xmm6[9],xmm4[10],xmm6[10],xmm4[11],xmm6[11],xmm4[12],xmm6[12],xmm4[13],xmm6[13],xmm4[14],xmm6[14],xmm4[15],xmm6[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm6[0],xmm3[1],xmm6[1],xmm3[2],xmm6[2],xmm3[3],xmm6[3],xmm3[4],xmm6[4],xmm3[5],xmm6[5],xmm3[6],xmm6[6],xmm3[7],xmm6[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[2,3],xmm4[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm4, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
+; SSE-NEXT:    pand %xmm4, %xmm3
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,1,2,2,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm2, 32(%rcx)
-; SSE-NEXT:    movdqa %xmm7, (%rcx)
-; SSE-NEXT:    movdqa %xmm3, 16(%rcx)
+; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    por %xmm3, %xmm4
+; SSE-NEXT:    movdqa %xmm4, 32(%rcx)
+; SSE-NEXT:    movdqa %xmm0, (%rcx)
+; SSE-NEXT:    movdqa %xmm2, 16(%rcx)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: store_i8_stride3_vf16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX1-NEXT:    vmovdqa (%rsi), %xmm1
 ; AVX1-NEXT:    vmovdqa (%rdx), %xmm2
 ; AVX1-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15,0,1,2,3,4,5]
; AVX512-NEXT:    retq
   store <48 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride3_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride3_vf32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa (%rdi), %xmm2
-; SSE-NEXT:    movdqa 16(%rdi), %xmm8
-; SSE-NEXT:    movdqa (%rsi), %xmm4
-; SSE-NEXT:    movdqa 16(%rsi), %xmm10
+; SSE-NEXT:    movdqa 16(%rdi), %xmm7
+; SSE-NEXT:    movdqa (%rsi), %xmm5
+; SSE-NEXT:    movdqa 16(%rsi), %xmm11
 ; SSE-NEXT:    movdqa (%rdx), %xmm1
-; SSE-NEXT:    movdqa 16(%rdx), %xmm7
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa 16(%rdx), %xmm6
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm7[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,4,4,6,5]
 ; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm10[2,1,2,3]
+; SSE-NEXT:    movdqa %xmm0, %xmm4
+; SSE-NEXT:    pandn %xmm3, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm11[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm3[0,1,2,3,4,5,5,6]
-; SSE-NEXT:    pand %xmm0, %xmm6
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255]
-; SSE-NEXT:    pand %xmm5, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm7[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm5, %xmm3
-; SSE-NEXT:    pandn %xmm9, %xmm3
-; SSE-NEXT:    por %xmm6, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm2[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm3[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pand %xmm0, %xmm8
+; SSE-NEXT:    por %xmm4, %xmm8
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm6[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm4[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    pandn %xmm9, %xmm4
+; SSE-NEXT:    por %xmm8, %xmm4
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,4,6,5]
 ; SSE-NEXT:    movdqa %xmm0, %xmm9
-; SSE-NEXT:    pandn %xmm6, %xmm9
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm4[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm6[0,1,2,3,4,5,5,6]
-; SSE-NEXT:    pand %xmm0, %xmm11
-; SSE-NEXT:    por %xmm9, %xmm11
-; SSE-NEXT:    pand %xmm5, %xmm11
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm1[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm6[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm5, %xmm6
-; SSE-NEXT:    pandn %xmm9, %xmm6
-; SSE-NEXT:    por %xmm11, %xmm6
-; SSE-NEXT:    movdqa %xmm10, %xmm9
+; SSE-NEXT:    pandn %xmm8, %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm5[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,3,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pand %xmm0, %xmm8
+; SSE-NEXT:    por %xmm9, %xmm8
+; SSE-NEXT:    pand %xmm3, %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm1[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm9, %xmm3
+; SSE-NEXT:    por %xmm8, %xmm3
+; SSE-NEXT:    movdqa %xmm7, %xmm8
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm11[8],xmm8[9],xmm11[9],xmm8[10],xmm11[10],xmm8[11],xmm11[11],xmm8[12],xmm11[12],xmm8[13],xmm11[13],xmm8[14],xmm11[14],xmm8[15],xmm11[15]
+; SSE-NEXT:    pxor %xmm10, %xmm10
+; SSE-NEXT:    movdqa %xmm8, %xmm9
 ; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm10[8],xmm9[9],xmm10[9],xmm9[10],xmm10[10],xmm9[11],xmm10[11],xmm9[12],xmm10[12],xmm9[13],xmm10[13],xmm9[14],xmm10[14],xmm9[15],xmm10[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm0, %xmm11
-; SSE-NEXT:    pandn %xmm9, %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm8[2,3,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm10[0],xmm8[1],xmm10[1],xmm8[2],xmm10[2],xmm8[3],xmm10[3],xmm8[4],xmm10[4],xmm8[5],xmm10[5],xmm8[6],xmm10[6],xmm8[7],xmm10[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[2,3],xmm9[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm8[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm9[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm8, %xmm12
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
+; SSE-NEXT:    pand %xmm8, %xmm12
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm6[2,3,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm9[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm0, %xmm12
-; SSE-NEXT:    por %xmm11, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
-; SSE-NEXT:    pand %xmm9, %xmm12
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm7[2,3,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm9[0,1,2,3,6,5,7,7]
+; SSE-NEXT:    movdqa %xmm8, %xmm9
+; SSE-NEXT:    pandn %xmm13, %xmm9
+; SSE-NEXT:    por %xmm12, %xmm9
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm11[0],xmm7[1],xmm11[1],xmm7[2],xmm11[2],xmm7[3],xmm11[3],xmm7[4],xmm11[4],xmm7[5],xmm11[5],xmm7[6],xmm11[6],xmm7[7],xmm11[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm10[8],xmm11[9],xmm10[9],xmm11[10],xmm10[10],xmm11[11],xmm10[11],xmm11[12],xmm10[12],xmm11[13],xmm10[13],xmm11[14],xmm10[14],xmm11[15],xmm10[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[3],xmm10[3],xmm7[4],xmm10[4],xmm7[5],xmm10[5],xmm7[6],xmm10[6],xmm7[7],xmm10[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm7[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[2,3],xmm11[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm12[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm7, %xmm11
+; SSE-NEXT:    pand %xmm0, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,1,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm6[0,1,2,3,5,5,6,6]
+; SSE-NEXT:    movdqa %xmm0, %xmm6
+; SSE-NEXT:    pandn %xmm7, %xmm6
+; SSE-NEXT:    por %xmm11, %xmm6
+; SSE-NEXT:    movdqa %xmm2, %xmm7
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm5[8],xmm7[9],xmm5[9],xmm7[10],xmm5[10],xmm7[11],xmm5[11],xmm7[12],xmm5[12],xmm7[13],xmm5[13],xmm7[14],xmm5[14],xmm7[15],xmm5[15]
+; SSE-NEXT:    movdqa %xmm7, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm10[8],xmm11[9],xmm10[9],xmm11[10],xmm10[10],xmm11[11],xmm10[11],xmm11[12],xmm10[12],xmm11[13],xmm10[13],xmm11[14],xmm10[14],xmm11[15],xmm10[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm10[0],xmm7[1],xmm10[1],xmm7[2],xmm10[2],xmm7[3],xmm10[3],xmm7[4],xmm10[4],xmm7[5],xmm10[5],xmm7[6],xmm10[6],xmm7[7],xmm10[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[2,3],xmm11[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm11, %xmm7
+; SSE-NEXT:    pand %xmm8, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm1[2,3,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm11[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    movdqa %xmm9, %xmm11
-; SSE-NEXT:    pandn %xmm13, %xmm11
-; SSE-NEXT:    por %xmm12, %xmm11
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm10[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm5, %xmm12
-; SSE-NEXT:    pandn %xmm10, %xmm12
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm5, %xmm8
-; SSE-NEXT:    por %xmm12, %xmm8
-; SSE-NEXT:    pand %xmm0, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm7[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    movdqa %xmm0, %xmm7
-; SSE-NEXT:    pandn %xmm10, %xmm7
-; SSE-NEXT:    por %xmm8, %xmm7
-; SSE-NEXT:    movdqa %xmm4, %xmm8
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm4[8],xmm8[9],xmm4[9],xmm8[10],xmm4[10],xmm8[11],xmm4[11],xmm8[12],xmm4[12],xmm8[13],xmm4[13],xmm8[14],xmm4[14],xmm8[15],xmm4[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm0, %xmm10
-; SSE-NEXT:    pandn %xmm8, %xmm10
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm2[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm0, %xmm8
-; SSE-NEXT:    por %xmm10, %xmm8
-; SSE-NEXT:    pand %xmm9, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm1[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm10[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pandn %xmm10, %xmm9
-; SSE-NEXT:    por %xmm8, %xmm9
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm5, %xmm2
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    por %xmm2, %xmm5
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,6,5,7,7]
+; SSE-NEXT:    pandn %xmm11, %xmm8
+; SSE-NEXT:    por %xmm7, %xmm8
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm5[0],xmm2[1],xmm5[1],xmm2[2],xmm5[2],xmm2[3],xmm5[3],xmm2[4],xmm5[4],xmm2[5],xmm5[5],xmm2[6],xmm5[6],xmm2[7],xmm5[7]
+; SSE-NEXT:    movdqa %xmm2, %xmm5
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm10[8],xmm5[9],xmm10[9],xmm5[10],xmm10[10],xmm5[11],xmm10[11],xmm5[12],xmm10[12],xmm5[13],xmm10[13],xmm5[14],xmm10[14],xmm5[15],xmm10[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,3],xmm5[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm7[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm2, %xmm5
 ; SSE-NEXT:    pand %xmm0, %xmm5
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,6]
 ; SSE-NEXT:    pandn %xmm1, %xmm0
 ; SSE-NEXT:    por %xmm5, %xmm0
 ; SSE-NEXT:    movdqa %xmm0, (%rcx)
-; SSE-NEXT:    movdqa %xmm9, 32(%rcx)
-; SSE-NEXT:    movdqa %xmm7, 48(%rcx)
-; SSE-NEXT:    movdqa %xmm11, 80(%rcx)
-; SSE-NEXT:    movdqa %xmm6, 16(%rcx)
-; SSE-NEXT:    movdqa %xmm3, 64(%rcx)
+; SSE-NEXT:    movdqa %xmm8, 32(%rcx)
+; SSE-NEXT:    movdqa %xmm6, 48(%rcx)
+; SSE-NEXT:    movdqa %xmm9, 80(%rcx)
+; SSE-NEXT:    movdqa %xmm3, 16(%rcx)
+; SSE-NEXT:    movdqa %xmm4, 64(%rcx)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride3_vf32:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX1-ONLY-NEXT:    vmovdqa 16(%rdi), %xmm1
 ; AVX1-ONLY-NEXT:    vpalignr {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15,0,1,2,3,4,5]
 ; AVX1-ONLY-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15,0,1,2,3,4,5]
; AVX512BW-NEXT:    retq
   %interleaved.vec = shufflevector <96 x i8> %3, <96 x i8> poison, <96 x i32> <i32 0, i32 32, i32 64, i32 1, i32 33, i32 65, i32 2, i32 34, i32 66, i32 3, i32 35, i32 67, i32 4, i32 36, i32 68, i32 5, i32 37, i32 69, i32 6, i32 38, i32 70, i32 7, i32 39, i32 71, i32 8, i32 40, i32 72, i32 9, i32 41, i32 73, i32 10, i32 42, i32 74, i32 11, i32 43, i32 75, i32 12, i32 44, i32 76, i32 13, i32 45, i32 77, i32 14, i32 46, i32 78, i32 15, i32 47, i32 79, i32 16, i32 48, i32 80, i32 17, i32 49, i32 81, i32 18, i32 50, i32 82, i32 19, i32 51, i32 83, i32 20, i32 52, i32 84, i32 21, i32 53, i32 85, i32 22, i32 54, i32 86, i32 23, i32 55, i32 87, i32 24, i32 56, i32 88, i32 25, i32 57, i32 89, i32 26, i32 58, i32 90, i32 27, i32 59, i32 91, i32 28, i32 60, i32 92, i32 29, i32 61, i32 93, i32 30, i32 62, i32 94, i32 31, i32 63, i32 95>
   store <96 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride3_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride3_vf64:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa 16(%rdi), %xmm10
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 16(%rdi), %xmm5
 ; SSE-NEXT:    movdqa 32(%rdi), %xmm12
-; SSE-NEXT:    movdqa 48(%rdi), %xmm5
-; SSE-NEXT:    movdqa 16(%rsi), %xmm9
-; SSE-NEXT:    movdqa 32(%rsi), %xmm15
-; SSE-NEXT:    movdqa 48(%rsi), %xmm4
-; SSE-NEXT:    movdqa 16(%rdx), %xmm8
+; SSE-NEXT:    movdqa 48(%rdi), %xmm2
+; SSE-NEXT:    movdqa 16(%rsi), %xmm13
+; SSE-NEXT:    movdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 32(%rsi), %xmm10
+; SSE-NEXT:    movdqa 48(%rsi), %xmm3
+; SSE-NEXT:    movdqa 16(%rdx), %xmm14
+; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 32(%rdx), %xmm8
 ; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa 32(%rdx), %xmm11
-; SSE-NEXT:    movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa 48(%rdx), %xmm7
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm0[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm2[0,1,2,3,4,5,5,6]
-; SSE-NEXT:    pand %xmm1, %xmm6
-; SSE-NEXT:    por %xmm3, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255]
-; SSE-NEXT:    pand %xmm2, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm7[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm2, %xmm0
-; SSE-NEXT:    pandn %xmm3, %xmm0
-; SSE-NEXT:    por %xmm6, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm12[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm15[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,5,6]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm6, %xmm3
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm11[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa 48(%rdx), %xmm9
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm6
+; SSE-NEXT:    pandn %xmm1, %xmm6
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm1[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pand %xmm4, %xmm7
+; SSE-NEXT:    por %xmm6, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm11, %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm9[2,1,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm2, %xmm0
+; SSE-NEXT:    movdqa %xmm11, %xmm0
 ; SSE-NEXT:    pandn %xmm6, %xmm0
-; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    por %xmm7, %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm10[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm9[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,5,6]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm6, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm8[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm12[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    movdqa %xmm4, %xmm7
+; SSE-NEXT:    pandn %xmm6, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm10[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pand %xmm4, %xmm6
+; SSE-NEXT:    por %xmm7, %xmm6
+; SSE-NEXT:    pand %xmm11, %xmm6
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm8[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm7[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm11, %xmm0
+; SSE-NEXT:    pandn %xmm8, %xmm0
+; SSE-NEXT:    por %xmm6, %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm5[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    movdqa %xmm4, %xmm8
+; SSE-NEXT:    pandn %xmm6, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm13[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm6[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pand %xmm4, %xmm13
+; SSE-NEXT:    por %xmm8, %xmm13
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm14[2,1,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm6[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm2, %xmm13
-; SSE-NEXT:    pandn %xmm8, %xmm13
-; SSE-NEXT:    movdqa (%rdi), %xmm0
+; SSE-NEXT:    movdqa %xmm11, %xmm7
+; SSE-NEXT:    pandn %xmm8, %xmm7
+; SSE-NEXT:    movdqa (%rdi), %xmm8
+; SSE-NEXT:    pand %xmm11, %xmm13
+; SSE-NEXT:    por %xmm13, %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    movdqa %xmm4, %xmm14
+; SSE-NEXT:    pandn %xmm13, %xmm14
+; SSE-NEXT:    movdqa (%rsi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm13
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm0[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    movdqa (%rsi), %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm11[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,5,5,6]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm8, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm0[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm13[2,3,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm13[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,4,5,5,6]
+; SSE-NEXT:    pand %xmm4, %xmm13
+; SSE-NEXT:    por %xmm14, %xmm13
+; SSE-NEXT:    pand %xmm11, %xmm13
 ; SSE-NEXT:    movdqa (%rdx), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm0[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm8[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm2, %xmm10
-; SSE-NEXT:    pandn %xmm14, %xmm10
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm10
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm14
-; SSE-NEXT:    pandn %xmm3, %xmm14
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm3[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    por %xmm14, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm7[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,6,5,7,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm0[2,1,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm14[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm14, %xmm11
+; SSE-NEXT:    por %xmm13, %xmm11
+; SSE-NEXT:    movdqa %xmm2, %xmm13
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8],xmm3[8],xmm13[9],xmm3[9],xmm13[10],xmm3[10],xmm13[11],xmm3[11],xmm13[12],xmm3[12],xmm13[13],xmm3[13],xmm13[14],xmm3[14],xmm13[15],xmm3[15]
+; SSE-NEXT:    pxor %xmm15, %xmm15
+; SSE-NEXT:    movdqa %xmm13, %xmm14
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8],xmm15[8],xmm14[9],xmm15[9],xmm14[10],xmm15[10],xmm14[11],xmm15[11],xmm14[12],xmm15[12],xmm14[13],xmm15[13],xmm14[14],xmm15[14],xmm14[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0],xmm15[0],xmm13[1],xmm15[1],xmm13[2],xmm15[2],xmm13[3],xmm15[3],xmm13[4],xmm15[4],xmm13[5],xmm15[5],xmm13[6],xmm15[6],xmm13[7],xmm15[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm13 = xmm13[2,3],xmm14[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm13[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm13[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm14[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm13[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm13, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm9[2,3,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm13[1,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm13[0,1,2,3,6,5,7,7]
 ; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
-; SSE-NEXT:    movdqa %xmm14, %xmm8
-; SSE-NEXT:    pandn %xmm3, %xmm8
+; SSE-NEXT:    movdqa %xmm14, %xmm13
+; SSE-NEXT:    pandn %xmm1, %xmm13
 ; SSE-NEXT:    pand %xmm14, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm2, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm0[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm4
-; SSE-NEXT:    por %xmm3, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[0,1,0,1]
+; SSE-NEXT:    por %xmm0, %xmm13
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
+; SSE-NEXT:    movdqa %xmm2, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm15[8],xmm0[9],xmm15[9],xmm0[10],xmm15[10],xmm0[11],xmm15[11],xmm0[12],xmm15[12],xmm0[13],xmm15[13],xmm0[14],xmm15[14],xmm0[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm15[0],xmm2[1],xmm15[1],xmm2[2],xmm15[2],xmm2[3],xmm15[3],xmm2[4],xmm15[4],xmm2[5],xmm15[5],xmm2[6],xmm15[6],xmm2[7],xmm15[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,3],xmm0[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[0,1,0,1]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm6
-; SSE-NEXT:    movdqa %xmm15, %xmm3
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm15[8],xmm3[9],xmm15[9],xmm3[10],xmm15[10],xmm3[11],xmm15[11],xmm3[12],xmm15[12],xmm3[13],xmm15[13],xmm3[14],xmm15[14],xmm3[15],xmm15[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm4
-; SSE-NEXT:    pandn %xmm3, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm12[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm4, %xmm3
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm0[0,1,2,3,5,5,6,6]
+; SSE-NEXT:    movdqa %xmm4, %xmm6
+; SSE-NEXT:    pandn %xmm2, %xmm6
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm6
+; SSE-NEXT:    movdqa %xmm12, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm10[8],xmm1[9],xmm10[9],xmm1[10],xmm10[10],xmm1[11],xmm10[11],xmm1[12],xmm10[12],xmm1[13],xmm10[13],xmm1[14],xmm10[14],xmm1[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm15[8],xmm2[9],xmm15[9],xmm2[10],xmm15[10],xmm2[11],xmm15[11],xmm2[12],xmm15[12],xmm2[13],xmm15[13],xmm2[14],xmm15[14],xmm2[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm15[0],xmm1[1],xmm15[1],xmm1[2],xmm15[2],xmm1[3],xmm15[3],xmm1[4],xmm15[4],xmm1[5],xmm15[5],xmm1[6],xmm15[6],xmm1[7],xmm15[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,3],xmm2[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm2, %xmm1
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm4[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    movdqa %xmm14, %xmm4
-; SSE-NEXT:    pandn %xmm5, %xmm4
-; SSE-NEXT:    pand %xmm14, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm4
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm15[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm12[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm5, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm5[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    pandn %xmm7, %xmm5
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm5
-; SSE-NEXT:    movdqa %xmm9, %xmm3
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm9[8],xmm3[9],xmm9[9],xmm3[10],xmm9[10],xmm3[11],xmm9[11],xmm3[12],xmm9[12],xmm3[13],xmm9[13],xmm3[14],xmm9[14],xmm3[15],xmm9[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm7
-; SSE-NEXT:    pandn %xmm3, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,5,7,7]
+; SSE-NEXT:    movdqa %xmm14, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    pand %xmm14, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0],xmm10[0],xmm12[1],xmm10[1],xmm12[2],xmm10[2],xmm12[3],xmm10[3],xmm12[4],xmm10[4],xmm12[5],xmm10[5],xmm12[6],xmm10[6],xmm12[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm12, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm15[8],xmm1[9],xmm15[9],xmm1[10],xmm15[10],xmm1[11],xmm15[11],xmm1[12],xmm15[12],xmm1[13],xmm15[13],xmm1[14],xmm15[14],xmm1[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0],xmm15[0],xmm12[1],xmm15[1],xmm12[2],xmm15[2],xmm12[3],xmm15[3],xmm12[4],xmm15[4],xmm12[5],xmm15[5],xmm12[6],xmm15[6],xmm12[7],xmm15[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm12[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm12 = xmm12[2,3],xmm1[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm12[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm2[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm1, %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,6]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    pand %xmm4, %xmm9
+; SSE-NEXT:    por %xmm9, %xmm2
+; SSE-NEXT:    movdqa %xmm5, %xmm1
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm7, %xmm3
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm15[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    movdqa %xmm14, %xmm12
-; SSE-NEXT:    pandn %xmm7, %xmm12
-; SSE-NEXT:    pand %xmm14, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm12
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm9[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm2, %xmm7
-; SSE-NEXT:    pandn %xmm3, %xmm7
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm7, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm15[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm7[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    movdqa %xmm1, %xmm7
-; SSE-NEXT:    pandn %xmm9, %xmm7
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm7
-; SSE-NEXT:    movdqa %xmm11, %xmm3
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm11[8],xmm3[9],xmm11[9],xmm3[10],xmm11[10],xmm3[11],xmm11[11],xmm3[12],xmm11[12],xmm3[13],xmm11[13],xmm3[14],xmm11[14],xmm3[15],xmm11[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,2,2,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,6,6,7]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = 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]
 ; SSE-NEXT:    movdqa %xmm1, %xmm9
-; SSE-NEXT:    pandn %xmm3, %xmm9
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm9, %xmm3
-; SSE-NEXT:    pand %xmm14, %xmm3
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm15[2,3,2,3]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm15[8],xmm9[9],xmm15[9],xmm9[10],xmm15[10],xmm9[11],xmm15[11],xmm9[12],xmm15[12],xmm9[13],xmm15[13],xmm9[14],xmm15[14],xmm9[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm15[0],xmm1[1],xmm15[1],xmm1[2],xmm15[2],xmm1[3],xmm15[3],xmm1[4],xmm15[4],xmm1[5],xmm15[5],xmm1[6],xmm15[6],xmm1[7],xmm15[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,3],xmm9[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm9, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm12[2,3,2,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pandn %xmm9, %xmm14
-; SSE-NEXT:    por %xmm3, %xmm14
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm11[0,1,1,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,1,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm0[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm3, %xmm2
-; SSE-NEXT:    por %xmm9, %xmm2
-; SSE-NEXT:    pand %xmm1, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm15[0,1,0,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,5,5,6,6]
-; SSE-NEXT:    pandn %xmm3, %xmm1
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    movdqa %xmm1, (%rcx)
+; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm9[0,1,2,3,6,5,7,7]
+; SSE-NEXT:    movdqa %xmm14, %xmm9
+; SSE-NEXT:    pandn %xmm10, %xmm9
+; SSE-NEXT:    pand %xmm14, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm9
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm0[0],xmm5[1],xmm0[1],xmm5[2],xmm0[2],xmm5[3],xmm0[3],xmm5[4],xmm0[4],xmm5[5],xmm0[5],xmm5[6],xmm0[6],xmm5[7],xmm0[7]
+; SSE-NEXT:    movdqa %xmm5, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm15[8],xmm1[9],xmm15[9],xmm1[10],xmm15[10],xmm1[11],xmm15[11],xmm1[12],xmm15[12],xmm1[13],xmm15[13],xmm1[14],xmm15[14],xmm1[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm15[0],xmm5[1],xmm15[1],xmm5[2],xmm15[2],xmm5[3],xmm15[3],xmm5[4],xmm15[4],xmm5[5],xmm15[5],xmm5[6],xmm15[6],xmm5[7],xmm15[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm5[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[2,3],xmm1[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm5[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm10[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm5[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm1, %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm12[0,1,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,6]
+; SSE-NEXT:    movdqa %xmm4, %xmm5
+; SSE-NEXT:    pandn %xmm1, %xmm5
+; SSE-NEXT:    pand %xmm4, %xmm10
+; SSE-NEXT:    por %xmm10, %xmm5
+; SSE-NEXT:    movdqa %xmm8, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm12[8],xmm1[9],xmm12[9],xmm1[10],xmm12[10],xmm1[11],xmm12[11],xmm1[12],xmm12[12],xmm1[13],xmm12[13],xmm1[14],xmm12[14],xmm1[15],xmm12[15]
+; SSE-NEXT:    movdqa %xmm1, %xmm10
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm15[8],xmm10[9],xmm15[9],xmm10[10],xmm15[10],xmm10[11],xmm15[11],xmm10[12],xmm15[12],xmm10[13],xmm15[13],xmm10[14],xmm15[14],xmm10[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm15[0],xmm1[1],xmm15[1],xmm1[2],xmm15[2],xmm1[3],xmm15[3],xmm1[4],xmm15[4],xmm1[5],xmm15[5],xmm1[6],xmm15[6],xmm1[7],xmm15[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,3],xmm10[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[2,1,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm10[3,1,0,1,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,4,6,7,7]
+; SSE-NEXT:    packuswb %xmm10, %xmm1
+; SSE-NEXT:    pand %xmm14, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm0[2,3,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm10[1,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,6,5,7,7]
+; SSE-NEXT:    pandn %xmm10, %xmm14
+; SSE-NEXT:    por %xmm1, %xmm14
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm12[0],xmm8[1],xmm12[1],xmm8[2],xmm12[2],xmm8[3],xmm12[3],xmm8[4],xmm12[4],xmm8[5],xmm12[5],xmm8[6],xmm12[6],xmm8[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm8, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm15[8],xmm1[9],xmm15[9],xmm1[10],xmm15[10],xmm1[11],xmm15[11],xmm1[12],xmm15[12],xmm1[13],xmm15[13],xmm1[14],xmm15[14],xmm1[15],xmm15[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm15[0],xmm8[1],xmm15[1],xmm8[2],xmm15[2],xmm8[3],xmm15[3],xmm8[4],xmm15[4],xmm8[5],xmm15[5],xmm8[6],xmm15[6],xmm8[7],xmm15[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm8[0,1,2,1]
+; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[2,3],xmm1[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm8[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm10[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,7,5,4,5]
+; SSE-NEXT:    packuswb %xmm1, %xmm8
+; SSE-NEXT:    pand %xmm4, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,6,6]
+; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    por %xmm8, %xmm4
+; SSE-NEXT:    movdqa %xmm4, (%rcx)
 ; SSE-NEXT:    movdqa %xmm14, 32(%rcx)
-; SSE-NEXT:    movdqa %xmm7, 48(%rcx)
-; SSE-NEXT:    movdqa %xmm12, 80(%rcx)
-; SSE-NEXT:    movdqa %xmm5, 96(%rcx)
-; SSE-NEXT:    movdqa %xmm4, 128(%rcx)
+; SSE-NEXT:    movdqa %xmm5, 48(%rcx)
+; SSE-NEXT:    movdqa %xmm9, 80(%rcx)
+; SSE-NEXT:    movdqa %xmm2, 96(%rcx)
+; SSE-NEXT:    movdqa %xmm3, 128(%rcx)
 ; SSE-NEXT:    movdqa %xmm6, 144(%rcx)
-; SSE-NEXT:    movdqa %xmm8, 176(%rcx)
-; SSE-NEXT:    movdqa %xmm10, 16(%rcx)
-; SSE-NEXT:    movdqa %xmm13, 64(%rcx)
+; SSE-NEXT:    movdqa %xmm13, 176(%rcx)
+; SSE-NEXT:    movdqa %xmm11, 16(%rcx)
+; SSE-NEXT:    movdqa %xmm7, 64(%rcx)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 112(%rcx)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 160(%rcx)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride3_vf64:
 ; AVX1-ONLY:       # %bb.0:

llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-5.ll

; AVX512BW-NEXT:    retq
   store <20 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride5_vf8(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride5_vf8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero
-; SSE-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
 ; SSE-NEXT:    movq {{.*#+}} xmm4 = mem[0],zero
+; SSE-NEXT:    movq {{.*#+}} xmm6 = mem[0],zero
 ; SSE-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero
 ; SSE-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm2[1,1,1,1]
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [65535,65535,0,65535,0,65535,65535,0]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm8, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    por %xmm6, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,0,255,0,255,255,0,255,0,255,255,0,255,0,255]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm4[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm6, %xmm10
-; SSE-NEXT:    pandn %xmm5, %xmm10
-; SSE-NEXT:    movdqa %xmm1, %xmm7
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm1[0],xmm7[1],xmm1[1],xmm7[2],xmm1[2],xmm7[3],xmm1[3],xmm7[4],xmm1[4],xmm7[5],xmm1[5],xmm7[6],xmm1[6],xmm7[7],xmm1[7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm7[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm6, %xmm5
-; SSE-NEXT:    por %xmm10, %xmm5
-; SSE-NEXT:    pand %xmm9, %xmm5
-; SSE-NEXT:    pandn %xmm8, %xmm9
-; SSE-NEXT:    por %xmm5, %xmm9
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
-; SSE-NEXT:    pand %xmm8, %xmm9
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm0[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm8, %xmm5
-; SSE-NEXT:    pandn %xmm10, %xmm5
-; SSE-NEXT:    por %xmm9, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm4[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[0,0,0,0]
+; SSE-NEXT:    movdqa %xmm2, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1],xmm4[2],xmm6[2],xmm4[3],xmm6[3],xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3],xmm6[4],xmm3[4],xmm6[5],xmm3[5],xmm6[6],xmm3[6],xmm6[7],xmm3[7]
+; SSE-NEXT:    pxor %xmm7, %xmm7
+; SSE-NEXT:    movdqa %xmm6, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm7[0],xmm5[1],xmm7[1],xmm5[2],xmm7[2],xmm5[3],xmm7[3],xmm5[4],xmm7[4],xmm5[5],xmm7[5],xmm5[6],xmm7[6],xmm5[7],xmm7[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm5[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm8, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm9
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm7[0],xmm9[1],xmm7[1],xmm9[2],xmm7[2],xmm9[3],xmm7[3],xmm9[4],xmm7[4],xmm9[5],xmm7[5],xmm9[6],xmm7[6],xmm9[7],xmm7[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm9[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm10, %xmm9
 ; SSE-NEXT:    pand %xmm8, %xmm9
-; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm3[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,1,0,1]
-; SSE-NEXT:    pandn %xmm10, %xmm8
+; SSE-NEXT:    pandn %xmm5, %xmm8
 ; SSE-NEXT:    por %xmm9, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm2[0,0,0,0]
-; SSE-NEXT:    pand %xmm6, %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,0,1,3]
-; SSE-NEXT:    pandn %xmm7, %xmm6
-; SSE-NEXT:    por %xmm10, %xmm6
-; SSE-NEXT:    pand %xmm9, %xmm6
-; SSE-NEXT:    pandn %xmm8, %xmm9
-; SSE-NEXT:    por %xmm6, %xmm9
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
-; SSE-NEXT:    pand %xmm6, %xmm9
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm0[0,0,0,0]
-; SSE-NEXT:    pandn %xmm7, %xmm6
-; SSE-NEXT:    por %xmm9, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,255,255,255,255,0,255,255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm7, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[3,3,3,3]
-; SSE-NEXT:    pandn %xmm3, %xmm7
-; SSE-NEXT:    por %xmm4, %xmm7
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm3, %xmm7
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
+; SSE-NEXT:    pand %xmm5, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm0[0,0,0,0]
+; SSE-NEXT:    pandn %xmm9, %xmm5
+; SSE-NEXT:    por %xmm8, %xmm5
+; SSE-NEXT:    movdqa %xmm4, %xmm8
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm7[8],xmm8[9],xmm7[9],xmm8[10],xmm7[10],xmm8[11],xmm7[11],xmm8[12],xmm7[12],xmm8[13],xmm7[13],xmm8[14],xmm7[14],xmm8[15],xmm7[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm8[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    psllq $16, %xmm8
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm7[0],xmm4[1],xmm7[1],xmm4[2],xmm7[2],xmm4[3],xmm7[3],xmm4[4],xmm7[4],xmm4[5],xmm7[5],xmm4[6],xmm7[6],xmm4[7],xmm7[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm8[0]
+; SSE-NEXT:    packuswb %xmm9, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,0,255,0,255,255,0,255,0,255,255,0,255,0,255]
+; SSE-NEXT:    pand %xmm8, %xmm4
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[3,1,2,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,3,2,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,0]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,2,1,0,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,5,6,5]
+; SSE-NEXT:    pandn %xmm2, %xmm8
+; SSE-NEXT:    por %xmm4, %xmm8
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm2, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[0,0,1,1]
+; SSE-NEXT:    pandn %xmm3, %xmm2
+; SSE-NEXT:    por %xmm8, %xmm2
 ; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm7[8],xmm6[9],xmm7[9],xmm6[10],xmm7[10],xmm6[11],xmm7[11],xmm6[12],xmm7[12],xmm6[13],xmm7[13],xmm6[14],xmm7[14],xmm6[15],xmm7[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm6[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm4, %xmm4
+; SSE-NEXT:    pand %xmm3, %xmm4
 ; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    por %xmm7, %xmm3
+; SSE-NEXT:    por %xmm4, %xmm3
 ; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,0,255,255,255,255,255,255,255,255]
 ; SSE-NEXT:    pand %xmm1, %xmm3
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; SSE-NEXT:    pandn %xmm0, %xmm1
 ; SSE-NEXT:    por %xmm3, %xmm1
 ; SSE-NEXT:    movq %xmm1, 32(%r9)
-; SSE-NEXT:    movdqa %xmm6, (%r9)
-; SSE-NEXT:    movdqa %xmm5, 16(%r9)
+; SSE-NEXT:    movdqa %xmm5, (%r9)
+; SSE-NEXT:    movdqa %xmm2, 16(%r9)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride5_vf8:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
 ; AVX1-ONLY-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
 ; AVX1-ONLY-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
 ; AVX1-ONLY-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
; AVX512BW-FAST-NEXT:    retq
   %interleaved.vec = shufflevector <40 x i8> %5, <40 x i8> poison, <40 x i32> <i32 0, i32 8, i32 16, i32 24, i32 32, i32 1, i32 9, i32 17, i32 25, i32 33, i32 2, i32 10, i32 18, i32 26, i32 34, i32 3, i32 11, i32 19, i32 27, i32 35, i32 4, i32 12, i32 20, i32 28, i32 36, i32 5, i32 13, i32 21, i32 29, i32 37, i32 6, i32 14, i32 22, i32 30, i32 38, i32 7, i32 15, i32 23, i32 31, i32 39>
   store <40 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride5_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride5_vf16:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa (%rdi), %xmm10
-; SSE-NEXT:    movdqa (%rsi), %xmm11
-; SSE-NEXT:    movdqa (%rdx), %xmm2
-; SSE-NEXT:    movdqa (%rcx), %xmm8
-; SSE-NEXT:    movdqa (%r8), %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm6, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm8[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[3,3,0,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm6, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    por %xmm1, %xmm5
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    pand %xmm4, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm10[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm12
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
-; SSE-NEXT:    pand %xmm2, %xmm7
-; SSE-NEXT:    movdqa %xmm11, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm11[0,1,2,1]
+; SSE-NEXT:    movdqa (%rdi), %xmm5
+; SSE-NEXT:    movdqa (%rsi), %xmm8
+; SSE-NEXT:    movdqa (%rdx), %xmm4
+; SSE-NEXT:    movdqa (%rcx), %xmm2
+; SSE-NEXT:    movdqa (%r8), %xmm1
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm2[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[3,3,0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm0, %xmm9
+; SSE-NEXT:    pandn %xmm6, %xmm9
+; SSE-NEXT:    por %xmm3, %xmm9
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
+; SSE-NEXT:    pand %xmm7, %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[1,1,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
+; SSE-NEXT:    pand %xmm3, %xmm6
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm8[0,1,2,1]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,4,7,6,7]
 ; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm10[3,1,0,3,4,5,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm11
+; SSE-NEXT:    movdqa %xmm3, %xmm11
 ; SSE-NEXT:    pandn %xmm10, %xmm11
-; SSE-NEXT:    por %xmm7, %xmm11
-; SSE-NEXT:    movdqa %xmm4, %xmm10
+; SSE-NEXT:    por %xmm6, %xmm11
+; SSE-NEXT:    movdqa %xmm7, %xmm10
 ; SSE-NEXT:    pandn %xmm11, %xmm10
-; SSE-NEXT:    por %xmm5, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm7, %xmm10
-; SSE-NEXT:    movdqa %xmm0, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm0[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    pandn %xmm11, %xmm0
-; SSE-NEXT:    por %xmm10, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm12[2,2,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
+; SSE-NEXT:    por %xmm9, %xmm10
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pand %xmm6, %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm1[1,1,2,2]
+; SSE-NEXT:    pandn %xmm9, %xmm6
+; SSE-NEXT:    por %xmm10, %xmm6
+; SSE-NEXT:    movdqa %xmm4, %xmm10
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm2[0],xmm10[1],xmm2[1],xmm10[2],xmm2[2],xmm10[3],xmm2[3],xmm10[4],xmm2[4],xmm10[5],xmm2[5],xmm10[6],xmm2[6],xmm10[7],xmm2[7]
+; SSE-NEXT:    pxor %xmm9, %xmm9
+; SSE-NEXT:    movdqa %xmm10, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm9[8],xmm11[9],xmm9[9],xmm11[10],xmm9[10],xmm11[11],xmm9[11],xmm11[12],xmm9[12],xmm11[13],xmm9[13],xmm11[14],xmm9[14],xmm11[15],xmm9[15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm11[0,0,1,1]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm9[0],xmm10[1],xmm9[1],xmm10[2],xmm9[2],xmm10[3],xmm9[3],xmm10[4],xmm9[4],xmm10[5],xmm9[5],xmm10[6],xmm9[6],xmm10[7],xmm9[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm10[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm10[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm10 = xmm10[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm10 = xmm10[0],xmm12[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm11 = zero,zero,xmm11[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm11, %xmm10
+; SSE-NEXT:    movdqa %xmm8, %xmm11
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm5[0],xmm11[1],xmm5[1],xmm11[2],xmm5[2],xmm11[3],xmm5[3],xmm11[4],xmm5[4],xmm11[5],xmm5[5],xmm11[6],xmm5[6],xmm11[7],xmm5[7]
+; SSE-NEXT:    movdqa %xmm11, %xmm12
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm9[8],xmm12[9],xmm9[9],xmm12[10],xmm9[10],xmm12[11],xmm9[11],xmm12[12],xmm9[12],xmm12[13],xmm9[13],xmm12[14],xmm9[14],xmm12[15],xmm9[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm9[0],xmm11[1],xmm9[1],xmm11[2],xmm9[2],xmm11[3],xmm9[3],xmm11[4],xmm9[4],xmm11[5],xmm9[5],xmm11[6],xmm9[6],xmm11[7],xmm9[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[2,3],xmm12[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm12[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm12, %xmm11
+; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,0,0,255,255,255,0,0,255,255,255,0,0,255,255,255]
 ; SSE-NEXT:    pand %xmm12, %xmm11
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm0[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm13[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa %xmm12, %xmm14
-; SSE-NEXT:    pandn %xmm13, %xmm14
-; SSE-NEXT:    por %xmm11, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm13, %xmm11
-; SSE-NEXT:    pandn %xmm14, %xmm11
-; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm9[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm9, %xmm10
-; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm14[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm15
-; SSE-NEXT:    pandn %xmm14, %xmm15
-; SSE-NEXT:    movdqa %xmm8, %xmm14
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8],xmm8[8],xmm14[9],xmm8[9],xmm14[10],xmm8[10],xmm14[11],xmm8[11],xmm14[12],xmm8[12],xmm14[13],xmm8[13],xmm14[14],xmm8[14],xmm14[15],xmm8[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm14[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm15, %xmm0
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    por %xmm11, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm5[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm6, %xmm11
-; SSE-NEXT:    pandn %xmm15, %xmm11
-; SSE-NEXT:    pand %xmm6, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm11
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm8[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa %xmm7, %xmm15
-; SSE-NEXT:    pandn %xmm0, %xmm15
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm9[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm15
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm2, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm3, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,0,255,255,255,0,0,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    pandn %xmm15, %xmm3
-; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm12, %xmm15
-; SSE-NEXT:    pandn %xmm0, %xmm15
-; SSE-NEXT:    pand %xmm12, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm15
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm10[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm12, %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm8[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,0,1]
-; SSE-NEXT:    pandn %xmm3, %xmm12
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[0,0,0,0]
-; SSE-NEXT:    pand %xmm6, %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,1,3]
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    por %xmm0, %xmm6
-; SSE-NEXT:    pand %xmm13, %xmm6
+; SSE-NEXT:    pandn %xmm10, %xmm12
+; SSE-NEXT:    por %xmm11, %xmm12
+; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm10, %xmm12
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm1[0,0,1,1]
+; SSE-NEXT:    pandn %xmm11, %xmm10
+; SSE-NEXT:    por %xmm12, %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm14[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm13, %xmm12
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm13
 ; SSE-NEXT:    pandn %xmm12, %xmm13
-; SSE-NEXT:    por %xmm6, %xmm13
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm6
-; SSE-NEXT:    pandn %xmm0, %xmm6
-; SSE-NEXT:    pand %xmm2, %xmm13
-; SSE-NEXT:    por %xmm13, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm10[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm14[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,3,2]
-; SSE-NEXT:    pandn %xmm3, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[3,3,3,3]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,2,2]
-; SSE-NEXT:    pandn %xmm3, %xmm7
-; SSE-NEXT:    por %xmm0, %xmm7
-; SSE-NEXT:    pand %xmm4, %xmm7
-; SSE-NEXT:    pandn %xmm2, %xmm4
-; SSE-NEXT:    por %xmm7, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [0,255,255,255,255,0,255,255,255,255,0,255,255,255,255,0]
-; SSE-NEXT:    pand %xmm0, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[3,3,3,3]
+; SSE-NEXT:    movdqa %xmm5, %xmm12
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0],xmm8[0],xmm12[1],xmm8[1],xmm12[2],xmm8[2],xmm12[3],xmm8[3],xmm12[4],xmm8[4],xmm12[5],xmm8[5],xmm12[6],xmm8[6],xmm12[7],xmm8[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0],xmm9[0],xmm12[1],xmm9[1],xmm12[2],xmm9[2],xmm12[3],xmm9[3],xmm12[4],xmm9[4],xmm12[5],xmm9[5],xmm12[6],xmm9[6],xmm12[7],xmm9[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm12[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm14[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm12 = xmm12[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm14, %xmm12
+; SSE-NEXT:    pand %xmm11, %xmm12
+; SSE-NEXT:    por %xmm13, %xmm12
+; SSE-NEXT:    pand %xmm3, %xmm12
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm1[0,0,0,0]
+; SSE-NEXT:    pandn %xmm13, %xmm3
+; SSE-NEXT:    por %xmm12, %xmm3
+; SSE-NEXT:    movdqa %xmm4, %xmm12
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm2[8],xmm12[9],xmm2[9],xmm12[10],xmm2[10],xmm12[11],xmm2[11],xmm12[12],xmm2[12],xmm12[13],xmm2[13],xmm12[14],xmm2[14],xmm12[15],xmm2[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm9[8],xmm12[9],xmm9[9],xmm12[10],xmm9[10],xmm12[11],xmm9[11],xmm12[12],xmm9[12],xmm12[13],xmm9[13],xmm12[14],xmm9[14],xmm12[15],xmm9[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm12[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm12 = zero,zero,xmm12[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm13[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm13, %xmm12
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm8[8],xmm5[9],xmm8[9],xmm5[10],xmm8[10],xmm5[11],xmm8[11],xmm5[12],xmm8[12],xmm5[13],xmm8[13],xmm5[14],xmm8[14],xmm5[15],xmm8[15]
+; SSE-NEXT:    movdqa %xmm5, %xmm8
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm9[8],xmm8[9],xmm9[9],xmm8[10],xmm9[10],xmm8[11],xmm9[11],xmm8[12],xmm9[12],xmm8[13],xmm9[13],xmm8[14],xmm9[14],xmm8[15],xmm9[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm8[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm9[0],xmm5[1],xmm9[1],xmm5[2],xmm9[2],xmm5[3],xmm9[3],xmm5[4],xmm9[4],xmm5[5],xmm9[5],xmm5[6],xmm9[6],xmm5[7],xmm9[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm5[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[2,3],xmm8[0,1]
+; SSE-NEXT:    psrldq {{.*#+}} xmm8 = xmm8[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm13[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm8, %xmm13
+; SSE-NEXT:    pand %xmm7, %xmm13
+; SSE-NEXT:    pandn %xmm12, %xmm7
+; SSE-NEXT:    por %xmm13, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [0,255,255,255,255,0,255,255,255,255,0,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm8, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm1[3,3,3,3]
+; SSE-NEXT:    pandn %xmm12, %xmm8
+; SSE-NEXT:    por %xmm7, %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm14[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm5, %xmm7
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm4
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm9[8],xmm4[9],xmm9[9],xmm4[10],xmm9[10],xmm4[11],xmm9[11],xmm4[12],xmm9[12],xmm4[13],xmm9[13],xmm4[14],xmm9[14],xmm4[15],xmm9[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm9[0],xmm2[1],xmm9[1],xmm2[2],xmm9[2],xmm2[3],xmm9[3],xmm2[4],xmm9[4],xmm2[5],xmm9[5],xmm2[6],xmm9[6],xmm2[7],xmm9[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm2[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm4[0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm5, %xmm2
+; SSE-NEXT:    pand %xmm11, %xmm2
+; SSE-NEXT:    pandn %xmm7, %xmm11
+; SSE-NEXT:    por %xmm2, %xmm11
+; SSE-NEXT:    pand %xmm0, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
 ; SSE-NEXT:    pandn %xmm1, %xmm0
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa %xmm0, 64(%r9)
-; SSE-NEXT:    movdqa %xmm6, (%r9)
-; SSE-NEXT:    movdqa %xmm15, 16(%r9)
-; SSE-NEXT:    movdqa %xmm11, 48(%r9)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 32(%r9)
+; SSE-NEXT:    por %xmm11, %xmm0
+; SSE-NEXT:    movdqa %xmm0, 48(%r9)
+; SSE-NEXT:    movdqa %xmm8, 64(%r9)
+; SSE-NEXT:    movdqa %xmm3, (%r9)
+; SSE-NEXT:    movdqa %xmm10, 16(%r9)
+; SSE-NEXT:    movdqa %xmm6, 32(%r9)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride5_vf16:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    vmovdqa (%rdi), %xmm2
 ; AVX1-ONLY-NEXT:    vmovdqa (%rsi), %xmm3
 ; AVX1-ONLY-NEXT:    vmovdqa (%rdx), %xmm1
 ; AVX1-ONLY-NEXT:    vmovdqa (%rcx), %xmm4
; AVX512BW-FAST-NEXT:    retq
   %interleaved.vec = shufflevector <80 x i8> %5, <80 x i8> poison, <80 x i32> <i32 0, i32 16, i32 32, i32 48, i32 64, i32 1, i32 17, i32 33, i32 49, i32 65, i32 2, i32 18, i32 34, i32 50, i32 66, i32 3, i32 19, i32 35, i32 51, i32 67, i32 4, i32 20, i32 36, i32 52, i32 68, i32 5, i32 21, i32 37, i32 53, i32 69, i32 6, i32 22, i32 38, i32 54, i32 70, i32 7, i32 23, i32 39, i32 55, i32 71, i32 8, i32 24, i32 40, i32 56, i32 72, i32 9, i32 25, i32 41, i32 57, i32 73, i32 10, i32 26, i32 42, i32 58, i32 74, i32 11, i32 27, i32 43, i32 59, i32 75, i32 12, i32 28, i32 44, i32 60, i32 76, i32 13, i32 29, i32 45, i32 61, i32 77, i32 14, i32 30, i32 46, i32 62, i32 78, i32 15, i32 31, i32 47, i32 63, i32 79>
   store <80 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride5_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride5_vf32:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    subq $152, %rsp
-; SSE-NEXT:    movdqa 16(%rdi), %xmm3
-; SSE-NEXT:    movdqa (%rsi), %xmm10
-; SSE-NEXT:    movdqa 16(%rsi), %xmm12
-; SSE-NEXT:    movdqa (%rdx), %xmm2
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa 16(%rdx), %xmm1
-; SSE-NEXT:    movdqa (%rcx), %xmm15
-; SSE-NEXT:    movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa 16(%rcx), %xmm7
-; SSE-NEXT:    movdqa 16(%r8), %xmm9
+; SSE-NEXT:    subq $72, %rsp
+; SSE-NEXT:    movdqa 16(%rdi), %xmm11
+; SSE-NEXT:    movdqa (%rsi), %xmm1
+; SSE-NEXT:    movdqa 16(%rsi), %xmm5
+; SSE-NEXT:    movdqa (%rdx), %xmm6
+; SSE-NEXT:    movdqa 16(%rdx), %xmm2
+; SSE-NEXT:    movdqa (%rcx), %xmm9
 ; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm14
+; SSE-NEXT:    movdqa 16(%rcx), %xmm3
+; SSE-NEXT:    movdqa 16(%r8), %xmm12
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm2, %xmm14
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm7[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,3,0,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm13, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm7, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[3,3,0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm7, %xmm4
+; SSE-NEXT:    movdqa %xmm7, %xmm13
+; SSE-NEXT:    pandn %xmm2, %xmm4
 ; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    pand %xmm8, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm3, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm12[0,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
+; SSE-NEXT:    pand %xmm7, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm11[1,1,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[0,1,2,1]
+; SSE-NEXT:    movdqa %xmm5, %xmm8
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,7,6,7]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,1,0,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,1,0]
+; SSE-NEXT:    movdqa %xmm15, %xmm5
+; SSE-NEXT:    pandn %xmm2, %xmm5
 ; SSE-NEXT:    por %xmm0, %xmm5
-; SSE-NEXT:    movdqa %xmm8, %xmm0
+; SSE-NEXT:    movdqa %xmm7, %xmm0
 ; SSE-NEXT:    pandn %xmm5, %xmm0
 ; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm9[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pand %xmm5, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm12[1,1,2,2]
+; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm5, %xmm4
+; SSE-NEXT:    pandn %xmm2, %xmm4
 ; SSE-NEXT:    por %xmm0, %xmm4
 ; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm6, %xmm10
+; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm15[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,3,0,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
 ; SSE-NEXT:    movdqa %xmm13, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    pand %xmm13, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm9[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[3,3,0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm2, %xmm4
 ; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm10[0,1,2,1]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,2,1]
+; SSE-NEXT:    movdqa %xmm1, %xmm13
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
 ; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,0,3,4,5,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa (%rdi), %xmm15
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm15[1,1,2,2]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm8, %xmm0
-; SSE-NEXT:    pandn %xmm1, %xmm0
-; SSE-NEXT:    pand %xmm8, %xmm4
+; SSE-NEXT:    movdqa %xmm15, %xmm2
+; SSE-NEXT:    pandn %xmm0, %xmm2
+; SSE-NEXT:    movdqa (%rdi), %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,2]
+; SSE-NEXT:    movdqa %xmm1, %xmm6
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    movdqa %xmm7, %xmm0
+; SSE-NEXT:    pandn %xmm2, %xmm0
+; SSE-NEXT:    pand %xmm7, %xmm4
 ; SSE-NEXT:    por %xmm4, %xmm0
 ; SSE-NEXT:    movdqa (%r8), %xmm1
 ; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm12, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm12[8],xmm0[9],xmm12[9],xmm0[10],xmm12[10],xmm0[11],xmm12[11],xmm0[12],xmm12[12],xmm0[13],xmm12[13],xmm0[14],xmm12[14],xmm0[15],xmm12[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
-; SSE-NEXT:    movdqa %xmm9, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pand %xmm9, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm0
-; SSE-NEXT:    pandn %xmm4, %xmm0
-; SSE-NEXT:    movdqa %xmm14, %xmm1
-; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm14[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    movdqa %xmm7, %xmm14
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8],xmm7[8],xmm14[9],xmm7[9],xmm14[10],xmm7[10],xmm14[11],xmm7[11],xmm14[12],xmm7[12],xmm14[13],xmm7[13],xmm14[14],xmm7[14],xmm14[15],xmm7[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm14[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm3, %xmm4
-; SSE-NEXT:    por %xmm5, %xmm4
-; SSE-NEXT:    pand %xmm8, %xmm4
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm13, %xmm5
-; SSE-NEXT:    pandn %xmm0, %xmm5
-; SSE-NEXT:    pand %xmm13, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm5
+; SSE-NEXT:    pand %xmm5, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,2,2]
+; SSE-NEXT:    pandn %xmm2, %xmm5
+; SSE-NEXT:    por %xmm0, %xmm5
 ; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa %xmm11, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,255,255,255,0,0,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm6[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm3, %xmm7
-; SSE-NEXT:    pandn %xmm4, %xmm7
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm12, (%rsp) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm12[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm3, %xmm4
-; SSE-NEXT:    por %xmm7, %xmm4
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    por %xmm5, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm9, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm9, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm10, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm10[8],xmm0[9],xmm10[9],xmm0[10],xmm10[10],xmm0[11],xmm10[11],xmm0[12],xmm10[12],xmm0[13],xmm10[13],xmm0[14],xmm10[14],xmm0[15],xmm10[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa %xmm9, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    movdqa %xmm15, %xmm6
-; SSE-NEXT:    movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm15[2,2,3,3]
-; SSE-NEXT:    pand %xmm9, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm5
-; SSE-NEXT:    movdqa %xmm8, %xmm4
-; SSE-NEXT:    pandn %xmm5, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm0[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm3, %xmm7
-; SSE-NEXT:    pandn %xmm5, %xmm7
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    movdqa %xmm2, %xmm12
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm2[8],xmm12[9],xmm2[9],xmm12[10],xmm2[10],xmm12[11],xmm2[11],xmm12[12],xmm2[12],xmm12[13],xmm2[13],xmm12[14],xmm2[14],xmm12[15],xmm2[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm12[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm15 = xmm15[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm15 = xmm15[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm3, %xmm15
-; SSE-NEXT:    por %xmm7, %xmm15
-; SSE-NEXT:    pand %xmm8, %xmm15
-; SSE-NEXT:    por %xmm4, %xmm15
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm5[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm13, %xmm7
-; SSE-NEXT:    pandn %xmm4, %xmm7
-; SSE-NEXT:    pand %xmm13, %xmm15
-; SSE-NEXT:    por %xmm15, %xmm7
-; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa %xmm11, %xmm15
-; SSE-NEXT:    pandn %xmm4, %xmm15
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm0[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE-NEXT:    pand %xmm11, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm15
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm6[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm3, %xmm2
-; SSE-NEXT:    pandn %xmm4, %xmm2
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm10[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm3, %xmm4
-; SSE-NEXT:    por %xmm2, %xmm4
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    pandn %xmm15, %xmm1
-; SSE-NEXT:    por %xmm4, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm9, %xmm7
-; SSE-NEXT:    pandn %xmm2, %xmm7
-; SSE-NEXT:    pand %xmm9, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm7
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm14[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,3,2]
-; SSE-NEXT:    movdqa %xmm3, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    movdqa %xmm6, %xmm0
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm2 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,3,2,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm15
-; SSE-NEXT:    pandn %xmm2, %xmm15
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[3,3,3,3]
-; SSE-NEXT:    pand %xmm11, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm15
-; SSE-NEXT:    pand %xmm6, %xmm15
-; SSE-NEXT:    por %xmm0, %xmm15
-; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [0,255,255,255,255,0,255,255,255,255,0,255,255,255,255,0]
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[3,3,3,3]
-; SSE-NEXT:    movdqa %xmm10, %xmm14
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    pand %xmm10, %xmm15
-; SSE-NEXT:    por %xmm15, %xmm14
-; SSE-NEXT:    pshuflw $225, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
-; SSE-NEXT:    movdqa %xmm9, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm9, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm8, %xmm0
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    pshuflw $164, (%rsp), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm2 = mem[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,1,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm15
-; SSE-NEXT:    pandn %xmm2, %xmm15
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm15
-; SSE-NEXT:    pand %xmm8, %xmm15
-; SSE-NEXT:    por %xmm0, %xmm15
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm0
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    pand %xmm3, %xmm15
-; SSE-NEXT:    por %xmm15, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm12[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,3,2]
-; SSE-NEXT:    movdqa %xmm3, %xmm5
+; SSE-NEXT:    movdqa %xmm14, %xmm4
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; SSE-NEXT:    pxor %xmm14, %xmm14
+; SSE-NEXT:    movdqa %xmm4, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm14[8],xmm0[9],xmm14[9],xmm0[10],xmm14[10],xmm0[11],xmm14[11],xmm0[12],xmm14[12],xmm0[13],xmm14[13],xmm0[14],xmm14[14],xmm0[15],xmm14[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm14[0],xmm4[1],xmm14[1],xmm4[2],xmm14[2],xmm4[3],xmm14[3],xmm4[4],xmm14[4],xmm4[5],xmm14[5],xmm4[6],xmm14[6],xmm4[7],xmm14[7]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[0,0,1,1]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm5[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm0, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,0,255,255,255,0,0,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm5
 ; SSE-NEXT:    pandn %xmm2, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm4[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,2,2,2]
-; SSE-NEXT:    pand %xmm3, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm15[3,3,3,3]
-; SSE-NEXT:    pand %xmm11, %xmm2
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm12 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[2,3,2,2]
-; SSE-NEXT:    pandn %xmm12, %xmm11
-; SSE-NEXT:    por %xmm2, %xmm11
-; SSE-NEXT:    movdqa %xmm6, %xmm1
-; SSE-NEXT:    pand %xmm6, %xmm11
+; SSE-NEXT:    movdqa %xmm8, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm11[0],xmm2[1],xmm11[1],xmm2[2],xmm11[2],xmm2[3],xmm11[3],xmm2[4],xmm11[4],xmm2[5],xmm11[5],xmm2[6],xmm11[6],xmm2[7],xmm11[7]
+; SSE-NEXT:    movdqa %xmm2, %xmm9
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm14[8],xmm9[9],xmm14[9],xmm9[10],xmm14[10],xmm9[11],xmm14[11],xmm9[12],xmm14[12],xmm9[13],xmm14[13],xmm9[14],xmm14[14],xmm9[15],xmm14[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm14[0],xmm2[1],xmm14[1],xmm2[2],xmm14[2],xmm2[3],xmm14[3],xmm2[4],xmm14[4],xmm2[5],xmm14[5],xmm2[6],xmm14[6],xmm2[7],xmm14[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,3],xmm9[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm9, %xmm2
+; SSE-NEXT:    pand %xmm0, %xmm2
+; SSE-NEXT:    por %xmm5, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm12[0,0,1,1]
+; SSE-NEXT:    movdqa %xmm1, %xmm8
+; SSE-NEXT:    pandn %xmm5, %xmm8
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm8
+; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm10, %xmm12
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm12 = xmm12[0],mem[0],xmm12[1],mem[1],xmm12[2],mem[2],xmm12[3],mem[3],xmm12[4],mem[4],xmm12[5],mem[5],xmm12[6],mem[6],xmm12[7],mem[7]
+; SSE-NEXT:    movdqa %xmm12, %xmm5
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm14[8],xmm5[9],xmm14[9],xmm5[10],xmm14[10],xmm5[11],xmm14[11],xmm5[12],xmm14[12],xmm5[13],xmm14[13],xmm5[14],xmm14[14],xmm5[15],xmm14[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
+; SSE-NEXT:    movdqa %xmm12, %xmm9
+; SSE-NEXT:    psrldq {{.*#+}} xmm9 = xmm9[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm5[0,0,1,1]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm10[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm5 = zero,zero,xmm5[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm5, %xmm9
+; SSE-NEXT:    movdqa %xmm13, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3],xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
+; SSE-NEXT:    movdqa %xmm5, %xmm10
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm14[8],xmm10[9],xmm14[9],xmm10[10],xmm14[10],xmm10[11],xmm14[11],xmm10[12],xmm14[12],xmm10[13],xmm14[13],xmm10[14],xmm14[14],xmm10[15],xmm14[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm14[0],xmm5[1],xmm14[1],xmm5[2],xmm14[2],xmm5[3],xmm14[3],xmm5[4],xmm14[4],xmm5[5],xmm14[5],xmm5[6],xmm14[6],xmm5[7],xmm14[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[2,3],xmm10[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm10, %xmm5
+; SSE-NEXT:    pand %xmm0, %xmm5
+; SSE-NEXT:    pandn %xmm9, %xmm0
+; SSE-NEXT:    por %xmm5, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pshufd $80, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm5 = mem[0,0,1,1]
 ; SSE-NEXT:    pandn %xmm5, %xmm1
-; SSE-NEXT:    por %xmm11, %xmm1
-; SSE-NEXT:    pand %xmm10, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[3,3,3,3]
-; SSE-NEXT:    pandn %xmm2, %xmm10
-; SSE-NEXT:    por %xmm5, %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    pand %xmm9, %xmm2
-; SSE-NEXT:    pshuflw $225, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,0,1]
-; SSE-NEXT:    pandn %xmm5, %xmm9
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm13, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm14[8],xmm0[9],xmm14[9],xmm0[10],xmm14[10],xmm0[11],xmm14[11],xmm0[12],xmm14[12],xmm0[13],xmm14[13],xmm0[14],xmm14[14],xmm0[15],xmm14[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm0[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm5, %xmm0
+; SSE-NEXT:    movdqa %xmm7, %xmm5
+; SSE-NEXT:    pandn %xmm0, %xmm5
+; SSE-NEXT:    movdqa %xmm11, %xmm9
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm8[8],xmm9[9],xmm8[9],xmm9[10],xmm8[10],xmm9[11],xmm8[11],xmm9[12],xmm8[12],xmm9[13],xmm8[13],xmm9[14],xmm8[14],xmm9[15],xmm8[15]
+; SSE-NEXT:    movdqa %xmm9, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm14[8],xmm1[9],xmm14[9],xmm1[10],xmm14[10],xmm1[11],xmm14[11],xmm1[12],xmm14[12],xmm1[13],xmm14[13],xmm1[14],xmm14[14],xmm1[15],xmm14[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm14[0],xmm9[1],xmm14[1],xmm9[2],xmm14[2],xmm9[3],xmm14[3],xmm9[4],xmm14[4],xmm9[5],xmm14[5],xmm9[6],xmm14[6],xmm9[7],xmm14[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm9[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[2,3],xmm1[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm1[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm10[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    pand %xmm7, %xmm0
+; SSE-NEXT:    por %xmm5, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [0,255,255,255,255,0,255,255,255,255,0,255,255,255,255,0]
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm6[3,3,3,3]
+; SSE-NEXT:    movdqa %xmm5, %xmm10
+; SSE-NEXT:    pandn %xmm1, %xmm10
+; SSE-NEXT:    pand %xmm5, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm10
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm9[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm13[8],xmm3[9],xmm13[9],xmm3[10],xmm13[10],xmm3[11],xmm13[11],xmm3[12],xmm13[12],xmm3[13],xmm13[13],xmm3[14],xmm13[14],xmm3[15],xmm13[15]
+; SSE-NEXT:    movdqa %xmm3, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm14[8],xmm1[9],xmm14[9],xmm1[10],xmm14[10],xmm1[11],xmm14[11],xmm1[12],xmm14[12],xmm1[13],xmm14[13],xmm1[14],xmm14[14],xmm1[15],xmm14[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm14[0],xmm3[1],xmm14[1],xmm3[2],xmm14[2],xmm3[3],xmm14[3],xmm3[4],xmm14[4],xmm3[5],xmm14[5],xmm3[6],xmm14[6],xmm3[7],xmm14[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm3[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm9 = xmm9[0],xmm1[0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm1[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm9, %xmm3
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa %xmm6, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm6[2,2,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm6, %xmm13
+; SSE-NEXT:    pandn %xmm2, %xmm13
+; SSE-NEXT:    pand %xmm6, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm13
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm4[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm3
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    pandn %xmm3, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm8[0],xmm11[1],xmm8[1],xmm11[2],xmm8[2],xmm11[3],xmm8[3],xmm11[4],xmm8[4],xmm11[5],xmm8[5],xmm11[6],xmm8[6],xmm11[7],xmm8[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm14[0],xmm11[1],xmm14[1],xmm11[2],xmm14[2],xmm11[3],xmm14[3],xmm11[4],xmm14[4],xmm11[5],xmm14[5],xmm11[6],xmm14[6],xmm11[7],xmm14[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm11[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm11[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm4[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm3, %xmm9
+; SSE-NEXT:    pand %xmm0, %xmm9
 ; SSE-NEXT:    por %xmm2, %xmm9
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm15[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm2
-; SSE-NEXT:    pshuflw $164, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,1,3]
-; SSE-NEXT:    pandn %xmm5, %xmm13
-; SSE-NEXT:    por %xmm2, %xmm13
-; SSE-NEXT:    pand %xmm8, %xmm13
-; SSE-NEXT:    pandn %xmm9, %xmm8
-; SSE-NEXT:    por %xmm13, %xmm8
-; SSE-NEXT:    pand %xmm3, %xmm8
 ; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,0,0,0]
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    por %xmm8, %xmm3
-; SSE-NEXT:    movdqa %xmm3, (%r9)
-; SSE-NEXT:    movdqa %xmm10, 64(%r9)
-; SSE-NEXT:    movdqa %xmm0, 80(%r9)
-; SSE-NEXT:    movdqa %xmm14, 144(%r9)
-; SSE-NEXT:    movdqa %xmm7, 16(%r9)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 48(%r9)
+; SSE-NEXT:    movdqa %xmm15, %xmm4
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    pand %xmm15, %xmm9
+; SSE-NEXT:    por %xmm9, %xmm4
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm8, %xmm11
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm6[8],xmm11[9],xmm6[9],xmm11[10],xmm6[10],xmm11[11],xmm6[11],xmm11[12],xmm6[12],xmm11[13],xmm6[13],xmm11[14],xmm6[14],xmm11[15],xmm6[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm14[8],xmm11[9],xmm14[9],xmm11[10],xmm14[10],xmm11[11],xmm14[11],xmm11[12],xmm14[12],xmm11[13],xmm14[13],xmm11[14],xmm14[14],xmm11[15],xmm14[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm11[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm11 = zero,zero,xmm11[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm11
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm9 = xmm9[8],mem[8],xmm9[9],mem[9],xmm9[10],mem[10],xmm9[11],mem[11],xmm9[12],mem[12],xmm9[13],mem[13],xmm9[14],mem[14],xmm9[15],mem[15]
+; SSE-NEXT:    movdqa %xmm9, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm14[8],xmm2[9],xmm14[9],xmm2[10],xmm14[10],xmm2[11],xmm14[11],xmm2[12],xmm14[12],xmm2[13],xmm14[13],xmm2[14],xmm14[14],xmm2[15],xmm14[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm14[0],xmm9[1],xmm14[1],xmm9[2],xmm14[2],xmm9[3],xmm14[3],xmm9[4],xmm14[4],xmm9[5],xmm14[5],xmm9[6],xmm14[6],xmm9[7],xmm14[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm9[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[2,3],xmm2[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm2, %xmm1
+; SSE-NEXT:    pand %xmm7, %xmm1
+; SSE-NEXT:    pandn %xmm11, %xmm7
+; SSE-NEXT:    por %xmm1, %xmm7
+; SSE-NEXT:    pand %xmm5, %xmm7
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm11[3,3,3,3]
+; SSE-NEXT:    pandn %xmm1, %xmm5
+; SSE-NEXT:    por %xmm7, %xmm5
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    pandn %xmm2, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm8[8],xmm6[9],xmm8[9],xmm6[10],xmm8[10],xmm6[11],xmm8[11],xmm6[12],xmm8[12],xmm6[13],xmm8[13],xmm6[14],xmm8[14],xmm6[15],xmm8[15]
+; SSE-NEXT:    movdqa %xmm6, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm14[8],xmm2[9],xmm14[9],xmm2[10],xmm14[10],xmm2[11],xmm14[11],xmm2[12],xmm14[12],xmm2[13],xmm14[13],xmm2[14],xmm14[14],xmm2[15],xmm14[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm14[0],xmm6[1],xmm14[1],xmm6[2],xmm14[2],xmm6[3],xmm14[3],xmm6[4],xmm14[4],xmm6[5],xmm14[5],xmm6[6],xmm14[6],xmm6[7],xmm14[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm2[0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm6[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm3, %xmm2
+; SSE-NEXT:    pand %xmm0, %xmm2
+; SSE-NEXT:    por %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm11[2,2,2,2]
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa %xmm3, %xmm6
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm12[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm12[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm3 = xmm3[0],mem[0],xmm3[1],mem[1],xmm3[2],mem[2],xmm3[3],mem[3],xmm3[4],mem[4],xmm3[5],mem[5],xmm3[6],mem[6],xmm3[7],mem[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm14[0],xmm3[1],xmm14[1],xmm3[2],xmm14[2],xmm3[3],xmm14[3],xmm3[4],xmm14[4],xmm3[5],xmm14[5],xmm3[6],xmm14[6],xmm3[7],xmm14[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm0
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm11[0,0,0,0]
+; SSE-NEXT:    pandn %xmm1, %xmm15
+; SSE-NEXT:    por %xmm0, %xmm15
+; SSE-NEXT:    movdqa %xmm15, (%r9)
+; SSE-NEXT:    movdqa %xmm6, 48(%r9)
+; SSE-NEXT:    movdqa %xmm5, 64(%r9)
+; SSE-NEXT:    movdqa %xmm4, 80(%r9)
+; SSE-NEXT:    movdqa %xmm13, 128(%r9)
+; SSE-NEXT:    movdqa %xmm10, 144(%r9)
+; SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movaps %xmm0, 16(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 96(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 128(%r9)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 32(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 112(%r9)
-; SSE-NEXT:    addq $152, %rsp
+; SSE-NEXT:    addq $72, %rsp
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride5_vf32:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    vmovdqa 16(%rsi), %xmm11
 ; AVX1-ONLY-NEXT:    vmovdqa 16(%rdi), %xmm13
 ; AVX1-ONLY-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm13[8],xmm11[8],xmm13[9],xmm11[9],xmm13[10],xmm11[10],xmm13[11],xmm11[11],xmm13[12],xmm11[12],xmm13[13],xmm11[13],xmm13[14],xmm11[14],xmm13[15],xmm11[15]
 ; AVX1-ONLY-NEXT:    vmovdqa {{.*#+}} xmm12 = <u,10,11,u,u,u,12,13,u,u,u,14,15,u,u,u>
; AVX512BW-FAST-NEXT:    retq
   %interleaved.vec = shufflevector <160 x i8> %5, <160 x i8> poison, <160 x i32> <i32 0, i32 32, i32 64, i32 96, i32 128, i32 1, i32 33, i32 65, i32 97, i32 129, i32 2, i32 34, i32 66, i32 98, i32 130, i32 3, i32 35, i32 67, i32 99, i32 131, i32 4, i32 36, i32 68, i32 100, i32 132, i32 5, i32 37, i32 69, i32 101, i32 133, i32 6, i32 38, i32 70, i32 102, i32 134, i32 7, i32 39, i32 71, i32 103, i32 135, i32 8, i32 40, i32 72, i32 104, i32 136, i32 9, i32 41, i32 73, i32 105, i32 137, i32 10, i32 42, i32 74, i32 106, i32 138, i32 11, i32 43, i32 75, i32 107, i32 139, i32 12, i32 44, i32 76, i32 108, i32 140, i32 13, i32 45, i32 77, i32 109, i32 141, i32 14, i32 46, i32 78, i32 110, i32 142, i32 15, i32 47, i32 79, i32 111, i32 143, i32 16, i32 48, i32 80, i32 112, i32 144, i32 17, i32 49, i32 81, i32 113, i32 145, i32 18, i32 50, i32 82, i32 114, i32 146, i32 19, i32 51, i32 83, i32 115, i32 147, i32 20, i32 52, i32 84, i32 116, i32 148, i32 21, i32 53, i32 85, i32 117, i32 149, i32 22, i32 54, i32 86, i32 118, i32 150, i32 23, i32 55, i32 87, i32 119, i32 151, i32 24, i32 56, i32 88, i32 120, i32 152, i32 25, i32 57, i32 89, i32 121, i32 153, i32 26, i32 58, i32 90, i32 122, i32 154, i32 27, i32 59, i32 91, i32 123, i32 155, i32 28, i32 60, i32 92, i32 124, i32 156, i32 29, i32 61, i32 93, i32 125, i32 157, i32 30, i32 62, i32 94, i32 126, i32 158, i32 31, i32 63, i32 95, i32 127, i32 159>
   store <160 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride5_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride5_vf64:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    subq $504, %rsp # imm = 0x1F8
-; SSE-NEXT:    movdqa (%rdi), %xmm8
-; SSE-NEXT:    movdqa (%rsi), %xmm9
-; SSE-NEXT:    movdqa 16(%rsi), %xmm14
-; SSE-NEXT:    movdqa (%rdx), %xmm6
-; SSE-NEXT:    movdqa 16(%rdx), %xmm15
-; SSE-NEXT:    movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa (%rcx), %xmm10
-; SSE-NEXT:    movdqa 16(%rcx), %xmm13
-; SSE-NEXT:    movdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa (%r8), %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    subq $312, %rsp # imm = 0x138
+; SSE-NEXT:    movdqa (%rdi), %xmm2
+; SSE-NEXT:    movdqa (%rsi), %xmm5
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 16(%rsi), %xmm9
+; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa (%rdx), %xmm1
+; SSE-NEXT:    movdqa 16(%rdx), %xmm6
+; SSE-NEXT:    movdqa (%rcx), %xmm3
+; SSE-NEXT:    movdqa 16(%rcx), %xmm8
+; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa (%r8), %xmm10
+; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm1, %xmm12
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm10[2,1,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[2,1,2,3]
+; SSE-NEXT:    movdqa %xmm3, %xmm13
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,3,0,3]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm11, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    pand %xmm7, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm8[1,1,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm9[0,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,7,6,7]
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
-; SSE-NEXT:    por %xmm0, %xmm5
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm12[1,1,2,2]
 ; SSE-NEXT:    movdqa %xmm4, %xmm3
 ; SSE-NEXT:    pandn %xmm1, %xmm3
 ; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm15[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm13[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,3,0,3]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm11, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
+; SSE-NEXT:    pand %xmm14, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[1,1,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,7,6,7]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[3,1,0,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,1,0]
+; SSE-NEXT:    movdqa %xmm15, %xmm7
+; SSE-NEXT:    pandn %xmm5, %xmm7
+; SSE-NEXT:    por %xmm0, %xmm7
+; SSE-NEXT:    movdqa %xmm14, %xmm0
+; SSE-NEXT:    pandn %xmm7, %xmm0
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm10[1,1,2,2]
+; SSE-NEXT:    movdqa %xmm1, %xmm5
+; SSE-NEXT:    pandn %xmm3, %xmm5
 ; SSE-NEXT:    por %xmm0, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm14[0,1,2,1]
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm6, %xmm11
+; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm8[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[3,3,0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm4, %xmm7
+; SSE-NEXT:    pandn %xmm3, %xmm7
+; SSE-NEXT:    por %xmm0, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[0,1,2,1]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
 ; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,0,3,4,5,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa 16(%rdi), %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    pandn %xmm1, %xmm0
-; SSE-NEXT:    pand %xmm7, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm0
-; SSE-NEXT:    movdqa 16(%r8), %xmm1
-; SSE-NEXT:    movdqa %xmm1, (%rsp) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    movdqa %xmm15, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    movdqa 16(%rdi), %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm8[1,1,2,2]
+; SSE-NEXT:    pand %xmm15, %xmm0
 ; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm14, %xmm0
+; SSE-NEXT:    pandn %xmm3, %xmm0
+; SSE-NEXT:    pand %xmm14, %xmm7
+; SSE-NEXT:    por %xmm7, %xmm0
+; SSE-NEXT:    movdqa 16(%r8), %xmm3
 ; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,2]
+; SSE-NEXT:    movdqa %xmm1, %xmm7
+; SSE-NEXT:    pandn %xmm3, %xmm7
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm7
+; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa 32(%rcx), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,3,0,3]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm11, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa 32(%rdx), %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    movdqa 32(%rdx), %xmm5
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa 32(%rsi), %xmm13
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm13[0,1,2,1]
+; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    movdqa %xmm4, %xmm6
+; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa 32(%rsi), %xmm0
+; SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
 ; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,0,3,4,5,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm0, %xmm5
-; SSE-NEXT:    movdqa 32(%rdi), %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm5
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa 32(%r8), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm15, %xmm7
+; SSE-NEXT:    pandn %xmm0, %xmm7
+; SSE-NEXT:    movdqa 32(%rdi), %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm10[1,1,2,2]
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm7
+; SSE-NEXT:    movdqa %xmm14, %xmm0
+; SSE-NEXT:    pandn %xmm7, %xmm0
+; SSE-NEXT:    pand %xmm14, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    movdqa 32(%r8), %xmm3
 ; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,2]
+; SSE-NEXT:    movdqa %xmm1, %xmm7
+; SSE-NEXT:    pandn %xmm3, %xmm7
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm7
+; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa 48(%rcx), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,3,0,3]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm11, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa 48(%rdx), %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    movdqa 48(%rdx), %xmm4
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    pand %xmm6, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm3
 ; SSE-NEXT:    movdqa 48(%rsi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,1]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,7,6,7]
 ; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,0,3,4,5,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm0, %xmm5
+; SSE-NEXT:    movdqa %xmm15, %xmm7
+; SSE-NEXT:    pandn %xmm0, %xmm7
 ; SSE-NEXT:    movdqa 48(%rdi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm5
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa 48(%r8), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,2]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm7
+; SSE-NEXT:    movdqa %xmm14, %xmm0
+; SSE-NEXT:    pandn %xmm7, %xmm0
+; SSE-NEXT:    pand %xmm14, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa 48(%r8), %xmm3
 ; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm10, %xmm0
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm10[0],xmm0[1],xmm10[1],xmm0[2],xmm10[2],xmm0[3],xmm10[3],xmm0[4],xmm10[4],xmm0[5],xmm10[5],xmm0[6],xmm10[6],xmm0[7],xmm10[7]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa %xmm4, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[1,1,2,2]
+; SSE-NEXT:    pandn %xmm3, %xmm1
 ; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,0,255,255,255,0,0,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    pandn %xmm1, %xmm0
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm8[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
-; SSE-NEXT:    movdqa %xmm9, %xmm1
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm9[0],xmm1[1],xmm9[1],xmm1[2],xmm9[2],xmm1[3],xmm9[3],xmm1[4],xmm9[4],xmm1[5],xmm9[5],xmm1[6],xmm9[6],xmm1[7],xmm9[7]
 ; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm1
-; SSE-NEXT:    por %xmm5, %xmm1
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
-; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm12[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm4, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
+; SSE-NEXT:    movdqa %xmm12, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm13[0],xmm3[1],xmm13[1],xmm3[2],xmm13[2],xmm3[3],xmm13[3],xmm3[4],xmm13[4],xmm3[5],xmm13[5],xmm3[6],xmm13[6],xmm3[7],xmm13[7]
+; SSE-NEXT:    pxor %xmm12, %xmm12
+; SSE-NEXT:    movdqa %xmm3, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm12[8],xmm0[9],xmm12[9],xmm0[10],xmm12[10],xmm0[11],xmm12[11],xmm0[12],xmm12[12],xmm0[13],xmm12[13],xmm0[14],xmm12[14],xmm0[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm12[0],xmm3[1],xmm12[1],xmm3[2],xmm12[2],xmm3[3],xmm12[3],xmm3[4],xmm12[4],xmm3[5],xmm12[5],xmm3[6],xmm12[6],xmm3[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm3, %xmm7
+; SSE-NEXT:    psrldq {{.*#+}} xmm7 = xmm7[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm0[0,0,1,1]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm7 = xmm7[0],xmm13[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm0, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,0,255,255,255,0,0,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm13
+; SSE-NEXT:    pandn %xmm7, %xmm13
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3],xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm15
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm15 = xmm15[8],xmm12[8],xmm15[9],xmm12[9],xmm15[10],xmm12[10],xmm15[11],xmm12[11],xmm15[12],xmm12[12],xmm15[13],xmm12[13],xmm15[14],xmm12[14],xmm15[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm12[0],xmm7[1],xmm12[1],xmm7[2],xmm12[2],xmm7[3],xmm12[3],xmm7[4],xmm12[4],xmm7[5],xmm12[5],xmm7[6],xmm12[6],xmm7[7],xmm12[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[2,3],xmm15[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm15[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm15 = xmm15[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm15, %xmm7
+; SSE-NEXT:    pand %xmm0, %xmm7
+; SSE-NEXT:    por %xmm13, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,255,255,255,255,0,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm6[0,0,1,1]
+; SSE-NEXT:    movdqa %xmm1, %xmm9
+; SSE-NEXT:    pandn %xmm13, %xmm9
+; SSE-NEXT:    pand %xmm1, %xmm7
+; SSE-NEXT:    por %xmm7, %xmm9
 ; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm9[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm8[2,2,3,3]
-; SSE-NEXT:    pand %xmm4, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm5
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm15, %xmm8
-; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm6[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm5, %xmm9
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm10[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm2, %xmm5
+; SSE-NEXT:    movdqa %xmm11, %xmm13
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm13 = xmm13[0],mem[0],xmm13[1],mem[1],xmm13[2],mem[2],xmm13[3],mem[3],xmm13[4],mem[4],xmm13[5],mem[5],xmm13[6],mem[6],xmm13[7],mem[7]
+; SSE-NEXT:    movdqa %xmm13, %xmm7
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm12[8],xmm7[9],xmm12[9],xmm7[10],xmm12[10],xmm7[11],xmm12[11],xmm7[12],xmm12[12],xmm7[13],xmm12[13],xmm7[14],xmm12[14],xmm7[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3],xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm13, %xmm15
+; SSE-NEXT:    psrldq {{.*#+}} xmm15 = xmm15[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm7[0,0,1,1]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm15 = xmm15[0],xmm11[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm7 = zero,zero,xmm7[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm7, %xmm15
+; SSE-NEXT:    movdqa %xmm0, %xmm7
+; SSE-NEXT:    pandn %xmm15, %xmm7
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm8[0],xmm11[1],xmm8[1],xmm11[2],xmm8[2],xmm11[3],xmm8[3],xmm11[4],xmm8[4],xmm11[5],xmm8[5],xmm11[6],xmm8[6],xmm11[7],xmm8[7]
+; SSE-NEXT:    movdqa %xmm11, %xmm15
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm15 = xmm15[8],xmm12[8],xmm15[9],xmm12[9],xmm15[10],xmm12[10],xmm15[11],xmm12[11],xmm15[12],xmm12[12],xmm15[13],xmm12[13],xmm15[14],xmm12[14],xmm15[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm12[0],xmm11[1],xmm12[1],xmm11[2],xmm12[2],xmm11[3],xmm12[3],xmm11[4],xmm12[4],xmm11[5],xmm12[5],xmm11[6],xmm12[6],xmm11[7],xmm12[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[2,3],xmm15[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm15[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm15 = xmm15[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm15, %xmm11
+; SSE-NEXT:    pand %xmm0, %xmm11
+; SSE-NEXT:    por %xmm7, %xmm11
+; SSE-NEXT:    pshufd $80, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm7 = mem[0,0,1,1]
+; SSE-NEXT:    movdqa %xmm1, %xmm9
+; SSE-NEXT:    pandn %xmm7, %xmm9
+; SSE-NEXT:    pand %xmm1, %xmm11
+; SSE-NEXT:    por %xmm11, %xmm9
+; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm5, %xmm7
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm7 = xmm7[0],mem[0],xmm7[1],mem[1],xmm7[2],mem[2],xmm7[3],mem[3],xmm7[4],mem[4],xmm7[5],mem[5],xmm7[6],mem[6],xmm7[7],mem[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm12[8],xmm11[9],xmm12[9],xmm11[10],xmm12[10],xmm11[11],xmm12[11],xmm11[12],xmm12[12],xmm11[13],xmm12[13],xmm11[14],xmm12[14],xmm11[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm12[0],xmm7[1],xmm12[1],xmm7[2],xmm12[2],xmm7[3],xmm12[3],xmm7[4],xmm12[4],xmm7[5],xmm12[5],xmm7[6],xmm12[6],xmm7[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm15
+; SSE-NEXT:    psrldq {{.*#+}} xmm15 = xmm15[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm11[0,0,1,1]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm15 = xmm15[0],xmm9[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm11 = zero,zero,xmm11[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm11, %xmm15
+; SSE-NEXT:    movdqa %xmm0, %xmm9
+; SSE-NEXT:    pandn %xmm15, %xmm9
+; SSE-NEXT:    movdqa (%rsp), %xmm11 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm10[0],xmm11[1],xmm10[1],xmm11[2],xmm10[2],xmm11[3],xmm10[3],xmm11[4],xmm10[4],xmm11[5],xmm10[5],xmm11[6],xmm10[6],xmm11[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm11, %xmm15
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm15 = xmm15[8],xmm12[8],xmm15[9],xmm12[9],xmm15[10],xmm12[10],xmm15[11],xmm12[11],xmm15[12],xmm12[12],xmm15[13],xmm12[13],xmm15[14],xmm12[14],xmm15[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm12[0],xmm11[1],xmm12[1],xmm11[2],xmm12[2],xmm11[3],xmm12[3],xmm11[4],xmm12[4],xmm11[5],xmm12[5],xmm11[6],xmm12[6],xmm11[7],xmm12[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[2,3],xmm15[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm11[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm15[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm15 = xmm15[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm15, %xmm11
+; SSE-NEXT:    pand %xmm0, %xmm11
+; SSE-NEXT:    por %xmm9, %xmm11
+; SSE-NEXT:    pshufd $80, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm9 = mem[0,0,1,1]
+; SSE-NEXT:    movdqa %xmm1, %xmm15
+; SSE-NEXT:    pandn %xmm9, %xmm15
+; SSE-NEXT:    pand %xmm1, %xmm11
+; SSE-NEXT:    por %xmm11, %xmm15
+; SSE-NEXT:    movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm4 = xmm4[0],mem[0],xmm4[1],mem[1],xmm4[2],mem[2],xmm4[3],mem[3],xmm4[4],mem[4],xmm4[5],mem[5],xmm4[6],mem[6],xmm4[7],mem[7]
+; SSE-NEXT:    movdqa %xmm4, %xmm9
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm12[8],xmm9[9],xmm12[9],xmm9[10],xmm12[10],xmm9[11],xmm12[11],xmm9[12],xmm12[12],xmm9[13],xmm12[13],xmm9[14],xmm12[14],xmm9[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm12[0],xmm4[1],xmm12[1],xmm4[2],xmm12[2],xmm4[3],xmm12[3],xmm4[4],xmm12[4],xmm4[5],xmm12[5],xmm4[6],xmm12[6],xmm4[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm9[0,0,1,1]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm15[0]
+; SSE-NEXT:    pslldq {{.*#+}} xmm9 = zero,zero,xmm9[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm9, %xmm4
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm9 = xmm9[0],mem[0],xmm9[1],mem[1],xmm9[2],mem[2],xmm9[3],mem[3],xmm9[4],mem[4],xmm9[5],mem[5],xmm9[6],mem[6],xmm9[7],mem[7]
+; SSE-NEXT:    movdqa %xmm9, %xmm15
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm15 = xmm15[8],xmm12[8],xmm15[9],xmm12[9],xmm15[10],xmm12[10],xmm15[11],xmm12[11],xmm15[12],xmm12[12],xmm15[13],xmm12[13],xmm15[14],xmm12[14],xmm15[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm12[0],xmm9[1],xmm12[1],xmm9[2],xmm12[2],xmm9[3],xmm12[3],xmm9[4],xmm12[4],xmm9[5],xmm12[5],xmm9[6],xmm12[6],xmm9[7],xmm12[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[2,3],xmm15[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm15[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm15 = xmm15[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm15, %xmm9
+; SSE-NEXT:    pand %xmm0, %xmm9
+; SSE-NEXT:    pandn %xmm4, %xmm0
+; SSE-NEXT:    por %xmm9, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pshufd $80, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm9 = mem[0,0,1,1]
+; SSE-NEXT:    pandn %xmm9, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm0, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,0,255,255,255,0,0,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm9
+; SSE-NEXT:    pandn %xmm3, %xmm9
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm12[0],xmm3[1],xmm12[1],xmm3[2],xmm12[2],xmm3[3],xmm12[3],xmm3[4],xmm12[4],xmm3[5],xmm12[5],xmm3[6],xmm12[6],xmm3[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm3[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm11, %xmm3
+; SSE-NEXT:    pand %xmm1, %xmm3
+; SSE-NEXT:    por %xmm9, %xmm3
+; SSE-NEXT:    movdqa %xmm6, %xmm5
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm6[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm6
+; SSE-NEXT:    pandn %xmm9, %xmm6
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm6
+; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm12[8],xmm3[9],xmm12[9],xmm3[10],xmm12[10],xmm3[11],xmm12[11],xmm3[12],xmm12[12],xmm3[13],xmm12[13],xmm3[14],xmm12[14],xmm3[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm12[0],xmm2[1],xmm12[1],xmm2[2],xmm12[2],xmm2[3],xmm12[3],xmm2[4],xmm12[4],xmm2[5],xmm12[5],xmm2[6],xmm12[6],xmm2[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm2[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,3],xmm3[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm9
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm9, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm4, %xmm9
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm6[8],xmm9[9],xmm6[9],xmm9[10],xmm6[10],xmm9[11],xmm6[11],xmm9[12],xmm6[12],xmm9[13],xmm6[13],xmm9[14],xmm6[14],xmm9[15],xmm6[15]
+; SSE-NEXT:    movdqa %xmm9, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm12[8],xmm11[9],xmm12[9],xmm11[10],xmm12[10],xmm11[11],xmm12[11],xmm11[12],xmm12[12],xmm11[13],xmm12[13],xmm11[14],xmm12[14],xmm11[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm12[0],xmm9[1],xmm12[1],xmm9[2],xmm12[2],xmm9[3],xmm12[3],xmm9[4],xmm12[4],xmm9[5],xmm12[5],xmm9[6],xmm12[6],xmm9[7],xmm12[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm9[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm15 = xmm15[0],xmm11[0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm15, %xmm9
+; SSE-NEXT:    pand %xmm1, %xmm9
+; SSE-NEXT:    por %xmm2, %xmm9
+; SSE-NEXT:    movdqa %xmm5, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[2,2,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm5
+; SSE-NEXT:    pandn %xmm2, %xmm5
+; SSE-NEXT:    pand %xmm0, %xmm9
 ; SSE-NEXT:    por %xmm9, %xmm5
-; SSE-NEXT:    pand %xmm15, %xmm5
-; SSE-NEXT:    por %xmm8, %xmm5
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm12[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm0
-; SSE-NEXT:    pand %xmm11, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm0
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm4[8],xmm6[9],xmm4[9],xmm6[10],xmm4[10],xmm6[11],xmm4[11],xmm6[12],xmm4[12],xmm6[13],xmm4[13],xmm6[14],xmm4[14],xmm6[15],xmm4[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm12[8],xmm6[9],xmm12[9],xmm6[10],xmm12[10],xmm6[11],xmm12[11],xmm6[12],xmm12[12],xmm6[13],xmm12[13],xmm6[14],xmm12[14],xmm6[15],xmm12[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm6[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm6 = zero,zero,xmm6[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm6
+; SSE-NEXT:    movdqa %xmm14, %xmm2
+; SSE-NEXT:    pandn %xmm6, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm3, %xmm4
+; SSE-NEXT:    pand %xmm14, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm11[3,3,3,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [0,255,255,255,255,0,255,255,255,255,0,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm6, %xmm0
+; SSE-NEXT:    pandn %xmm2, %xmm0
+; SSE-NEXT:    pand %xmm6, %xmm4
+; SSE-NEXT:    por %xmm4, %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm13[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm13[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm4
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm4, %xmm2
+; SSE-NEXT:    movdqa %xmm8, %xmm4
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    movdqa %xmm3, %xmm0
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm8
-; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm10[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm0, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa %xmm7, %xmm5
-; SSE-NEXT:    pandn %xmm8, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm1[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm8, %xmm9
-; SSE-NEXT:    movdqa %xmm14, %xmm0
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm14[0],xmm0[1],xmm14[1],xmm0[2],xmm14[2],xmm0[3],xmm14[3],xmm0[4],xmm14[4],xmm0[5],xmm14[5],xmm0[6],xmm14[6],xmm0[7],xmm14[7]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm0[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm9, %xmm8
-; SSE-NEXT:    pand %xmm7, %xmm8
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pand %xmm4, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm6
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm14[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm8
-; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm1[2,2,3,3]
-; SSE-NEXT:    pand %xmm4, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa %xmm15, %xmm5
-; SSE-NEXT:    pandn %xmm8, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm10[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm8, %xmm9
-; SSE-NEXT:    movdqa %xmm3, %xmm1
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm1[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm9, %xmm8
-; SSE-NEXT:    pand %xmm15, %xmm8
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm12
-; SSE-NEXT:    movdqa %xmm11, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm11, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    movdqa %xmm11, %xmm0
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm11[0],xmm0[1],xmm11[1],xmm0[2],xmm11[2],xmm0[3],xmm11[3],xmm0[4],xmm11[4],xmm0[5],xmm11[5],xmm0[6],xmm11[6],xmm0[7],xmm11[7]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm5, %xmm8
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm12[0],xmm4[1],xmm12[1],xmm4[2],xmm12[2],xmm4[3],xmm12[3],xmm4[4],xmm12[4],xmm4[5],xmm12[5],xmm4[6],xmm12[6],xmm4[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm4[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm9, %xmm4
+; SSE-NEXT:    pand %xmm1, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm0[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm1, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa %xmm7, %xmm5
-; SSE-NEXT:    pandn %xmm8, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm10[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm8, %xmm9
-; SSE-NEXT:    movdqa %xmm13, %xmm14
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm14 = xmm14[0],xmm13[0],xmm14[1],xmm13[1],xmm14[2],xmm13[2],xmm14[3],xmm13[3],xmm14[4],xmm13[4],xmm14[5],xmm13[5],xmm14[6],xmm13[6],xmm14[7],xmm13[7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm14[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm9, %xmm8
-; SSE-NEXT:    pand %xmm7, %xmm8
-; SSE-NEXT:    por %xmm5, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm0[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,0,255,255,255,255,0,255,255,255,255,0,255]
+; SSE-NEXT:    movdqa %xmm2, %xmm5
+; SSE-NEXT:    pandn %xmm9, %xmm5
+; SSE-NEXT:    pand %xmm2, %xmm4
+; SSE-NEXT:    por %xmm4, %xmm5
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm3[8],xmm8[9],xmm3[9],xmm8[10],xmm3[10],xmm8[11],xmm3[11],xmm8[12],xmm3[12],xmm8[13],xmm3[13],xmm8[14],xmm3[14],xmm8[15],xmm3[15]
+; SSE-NEXT:    movdqa %xmm8, %xmm13
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8],xmm12[8],xmm13[9],xmm12[9],xmm13[10],xmm12[10],xmm13[11],xmm12[11],xmm13[12],xmm12[12],xmm13[13],xmm12[13],xmm13[14],xmm12[14],xmm13[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm12[0],xmm8[1],xmm12[1],xmm8[2],xmm12[2],xmm8[3],xmm12[3],xmm8[4],xmm12[4],xmm8[5],xmm12[5],xmm8[6],xmm12[6],xmm8[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm8[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[2,3],xmm13[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm5, %xmm4
+; SSE-NEXT:    movdqa %xmm1, %xmm5
+; SSE-NEXT:    pandn %xmm4, %xmm5
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm15, %xmm4
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pand %xmm4, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm6
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm13[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm8
-; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm10[2,2,3,3]
-; SSE-NEXT:    pand %xmm4, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa %xmm15, %xmm5
-; SSE-NEXT:    pandn %xmm8, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm0[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm8, %xmm9
-; SSE-NEXT:    movdqa %xmm11, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm11[8],xmm0[9],xmm11[9],xmm0[10],xmm11[10],xmm0[11],xmm11[11],xmm0[12],xmm11[12],xmm0[13],xmm11[13],xmm0[14],xmm11[14],xmm0[15],xmm11[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm0[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm9, %xmm8
-; SSE-NEXT:    pand %xmm15, %xmm8
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm12, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
+; SSE-NEXT:    movdqa %xmm4, %xmm9
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm12[8],xmm9[9],xmm12[9],xmm9[10],xmm12[10],xmm9[11],xmm12[11],xmm9[12],xmm12[12],xmm9[13],xmm12[13],xmm9[14],xmm12[14],xmm9[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm12[0],xmm4[1],xmm12[1],xmm4[2],xmm12[2],xmm4[3],xmm12[3],xmm4[4],xmm12[4],xmm4[5],xmm12[5],xmm4[6],xmm12[6],xmm4[7],xmm12[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm4[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm11 = xmm11[0],xmm9[0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm4[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm11, %xmm9
+; SSE-NEXT:    pand %xmm1, %xmm9
+; SSE-NEXT:    por %xmm5, %xmm9
+; SSE-NEXT:    movdqa %xmm0, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,2,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm8, %xmm0
 ; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm12, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; SSE-NEXT:    movdqa %xmm13, %xmm0
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm13[0],xmm0[1],xmm13[1],xmm0[2],xmm13[2],xmm0[3],xmm13[3],xmm0[4],xmm13[4],xmm0[5],xmm13[5],xmm0[6],xmm13[6],xmm0[7],xmm13[7]
+; SSE-NEXT:    pand %xmm8, %xmm9
+; SSE-NEXT:    por %xmm9, %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,3,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,5,5,5]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm15[8],xmm3[9],xmm15[9],xmm3[10],xmm15[10],xmm3[11],xmm15[11],xmm3[12],xmm15[12],xmm3[13],xmm15[13],xmm3[14],xmm15[14],xmm3[15],xmm15[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm12[8],xmm3[9],xmm12[9],xmm3[10],xmm12[10],xmm3[11],xmm12[11],xmm3[12],xmm12[12],xmm3[13],xmm12[13],xmm3[14],xmm12[14],xmm3[15],xmm12[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm3[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,xmm3[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm5, %xmm3
+; SSE-NEXT:    movdqa %xmm14, %xmm5
+; SSE-NEXT:    pandn %xmm3, %xmm5
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm13[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm13 = xmm13[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm9[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm13, %xmm11
+; SSE-NEXT:    pand %xmm14, %xmm11
+; SSE-NEXT:    por %xmm5, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[3,3,3,3]
+; SSE-NEXT:    movdqa %xmm6, %xmm9
+; SSE-NEXT:    pandn %xmm5, %xmm9
+; SSE-NEXT:    pand %xmm6, %xmm11
+; SSE-NEXT:    por %xmm11, %xmm9
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm7[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm5, %xmm7
+; SSE-NEXT:    movdqa %xmm1, %xmm5
+; SSE-NEXT:    pandn %xmm7, %xmm5
+; SSE-NEXT:    movdqa %xmm10, %xmm7
+; SSE-NEXT:    movdqa (%rsp), %xmm3 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3],xmm7[4],xmm3[4],xmm7[5],xmm3[5],xmm7[6],xmm3[6],xmm7[7],xmm3[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm12[0],xmm7[1],xmm12[1],xmm7[2],xmm12[2],xmm7[3],xmm12[3],xmm7[4],xmm12[4],xmm7[5],xmm12[5],xmm7[6],xmm12[6],xmm7[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm7[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm7[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm7[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm11, %xmm13
+; SSE-NEXT:    pand %xmm1, %xmm13
+; SSE-NEXT:    por %xmm5, %xmm13
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[0,0,0,0]
+; SSE-NEXT:    movdqa %xmm2, %xmm7
+; SSE-NEXT:    pandn %xmm5, %xmm7
+; SSE-NEXT:    pand %xmm2, %xmm13
+; SSE-NEXT:    por %xmm13, %xmm7
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm3[8],xmm10[9],xmm3[9],xmm10[10],xmm3[10],xmm10[11],xmm3[11],xmm10[12],xmm3[12],xmm10[13],xmm3[13],xmm10[14],xmm3[14],xmm10[15],xmm3[15]
+; SSE-NEXT:    movdqa %xmm10, %xmm13
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8],xmm12[8],xmm13[9],xmm12[9],xmm13[10],xmm12[10],xmm13[11],xmm12[11],xmm13[12],xmm12[12],xmm13[13],xmm12[13],xmm13[14],xmm12[14],xmm13[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm12[0],xmm10[1],xmm12[1],xmm10[2],xmm12[2],xmm10[3],xmm12[3],xmm10[4],xmm12[4],xmm10[5],xmm12[5],xmm10[6],xmm12[6],xmm10[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm10[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[2,3],xmm13[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm10[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm8, %xmm5
 ; SSE-NEXT:    movdqa %xmm1, %xmm8
 ; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm10[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm1, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[1,1,1,1]
-; SSE-NEXT:    movdqa %xmm2, %xmm9
-; SSE-NEXT:    pandn %xmm5, %xmm9
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    movdqa %xmm11, %xmm3
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm11[0],xmm3[1],xmm11[1],xmm3[2],xmm11[2],xmm3[3],xmm11[3],xmm3[4],xmm11[4],xmm3[5],xmm11[5],xmm3[6],xmm11[6],xmm3[7],xmm11[7]
-; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[3,1,0,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pand %xmm2, %xmm5
-; SSE-NEXT:    por %xmm9, %xmm5
-; SSE-NEXT:    pand %xmm7, %xmm5
-; SSE-NEXT:    pandn %xmm8, %xmm7
-; SSE-NEXT:    por %xmm5, %xmm7
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pand %xmm4, %xmm7
-; SSE-NEXT:    por %xmm7, %xmm6
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm11, %xmm5
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm11[8],xmm5[9],xmm11[9],xmm5[10],xmm11[10],xmm5[11],xmm11[11],xmm5[12],xmm11[12],xmm5[13],xmm11[13],xmm5[14],xmm11[14],xmm5[15],xmm11[15]
-; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
+; SSE-NEXT:    movdqa %xmm3, %xmm5
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm10[8],xmm5[9],xmm10[9],xmm5[10],xmm10[10],xmm5[11],xmm10[11],xmm5[12],xmm10[12],xmm5[13],xmm10[13],xmm5[14],xmm10[14],xmm5[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm5, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm12[8],xmm11[9],xmm12[9],xmm11[10],xmm12[10],xmm11[11],xmm12[11],xmm11[12],xmm12[12],xmm11[13],xmm12[13],xmm11[14],xmm12[14],xmm11[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm12[0],xmm5[1],xmm12[1],xmm5[2],xmm12[2],xmm5[3],xmm12[3],xmm5[4],xmm12[4],xmm5[5],xmm12[5],xmm5[6],xmm12[6],xmm5[7],xmm12[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm5[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm15 = xmm15[0],xmm11[0]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,7,5,4,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,2,3,3]
-; SSE-NEXT:    pand %xmm4, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa %xmm15, %xmm5
-; SSE-NEXT:    pandn %xmm6, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm10[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
-; SSE-NEXT:    movdqa %xmm13, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm13[8],xmm0[9],xmm13[9],xmm0[10],xmm13[10],xmm0[11],xmm13[11],xmm0[12],xmm13[12],xmm0[13],xmm13[13],xmm0[14],xmm13[14],xmm0[15],xmm13[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[0,1,2,1]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm6[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,7,6,4]
-; SSE-NEXT:    pand %xmm2, %xmm6
-; SSE-NEXT:    por %xmm8, %xmm6
-; SSE-NEXT:    pand %xmm15, %xmm6
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm12, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm12, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw $225, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,0,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm5[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm15, %xmm11
+; SSE-NEXT:    pand %xmm1, %xmm11
+; SSE-NEXT:    por %xmm8, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm4[2,2,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm5
+; SSE-NEXT:    pandn %xmm8, %xmm5
+; SSE-NEXT:    pand %xmm0, %xmm11
+; SSE-NEXT:    por %xmm11, %xmm5
+; SSE-NEXT:    movdqa %xmm10, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm12[8],xmm0[9],xmm12[9],xmm0[10],xmm12[10],xmm0[11],xmm12[11],xmm0[12],xmm12[12],xmm0[13],xmm12[13],xmm0[14],xmm12[14],xmm0[15],xmm12[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm0[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm8, %xmm0
+; SSE-NEXT:    movdqa %xmm14, %xmm8
+; SSE-NEXT:    pandn %xmm0, %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm13[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm13 = xmm13[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm13, %xmm10
+; SSE-NEXT:    pand %xmm14, %xmm10
+; SSE-NEXT:    por %xmm8, %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm4[3,3,3,3]
+; SSE-NEXT:    movdqa %xmm6, %xmm8
+; SSE-NEXT:    pandn %xmm11, %xmm8
+; SSE-NEXT:    pand %xmm6, %xmm10
+; SSE-NEXT:    por %xmm10, %xmm8
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm0[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm4, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa %xmm15, %xmm5
-; SSE-NEXT:    pandn %xmm6, %xmm5
-; SSE-NEXT:    pshuflw $164, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm6 = mem[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,1,3]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm0[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,0,2,1]
+; SSE-NEXT:    packuswb %xmm10, %xmm11
+; SSE-NEXT:    movdqa %xmm1, %xmm10
+; SSE-NEXT:    pandn %xmm11, %xmm10
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm15, %xmm11
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[0,0,0,0]
-; SSE-NEXT:    pand %xmm12, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm8
-; SSE-NEXT:    pand %xmm15, %xmm8
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm9[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm6
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,3,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm0[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,2,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    movdqa %xmm7, %xmm5
-; SSE-NEXT:    pandn %xmm6, %xmm5
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm6 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,2,2]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[3,3,3,3]
-; SSE-NEXT:    pand %xmm1, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm8
-; SSE-NEXT:    pand %xmm7, %xmm8
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm9[3,3,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [0,255,255,255,255,0,255,255,255,255,0,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm9, %xmm13
-; SSE-NEXT:    pandn %xmm5, %xmm13
-; SSE-NEXT:    pand %xmm9, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm13
-; SSE-NEXT:    pshuflw $225, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,0,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm1[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm4, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa %xmm15, %xmm5
-; SSE-NEXT:    pandn %xmm6, %xmm5
-; SSE-NEXT:    pshuflw $164, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm6 = mem[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,1,3]
-; SSE-NEXT:    movdqa %xmm12, %xmm3
-; SSE-NEXT:    pandn %xmm6, %xmm3
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm12, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm3
-; SSE-NEXT:    pand %xmm15, %xmm3
-; SSE-NEXT:    por %xmm5, %xmm3
-; SSE-NEXT:    movdqa (%rsp), %xmm11 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm11[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm10
-; SSE-NEXT:    pandn %xmm5, %xmm10
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm10
-; SSE-NEXT:    pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,3,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm1[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,2,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm5
-; SSE-NEXT:    movdqa %xmm7, %xmm3
-; SSE-NEXT:    pandn %xmm5, %xmm3
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,3,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[3,3,3,3]
-; SSE-NEXT:    pand %xmm1, %xmm5
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    pand %xmm7, %xmm6
-; SSE-NEXT:    por %xmm3, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm11[3,3,3,3]
-; SSE-NEXT:    movdqa %xmm9, %xmm8
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    pand %xmm9, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm8
-; SSE-NEXT:    pshuflw $225, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,0,1]
-; SSE-NEXT:    movdqa %xmm4, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; SSE-NEXT:    pand %xmm4, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm6
-; SSE-NEXT:    movdqa %xmm15, %xmm3
-; SSE-NEXT:    pandn %xmm6, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm14[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,1,3]
-; SSE-NEXT:    movdqa %xmm12, %xmm11
-; SSE-NEXT:    pandn %xmm6, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm12, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm11
-; SSE-NEXT:    pand %xmm15, %xmm11
-; SSE-NEXT:    por %xmm3, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm7[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm14
-; SSE-NEXT:    pandn %xmm3, %xmm14
-; SSE-NEXT:    pand %xmm2, %xmm11
-; SSE-NEXT:    por %xmm11, %xmm14
-; SSE-NEXT:    pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,3,2]
-; SSE-NEXT:    movdqa %xmm2, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm1[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,2,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm3
-; SSE-NEXT:    pandn %xmm6, %xmm3
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm6 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,3,2,2]
-; SSE-NEXT:    movdqa %xmm0, %xmm11
-; SSE-NEXT:    pandn %xmm6, %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[3,3,3,3]
-; SSE-NEXT:    pand %xmm0, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm11
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm3[0],xmm11[1],xmm3[1],xmm11[2],xmm3[2],xmm11[3],xmm3[3],xmm11[4],xmm3[4],xmm11[5],xmm3[5],xmm11[6],xmm3[6],xmm11[7],xmm3[7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm12[0],xmm11[1],xmm12[1],xmm11[2],xmm12[2],xmm11[3],xmm12[3],xmm11[4],xmm12[4],xmm11[5],xmm12[5],xmm11[6],xmm12[6],xmm11[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm11[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm13[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm11[0,1,3,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,1,1,1]
+; SSE-NEXT:    packuswb %xmm13, %xmm11
 ; SSE-NEXT:    pand %xmm1, %xmm11
-; SSE-NEXT:    por %xmm3, %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm7[3,3,3,3]
-; SSE-NEXT:    movdqa %xmm9, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pand %xmm9, %xmm11
-; SSE-NEXT:    por %xmm11, %xmm6
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; SSE-NEXT:    pand %xmm4, %xmm3
-; SSE-NEXT:    pshuflw $225, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm11 = mem[1,0,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,1,0,1]
-; SSE-NEXT:    pandn %xmm11, %xmm4
-; SSE-NEXT:    por %xmm3, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm5[0,0,0,0]
-; SSE-NEXT:    pand %xmm12, %xmm3
-; SSE-NEXT:    pshuflw $164, {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm11 = mem[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,0,1,3]
-; SSE-NEXT:    pandn %xmm11, %xmm12
-; SSE-NEXT:    por %xmm3, %xmm12
-; SSE-NEXT:    pand %xmm15, %xmm12
-; SSE-NEXT:    pandn %xmm4, %xmm15
-; SSE-NEXT:    por %xmm12, %xmm15
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm11[0,0,0,0]
+; SSE-NEXT:    por %xmm10, %xmm11
 ; SSE-NEXT:    movdqa %xmm2, %xmm0
-; SSE-NEXT:    pandn %xmm3, %xmm0
-; SSE-NEXT:    pand %xmm2, %xmm15
-; SSE-NEXT:    por %xmm15, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm1
-; SSE-NEXT:    pshufhw $237, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,5,7,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,3,2]
-; SSE-NEXT:    pandn %xmm3, %xmm2
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[3,3,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,255,255,0,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm4, %xmm1
-; SSE-NEXT:    pshufhw $167, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,7,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,3,2,2]
-; SSE-NEXT:    pandn %xmm3, %xmm4
-; SSE-NEXT:    por %xmm1, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,255,255,255,0,0,255,255,255,0,0,255]
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    pandn %xmm2, %xmm1
-; SSE-NEXT:    por %xmm4, %xmm1
-; SSE-NEXT:    pand %xmm9, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm11[3,3,3,3]
-; SSE-NEXT:    pandn %xmm1, %xmm9
-; SSE-NEXT:    por %xmm2, %xmm9
-; SSE-NEXT:    movdqa %xmm9, 304(%r9)
-; SSE-NEXT:    movdqa %xmm0, 240(%r9)
-; SSE-NEXT:    movdqa %xmm6, 224(%r9)
-; SSE-NEXT:    movdqa %xmm14, 160(%r9)
-; SSE-NEXT:    movdqa %xmm8, 144(%r9)
-; SSE-NEXT:    movdqa %xmm10, 80(%r9)
-; SSE-NEXT:    movdqa %xmm13, 64(%r9)
+; SSE-NEXT:    pand %xmm2, %xmm11
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm4[0,0,0,0]
+; SSE-NEXT:    pandn %xmm10, %xmm0
+; SSE-NEXT:    por %xmm11, %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm15, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm10
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm12[8],xmm10[9],xmm12[9],xmm10[10],xmm12[10],xmm10[11],xmm12[11],xmm10[12],xmm12[12],xmm10[13],xmm12[13],xmm10[14],xmm12[14],xmm10[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm11 = xmm0[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[2,3],xmm10[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,2,2,3]
+; SSE-NEXT:    packuswb %xmm13, %xmm11
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm3, %xmm13
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8],xmm2[8],xmm13[9],xmm2[9],xmm13[10],xmm2[10],xmm13[11],xmm2[11],xmm13[12],xmm2[12],xmm13[13],xmm2[13],xmm13[14],xmm2[14],xmm13[15],xmm2[15]
+; SSE-NEXT:    movdqa %xmm13, %xmm15
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm15 = xmm15[8],xmm12[8],xmm15[9],xmm12[9],xmm15[10],xmm12[10],xmm15[11],xmm12[11],xmm15[12],xmm12[12],xmm15[13],xmm12[13],xmm15[14],xmm12[14],xmm15[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm15 = xmm15[0,1,1,0,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3],xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm13[3,3,3,3]
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm15[0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm13[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm13[0,1,2,3,5,4,6,7]
+; SSE-NEXT:    packuswb %xmm0, %xmm13
+; SSE-NEXT:    pand %xmm1, %xmm13
+; SSE-NEXT:    pandn %xmm11, %xmm1
+; SSE-NEXT:    por %xmm13, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,0,255,255,255,255,0,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm11, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[2,2,2,2]
+; SSE-NEXT:    pandn %xmm0, %xmm11
+; SSE-NEXT:    por %xmm1, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm12[8],xmm2[9],xmm12[9],xmm2[10],xmm12[10],xmm2[11],xmm12[11],xmm2[12],xmm12[12],xmm2[13],xmm12[13],xmm2[14],xmm12[14],xmm2[15],xmm12[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,4,5,7,6]
+; SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,xmm2[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]
+; SSE-NEXT:    packuswb %xmm0, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm10[0,2,3,3,4,5,6,7]
+; SSE-NEXT:    psrldq {{.*#+}} xmm10 = xmm10[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
+; SSE-NEXT:    packuswb %xmm10, %xmm0
+; SSE-NEXT:    pand %xmm14, %xmm0
+; SSE-NEXT:    pandn %xmm2, %xmm14
+; SSE-NEXT:    por %xmm0, %xmm14
+; SSE-NEXT:    pand %xmm6, %xmm14
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[3,3,3,3]
+; SSE-NEXT:    pandn %xmm0, %xmm6
+; SSE-NEXT:    por %xmm14, %xmm6
+; SSE-NEXT:    movdqa %xmm6, 304(%r9)
+; SSE-NEXT:    movdqa %xmm11, 288(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, (%r9)
+; SSE-NEXT:    movaps %xmm0, 240(%r9)
+; SSE-NEXT:    movdqa %xmm8, 224(%r9)
+; SSE-NEXT:    movdqa %xmm5, 208(%r9)
+; SSE-NEXT:    movdqa %xmm7, 160(%r9)
+; SSE-NEXT:    movdqa %xmm9, 144(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 288(%r9)
+; SSE-NEXT:    movaps %xmm0, 128(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 256(%r9)
+; SSE-NEXT:    movaps %xmm0, 80(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 208(%r9)
+; SSE-NEXT:    movaps %xmm0, 64(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 176(%r9)
+; SSE-NEXT:    movaps %xmm0, 48(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 128(%r9)
+; SSE-NEXT:    movaps %xmm0, (%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 96(%r9)
+; SSE-NEXT:    movaps %xmm0, 256(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 48(%r9)
+; SSE-NEXT:    movaps %xmm0, 176(%r9)
+; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movaps %xmm0, 96(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 16(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 272(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 192(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 112(%r9)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 32(%r9)
-; SSE-NEXT:    addq $504, %rsp # imm = 0x1F8
+; SSE-NEXT:    addq $312, %rsp # imm = 0x138
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride5_vf64:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    subq $104, %rsp
 ; AVX1-ONLY-NEXT:    vmovdqa 48(%rcx), %xmm0
 ; AVX1-ONLY-NEXT:    vpshufb {{.*#+}} xmm1 = zero,xmm0[6,u,u,u],zero,xmm0[7,u,u,u],zero,xmm0[8,u,u,u],zero
 ; AVX1-ONLY-NEXT:    vmovdqa 48(%rdx), %xmm2

llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-7.ll

; AVX512BW-NEXT:    retq
   store <28 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride7_vf8(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %in.vecptr5, ptr %in.vecptr6, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride7_vf8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %r10
 ; SSE-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movq {{.*#+}} xmm10 = mem[0],zero
 ; SSE-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movq {{.*#+}} xmm5 = mem[0],zero
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE-NEXT:    movq {{.*#+}} xmm1 = mem[0],zero
 ; SSE-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movq {{.*#+}} xmm6 = mem[0],zero
-; SSE-NEXT:    movq {{.*#+}} xmm4 = mem[0],zero
-; SSE-NEXT:    punpcklqdq {{.*#+}} xmm4 = xmm4[0],xmm6[0]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm1[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm11, %xmm6
-; SSE-NEXT:    movdqa %xmm5, %xmm7
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm5[0],xmm7[1],xmm5[1],xmm7[2],xmm5[2],xmm7[3],xmm5[3],xmm7[4],xmm5[4],xmm7[5],xmm5[5],xmm7[6],xmm5[6],xmm7[7],xmm5[7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm7[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,1,1,3]
-; SSE-NEXT:    pandn %xmm8, %xmm11
-; SSE-NEXT:    por %xmm6, %xmm11
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm0[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,2,1]
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    pand %xmm8, %xmm6
-; SSE-NEXT:    movdqa %xmm10, %xmm3
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm10[0],xmm3[1],xmm10[1],xmm3[2],xmm10[2],xmm3[3],xmm10[3],xmm3[4],xmm10[4],xmm3[5],xmm10[5],xmm3[6],xmm10[6],xmm3[7],xmm10[7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm12 = xmm3[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[0,1,1,0]
-; SSE-NEXT:    movdqa %xmm8, %xmm13
-; SSE-NEXT:    pandn %xmm12, %xmm13
-; SSE-NEXT:    por %xmm6, %xmm13
-; SSE-NEXT:    pand %xmm9, %xmm13
-; SSE-NEXT:    pandn %xmm11, %xmm9
-; SSE-NEXT:    por %xmm13, %xmm9
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3],xmm1[4],xmm2[4],xmm1[5],xmm2[5],xmm1[6],xmm2[6],xmm1[7],xmm2[7]
+; SSE-NEXT:    movq {{.*#+}} xmm2 = mem[0],zero
+; SSE-NEXT:    movq {{.*#+}} xmm3 = mem[0],zero
+; SSE-NEXT:    movq {{.*#+}} xmm7 = mem[0],zero
+; SSE-NEXT:    punpcklqdq {{.*#+}} xmm7 = xmm7[0],xmm3[0]
 ; SSE-NEXT:    pxor %xmm6, %xmm6
-; SSE-NEXT:    movdqa %xmm4, %xmm15
-; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm4, %xmm12
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm15 = xmm15[8],xmm6[8],xmm15[9],xmm6[9],xmm15[10],xmm6[10],xmm15[11],xmm6[11],xmm15[12],xmm6[12],xmm15[13],xmm6[13],xmm15[14],xmm6[14],xmm15[15],xmm6[15]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0],xmm6[0],xmm12[1],xmm6[1],xmm12[2],xmm6[2],xmm12[3],xmm6[3],xmm12[4],xmm6[4],xmm12[5],xmm6[5],xmm12[6],xmm6[6],xmm12[7],xmm6[7]
+; SSE-NEXT:    movdqa %xmm1, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm6[0],xmm3[1],xmm6[1],xmm3[2],xmm6[2],xmm3[3],xmm6[3],xmm3[4],xmm6[4],xmm3[5],xmm6[5],xmm3[6],xmm6[6],xmm3[7],xmm6[7]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    psrlq $16, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm3[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm4, %xmm9
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm6[0],xmm5[1],xmm6[1],xmm5[2],xmm6[2],xmm5[3],xmm6[3],xmm5[4],xmm6[4],xmm5[5],xmm6[5],xmm5[6],xmm6[6],xmm5[7],xmm6[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm5[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm5[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm4, %xmm10
+; SSE-NEXT:    pand %xmm8, %xmm10
+; SSE-NEXT:    pandn %xmm9, %xmm8
+; SSE-NEXT:    por %xmm10, %xmm8
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm8
+; SSE-NEXT:    movdqa %xmm7, %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm7[3,1,2,3]
+; SSE-NEXT:    movdqa %xmm7, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8],xmm6[8],xmm11[9],xmm6[9],xmm11[10],xmm6[10],xmm11[11],xmm6[11],xmm11[12],xmm6[12],xmm11[13],xmm6[13],xmm11[14],xmm6[14],xmm11[15],xmm6[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm6[0],xmm9[1],xmm6[1],xmm9[2],xmm6[2],xmm9[3],xmm6[3],xmm9[4],xmm6[4],xmm9[5],xmm6[5],xmm9[6],xmm6[6],xmm9[7],xmm6[7]
+; SSE-NEXT:    movdqa %xmm9, %xmm12
+; SSE-NEXT:    punpcklwd {{.*#+}} xmm12 = xmm12[0],xmm11[0],xmm12[1],xmm11[1],xmm12[2],xmm11[2],xmm12[3],xmm11[3]
+; SSE-NEXT:    movdqa %xmm12, %xmm7
+; SSE-NEXT:    pslldq {{.*#+}} xmm7 = zero,zero,xmm7[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
 ; SSE-NEXT:    movdqa %xmm12, %xmm13
-; SSE-NEXT:    punpcklwd {{.*#+}} xmm13 = xmm13[0],xmm15[0],xmm13[1],xmm15[1],xmm13[2],xmm15[2],xmm13[3],xmm15[3]
-; SSE-NEXT:    movdqa %xmm13, %xmm6
-; SSE-NEXT:    pslldq {{.*#+}} xmm6 = zero,zero,xmm6[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
-; SSE-NEXT:    movdqa %xmm13, %xmm14
-; SSE-NEXT:    packuswb %xmm6, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm2[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    pandn %xmm6, %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm14[0,0,2,3]
-; SSE-NEXT:    pand %xmm1, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm11
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; SSE-NEXT:    pand %xmm6, %xmm9
-; SSE-NEXT:    pandn %xmm11, %xmm6
-; SSE-NEXT:    por %xmm9, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm7[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    pandn %xmm9, %xmm11
-; SSE-NEXT:    pshuflw $246, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm9 = mem[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm9[0,0,2,1]
-; SSE-NEXT:    pand %xmm1, %xmm14
-; SSE-NEXT:    por %xmm11, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm9, %xmm11
+; SSE-NEXT:    packuswb %xmm7, %xmm13
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm13[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm13, %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm2[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm14[0,0,0,0]
+; SSE-NEXT:    pandn %xmm14, %xmm13
+; SSE-NEXT:    por %xmm7, %xmm13
+; SSE-NEXT:    pandn %xmm13, %xmm4
+; SSE-NEXT:    por %xmm8, %xmm4
+; SSE-NEXT:    movdqa %xmm1, %xmm8
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm6[8],xmm8[9],xmm6[9],xmm8[10],xmm6[10],xmm8[11],xmm6[11],xmm8[12],xmm6[12],xmm8[13],xmm6[13],xmm8[14],xmm6[14],xmm8[15],xmm6[15]
+; SSE-NEXT:    movdqa %xmm8, %xmm13
+; SSE-NEXT:    pslldq {{.*#+}} xmm13 = zero,zero,zero,zero,zero,zero,xmm13[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm8, %xmm13
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm7, %xmm14
+; SSE-NEXT:    pandn %xmm13, %xmm14
+; SSE-NEXT:    movdqa %xmm0, %xmm13
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8],xmm6[8],xmm13[9],xmm6[9],xmm13[10],xmm6[10],xmm13[11],xmm6[11],xmm13[12],xmm6[12],xmm13[13],xmm6[13],xmm13[14],xmm6[14],xmm13[15],xmm6[15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm13[2,2,3,3]
+; SSE-NEXT:    psrlq $16, %xmm5
+; SSE-NEXT:    packuswb %xmm13, %xmm5
+; SSE-NEXT:    pslldq {{.*#+}} xmm13 = zero,zero,xmm13[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm6, %xmm13
+; SSE-NEXT:    pand %xmm7, %xmm13
+; SSE-NEXT:    por %xmm14, %xmm13
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm14
+; SSE-NEXT:    punpckhwd {{.*#+}} xmm14 = xmm14[4],xmm9[4],xmm14[5],xmm9[5],xmm14[6],xmm9[6],xmm14[7],xmm9[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm14[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm14[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    punpckhwd {{.*#+}} xmm9 = xmm9[4],xmm11[4],xmm9[5],xmm11[5],xmm9[6],xmm11[6],xmm9[7],xmm11[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm9 = xmm9[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm14, %xmm9
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm11, %xmm9
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm14[0,0,0,0]
 ; SSE-NEXT:    pandn %xmm14, %xmm11
-; SSE-NEXT:    pshufhw {{.*#+}} xmm14 = xmm3[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm14[2,2,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm14
-; SSE-NEXT:    pandn %xmm2, %xmm14
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm14
-; SSE-NEXT:    pand %xmm9, %xmm14
-; SSE-NEXT:    por %xmm11, %xmm14
-; SSE-NEXT:    movdqa %xmm15, %xmm2
-; SSE-NEXT:    punpckhwd {{.*#+}} xmm2 = xmm2[4],xmm12[4],xmm2[5],xmm12[5],xmm2[6],xmm12[6],xmm2[7],xmm12[7]
-; SSE-NEXT:    punpckhwd {{.*#+}} xmm12 = xmm12[4],xmm15[4],xmm12[5],xmm15[5],xmm12[6],xmm15[6],xmm12[7],xmm15[7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm12[0,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,1,2,1]
-; SSE-NEXT:    packuswb %xmm2, %xmm11
-; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm12, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm15[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    pandn %xmm2, %xmm12
-; SSE-NEXT:    por %xmm11, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    pand %xmm2, %xmm12
-; SSE-NEXT:    pandn %xmm14, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm12
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm13[2,2,3,3]
-; SSE-NEXT:    psrldq {{.*#+}} xmm13 = xmm13[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT:    packuswb %xmm13, %xmm2
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm11 = xmm15[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm15, %xmm13
-; SSE-NEXT:    pshufd {{.*#+}} xmm11 = xmm11[0,0,0,0]
-; SSE-NEXT:    pandn %xmm11, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,1,2,2]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm4[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,2,1]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm1
-; SSE-NEXT:    por %xmm3, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm11[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    pand %xmm8, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm7[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    por %xmm2, %xmm8
-; SSE-NEXT:    pand %xmm9, %xmm8
-; SSE-NEXT:    pandn %xmm1, %xmm9
-; SSE-NEXT:    por %xmm8, %xmm9
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm1, %xmm9
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm9
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm11[0],xmm5[1],xmm11[1],xmm5[2],xmm11[2],xmm5[3],xmm11[3],xmm5[4],xmm11[4],xmm5[5],xmm11[5],xmm5[6],xmm11[6],xmm5[7],xmm11[7]
-; SSE-NEXT:    psrldq {{.*#+}} xmm5 = xmm5[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm4[0],xmm10[1],xmm4[1],xmm10[2],xmm4[2],xmm10[3],xmm4[3],xmm10[4],xmm4[4],xmm10[5],xmm4[5],xmm10[6],xmm4[6],xmm10[7],xmm4[7]
-; SSE-NEXT:    pshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[3,1,2,3]
-; SSE-NEXT:    psrldq {{.*#+}} xmm10 = xmm10[13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; SSE-NEXT:    por %xmm0, %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,3,1,4,5,6,7]
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [0,255,255,255,255,0,0,0,255,255,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm10
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    por %xmm10, %xmm0
-; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    movq %xmm0, 48(%rax)
-; SSE-NEXT:    movdqa %xmm9, 16(%rax)
-; SSE-NEXT:    movdqa %xmm12, 32(%rax)
-; SSE-NEXT:    movdqa %xmm6, (%rax)
+; SSE-NEXT:    por %xmm9, %xmm11
+; SSE-NEXT:    pand %xmm6, %xmm11
+; SSE-NEXT:    pandn %xmm13, %xmm6
+; SSE-NEXT:    por %xmm6, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm12[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm12 = xmm12[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm12, %xmm6
+; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm9, %xmm6
+; SSE-NEXT:    pshuflw {{.*#+}} xmm12 = xmm2[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[0,0,0,0]
+; SSE-NEXT:    pandn %xmm12, %xmm9
+; SSE-NEXT:    por %xmm6, %xmm9
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,2,2]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm3[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm3 = xmm3[0,3],xmm8[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm12[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm3, %xmm8
+; SSE-NEXT:    pand %xmm7, %xmm8
+; SSE-NEXT:    pandn %xmm5, %xmm7
+; SSE-NEXT:    por %xmm8, %xmm7
+; SSE-NEXT:    pand %xmm6, %xmm7
+; SSE-NEXT:    pandn %xmm9, %xmm6
+; SSE-NEXT:    por %xmm6, %xmm7
+; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,0,255,255,255,255,255,255,255,255,255,255,255]
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [0,255,255,255,255,0,0,0,255,255,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm10[1,3,2,3,4,5,6,7]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[0,1,3,1,4,5,6,7]
+; SSE-NEXT:    pandn %xmm5, %xmm3
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    pandn %xmm3, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movq %xmm1, 48(%rax)
+; SSE-NEXT:    movdqa %xmm7, 16(%rax)
+; SSE-NEXT:    movdqa %xmm11, 32(%rax)
+; SSE-NEXT:    movdqa %xmm4, (%rax)
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride7_vf8:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX1-ONLY-NEXT:    movq {{[0-9]+}}(%rsp), %r10
 ; AVX1-ONLY-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
 ; AVX1-ONLY-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
; AVX512BW-FAST-NEXT:    retq
   %interleaved.vec = shufflevector <56 x i8> %8, <56 x i8> poison, <56 x i32> <i32 0, i32 8, i32 16, i32 24, i32 32, i32 40, i32 48, i32 1, i32 9, i32 17, i32 25, i32 33, i32 41, i32 49, i32 2, i32 10, i32 18, i32 26, i32 34, i32 42, i32 50, i32 3, i32 11, i32 19, i32 27, i32 35, i32 43, i32 51, i32 4, i32 12, i32 20, i32 28, i32 36, i32 44, i32 52, i32 5, i32 13, i32 21, i32 29, i32 37, i32 45, i32 53, i32 6, i32 14, i32 22, i32 30, i32 38, i32 46, i32 54, i32 7, i32 15, i32 23, i32 31, i32 39, i32 47, i32 55>
   store <56 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride7_vf16(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %in.vecptr5, ptr %in.vecptr6, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride7_vf16:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    subq $72, %rsp
+; SSE-NEXT:    pushq %rax
 ; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    movdqa (%rdi), %xmm3
-; SSE-NEXT:    movdqa (%rsi), %xmm5
-; SSE-NEXT:    movdqa (%rdx), %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa (%rdi), %xmm1
+; SSE-NEXT:    movdqa (%rsi), %xmm15
+; SSE-NEXT:    movdqa (%rdx), %xmm3
 ; SSE-NEXT:    movdqa (%rcx), %xmm7
-; SSE-NEXT:    movdqa (%r8), %xmm4
-; SSE-NEXT:    movdqa %xmm4, (%rsp) # 16-byte Spill
-; SSE-NEXT:    movdqa (%r9), %xmm9
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    movdqa %xmm7, %xmm1
-; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm7[8],xmm1[9],xmm7[9],xmm1[10],xmm7[10],xmm1[11],xmm7[11],xmm1[12],xmm7[12],xmm1[13],xmm7[13],xmm1[14],xmm7[14],xmm1[15],xmm7[15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pandn %xmm2, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm3[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    movdqa %xmm3, %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[2,2,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm15, %xmm3
-; SSE-NEXT:    movdqa %xmm5, %xmm2
-; SSE-NEXT:    movdqa %xmm5, %xmm14
+; SSE-NEXT:    movdqa (%r8), %xmm5
 ; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm5[8],xmm2[9],xmm5[9],xmm2[10],xmm5[10],xmm2[11],xmm5[11],xmm2[12],xmm5[12],xmm2[13],xmm5[13],xmm2[14],xmm5[14],xmm2[15],xmm5[15]
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm2[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,1,3,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm10
-; SSE-NEXT:    pandn %xmm8, %xmm10
-; SSE-NEXT:    por %xmm3, %xmm10
-; SSE-NEXT:    pand %xmm0, %xmm10
-; SSE-NEXT:    por %xmm1, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm4[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
-; SSE-NEXT:    movdqa %xmm5, %xmm8
-; SSE-NEXT:    pandn %xmm1, %xmm8
-; SSE-NEXT:    movdqa %xmm9, %xmm1
-; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm9[8],xmm1[9],xmm9[9],xmm1[10],xmm9[10],xmm1[11],xmm9[11],xmm1[12],xmm9[12],xmm1[13],xmm9[13],xmm1[14],xmm9[14],xmm1[15],xmm9[15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,3]
-; SSE-NEXT:    pand %xmm5, %xmm1
-; SSE-NEXT:    por %xmm8, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; SSE-NEXT:    pand %xmm12, %xmm10
-; SSE-NEXT:    pandn %xmm1, %xmm12
-; SSE-NEXT:    movdqa (%rax), %xmm3
-; SSE-NEXT:    por %xmm10, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    movdqa %xmm3, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm11, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    pand %xmm11, %xmm12
-; SSE-NEXT:    por %xmm12, %xmm4
-; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm14[2,1,2,3]
+; SSE-NEXT:    movdqa (%r9), %xmm8
+; SSE-NEXT:    movdqa (%rax), %xmm6
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm1, %xmm12
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm15[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,0,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm10, %xmm14
-; SSE-NEXT:    pandn %xmm1, %xmm14
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm6[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm10, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm14
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm7[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,0,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm0, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm3, %xmm10
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
 ; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm4, %xmm12
-; SSE-NEXT:    pandn %xmm1, %xmm12
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
 ; SSE-NEXT:    pand %xmm4, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm1, %xmm14
-; SSE-NEXT:    pandn %xmm12, %xmm1
-; SSE-NEXT:    por %xmm14, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm9[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm12 = xmm12[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm5, %xmm14
-; SSE-NEXT:    pandn %xmm12, %xmm14
-; SSE-NEXT:    movdqa (%rsp), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm12 = xmm3[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm12[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm5, %xmm12
-; SSE-NEXT:    por %xmm12, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
-; SSE-NEXT:    pand %xmm12, %xmm14
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm8[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm7[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pandn %xmm7, %xmm12
-; SSE-NEXT:    por %xmm14, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    pandn %xmm12, %xmm7
-; SSE-NEXT:    por %xmm1, %xmm7
-; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm7
-; SSE-NEXT:    pandn %xmm1, %xmm7
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm6[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    movdqa %xmm6, %xmm9
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    pand %xmm13, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm7
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm2[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm5, %xmm12
-; SSE-NEXT:    pandn %xmm1, %xmm12
-; SSE-NEXT:    pshufd $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm5, %xmm1
-; SSE-NEXT:    por %xmm12, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pandn %xmm7, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pshufd $229, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[1,1,2,3]
-; SSE-NEXT:    movdqa %xmm10, %xmm7
-; SSE-NEXT:    pandn %xmm1, %xmm7
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm3[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm3, %xmm14
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
-; SSE-NEXT:    pand %xmm10, %xmm1
-; SSE-NEXT:    por %xmm7, %xmm1
-; SSE-NEXT:    movdqa %xmm8, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm8[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[2,1,3,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm12
-; SSE-NEXT:    pandn %xmm7, %xmm12
-; SSE-NEXT:    pand %xmm15, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pandn %xmm12, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm7[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,3,0,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm3, %xmm4
+; SSE-NEXT:    por %xmm1, %xmm4
+; SSE-NEXT:    pandn %xmm4, %xmm0
+; SSE-NEXT:    por %xmm2, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
+; SSE-NEXT:    pand %xmm9, %xmm0
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm5[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm5, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm8[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,3,2,0]
+; SSE-NEXT:    movdqa %xmm5, %xmm4
+; SSE-NEXT:    pandn %xmm3, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
+; SSE-NEXT:    pand %xmm2, %xmm4
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm6[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm3, %xmm2
+; SSE-NEXT:    por %xmm4, %xmm2
+; SSE-NEXT:    pandn %xmm2, %xmm9
+; SSE-NEXT:    por %xmm0, %xmm9
+; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm10, %xmm1
+; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm7[8],xmm1[9],xmm7[9],xmm1[10],xmm7[10],xmm1[11],xmm7[11],xmm1[12],xmm7[12],xmm1[13],xmm7[13],xmm1[14],xmm7[14],xmm1[15],xmm7[15]
+; SSE-NEXT:    pxor %xmm0, %xmm0
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} 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]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm3, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm13, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    movdqa %xmm12, %xmm11
+; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm15[8],xmm12[9],xmm15[9],xmm12[10],xmm15[10],xmm12[11],xmm15[11],xmm12[12],xmm15[12],xmm12[13],xmm15[13],xmm12[14],xmm15[14],xmm12[15],xmm15[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm0[8],xmm12[9],xmm0[9],xmm12[10],xmm0[10],xmm12[11],xmm0[11],xmm12[12],xmm0[12],xmm12[13],xmm0[13],xmm12[14],xmm0[14],xmm12[15],xmm0[15]
+; SSE-NEXT:    movdqa %xmm12, %xmm2
+; SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm12[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm4
+; SSE-NEXT:    pand %xmm13, %xmm4
+; SSE-NEXT:    por %xmm3, %xmm4
+; SSE-NEXT:    movdqa %xmm8, %xmm2
+; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm9[8],xmm2[9],xmm9[9],xmm2[10],xmm9[10],xmm2[11],xmm9[11],xmm2[12],xmm9[12],xmm2[13],xmm9[13],xmm2[14],xmm9[14],xmm2[15],xmm9[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} 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]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[2,2,3,3]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm3, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; SSE-NEXT:    pand %xmm3, %xmm4
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    por %xmm4, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm6[0,1,2,3,6,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
+; SSE-NEXT:    pandn %xmm2, %xmm0
+; SSE-NEXT:    por %xmm3, %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm7
-; SSE-NEXT:    pandn %xmm0, %xmm7
-; SSE-NEXT:    movdqa %xmm2, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm7
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    movdqa %xmm10, %xmm0
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3],xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,0]
-; SSE-NEXT:    movdqa %xmm11, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    movdqa %xmm9, %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm9[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
+; SSE-NEXT:    movdqa %xmm0, %xmm10
+; SSE-NEXT:    pxor %xmm0, %xmm0
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm0[0],xmm10[1],xmm0[1],xmm10[2],xmm0[2],xmm10[3],xmm0[3],xmm10[4],xmm0[4],xmm10[5],xmm0[5],xmm10[6],xmm0[6],xmm10[7],xmm0[7]
+; SSE-NEXT:    movdqa %xmm10, %xmm2
+; SSE-NEXT:    psrlq $16, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm10[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm3
+; SSE-NEXT:    movdqa %xmm11, %xmm6
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm15[0],xmm6[1],xmm15[1],xmm6[2],xmm15[2],xmm6[3],xmm15[3],xmm6[4],xmm15[4],xmm6[5],xmm15[5],xmm6[6],xmm15[6],xmm6[7],xmm15[7]
+; SSE-NEXT:    movdqa %xmm6, %xmm11
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0],xmm0[0],xmm11[1],xmm0[1],xmm11[2],xmm0[2],xmm11[3],xmm0[3],xmm11[4],xmm0[4],xmm11[5],xmm0[5],xmm11[6],xmm0[6],xmm11[7],xmm0[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm11[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm11[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm14[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm4, %xmm14
 ; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm12
-; SSE-NEXT:    pandn %xmm7, %xmm0
-; SSE-NEXT:    por %xmm12, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm1[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,0,2,1]
-; SSE-NEXT:    movdqa %xmm13, %xmm12
-; SSE-NEXT:    pandn %xmm7, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm14[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm14, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm7
-; SSE-NEXT:    por %xmm7, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm3[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm3, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm5, %xmm9
-; SSE-NEXT:    pandn %xmm7, %xmm9
-; SSE-NEXT:    pand %xmm5, %xmm12
-; SSE-NEXT:    por %xmm12, %xmm9
+; SSE-NEXT:    pand %xmm0, %xmm14
+; SSE-NEXT:    pandn %xmm3, %xmm0
+; SSE-NEXT:    por %xmm14, %xmm0
+; SSE-NEXT:    movdqa %xmm9, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3],xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; SSE-NEXT:    movdqa %xmm9, %xmm8
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],mem[0],xmm8[1],mem[1],xmm8[2],mem[2],xmm8[3],mem[3],xmm8[4],mem[4],xmm8[5],mem[5],xmm8[6],mem[6],xmm8[7],mem[7]
+; SSE-NEXT:    movdqa %xmm8, %xmm14
+; SSE-NEXT:    pslldq {{.*#+}} xmm14 = zero,zero,xmm14[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm8, %xmm4
+; SSE-NEXT:    packuswb %xmm14, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,2,3]
+; SSE-NEXT:    pand %xmm5, %xmm4
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    pshuflw {{.*#+}} xmm14 = xmm3[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm14 = xmm14[0,0,0,0]
+; SSE-NEXT:    pandn %xmm14, %xmm5
+; SSE-NEXT:    por %xmm4, %xmm5
 ; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
 ; SSE-NEXT:    pand %xmm14, %xmm0
-; SSE-NEXT:    pandn %xmm9, %xmm14
+; SSE-NEXT:    pandn %xmm5, %xmm14
 ; SSE-NEXT:    por %xmm0, %xmm14
-; SSE-NEXT:    pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm10, %xmm9
-; SSE-NEXT:    pandn %xmm0, %xmm9
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
-; SSE-NEXT:    pand %xmm10, %xmm0
-; SSE-NEXT:    por %xmm9, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm8[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa %xmm8, %xmm7
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm5, %xmm12
-; SSE-NEXT:    pandn %xmm9, %xmm12
-; SSE-NEXT:    pshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm6 = mem[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,2,1]
-; SSE-NEXT:    pand %xmm5, %xmm6
-; SSE-NEXT:    por %xmm12, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
-; SSE-NEXT:    pand %xmm9, %xmm6
-; SSE-NEXT:    pandn %xmm0, %xmm9
-; SSE-NEXT:    por %xmm6, %xmm9
-; SSE-NEXT:    pshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    pand %xmm13, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    pandn %xmm0, %xmm13
-; SSE-NEXT:    por %xmm3, %xmm13
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; SSE-NEXT:    pand %xmm3, %xmm9
-; SSE-NEXT:    pandn %xmm13, %xmm3
-; SSE-NEXT:    por %xmm9, %xmm3
-; SSE-NEXT:    pshufd $250, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm5, %xmm6
-; SSE-NEXT:    pandn %xmm0, %xmm6
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm5, %xmm0
-; SSE-NEXT:    por %xmm6, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm6, %xmm9
-; SSE-NEXT:    pandn %xmm0, %xmm9
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm15 = xmm15[8],mem[8],xmm15[9],mem[9],xmm15[10],mem[10],xmm15[11],mem[11],xmm15[12],mem[12],xmm15[13],mem[13],xmm15[14],mem[14],xmm15[15],mem[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm4[0],xmm15[1],xmm4[1],xmm15[2],xmm4[2],xmm15[3],xmm4[3],xmm15[4],xmm4[4],xmm15[5],xmm4[5],xmm15[6],xmm4[6],xmm15[7],xmm4[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm15[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm15[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm0, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm5, %xmm4
+; SSE-NEXT:    pandn %xmm1, %xmm5
+; SSE-NEXT:    por %xmm4, %xmm5
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm2 = xmm2[8],mem[8],xmm2[9],mem[9],xmm2[10],mem[10],xmm2[11],mem[11],xmm2[12],mem[12],xmm2[13],mem[13],xmm2[14],mem[14],xmm2[15],mem[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm0
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3],xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm4, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    pshufhw {{.*#+}} xmm15 = xmm3[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm15[0,2,2,3]
+; SSE-NEXT:    pandn %xmm15, %xmm4
+; SSE-NEXT:    por %xmm1, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; SSE-NEXT:    pand %xmm15, %xmm5
+; SSE-NEXT:    pandn %xmm4, %xmm15
+; SSE-NEXT:    por %xmm5, %xmm15
+; SSE-NEXT:    movdqa %xmm12, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm1, %xmm12
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm7 = xmm7[8],mem[8],xmm7[9],mem[9],xmm7[10],mem[10],xmm7[11],mem[11],xmm7[12],mem[12],xmm7[13],mem[13],xmm7[14],mem[14],xmm7[15],mem[15]
+; SSE-NEXT:    movdqa %xmm7, %xmm1
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm4[8],xmm1[9],xmm4[9],xmm1[10],xmm4[10],xmm1[11],xmm4[11],xmm1[12],xmm4[12],xmm1[13],xmm4[13],xmm1[14],xmm4[14],xmm1[15],xmm4[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm4[0],xmm7[1],xmm4[1],xmm7[2],xmm4[2],xmm7[3],xmm4[3],xmm7[4],xmm4[4],xmm7[5],xmm4[5],xmm7[6],xmm4[6],xmm7[7],xmm4[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[3,1],xmm1[0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm7[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm12[0,0,2,3]
+; SSE-NEXT:    pand %xmm13, %xmm4
+; SSE-NEXT:    pandn %xmm1, %xmm13
+; SSE-NEXT:    por %xmm4, %xmm13
+; SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pxor %xmm12, %xmm12
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm12[8],xmm2[9],xmm12[9],xmm2[10],xmm12[10],xmm2[11],xmm12[11],xmm2[12],xmm12[12],xmm2[13],xmm12[13],xmm2[14],xmm12[14],xmm2[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm3, %xmm7
+; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm3[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,1,3,3]
+; SSE-NEXT:    pandn %xmm4, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm0, %xmm13
+; SSE-NEXT:    pandn %xmm1, %xmm0
+; SSE-NEXT:    por %xmm13, %xmm0
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm4, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm7[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    pand %xmm6, %xmm12
-; SSE-NEXT:    por %xmm9, %xmm12
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm7[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm9
-; SSE-NEXT:    pandn %xmm0, %xmm9
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm9
-; SSE-NEXT:    pand %xmm10, %xmm9
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pandn %xmm0, %xmm10
-; SSE-NEXT:    por %xmm9, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm10
-; SSE-NEXT:    pandn %xmm12, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm10
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,1,2,2]
-; SSE-NEXT:    pshuflw $229, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm9 = mem[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm9 = xmm9[0,0,2,1]
-; SSE-NEXT:    pand %xmm5, %xmm9
-; SSE-NEXT:    pandn %xmm0, %xmm5
-; SSE-NEXT:    por %xmm9, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    pshufd $101, {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm9 = mem[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    pandn %xmm9, %xmm11
-; SSE-NEXT:    por %xmm0, %xmm11
-; SSE-NEXT:    pand %xmm6, %xmm11
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    por %xmm11, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm7[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
-; SSE-NEXT:    pandn %xmm2, %xmm15
-; SSE-NEXT:    por %xmm0, %xmm15
-; SSE-NEXT:    pand %xmm4, %xmm15
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    por %xmm15, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm6
-; SSE-NEXT:    pandn %xmm4, %xmm0
-; SSE-NEXT:    por %xmm6, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm12[8],xmm1[9],xmm12[9],xmm1[10],xmm12[10],xmm1[11],xmm12[11],xmm1[12],xmm12[12],xmm1[13],xmm12[13],xmm1[14],xmm12[14],xmm1[15],xmm12[15]
+; SSE-NEXT:    movdqa %xmm1, %xmm4
+; SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,zero,zero,zero,zero,xmm4[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm1, %xmm4
+; SSE-NEXT:    movdqa %xmm1, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm5
+; SSE-NEXT:    pandn %xmm4, %xmm5
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm12[8],xmm6[9],xmm12[9],xmm6[10],xmm12[10],xmm6[11],xmm12[11],xmm6[12],xmm12[12],xmm6[13],xmm12[13],xmm6[14],xmm12[14],xmm6[15],xmm12[15]
+; SSE-NEXT:    psrlq $16, %xmm11
+; SSE-NEXT:    packuswb %xmm6, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm6[2,2,3,3]
+; SSE-NEXT:    movdqa %xmm6, %xmm2
+; SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,xmm2[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm4, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    por %xmm5, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm12[8],xmm9[9],xmm12[9],xmm9[10],xmm12[10],xmm9[11],xmm12[11],xmm9[12],xmm12[12],xmm9[13],xmm12[13],xmm9[14],xmm12[14],xmm9[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm9[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm9[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm4, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm6, %xmm5
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm7[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; SSE-NEXT:    pandn %xmm4, %xmm6
+; SSE-NEXT:    por %xmm5, %xmm6
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm6
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    por %xmm4, %xmm6
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm10[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,3],xmm3[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm10[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm4, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm11[0,1,2,2]
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm4, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm8[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm8 = xmm8[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm8, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm7[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE-NEXT:    pandn %xmm3, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm2, %xmm1
+; SSE-NEXT:    pandn %xmm4, %xmm2
+; SSE-NEXT:    por %xmm1, %xmm2
 ; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    movdqa %xmm0, 16(%rax)
-; SSE-NEXT:    movdqa %xmm10, 32(%rax)
-; SSE-NEXT:    movdqa %xmm3, 64(%rax)
+; SSE-NEXT:    movdqa %xmm2, 16(%rax)
+; SSE-NEXT:    movdqa %xmm6, 32(%rax)
+; SSE-NEXT:    movdqa %xmm0, 80(%rax)
+; SSE-NEXT:    movdqa %xmm15, 64(%rax)
 ; SSE-NEXT:    movdqa %xmm14, (%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 80(%rax)
+; SSE-NEXT:    movaps %xmm0, 96(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 48(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 96(%rax)
-; SSE-NEXT:    addq $72, %rsp
+; SSE-NEXT:    popq %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride7_vf16:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX1-ONLY-NEXT:    movq {{[0-9]+}}(%rsp), %r10
 ; AVX1-ONLY-NEXT:    vmovdqa (%rdi), %xmm4
 ; AVX1-ONLY-NEXT:    vmovdqa (%rsi), %xmm5
; AVX512BW-FAST-NEXT:    retq
   %interleaved.vec = shufflevector <112 x i8> %8, <112 x i8> poison, <112 x i32> <i32 0, i32 16, i32 32, i32 48, i32 64, i32 80, i32 96, i32 1, i32 17, i32 33, i32 49, i32 65, i32 81, i32 97, i32 2, i32 18, i32 34, i32 50, i32 66, i32 82, i32 98, i32 3, i32 19, i32 35, i32 51, i32 67, i32 83, i32 99, i32 4, i32 20, i32 36, i32 52, i32 68, i32 84, i32 100, i32 5, i32 21, i32 37, i32 53, i32 69, i32 85, i32 101, i32 6, i32 22, i32 38, i32 54, i32 70, i32 86, i32 102, i32 7, i32 23, i32 39, i32 55, i32 71, i32 87, i32 103, i32 8, i32 24, i32 40, i32 56, i32 72, i32 88, i32 104, i32 9, i32 25, i32 41, i32 57, i32 73, i32 89, i32 105, i32 10, i32 26, i32 42, i32 58, i32 74, i32 90, i32 106, i32 11, i32 27, i32 43, i32 59, i32 75, i32 91, i32 107, i32 12, i32 28, i32 44, i32 60, i32 76, i32 92, i32 108, i32 13, i32 29, i32 45, i32 61, i32 77, i32 93, i32 109, i32 14, i32 30, i32 46, i32 62, i32 78, i32 94, i32 110, i32 15, i32 31, i32 47, i32 63, i32 79, i32 95, i32 111>
   store <112 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride7_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %in.vecptr5, ptr %in.vecptr6, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride7_vf32:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    subq $344, %rsp # imm = 0x158
-; SSE-NEXT:    movdqa 16(%rdi), %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa 16(%rsi), %xmm6
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa 16(%rdx), %xmm8
-; SSE-NEXT:    movdqa 16(%rcx), %xmm12
-; SSE-NEXT:    movdqa 16(%r8), %xmm7
-; SSE-NEXT:    movdqa 16(%r9), %xmm14
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm9, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm6[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm9, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    subq $216, %rsp
+; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; SSE-NEXT:    movdqa 16(%rdi), %xmm1
+; SSE-NEXT:    movdqa 16(%rsi), %xmm2
+; SSE-NEXT:    movdqa 16(%rdx), %xmm15
+; SSE-NEXT:    movdqa 16(%rcx), %xmm6
+; SSE-NEXT:    movdqa 16(%r8), %xmm14
+; SSE-NEXT:    movdqa 16(%r9), %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 16(%rax), %xmm5
 ; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm8[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm8
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm12
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pxor %xmm11, %xmm11
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm11[8],xmm0[9],xmm11[9],xmm0[10],xmm11[10],xmm0[11],xmm11[11],xmm0[12],xmm11[12],xmm0[13],xmm11[13],xmm0[14],xmm11[14],xmm0[15],xmm11[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
 ; SSE-NEXT:    movdqa %xmm10, %xmm2
 ; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm12, %xmm4
-; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm12[8],xmm4[9],xmm12[9],xmm4[10],xmm12[10],xmm4[11],xmm12[11],xmm4[12],xmm12[12],xmm4[13],xmm12[13],xmm4[14],xmm12[14],xmm4[15],xmm12[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm4, %xmm5
-; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,5,5]
+; SSE-NEXT:    movdqa %xmm6, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm15[8],xmm1[9],xmm15[9],xmm1[10],xmm15[10],xmm1[11],xmm15[11],xmm1[12],xmm15[12],xmm1[13],xmm15[13],xmm1[14],xmm15[14],xmm1[15],xmm15[15]
+; SSE-NEXT:    movdqa %xmm1, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm11[8],xmm3[9],xmm11[9],xmm3[10],xmm11[10],xmm3[11],xmm11[11],xmm3[12],xmm11[12],xmm3[13],xmm11[13],xmm3[14],xmm11[14],xmm3[15],xmm11[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3],xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,3],xmm3[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm3, %xmm1
 ; SSE-NEXT:    pand %xmm10, %xmm1
 ; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    pand %xmm11, %xmm1
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    pand %xmm2, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm7[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm7, %xmm11
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm7, %xmm1
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm14[0,1,2,3,5,6,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
+; SSE-NEXT:    movdqa %xmm7, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    por %xmm1, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm9, %xmm3
+; SSE-NEXT:    movdqa %xmm4, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm4[8],xmm1[9],xmm4[9],xmm1[10],xmm4[10],xmm1[11],xmm4[11],xmm1[12],xmm4[12],xmm1[13],xmm4[13],xmm1[14],xmm4[14],xmm1[15],xmm4[15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,1,2,3]
+; SSE-NEXT:    movdqa %xmm9, %xmm4
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    por %xmm3, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm5[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,3]
+; SSE-NEXT:    movdqa %xmm13, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    pand %xmm13, %xmm4
+; SSE-NEXT:    por %xmm4, %xmm3
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm15, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm6[8],xmm2[9],xmm6[9],xmm2[10],xmm6[10],xmm2[11],xmm6[11],xmm2[12],xmm6[12],xmm2[13],xmm6[13],xmm2[14],xmm6[14],xmm2[15],xmm6[15]
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm11[8],xmm2[9],xmm11[9],xmm2[10],xmm11[10],xmm2[11],xmm11[11],xmm2[12],xmm11[12],xmm2[13],xmm11[13],xmm2[14],xmm11[14],xmm2[15],xmm11[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm3, %xmm2
+; SSE-NEXT:    movdqa %xmm10, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm0, %xmm2
+; SSE-NEXT:    pand %xmm10, %xmm2
+; SSE-NEXT:    por %xmm3, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm14[0,1,2,3,7,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,5,6,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm0, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm5[0,1,2,3,6,7,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
+; SSE-NEXT:    movdqa %xmm7, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    pand %xmm7, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm12[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,0,3]
+; SSE-NEXT:    movdqa %xmm9, %xmm2
 ; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm14, %xmm1
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm9, %xmm0
+; SSE-NEXT:    movdqa %xmm9, %xmm8
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[2,1,2,3]
+; SSE-NEXT:    movdqa %xmm6, %xmm12
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,0,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm15[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm11, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm13
+; SSE-NEXT:    pandn %xmm3, %xmm13
+; SSE-NEXT:    pand %xmm0, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm13
+; SSE-NEXT:    movdqa %xmm14, %xmm10
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm6[0],xmm10[1],xmm6[1],xmm10[2],xmm6[2],xmm10[3],xmm6[3],xmm10[4],xmm6[4],xmm10[5],xmm6[5],xmm10[6],xmm6[6],xmm10[7],xmm6[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm14[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8],xmm6[8],xmm14[9],xmm6[9],xmm14[10],xmm6[10],xmm14[11],xmm6[11],xmm14[12],xmm6[12],xmm14[13],xmm6[13],xmm14[14],xmm6[14],xmm14[15],xmm6[15]
 ; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm14[8],xmm1[9],xmm14[9],xmm1[10],xmm14[10],xmm1[11],xmm14[11],xmm1[12],xmm14[12],xmm1[13],xmm14[13],xmm1[14],xmm14[14],xmm1[15],xmm14[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,2,3]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm6[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm5[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    movdqa %xmm4, %xmm5
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm9[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
 ; SSE-NEXT:    movdqa %xmm1, %xmm4
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm15, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    pand %xmm15, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    movdqa 16(%rax), %xmm7
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm7[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm13, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pand %xmm13, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    pand %xmm1, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
+; SSE-NEXT:    movdqa %xmm14, %xmm2
+; SSE-NEXT:    pandn %xmm4, %xmm2
+; SSE-NEXT:    pand %xmm14, %xmm13
+; SSE-NEXT:    por %xmm13, %xmm2
 ; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm5[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm9, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm9, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm0
-; SSE-NEXT:    pandn %xmm1, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm6[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm2
+; SSE-NEXT:    movdqa (%rsi), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[3,1,0,3]
+; SSE-NEXT:    movdqa %xmm8, %xmm4
+; SSE-NEXT:    movdqa %xmm8, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    movdqa (%rdi), %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa (%rcx), %xmm13
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm13[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,0,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm11, %xmm6
+; SSE-NEXT:    pandn %xmm2, %xmm6
+; SSE-NEXT:    movdqa (%rdx), %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm7[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm11, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm6
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    pandn %xmm6, %xmm0
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    movdqa (%r9), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,0]
+; SSE-NEXT:    movdqa %xmm5, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    movdqa (%r8), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm5, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    pand %xmm1, %xmm3
+; SSE-NEXT:    movdqa (%rax), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm2, %xmm1
+; SSE-NEXT:    por %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm14, %xmm0
+; SSE-NEXT:    pandn %xmm1, %xmm14
+; SSE-NEXT:    por %xmm0, %xmm14
+; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm12[0],xmm15[1],xmm12[1],xmm15[2],xmm12[2],xmm15[3],xmm12[3],xmm15[4],xmm12[4],xmm15[5],xmm12[5],xmm15[6],xmm12[6],xmm15[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm15, %xmm0
+; SSE-NEXT:    pxor %xmm12, %xmm12
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm12[8],xmm0[9],xmm12[9],xmm0[10],xmm12[10],xmm0[11],xmm12[11],xmm0[12],xmm12[12],xmm0[13],xmm12[13],xmm0[14],xmm12[14],xmm0[15],xmm12[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm14, %xmm2
 ; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm9[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    pand %xmm13, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    pand %xmm3, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm11[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pand %xmm10, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm3 = xmm3[0],mem[0],xmm3[1],mem[1],xmm3[2],mem[2],xmm3[3],mem[3],xmm3[4],mem[4],xmm3[5],mem[5],xmm3[6],mem[6],xmm3[7],mem[7]
+; SSE-NEXT:    movdqa %xmm3, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm12[8],xmm1[9],xmm12[9],xmm1[10],xmm12[10],xmm1[11],xmm12[11],xmm1[12],xmm12[12],xmm1[13],xmm12[13],xmm1[14],xmm12[14],xmm1[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm12[0],xmm3[1],xmm12[1],xmm3[2],xmm12[2],xmm3[3],xmm12[3],xmm3[4],xmm12[4],xmm3[5],xmm12[5],xmm3[6],xmm12[6],xmm3[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    psrlq $16, %xmm3
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm6, %xmm1
+; SSE-NEXT:    pand %xmm14, %xmm1
 ; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
-; SSE-NEXT:    movdqa %xmm5, %xmm4
-; SSE-NEXT:    pandn %xmm2, %xmm4
-; SSE-NEXT:    pand %xmm5, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm4
-; SSE-NEXT:    movdqa %xmm7, %xmm0
-; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm7[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm6, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm2
 ; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    pand %xmm6, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm2
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd $230, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,0,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm15, %xmm1
+; SSE-NEXT:    movdqa %xmm10, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm12[8],xmm1[9],xmm12[9],xmm1[10],xmm12[10],xmm1[11],xmm12[11],xmm1[12],xmm12[12],xmm1[13],xmm12[13],xmm1[14],xmm12[14],xmm1[15],xmm12[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm6, %xmm1
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm9[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm5
+; SSE-NEXT:    pandn %xmm6, %xmm4
+; SSE-NEXT:    pand %xmm5, %xmm1
 ; SSE-NEXT:    por %xmm1, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm12[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,0,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm12, %xmm7
-; SSE-NEXT:    pandn %xmm1, %xmm7
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm8[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm12, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm7
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm7, %xmm8
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm14[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm10, %xmm7
-; SSE-NEXT:    pandn %xmm4, %xmm7
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm11[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm10, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm7
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm0[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm5, %xmm9
-; SSE-NEXT:    pandn %xmm4, %xmm9
-; SSE-NEXT:    pand %xmm5, %xmm7
-; SSE-NEXT:    por %xmm7, %xmm9
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    pandn %xmm9, %xmm4
-; SSE-NEXT:    pand %xmm3, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm4
+; SSE-NEXT:    pand %xmm11, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
 ; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa (%rsi), %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm0, %xmm14
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[3,1,0,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm0
-; SSE-NEXT:    movdqa %xmm15, %xmm7
-; SSE-NEXT:    pandn %xmm4, %xmm7
-; SSE-NEXT:    movdqa (%rdi), %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm8[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm8, %xmm15
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm0, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm7
-; SSE-NEXT:    movdqa (%rcx), %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm11[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,3,0,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm4, %xmm8
-; SSE-NEXT:    movdqa (%rdx), %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm2, (%rsp) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm12, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm8
-; SSE-NEXT:    pand %xmm1, %xmm7
-; SSE-NEXT:    pandn %xmm8, %xmm1
-; SSE-NEXT:    por %xmm7, %xmm1
-; SSE-NEXT:    movdqa (%r9), %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm0, %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,1,2,2]
+; SSE-NEXT:    movdqa %xmm14, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0],xmm12[0],xmm15[1],xmm12[1],xmm15[2],xmm12[2],xmm15[3],xmm12[3],xmm15[4],xmm12[4],xmm15[5],xmm12[5],xmm15[6],xmm12[6],xmm15[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm15, (%rsp) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm15, %xmm1
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm0[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm15[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    pand %xmm14, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm12[0],xmm10[1],xmm12[1],xmm10[2],xmm12[2],xmm10[3],xmm12[3],xmm10[4],xmm12[4],xmm10[5],xmm12[5],xmm10[6],xmm12[6],xmm10[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm10, %xmm0
+; SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm10[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm0, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm9[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    pand %xmm11, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm2, %xmm0
+; SSE-NEXT:    pandn %xmm3, %xmm0
+; SSE-NEXT:    pand %xmm2, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm10, %xmm7
-; SSE-NEXT:    pandn %xmm4, %xmm7
-; SSE-NEXT:    movdqa (%r8), %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm6[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm10, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm7
-; SSE-NEXT:    pand %xmm5, %xmm7
-; SSE-NEXT:    movdqa (%rax), %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm8[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm4[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    por %xmm7, %xmm5
-; SSE-NEXT:    pand %xmm3, %xmm1
-; SSE-NEXT:    pandn %xmm5, %xmm3
-; SSE-NEXT:    por %xmm1, %xmm3
-; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm8, %xmm14
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8],xmm5[8],xmm14[9],xmm5[9],xmm14[10],xmm5[10],xmm14[11],xmm5[11],xmm14[12],xmm5[12],xmm14[13],xmm5[13],xmm14[14],xmm5[14],xmm14[15],xmm5[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm14 = xmm14[8],xmm12[8],xmm14[9],xmm12[9],xmm14[10],xmm12[10],xmm14[11],xmm12[11],xmm14[12],xmm12[12],xmm14[13],xmm12[13],xmm14[14],xmm12[14],xmm14[15],xmm12[15]
 ; SSE-NEXT:    movdqa %xmm14, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm14[8],xmm0[9],xmm14[9],xmm0[10],xmm14[10],xmm0[11],xmm14[11],xmm0[12],xmm14[12],xmm0[13],xmm14[13],xmm0[14],xmm14[14],xmm0[15],xmm14[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm14, %xmm1
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
 ; SSE-NEXT:    movdqa %xmm3, %xmm1
 ; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm15[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
+; SSE-NEXT:    movdqa %xmm13, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm7[8],xmm0[9],xmm7[9],xmm0[10],xmm7[10],xmm0[11],xmm7[11],xmm0[12],xmm7[12],xmm0[13],xmm7[13],xmm0[14],xmm7[14],xmm0[15],xmm7[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm12[8],xmm2[9],xmm12[9],xmm2[10],xmm12[10],xmm2[11],xmm12[11],xmm2[12],xmm12[12],xmm2[13],xmm12[13],xmm2[14],xmm12[14],xmm2[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,1],xmm2[0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm11 = xmm11[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm11[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm10, %xmm0
 ; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa %xmm9, %xmm1
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm9[8],xmm1[9],xmm9[9],xmm1[10],xmm9[10],xmm1[11],xmm9[11],xmm1[12],xmm9[12],xmm1[13],xmm9[13],xmm1[14],xmm9[14],xmm1[15],xmm9[15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm11, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm6[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
-; SSE-NEXT:    pand %xmm11, %xmm1
-; SSE-NEXT:    movdqa %xmm11, %xmm14
-; SSE-NEXT:    por %xmm4, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm8[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,1,3,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm7
-; SSE-NEXT:    pandn %xmm4, %xmm7
-; SSE-NEXT:    pand %xmm13, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm7
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm15[8],xmm1[9],xmm15[9],xmm1[10],xmm15[10],xmm1[11],xmm15[11],xmm1[12],xmm15[12],xmm1[13],xmm15[13],xmm1[14],xmm15[14],xmm1[15],xmm15[15]
+; SSE-NEXT:    movdqa %xmm1, %xmm10
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm12[0],xmm10[1],xmm12[1],xmm10[2],xmm12[2],xmm10[3],xmm12[3],xmm10[4],xmm12[4],xmm10[5],xmm12[5],xmm10[6],xmm12[6],xmm10[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm10, %xmm2
+; SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm12[8],xmm1[9],xmm12[9],xmm1[10],xmm12[10],xmm1[11],xmm12[11],xmm1[12],xmm12[12],xmm1[13],xmm12[13],xmm1[14],xmm12[14],xmm1[15],xmm12[15]
+; SSE-NEXT:    pxor %xmm4, %xmm4
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm6[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    por %xmm2, %xmm3
 ; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
 ; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pandn %xmm7, %xmm1
+; SSE-NEXT:    pandn %xmm3, %xmm1
 ; SSE-NEXT:    por %xmm0, %xmm1
 ; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm10, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm10, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm11, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm15[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm12, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    pand %xmm12, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    pand %xmm11, %xmm4
-; SSE-NEXT:    por %xmm1, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm7, %xmm9
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm13[8],xmm9[9],xmm13[9],xmm9[10],xmm13[10],xmm9[11],xmm13[11],xmm9[12],xmm13[12],xmm9[13],xmm13[13],xmm9[14],xmm13[14],xmm9[15],xmm13[15]
 ; SSE-NEXT:    movdqa %xmm9, %xmm0
-; SSE-NEXT:    pandn %xmm4, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm7[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm4
-; SSE-NEXT:    pandn %xmm1, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm5[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,2,1]
-; SSE-NEXT:    pand %xmm2, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm6[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm14, %xmm1
-; SSE-NEXT:    pandn %xmm2, %xmm14
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    movdqa %xmm1, %xmm2
-; SSE-NEXT:    por %xmm4, %xmm14
-; SSE-NEXT:    pand %xmm9, %xmm14
-; SSE-NEXT:    por %xmm0, %xmm14
-; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm15[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm10, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa %xmm11, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshufd $101, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm3, %xmm2
+; SSE-NEXT:    pandn %xmm0, %xmm2
+; SSE-NEXT:    movdqa %xmm5, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm8[8],xmm0[9],xmm8[9],xmm0[10],xmm8[10],xmm0[11],xmm8[11],xmm0[12],xmm8[12],xmm0[13],xmm8[13],xmm0[14],xmm8[14],xmm0[15],xmm8[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    pand %xmm11, %xmm4
-; SSE-NEXT:    movdqa %xmm11, %xmm15
-; SSE-NEXT:    por %xmm1, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm7[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm0
+; SSE-NEXT:    por %xmm2, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm10[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm10 = zero,zero,xmm10[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm2, %xmm10
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm6[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    pand %xmm4, %xmm10
+; SSE-NEXT:    por %xmm10, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm12, %xmm1
+; SSE-NEXT:    pandn %xmm2, %xmm1
+; SSE-NEXT:    pand %xmm12, %xmm0
 ; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pand %xmm12, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm2, %xmm11
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm5[0],xmm8[1],xmm5[1],xmm8[2],xmm5[2],xmm8[3],xmm5[3],xmm8[4],xmm5[4],xmm8[5],xmm5[5],xmm8[6],xmm5[6],xmm8[7],xmm5[7]
+; SSE-NEXT:    movdqa %xmm8, %xmm6
+; SSE-NEXT:    pxor %xmm1, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm1[8],xmm6[9],xmm1[9],xmm6[10],xmm1[10],xmm6[11],xmm1[11],xmm6[12],xmm1[12],xmm6[13],xmm1[13],xmm6[14],xmm1[14],xmm6[15],xmm1[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm1[0],xmm8[1],xmm1[1],xmm8[2],xmm1[2],xmm8[3],xmm1[3],xmm8[4],xmm1[4],xmm8[5],xmm1[5],xmm8[6],xmm1[6],xmm8[7],xmm1[7]
+; SSE-NEXT:    movdqa %xmm8, %xmm0
+; SSE-NEXT:    psrlq $16, %xmm0
+; SSE-NEXT:    packuswb %xmm6, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm5, %xmm2
+; SSE-NEXT:    pandn %xmm0, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm13[0],xmm7[1],xmm13[1],xmm7[2],xmm13[2],xmm7[3],xmm13[3],xmm7[4],xmm13[4],xmm7[5],xmm13[5],xmm7[6],xmm13[6],xmm7[7],xmm13[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm10
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm1[8],xmm10[9],xmm1[9],xmm10[10],xmm1[10],xmm10[11],xmm1[11],xmm10[12],xmm1[12],xmm10[13],xmm1[13],xmm10[14],xmm1[14],xmm10[15],xmm1[15]
+; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm7 = xmm7[0],xmm1[0],xmm7[1],xmm1[1],xmm7[2],xmm1[2],xmm7[3],xmm1[3],xmm7[4],xmm1[4],xmm7[5],xmm1[5],xmm7[6],xmm1[6],xmm7[7],xmm1[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm0
+; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,3],xmm10[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm13 = xmm7[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm13[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm0, %xmm13
+; SSE-NEXT:    pand %xmm5, %xmm13
+; SSE-NEXT:    por %xmm2, %xmm13
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm15[0],xmm10[1],xmm15[1],xmm10[2],xmm15[2],xmm10[3],xmm15[3],xmm10[4],xmm15[4],xmm10[5],xmm15[5],xmm10[6],xmm15[6],xmm10[7],xmm15[7]
+; SSE-NEXT:    movdqa %xmm10, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
 ; SSE-NEXT:    movdqa %xmm2, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    movdqa (%rsp), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
+; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm1, %xmm0
 ; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm1, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm10, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    pshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm4 = mem[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm4, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm9[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[2,1,3,3]
-; SSE-NEXT:    pand %xmm12, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm4 = xmm11[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    pand %xmm3, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm5
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    pandn %xmm5, %xmm3
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm14 = xmm14[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm14[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm10, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm10, %xmm0
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa %xmm15, %xmm7
-; SSE-NEXT:    movdqa %xmm15, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm15[1,1,2,1]
-; SSE-NEXT:    movdqa %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm0[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    movdqa %xmm2, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm2[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,2,1]
-; SSE-NEXT:    pand %xmm3, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm4, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm9[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm11[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm4, %xmm8
-; SSE-NEXT:    pand %xmm12, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm4
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm15[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm14[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,1,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm6
-; SSE-NEXT:    pandn %xmm4, %xmm6
-; SSE-NEXT:    pshuflw $96, {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm4 = mem[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm4
-; SSE-NEXT:    por %xmm4, %xmm6
-; SSE-NEXT:    pand %xmm0, %xmm6
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm2[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,2,1]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm4, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm9[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE-NEXT:    pand %xmm2, %xmm4
-; SSE-NEXT:    movdqa %xmm2, %xmm14
-; SSE-NEXT:    por %xmm4, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm11[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm10, %xmm3
-; SSE-NEXT:    pandn %xmm4, %xmm3
-; SSE-NEXT:    pand %xmm10, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm3
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm15
-; SSE-NEXT:    pandn %xmm3, %xmm15
-; SSE-NEXT:    pand %xmm8, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm15
-; SSE-NEXT:    pshufd $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,1,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm2, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm7[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,2]
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    por %xmm5, %xmm3
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm9[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm5 = mem[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,2,1]
-; SSE-NEXT:    pand %xmm10, %xmm5
-; SSE-NEXT:    por %xmm6, %xmm5
-; SSE-NEXT:    pand %xmm1, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm1
-; SSE-NEXT:    por %xmm5, %xmm1
-; SSE-NEXT:    pshuflw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm12, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm2[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,3]
-; SSE-NEXT:    pand %xmm12, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm4[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm14, %xmm11
-; SSE-NEXT:    movdqa %xmm14, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pand %xmm14, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; SSE-NEXT:    pand %xmm14, %xmm1
-; SSE-NEXT:    pandn %xmm6, %xmm14
-; SSE-NEXT:    por %xmm1, %xmm14
-; SSE-NEXT:    pshuflw $96, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm7[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm15[1,1,1,1,4,5,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm3
-; SSE-NEXT:    pshuflw $216, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,1,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm5
-; SSE-NEXT:    pand %xmm0, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm0
-; SSE-NEXT:    por %xmm5, %xmm0
-; SSE-NEXT:    pshuflw $212, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,2,1]
-; SSE-NEXT:    movdqa %xmm11, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE-NEXT:    pand %xmm11, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm4[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm10, %xmm5
-; SSE-NEXT:    pandn %xmm1, %xmm5
-; SSE-NEXT:    pand %xmm10, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm5
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    pandn %xmm5, %xmm8
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    movdqa (%rsp), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm9[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    pshufhw $246, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
 ; SSE-NEXT:    pandn %xmm1, %xmm11
 ; SSE-NEXT:    por %xmm0, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    pshufhw $232, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
-; SSE-NEXT:    pandn %xmm1, %xmm13
-; SSE-NEXT:    por %xmm0, %xmm13
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
 ; SSE-NEXT:    pand %xmm0, %xmm13
 ; SSE-NEXT:    pandn %xmm11, %xmm0
 ; SSE-NEXT:    por %xmm13, %xmm0
-; SSE-NEXT:    movdqa %xmm0, %xmm5
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm10, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshufhw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm0 = mem[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
-; SSE-NEXT:    pand %xmm10, %xmm0
-; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm7[0,0,1,1]
+; SSE-NEXT:    psrlq $16, %xmm7
+; SSE-NEXT:    packuswb %xmm7, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm0, %xmm7
+; SSE-NEXT:    pandn %xmm1, %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,2]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm8
+; SSE-NEXT:    pand %xmm0, %xmm8
+; SSE-NEXT:    por %xmm7, %xmm8
+; SSE-NEXT:    movdqa %xmm2, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm15[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm13
+; SSE-NEXT:    pandn %xmm1, %xmm13
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,0,2,3]
+; SSE-NEXT:    pand %xmm11, %xmm1
+; SSE-NEXT:    movdqa %xmm11, %xmm5
+; SSE-NEXT:    por %xmm1, %xmm13
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm7
+; SSE-NEXT:    pandn %xmm13, %xmm7
+; SSE-NEXT:    pand %xmm1, %xmm8
+; SSE-NEXT:    por %xmm8, %xmm7
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT:    pxor %xmm11, %xmm11
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3],xmm8[4],xmm11[4],xmm8[5],xmm11[5],xmm8[6],xmm11[6],xmm8[7],xmm11[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm8[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm8
+; SSE-NEXT:    movdqa %xmm8, %xmm13
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm8 = xmm8[8],mem[8],xmm8[9],mem[9],xmm8[10],mem[10],xmm8[11],mem[11],xmm8[12],mem[12],xmm8[13],mem[13],xmm8[14],mem[14],xmm8[15],mem[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3],xmm8[4],xmm11[4],xmm8[5],xmm11[5],xmm8[6],xmm11[6],xmm8[7],xmm11[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm8[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm2, %xmm8
+; SSE-NEXT:    pand %xmm3, %xmm8
+; SSE-NEXT:    pandn %xmm13, %xmm3
+; SSE-NEXT:    por %xmm8, %xmm3
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm8 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3],xmm8[4],xmm11[4],xmm8[5],xmm11[5],xmm8[6],xmm11[6],xmm8[7],xmm11[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm8[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm8 = zero,zero,xmm8[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm2, %xmm8
+; SSE-NEXT:    pand %xmm4, %xmm8
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm13[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    por %xmm8, %xmm4
+; SSE-NEXT:    pand %xmm12, %xmm3
+; SSE-NEXT:    pandn %xmm4, %xmm12
+; SSE-NEXT:    por %xmm3, %xmm12
+; SSE-NEXT:    movdqa (%rsp), %xmm3 # 16-byte Reload
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,1,1]
+; SSE-NEXT:    psrlq $16, %xmm3
+; SSE-NEXT:    packuswb %xmm3, %xmm2
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm4[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm3
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm4[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,2,2]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm3, %xmm4
+; SSE-NEXT:    pand %xmm0, %xmm4
+; SSE-NEXT:    pandn %xmm2, %xmm0
+; SSE-NEXT:    por %xmm4, %xmm0
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm3, %xmm2
+; SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,xmm2[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm2, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,0,2,3]
+; SSE-NEXT:    movdqa %xmm5, %xmm4
+; SSE-NEXT:    pand %xmm5, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm13[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE-NEXT:    pandn %xmm3, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    pandn %xmm4, %xmm1
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm11[8],xmm9[9],xmm11[9],xmm9[10],xmm11[10],xmm9[11],xmm11[11],xmm9[12],xmm11[12],xmm9[13],xmm11[13],xmm9[14],xmm11[14],xmm9[15],xmm11[15]
+; SSE-NEXT:    movdqa %xmm9, %xmm0
+; SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm0, %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm14[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm14 = xmm14[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm14, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    pandn %xmm9, %xmm2
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm2 = xmm2[8],mem[8],xmm2[9],mem[9],xmm2[10],mem[10],xmm2[11],mem[11],xmm2[12],mem[12],xmm2[13],mem[13],xmm2[14],mem[14],xmm2[15],mem[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm11[8],xmm2[9],xmm11[9],xmm2[10],xmm11[10],xmm2[11],xmm11[11],xmm2[12],xmm11[12],xmm2[13],xmm11[13],xmm2[14],xmm11[14],xmm2[15],xmm11[15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm0, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm15[0,1,2,3,6,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
+; SSE-NEXT:    pandn %xmm0, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
 ; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
-; SSE-NEXT:    pand %xmm0, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm0
-; SSE-NEXT:    por %xmm5, %xmm0
-; SSE-NEXT:    pshufd $250, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[2,2,3,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm9[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,2,1]
-; SSE-NEXT:    pand %xmm10, %xmm3
-; SSE-NEXT:    pandn %xmm1, %xmm10
-; SSE-NEXT:    por %xmm3, %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT:    pand %xmm12, %xmm1
-; SSE-NEXT:    pshufhw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,2,2,2]
-; SSE-NEXT:    pandn %xmm3, %xmm12
-; SSE-NEXT:    por %xmm1, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    pand %xmm1, %xmm12
-; SSE-NEXT:    pandn %xmm10, %xmm1
-; SSE-NEXT:    por %xmm12, %xmm1
-; SSE-NEXT:    movdqa %xmm1, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm6[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,2,1]
-; SSE-NEXT:    movdqa %xmm7, %xmm5
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    pshufhw $216, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm2 = mem[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
-; SSE-NEXT:    pandn %xmm2, %xmm5
-; SSE-NEXT:    por %xmm1, %xmm5
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm2, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm4[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    por %xmm5, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    pand %xmm1, %xmm2
-; SSE-NEXT:    pandn %xmm3, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    pandn %xmm4, %xmm0
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm3, %xmm2
+; SSE-NEXT:    pslldq {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,xmm2[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm3, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm6[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm6 = zero,zero,xmm6[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm3, %xmm6
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm6
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    por %xmm6, %xmm3
+; SSE-NEXT:    movdqa %xmm3, %xmm6
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm11[8],xmm10[9],xmm11[9],xmm10[10],xmm11[10],xmm10[11],xmm11[11],xmm10[12],xmm11[12],xmm10[13],xmm11[13],xmm10[14],xmm11[14],xmm10[15],xmm11[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm10[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm10[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm3
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm15[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    por %xmm3, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    pand %xmm2, %xmm4
+; SSE-NEXT:    pandn %xmm6, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm4
 ; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    movdqa %xmm2, 32(%rax)
+; SSE-NEXT:    movdqa %xmm4, 32(%rax)
 ; SSE-NEXT:    movdqa %xmm0, 96(%rax)
-; SSE-NEXT:    movdqa %xmm8, 112(%rax)
-; SSE-NEXT:    movdqa %xmm14, 176(%rax)
-; SSE-NEXT:    movdqa %xmm15, (%rax)
+; SSE-NEXT:    movdqa %xmm1, 112(%rax)
+; SSE-NEXT:    movdqa %xmm12, 176(%rax)
+; SSE-NEXT:    movdqa %xmm7, (%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 16(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 64(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movaps %xmm0, 80(%rax)
+; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 128(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 144(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 80(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 48(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 160(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 208(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 192(%rax)
-; SSE-NEXT:    addq $344, %rsp # imm = 0x158
+; SSE-NEXT:    addq $216, %rsp
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride7_vf32:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    subq $168, %rsp
 ; AVX1-ONLY-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX1-ONLY-NEXT:    vmovdqa 16(%rax), %xmm12
 ; AVX1-ONLY-NEXT:    vpshufb {{.*#+}} xmm0 = xmm12[u,u,u],zero,zero,xmm12[9,u,u,u,u],zero,zero,xmm12[10,u,u,u]

 ; AVX1-ONLY-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; AVX1-ONLY-NEXT:    vmovaps %xmm0, 208(%rax)
 ; AVX1-ONLY-NEXT:    addq $168, %rsp
 ; AVX1-ONLY-NEXT:    vzeroupper
 ; AVX1-ONLY-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: store_i8_stride7_vf32:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    pushq %rax
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm4
-; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm6
-; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %ymm5
-; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm7
-; AVX2-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm2
-; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm8 = ymm3[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[0,2,3,3,4,6,7,7]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,u,u,u,0,255,0,255,u,u,u,0,255,0,255,u,255,u,u,u,0,255,0,255,u,u,u,0,255,0,255,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm9 = ymm4[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[2,2,3,3,6,6,7,7]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,0,255,0,u,u,u,255,0,255,0,u,u,u,u,u,255,0,255,0,u,u,u,255,0,255,0,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm9, %ymm8, %ymm8
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm7[27],zero,ymm7[27,28,29,30],zero,ymm7[28],zero,ymm7[26,27,30,31],zero,ymm7[29]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm2[27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm8, %ymm9, %ymm8
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm0, %ymm8, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm10
-; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %xmm11
-; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm11[8],xmm10[8],xmm11[9],xmm10[9],xmm11[10],xmm10[10],xmm11[11],xmm10[11],xmm11[12],xmm10[12],xmm11[13],xmm10[13],xmm11[14],xmm10[14],xmm11[15],xmm10[15]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm0[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm14
-; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm0
-; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm9 = xmm0[8],xmm14[8],xmm0[9],xmm14[9],xmm0[10],xmm14[10],xmm0[11],xmm14[11],xmm0[12],xmm14[12],xmm0[13],xmm14[13],xmm0[14],xmm14[14],xmm0[15],xmm14[15]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm8, %ymm9, %ymm8
-; AVX2-SLOW-NEXT:    vmovdqa (%r9), %xmm12
-; AVX2-SLOW-NEXT:    vmovdqa (%r8), %xmm13
-; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm9 = xmm13[8],xmm12[8],xmm13[9],xmm12[9],xmm13[10],xmm12[10],xmm13[11],xmm12[11],xmm13[12],xmm12[12],xmm13[13],xmm12[13],xmm13[14],xmm12[14],xmm13[15],xmm12[15]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqa (%rax), %xmm15
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm7 = xmm15[0,1,2,3,4,5,5,6]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[2,2,3,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm9, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm8, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[u],zero,xmm0[7],zero,xmm0[5,u,u,u],zero,xmm0[8],zero,xmm0[6,u,u,u],zero
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm14[u,7],zero,xmm14[5],zero,xmm14[u,u,u,8],zero,xmm14[6],zero,xmm14[u,u,u,9]
-; AVX2-SLOW-NEXT:    vpor %xmm1, %xmm7, %xmm1
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm11[u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6,u,u]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm10[u,u,u,7],zero,xmm10[5],zero,xmm10[u,u,u,8],zero,xmm10[6],zero,xmm10[u,u]
-; AVX2-SLOW-NEXT:    vpor %xmm7, %xmm9, %xmm7
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = zero,xmm12[4,u,u,u],zero,xmm12[7],zero,xmm12[5,u,u,u],zero,xmm12[8],zero,xmm12[6]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm13[4],zero,xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero
-; AVX2-SLOW-NEXT:    vpor %xmm7, %xmm9, %xmm7
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm15[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,0,1,0]
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm3
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %ymm2
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm9
+; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %xmm10
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,1]
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm13
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm15
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm15[8],xmm13[8],xmm15[9],xmm13[9],xmm15[10],xmm13[10],xmm15[11],xmm13[11],xmm15[12],xmm13[12],xmm15[13],xmm13[13],xmm15[14],xmm13[14],xmm15[15],xmm13[15]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vmovdqa (%r9), %xmm11
+; AVX2-SLOW-NEXT:    vmovdqa (%r8), %xmm12
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm12[8],xmm11[8],xmm12[9],xmm11[9],xmm12[10],xmm11[10],xmm12[11],xmm11[11],xmm12[12],xmm11[12],xmm12[13],xmm11[13],xmm12[14],xmm11[14],xmm12[15],xmm11[15]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-SLOW-NEXT:    vmovdqa (%rax), %xmm14
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} xmm6 = xmm14[0,1,2,3,4,5,5,6]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm6 = xmm6[2,2,3,3]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm15[u],zero,xmm15[7],zero,xmm15[5,u,u,u],zero,xmm15[8],zero,xmm15[6,u,u,u],zero
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm13[u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero,xmm13[u,u,u,9]
+; AVX2-SLOW-NEXT:    vpor %xmm5, %xmm6, %xmm5
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm10[u,u,u],zero,xmm10[7],zero,xmm10[5,u,u,u],zero,xmm10[8],zero,xmm10[6,u,u]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm9[u,u,u,7],zero,xmm9[5],zero,xmm9[u,u,u,8],zero,xmm9[6],zero,xmm9[u,u]
+; AVX2-SLOW-NEXT:    vpor %xmm6, %xmm7, %xmm6
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm14[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,1,0]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm12[4],zero,xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero
+; AVX2-SLOW-NEXT:    vpor %xmm7, %xmm8, %xmm7
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm7, %ymm9, %ymm7
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm7, %ymm6, %ymm6
+; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm7
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm1, %ymm7, %ymm9
-; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm14[0],xmm0[0],xmm14[1],xmm0[1],xmm14[2],xmm0[2],xmm14[3],xmm0[3],xmm14[4],xmm0[4],xmm14[5],xmm0[5],xmm14[6],xmm0[6],xmm14[7],xmm0[7]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm10[0],xmm11[0],xmm10[1],xmm11[1],xmm10[2],xmm11[2],xmm10[3],xmm11[3],xmm10[4],xmm11[4],xmm10[5],xmm11[5],xmm10[6],xmm11[6],xmm10[7],xmm11[7]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3],xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm15[1,1,0,0,4,5,6,7]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm7 = xmm7[0,1,2,0]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,0,1,0]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm10
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,1,1,4,4,5,5]
-; AVX2-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [0,0,0,255,0,255,0,0,0,0,255,0,255,0,0,0,0,0,0,255,0,255,0,0,0,0,255,0,255,0,0,0]
-; AVX2-SLOW-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm1, %ymm0, %ymm0
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm5[18],zero,zero,zero,zero,ymm5[21],zero,ymm5[19],zero,zero,zero,zero,ymm5[22],zero,ymm5[20]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zero,zero,zero,ymm3[21],zero,ymm3[19],zero,zero,zero,zero,ymm3[22],zero,ymm3[20],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm1, %ymm0, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[20],zero,ymm2[18],zero,zero,zero,zero,ymm2[21],zero,ymm2[19],zero,zero,zero,zero,ymm2[22]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm12[18],zero,zero,zero,zero,ymm12[21],zero,ymm12[19],zero,zero,zero,zero,ymm12[22],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm11[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,1,1,3,4,5,5,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,2,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm4[23],zero,ymm4[27,20,21,26],zero,ymm4[24],zero,ymm4[26,27,26,27],zero,ymm4[25]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero,ymm6[27],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm5[25],zero,ymm5[23],zero,zero,zero,zero,ymm5[26],zero,ymm5[24],zero,zero,zero,zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero,ymm3[27]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm12[25],zero,ymm12[23],zero,zero,zero,zero,ymm12[26],zero,ymm12[24],zero,zero,zero
-; AVX2-SLOW-NEXT:    vmovdqa %ymm12, %ymm13
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX2-SLOW-NEXT:    vmovdqa %ymm11, %ymm12
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm7, %ymm8, %ymm7
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm1, %ymm7, %ymm1
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero,zero,zero,ymm6[18]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = zero,zero,ymm4[14],zero,zero,zero,zero,zero,zero,ymm4[15],zero,zero,zero,zero,zero,zero,ymm4[16],zero,zero,zero,zero,zero,zero,ymm4[17],zero,zero,zero,zero,zero,zero,ymm4[18],zero
-; AVX2-SLOW-NEXT:    vpor %ymm6, %ymm4, %ymm4
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero,zero,zero
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero,zero
-; AVX2-SLOW-NEXT:    vpor %ymm5, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm4, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = zero,ymm13[1,2,3,0,1,14],zero,ymm13[0,1,0,1,14,15],zero,ymm13[15,16,17,18,19,16],zero,ymm13[30,31,16,17,16,17],zero,ymm13[31,30,31]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[13],zero,zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero
-; AVX2-SLOW-NEXT:    vpor %ymm4, %ymm2, %ymm2
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm12[12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm2, %ymm4, %ymm2
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm5, %ymm6, %ymm5
+; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm6
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm13[0],xmm15[0],xmm13[1],xmm15[1],xmm13[2],xmm15[2],xmm13[3],xmm15[3],xmm13[4],xmm15[4],xmm13[5],xmm15[5],xmm13[6],xmm15[6],xmm13[7],xmm15[7]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,0,1]
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm13, %ymm9, %ymm9
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm10 = xmm12[0],xmm11[0],xmm12[1],xmm11[1],xmm12[2],xmm11[2],xmm12[3],xmm11[3],xmm12[4],xmm11[4],xmm12[5],xmm11[5],xmm12[6],xmm11[6],xmm12[7],xmm11[7]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm11 = xmm14[1,1,0,0,4,5,6,7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm11 = xmm11[0,1,2,0]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,1,0]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm9, %ymm10, %ymm9
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm11 = ymm0[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm11 = ymm11[0,2,3,3,4,6,7,7]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,u,u,u,0,255,0,255,u,u,u,0,255,0,255,u,255,u,u,u,0,255,0,255,u,u,u,0,255,0,255,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm11 = 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]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm11 = ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15],ymm8[24],ymm7[24],ymm8[25],ymm7[25],ymm8[26],ymm7[26],ymm8[27],ymm7[27],ymm8[28],ymm7[28],ymm8[29],ymm7[29],ymm8[30],ymm7[30],ymm8[31],ymm7[31]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[23],zero,ymm1[27,20,21,26],zero,ymm1[24],zero,ymm1[26,27,26,27],zero,ymm1[25]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero,ymm3[27],zero
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpor %ymm11, %ymm12, %ymm11
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[25],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero,ymm0[27]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpor %ymm12, %ymm13, %ymm12
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm8[25],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpor %ymm12, %ymm13, %ymm12
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm13 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm12 = 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]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm13 = 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]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm13 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[16],ymm8[16],ymm7[17],ymm8[17],ymm7[18],ymm8[18],ymm7[19],ymm8[19],ymm7[20],ymm8[20],ymm7[21],ymm8[21],ymm7[22],ymm8[22],ymm7[23],ymm8[23]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm14 = ymm6[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm14 = ymm14[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,2,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm13, %ymm14, %ymm13
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero,zero,ymm3[18]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero,zero,zero,ymm1[18],zero
+; AVX2-SLOW-NEXT:    vpor %ymm3, %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,ymm0[14],zero,zero,zero,zero,zero,zero,ymm0[15],zero,zero,zero,zero,zero,zero,ymm0[16],zero,zero,zero,zero,zero,zero,ymm0[17],zero,zero,zero,zero,zero,zero
+; AVX2-SLOW-NEXT:    vpor %ymm2, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = zero,ymm7[1,2,3,0,1,14],zero,ymm7[0,1,0,1,14,15],zero,ymm7[15,16,17,18,19,16],zero,ymm7[30,31,16,17,16,17],zero,ymm7[31,30,31]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm8[13],zero,zero,zero,zero,zero,zero,ymm8[14],zero,zero,zero,zero,zero,zero,ymm8[15],zero,zero,zero,zero,zero,zero,ymm8[16],zero,zero,zero,zero,zero,zero,ymm8[17],zero,zero,zero
+; AVX2-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm6[12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-SLOW-NEXT:    vmovdqa %ymm2, 96(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 160(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm10, (%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 128(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm9, 32(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm0, 64(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm0, 192(%rax)
-; AVX2-SLOW-NEXT:    popq %rax
+; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 96(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm12, 128(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm11, 160(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm9, (%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm5, 32(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm4, 64(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm10, 192(%rax)
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: store_i8_stride7_vf32:
 ; AVX2-FAST:       # %bb.0:
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm3

 ; AVX2-FAST-NEXT:    vpermd %ymm10, %ymm11, %ymm10
 ; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm14[0],xmm12[0],xmm14[1],xmm12[1],xmm14[2],xmm12[2],xmm14[3],xmm12[3],xmm14[4],xmm12[4],xmm14[5],xmm12[5],xmm14[6],xmm12[6],xmm14[7],xmm12[7]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm11, %ymm10, %ymm10
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm9, %ymm10, %ymm9
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[18],zero,zero,zero,zero,ymm2[21],zero,ymm2[19],zero,zero,zero,zero,ymm2[22],zero,ymm2[20]
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm10 = 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]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zero,zero,zero,ymm0[21],zero,ymm0[19],zero,zero,zero,zero,ymm0[22],zero,ymm0[20],zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm10, %ymm11, %ymm10
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[21],zero,ymm3[19],zero,zero,zero,zero,ymm3[22],zero,ymm3[20],zero,zero
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm11 = 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]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22],zero,ymm1[20],zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm11, %ymm12, %ymm11
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
 ; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm6[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [4,5,4,5,5,7,4,5]
 ; AVX2-FAST-NEXT:    vpermd %ymm11, %ymm12, %ymm11
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm8[20],zero,ymm8[18],zero,zero,zero,zero,ymm8[21],zero,ymm8[19],zero,zero,zero,zero,ymm8[22]
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[16],ymm8[16],ymm7[17],ymm8[17],ymm7[18],ymm8[18],ymm7[19],ymm8[19],ymm7[20],ymm8[20],ymm7[21],ymm8[21],ymm7[22],ymm8[22],ymm7[23],ymm8[23]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm13 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm7[18],zero,zero,zero,zero,ymm7[21],zero,ymm7[19],zero,zero,zero,zero,ymm7[22],zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm12, %ymm13, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm12, %ymm11, %ymm11
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[17,18,19,30],zero,ymm0[28],zero,ymm0[28,29,30,31],zero,ymm0[29],zero,ymm0[31]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpor %ymm11, %ymm12, %ymm11
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero,zero,zero
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm12 = 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]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm13 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm12, %ymm13, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm7[27],zero,ymm7[27,28,29,30],zero,ymm7[28],zero,ymm7[26,27,30,31],zero,ymm7[29]
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm12 = ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15],ymm8[24],ymm7[24],ymm8[25],ymm7[25],ymm8[26],ymm7[26],ymm8[27],ymm7[27],ymm8[28],ymm7[28],ymm8[29],ymm7[29],ymm8[30],ymm7[30],ymm8[31],ymm7[31]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm13 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm8[27],zero,zero,zero,zero,ymm8[30],zero,ymm8[28],zero,zero,zero,zero,ymm8[31],zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm12, %ymm13, %ymm12
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[23],zero,ymm1[27,20,21,26],zero,ymm1[24],zero,ymm1[26,27,26,27],zero,ymm1[25]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]

 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm11, %ymm10, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm9, %ymm10, %ymm9
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm10 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[17,18,19,30],zero,ymm0[28],zero,ymm0[28,29,30,31],zero,ymm0[29],zero,ymm0[31]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm10, %ymm11, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} ymm11 = 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]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm11, %ymm12, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm7[27],zero,ymm7[27,28,29,30],zero,ymm7[28],zero,ymm7[26,27,30,31],zero,ymm7[29]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} ymm11 = ymm8[8],ymm7[8],ymm8[9],ymm7[9],ymm8[10],ymm7[10],ymm8[11],ymm7[11],ymm8[12],ymm7[12],ymm8[13],ymm7[13],ymm8[14],ymm7[14],ymm8[15],ymm7[15],ymm8[24],ymm7[24],ymm8[25],ymm7[25],ymm8[26],ymm7[26],ymm8[27],ymm7[27],ymm8[28],ymm7[28],ymm8[29],ymm7[29],ymm8[30],ymm7[30],ymm8[31],ymm7[31]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm12 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm8[27],zero,zero,zero,zero,ymm8[30],zero,ymm8[28],zero,zero,zero,zero,ymm8[31],zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm11, %ymm12, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u,255,255,255,255,0,0,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm11 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[23],zero,ymm1[27,20,21,26],zero,ymm1[24],zero,ymm1[26,27,26,27],zero,ymm1[25]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]

 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm12, %ymm13, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm13 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm11, %ymm12, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm12 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[18],zero,zero,zero,zero,ymm2[21],zero,ymm2[19],zero,zero,zero,zero,ymm2[22],zero,ymm2[20]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm12 = 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]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm12 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm13 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zero,zero,zero,ymm0[21],zero,ymm0[19],zero,zero,zero,zero,ymm0[22],zero,ymm0[20],zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm12, %ymm13, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm13 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[21],zero,ymm3[19],zero,zero,zero,zero,ymm3[22],zero,ymm3[20],zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm13 = 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]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm14 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22],zero,ymm1[20],zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm13, %ymm14, %ymm13
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm13 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm8[20],zero,ymm8[18],zero,zero,zero,zero,ymm8[21],zero,ymm8[19],zero,zero,zero,zero,ymm8[22]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm13 = ymm7[0],ymm8[0],ymm7[1],ymm8[1],ymm7[2],ymm8[2],ymm7[3],ymm8[3],ymm7[4],ymm8[4],ymm7[5],ymm8[5],ymm7[6],ymm8[6],ymm7[7],ymm8[7],ymm7[16],ymm8[16],ymm7[17],ymm8[17],ymm7[18],ymm8[18],ymm7[19],ymm8[19],ymm7[20],ymm8[20],ymm7[21],ymm8[21],ymm7[22],ymm8[22],ymm7[23],ymm8[23]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm13 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm14 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm7[18],zero,zero,zero,zero,ymm7[21],zero,ymm7[19],zero,zero,zero,zero,ymm7[22],zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm13, %ymm14, %ymm13
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm14 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,18,19,18,19,20,21,18,19,20,21,28,29,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,2,3,2]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm13, %ymm14, %ymm13
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm12, %ymm13, %ymm12
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero,zero,ymm3[18]
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero,zero,zero,ymm1[18],zero

 ; AVX2-FAST-PERLANE-NEXT:    vzeroupper
 ; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-SLOW-LABEL: store_i8_stride7_vf32:
 ; AVX512F-SLOW:       # %bb.0:
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %r10
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 (%rdx), %ymm18
-; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm3
-; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm5
-; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm6
-; AVX512F-SLOW-NEXT:    vmovdqa (%r10), %ymm4
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm4
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm5
+; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm6
+; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm3
+; AVX512F-SLOW-NEXT:    vmovdqa64 (%r10), %ymm17
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm8
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm8[u],zero,xmm8[7],zero,xmm8[5,u,u,u],zero,xmm8[8],zero,xmm8[6,u,u,u],zero
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm9
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm9[u,7],zero,xmm9[5],zero,xmm9[u,u,u,8],zero,xmm9[6],zero,xmm9[u,u,u,9]
 ; AVX512F-SLOW-NEXT:    vpor %xmm7, %xmm10, %xmm7
 ; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm10 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3],xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
 ; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm7, %zmm10, %zmm7

 ; AVX512F-SLOW-NEXT:    vmovdqa (%r10), %xmm12
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm12[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
 ; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm12[1,1,0,0,4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,0]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm7 = zmm0[0,0,1,0,4,4,5,4]
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm16, %zmm7
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm15, %zmm7
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm1[14,u,u],zero,zero,zero,zero,ymm1[15,u,u],zero,zero,zero,zero,ymm1[16,u,u],zero,zero,zero,zero,ymm1[17,u,u],zero,zero,zero,zero,ymm1[18]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm4[14,u,u],zero,zero,zero,zero,ymm4[15,u,u],zero,zero,zero,zero,ymm4[16,u,u],zero,zero,zero,zero,ymm4[17,u,u],zero,zero,zero,zero,ymm4[18]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm2[0,1,14],zero,ymm2[u,u,0,1,14,15],zero,ymm2[u,u,13,2,3,16],zero,ymm2[u,u,28,29,16,17],zero,ymm2[u,u,19,28,29,18],zero
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm16
 ; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm15, %ymm0
 ; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm13 = xmm13[8],xmm14[8],xmm13[9],xmm14[9],xmm13[10],xmm14[10],xmm13[11],xmm14[11],xmm13[12],xmm14[12],xmm13[13],xmm14[13],xmm13[14],xmm14[14],xmm13[15],xmm14[15]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,0,1]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm13, %zmm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm3[u,u,u,u],zero,ymm3[14,u,u,u,u,u],zero,ymm3[15,u,u,u,u,u],zero,ymm3[16,u,u,u,u,u],zero,ymm3[17,u,u,u,u,u]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm18, %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm14 = ymm2[u,u,u,u,14],zero,ymm2[u,u,u,u,u,15],zero,ymm2[u,u,u,u,u,16],zero,ymm2[u,u,u,u,u,17],zero,ymm2[u,u,u,u,u]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm6[u,u,u,u],zero,ymm6[14,u,u,u,u,u],zero,ymm6[15,u,u,u,u,u],zero,ymm6[16,u,u,u,u,u],zero,ymm6[17,u,u,u,u,u]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm14 = ymm5[u,u,u,u,14],zero,ymm5[u,u,u,u,u,15],zero,ymm5[u,u,u,u,u,16],zero,ymm5[u,u,u,u,u,17],zero,ymm5[u,u,u,u,u]
 ; AVX512F-SLOW-NEXT:    vpor %ymm13, %ymm14, %ymm13
 ; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm8[8],xmm9[8],xmm8[9],xmm9[9],xmm8[10],xmm9[10],xmm8[11],xmm9[11],xmm8[12],xmm9[12],xmm8[13],xmm9[13],xmm8[14],xmm9[14],xmm8[15],xmm9[15]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm8, %zmm9
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm9
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,ymm5[u,u,u,u,u,14],zero,ymm5[u,u,u,u,u,15],zero,ymm5[u,u,u,u,u,16],zero,ymm5[u,u,u,u,u,17],zero,ymm5[u,u,u]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm6[13,u,u,u,u,u],zero,ymm6[14,u,u,u,u,u],zero,ymm6[15,u,u,u,u,u],zero,ymm6[16,u,u,u,u,u],zero,ymm6[17,u,u,u]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,ymm1[u,u,u,u,u,14],zero,ymm1[u,u,u,u,u,15],zero,ymm1[u,u,u,u,u,16],zero,ymm1[u,u,u,u,u,17],zero,ymm1[u,u,u]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm3[13,u,u,u,u,u],zero,ymm3[14,u,u,u,u,u],zero,ymm3[15,u,u,u,u,u],zero,ymm3[16,u,u,u,u,u],zero,ymm3[17,u,u,u]
 ; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm8, %ymm0
 ; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm11[8],xmm10[8],xmm11[9],xmm10[9],xmm11[10],xmm10[10],xmm11[11],xmm10[11],xmm11[12],xmm10[12],xmm11[13],xmm10[13],xmm11[14],xmm10[14],xmm11[15],xmm10[15]
 ; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
 ; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm8 = xmm12[0,1,2,3,4,5,5,6]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm8 = xmm8[2,2,3,3]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
 ; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
 ; AVX512F-SLOW-NEXT:    vpandn %ymm8, %ymm10, %ymm8
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = zero,ymm4[13,u,u,u,u],zero,zero,ymm4[14,u,u,u,u],zero,zero,ymm4[15,u,u,u,u],zero,zero,ymm4[16,u,u,u,u],zero,zero,ymm4[17,u,u]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm17, %ymm12
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = zero,ymm12[13,u,u,u,u],zero,zero,ymm12[14,u,u,u,u],zero,zero,ymm12[15,u,u,u,u],zero,zero,ymm12[16,u,u,u,u],zero,zero,ymm12[17,u,u]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm8, %zmm8
 ; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm8
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25,u,23,u,u,u,u,26,u,24,u,u,u,u,27,u]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25,u,23,u,u,u,u,26,u,24,u,u,u,u,27,u]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero,ymm2[25]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm9
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm5[0],ymm6[0],ymm5[1],ymm6[1],ymm5[2],ymm6[2],ymm5[3],ymm6[3],ymm5[4],ymm6[4],ymm5[5],ymm6[5],ymm5[6],ymm6[6],ymm5[7],ymm6[7],ymm5[16],ymm6[16],ymm5[17],ymm6[17],ymm5[18],ymm6[18],ymm5[19],ymm6[19],ymm5[20],ymm6[20],ymm5[21],ymm6[21],ymm5[22],ymm6[22],ymm5[23],ymm6[23]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm5[23],zero,ymm5[21,22,23,26],zero,ymm5[24],zero,ymm5[28,29,26,27]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpor %ymm9, %ymm10, %ymm9
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm10 = 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]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm10, %zmm9
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm9
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [18374967954648269055,71777218572844800,18374967954648269055,71777218572844800]
-; AVX512F-SLOW-NEXT:    vpand %ymm0, %ymm9, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,ymm2[18,19,20,21],zero,ymm2[19],zero,ymm2[25,26,27,22],zero,ymm2[20],zero
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm1[23],zero,ymm1[23,24,25,26],zero,ymm1[24],zero,ymm1[30,31]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm10, %ymm0
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm10 = 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]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm10, %zmm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm3[18],zero,zero,zero,zero,ymm3[21],zero,ymm3[19],zero,zero,zero,zero,ymm3[22],zero,ymm3[20]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm16, %ymm14
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm14[23],zero,zero,zero,zero,ymm14[26],zero,ymm14[24],zero,zero,zero,zero,ymm14[27],zero,ymm14[25]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm12[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,2,3,2]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm10, %zmm10
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vporq %zmm10, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm14[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[0,0,1,1,4,4,5,5]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = [18374966859431673855,18446463693966278655,18374966859431673855,18446463693966278655]
-; AVX512F-SLOW-NEXT:    vpand %ymm11, %ymm10, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm2[23],zero,ymm2[21,22,23,26],zero,ymm2[24],zero,ymm2[28,29,26,27]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm10, %zmm10
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22],zero,ymm1[20],zero,zero
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm12, %zmm12
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vporq %zmm12, %zmm10, %zmm10
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm10
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm6[20],zero,ymm6[18],zero,zero,zero,zero,ymm6[21],zero,ymm6[19],zero,zero,zero,zero,ymm6[22]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm5[18],zero,ymm5[20,21,20,21],zero,ymm5[19],zero,ymm5[19,20,21,22],zero
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm5[23],zero,ymm5[23,24,25,26],zero,ymm5[24],zero,ymm5[30,31]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm12, %zmm12
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vporq %zmm0, %zmm12, %zmm0
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm4[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm12 = ymm12[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,2,3,2]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm12, %zmm12
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm12
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm10, %zmm12
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm10
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm5[27],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero,ymm5[29]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero,ymm5[29],zero,zero
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm9, %ymm0, %ymm5
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm5
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm0 = 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]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm11, %ymm3, %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero,zero
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15],ymm3[24],ymm1[24],ymm3[25],ymm1[25],ymm3[26],ymm1[26],ymm3[27],ymm1[27],ymm3[28],ymm1[28],ymm3[29],ymm1[29],ymm3[30],ymm1[30],ymm3[31],ymm1[31]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm14[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,2,3,3,6,6,7,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm3
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm3
-; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm3
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm3, 192(%rax)
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm2
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm2, 192(%rax)
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm7, (%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm12, 128(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm10, 128(%rax)
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm8, 64(%rax)
 ; AVX512F-SLOW-NEXT:    vzeroupper
 ; AVX512F-SLOW-NEXT:    retq
 ;
 ; AVX512F-FAST-LABEL: store_i8_stride7_vf32:
 ; AVX512F-FAST:       # %bb.0:
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %r10

 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm8, %zmm0
 ; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} xmm8 = xmm10[0,1,2,3,4,5,5,6]
 ; AVX512F-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm10 = [2,2,3,3,2,2,3,3]
 ; AVX512F-FAST-NEXT:    # ymm10 = mem[0,1,0,1]
 ; AVX512F-FAST-NEXT:    vpermd %ymm8, %ymm10, %ymm8
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
 ; AVX512F-FAST-NEXT:    vpandn %ymm8, %ymm10, %ymm8
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm13
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = zero,ymm13[13,u,u,u,u],zero,zero,ymm13[14,u,u,u,u],zero,zero,ymm13[15,u,u,u,u],zero,zero,ymm13[16,u,u,u,u],zero,zero,ymm13[17,u,u]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm12
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = zero,ymm12[13,u,u,u,u],zero,zero,ymm12[14,u,u,u,u],zero,zero,ymm12[15,u,u,u,u],zero,zero,ymm12[16,u,u,u,u],zero,zero,ymm12[17,u,u]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm8, %zmm8
 ; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm8
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm8
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25,u,23,u,u,u,u,26,u,24,u,u,u,u,27,u]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [18374967954648269055,71777218572844800,18374967954648269055,71777218572844800]
-; AVX512F-FAST-NEXT:    vpand %ymm0, %ymm9, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,ymm5[18,19,20,21],zero,ymm5[19],zero,ymm5[25,26,27,22],zero,ymm5[20],zero
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero,ymm2[25]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm9
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm5[0],ymm6[0],ymm5[1],ymm6[1],ymm5[2],ymm6[2],ymm5[3],ymm6[3],ymm5[4],ymm6[4],ymm5[5],ymm6[5],ymm5[6],ymm6[6],ymm5[7],ymm6[7],ymm5[16],ymm6[16],ymm5[17],ymm6[17],ymm5[18],ymm6[18],ymm5[19],ymm6[19],ymm5[20],ymm6[20],ymm5[21],ymm6[21],ymm5[22],ymm6[22],ymm5[23],ymm6[23]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm5[23],zero,ymm5[21,22,23,26],zero,ymm5[24],zero,ymm5[28,29,26,27]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpor %ymm9, %ymm10, %ymm9
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm10 = 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]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm10, %zmm9
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm9
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm1[23],zero,ymm1[23,24,25,26],zero,ymm1[24],zero,ymm1[30,31]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm10, %ymm0
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm10 = 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]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm10, %zmm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm6[18],zero,zero,zero,zero,ymm6[21],zero,ymm6[19],zero,zero,zero,zero,ymm6[22],zero,ymm6[20]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero,ymm2[25]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm10, %zmm10
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm10, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm4[21],zero,ymm4[19],zero,zero,zero,zero,ymm4[22],zero,ymm4[20],zero,zero
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm10 = ymm12[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,5,4,u,5,u,4,u>
+; AVX512F-FAST-NEXT:    vpermd %ymm10, %ymm11, %ymm10
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm10, %zmm10
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,ymm2[19],zero,ymm2[21,20,21,22],zero,ymm2[20],zero,ymm2[22,23]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm5[23],zero,ymm5[21,22,23,26],zero,ymm5[24],zero,ymm5[28,29,26,27]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm11, %zmm11
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm11 = zmm11[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm10, %zmm11, %zmm10
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm10
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm3[20],zero,ymm3[18],zero,zero,zero,zero,ymm3[21],zero,ymm3[19],zero,zero,zero,zero,ymm3[22]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm11 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,ymm1[18],zero,ymm1[20,21,20,21],zero,ymm1[19],zero,ymm1[19,20,21,22],zero
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm1[23],zero,ymm1[23,24,25,26],zero,ymm1[24],zero,ymm1[30,31]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm11, %zmm11
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm11 = zmm11[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm0, %zmm11, %zmm0
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm11 = ymm13[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,5,4,u,5,u,4,u>
-; AVX512F-FAST-NEXT:    vpermd %ymm11, %ymm12, %ymm11
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm11, %zmm11
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm11
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm10, %zmm11
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm10
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero,ymm5[29],zero,zero
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
 ; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm5
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm4[30],zero,ymm4[28],zero,zero,zero,zero,ymm4[31],zero,ymm4[29],zero,zero,zero
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm0 = 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]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm2[28],zero,ymm2[30,31,30,31],zero,ymm2[29],zero,ymm2[31,28,29]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm2, %ymm0
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[27],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29]
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm3[8],ymm1[8],ymm3[9],ymm1[9],ymm3[10],ymm1[10],ymm3[11],ymm1[11],ymm3[12],ymm1[12],ymm3[13],ymm1[13],ymm3[14],ymm1[14],ymm3[15],ymm1[15],ymm3[24],ymm1[24],ymm3[25],ymm1[25],ymm3[26],ymm1[26],ymm3[27],ymm1[27],ymm3[28],ymm1[28],ymm3[29],ymm1[29],ymm3[30],ymm1[30],ymm3[31],ymm1[31]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm9, %ymm2, %ymm1
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm2
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm2
 ; AVX512F-FAST-NEXT:    vmovdqa %ymm2, 192(%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm7, (%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm11, 128(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm10, 128(%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm8, 64(%rax)
 ; AVX512F-FAST-NEXT:    vzeroupper
 ; AVX512F-FAST-NEXT:    retq
 ;
 ; AVX512BW-SLOW-LABEL: store_i8_stride7_vf32:
 ; AVX512BW-SLOW:       # %bb.0:
 ; AVX512BW-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512BW-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rdi), %ymm4
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rsi), %ymm2
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rdx), %ymm1
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rsi), %ymm1
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdx), %ymm2
 ; AVX512BW-SLOW-NEXT:    vmovdqa (%rcx), %ymm3
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[0,1,0,1,14],zero,ymm1[14,15,0,1,14,15],zero,ymm1[13,14,15,16,17,16],zero,ymm1[30,31,30,31,16,17],zero,ymm1[31,28,29,30,31]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
-; AVX512BW-SLOW-NEXT:    vpor %ymm0, %ymm5, %ymm0
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rdi), %xmm9
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rsi), %xmm10
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm5
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[0,1,14],zero,ymm4[12,13,0,1,14,15],zero,ymm4[3,12,13,2,3,16],zero,ymm4[30,31,28,29,16,17],zero,ymm4[31,18,19,28,29,18],zero
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero,zero,ymm2[18]
-; AVX512BW-SLOW-NEXT:    vpor %ymm0, %ymm6, %ymm0
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rdx), %xmm12
+; AVX512BW-SLOW-NEXT:    vmovdqa (%r8), %ymm5
+; AVX512BW-SLOW-NEXT:    vmovdqa (%r9), %ymm6
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm2[0,1,0,1,14],zero,ymm2[14,15,0,1,14,15],zero,ymm2[13,14,15,16,17,16],zero,ymm2[30,31,30,31,16,17],zero,ymm2[31,28,29,30,31]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
+; AVX512BW-SLOW-NEXT:    vpor %ymm4, %ymm7, %ymm4
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdi), %xmm7
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rsi), %xmm8
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm9 = xmm8[8],xmm7[8],xmm8[9],xmm7[9],xmm8[10],xmm7[10],xmm8[11],xmm7[11],xmm8[12],xmm7[12],xmm8[13],xmm7[13],xmm8[14],xmm7[14],xmm8[15],xmm7[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm9, %zmm9
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm0[0,1,14],zero,ymm0[12,13,0,1,14,15],zero,ymm0[3,12,13,2,3,16],zero,ymm0[30,31,28,29,16,17],zero,ymm0[31,18,19,28,29,18],zero
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero,zero,zero,ymm1[18]
+; AVX512BW-SLOW-NEXT:    vpor %ymm4, %ymm10, %ymm4
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdx), %xmm13
 ; AVX512BW-SLOW-NEXT:    vmovdqa (%rcx), %xmm14
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm14[8],xmm12[8],xmm14[9],xmm12[9],xmm14[10],xmm12[10],xmm14[11],xmm12[11],xmm14[12],xmm12[12],xmm14[13],xmm12[13],xmm14[14],xmm12[14],xmm14[15],xmm12[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm6, %zmm0
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm14[8],xmm13[8],xmm14[9],xmm13[9],xmm14[10],xmm13[10],xmm14[11],xmm13[11],xmm14[12],xmm13[12],xmm14[13],xmm13[13],xmm14[14],xmm13[14],xmm14[15],xmm13[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm10, %zmm4
 ; AVX512BW-SLOW-NEXT:    movabsq $435749858791416001, %rcx # imm = 0x60C1830183060C1
 ; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm5, %zmm0 {%k1}
-; AVX512BW-SLOW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm5 = mem[0,1,2,3,0,1,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[13],zero,zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero
-; AVX512BW-SLOW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm6 = mem[0,1,2,3,0,1,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = zero,zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero
-; AVX512BW-SLOW-NEXT:    vpor %ymm7, %ymm8, %ymm7
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm9, %zmm4 {%k1}
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[13],zero,zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero,zero
+; AVX512BW-SLOW-NEXT:    vpor %ymm9, %ymm10, %ymm9
 ; AVX512BW-SLOW-NEXT:    vmovdqa (%r9), %xmm11
-; AVX512BW-SLOW-NEXT:    vmovdqa (%r8), %xmm13
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm13[8],xmm11[8],xmm13[9],xmm11[9],xmm13[10],xmm11[10],xmm13[11],xmm11[11],xmm13[12],xmm11[12],xmm13[13],xmm11[13],xmm13[14],xmm11[14],xmm13[15],xmm11[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm8, %zmm8
-; AVX512BW-SLOW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm7 = mem[0,1,2,3,0,1,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqa (%r8), %xmm12
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm12[8],xmm11[8],xmm12[9],xmm11[9],xmm12[10],xmm11[10],xmm12[11],xmm11[11],xmm12[12],xmm11[12],xmm12[13],xmm11[13],xmm12[14],xmm11[14],xmm12[15],xmm11[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm10, %zmm10
+; AVX512BW-SLOW-NEXT:    vbroadcasti64x4 {{.*#+}} zmm9 = mem[0,1,2,3,0,1,2,3]
 ; AVX512BW-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm15 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6]
 ; AVX512BW-SLOW-NEXT:    # ymm15 = mem[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vpermw %ymm7, %ymm15, %ymm15
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm7[12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX512BW-SLOW-NEXT:    vpermw %ymm9, %ymm15, %ymm15
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm9[12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
 ; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm15
 ; AVX512BW-SLOW-NEXT:    movabsq $2323999253380730912, %rcx # imm = 0x2040810204081020
 ; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm15, %zmm8 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm15, %zmm10 {%k1}
 ; AVX512BW-SLOW-NEXT:    movabsq $4066998693416279096, %rcx # imm = 0x3870E1C3870E1C38
 ; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm8, %zmm0 {%k1}
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm8
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm8 = zmm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm8[18,19,20,21],zero,zmm8[19],zero,zmm8[25,26,27,22],zero,zmm8[20],zero,zmm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm8[55],zero,zero,zero,zero,zmm8[58],zero,zmm8[56],zero,zero,zero,zero,zmm8[59],zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm15
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm15 = zmm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm15[18],zero,zero,zero,zero,zmm15[21],zero,zmm15[19],zero,zero,zero,zero,zmm15[22],zero,zmm15[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm15[55],zero,zero,zero,zero,zmm15[58],zero,zmm15[56],zero,zero,zero,zero,zmm15[59],zero,zmm15[57]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm15 = zmm15[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vporq %zmm8, %zmm15, %zmm8
-; AVX512BW-SLOW-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm4[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[0,0,1,1,4,4,5,5]
-; AVX512BW-SLOW-NEXT:    movl $676341840, %ecx # imm = 0x28502850
-; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm15 {%k1} = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm10, %zmm4 {%k1}
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm2[23],zero,ymm2[21,22,23,26],zero,ymm2[24],zero,ymm2[28,29,26,27]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpor %ymm10, %ymm15, %ymm10
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm15 = 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]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm1[23],zero,ymm1[21,22,23,26],zero,ymm1[24],zero,ymm1[28,29,26,27]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm15, %zmm15
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero,ymm1[27],zero
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero,ymm0[27],zero,ymm0[25]
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm17 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm17 = ymm17[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vporq %ymm16, %ymm17, %ymm16
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm15
+; AVX512BW-SLOW-NEXT:    vporq %ymm10, %ymm16, %ymm10
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm16 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[16],ymm3[16],ymm2[17],ymm3[17],ymm2[18],ymm3[18],ymm2[19],ymm3[19],ymm2[20],ymm3[20],ymm2[21],ymm3[21],ymm2[22],ymm3[22],ymm2[23],ymm3[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm16, %zmm10
 ; AVX512BW-SLOW-NEXT:    movabsq $-9005497107459067808, %rcx # imm = 0x83060C180C183060
-; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm15, %zmm8 {%k2}
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm15 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
-; AVX512BW-SLOW-NEXT:    vpermw %zmm7, %zmm15, %zmm15
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm16 = zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm6[18],zero,zmm6[20,21,20,21],zero,zmm6[19],zero,zmm6[19,20,21,22],zero,zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm6[55],zero,zmm6[55,56,57,58],zero,zmm6[56],zero,zmm6[62,63]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm16 = zmm16[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm17 = zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm5[20],zero,zmm5[18],zero,zero,zero,zero,zmm5[21],zero,zmm5[19],zero,zero,zero,zero,zmm5[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm5[57],zero,zmm5[55],zero,zero,zero,zero,zmm5[58],zero,zmm5[56],zero,zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm17 = zmm17[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vporq %zmm16, %zmm17, %zmm16
+; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm15, %zmm10 {%k1}
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm5[23],zero,ymm5[23,24,25,26],zero,ymm5[24],zero,ymm5[30,31]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vporq %ymm15, %ymm16, %ymm15
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm16 = ymm5[0],ymm6[0],ymm5[1],ymm6[1],ymm5[2],ymm6[2],ymm5[3],ymm6[3],ymm5[4],ymm6[4],ymm5[5],ymm6[5],ymm5[6],ymm6[6],ymm5[7],ymm6[7],ymm5[16],ymm6[16],ymm5[17],ymm6[17],ymm5[18],ymm6[18],ymm5[19],ymm6[19],ymm5[20],ymm6[20],ymm5[21],ymm6[21],ymm5[22],ymm6[22],ymm5[23],ymm6[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm16 = ymm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm15, %zmm16, %zmm15
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm16 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
+; AVX512BW-SLOW-NEXT:    vpermw %zmm9, %zmm16, %zmm16
 ; AVX512BW-SLOW-NEXT:    movabsq $1161999626690365456, %rcx # imm = 0x1020408102040810
-; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm15, %zmm16 {%k2}
+; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm16, %zmm15 {%k1}
 ; AVX512BW-SLOW-NEXT:    movabsq $2033499346708139548, %rcx # imm = 0x1C3870E1C3870E1C
-; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm16, %zmm8 {%k2}
+; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm15, %zmm10 {%k1}
 ; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm14[u,u,u],zero,xmm14[7],zero,xmm14[5,u,u,u],zero,xmm14[8],zero,xmm14[6,u,u]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm16 = xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero,xmm12[u,u]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm16 = xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero,xmm13[u,u]
 ; AVX512BW-SLOW-NEXT:    vporq %xmm15, %xmm16, %xmm15
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm15, %zmm12, %zmm12
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[0,1,0,1,4,5,4,5]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm14 = xmm10[u],zero,xmm10[7],zero,xmm10[5,u,u,u],zero,xmm10[8],zero,xmm10[6,u,u,u],zero
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm9[u,7],zero,xmm9[5],zero,xmm9[u,u,u,8],zero,xmm9[6],zero,xmm9[u,u,u,9]
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm13[0],xmm14[0],xmm13[1],xmm14[1],xmm13[2],xmm14[2],xmm13[3],xmm14[3],xmm13[4],xmm14[4],xmm13[5],xmm14[5],xmm13[6],xmm14[6],xmm13[7],xmm14[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm15, %zmm13, %zmm13
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm13 = zmm13[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm14 = xmm8[u],zero,xmm8[7],zero,xmm8[5,u,u,u],zero,xmm8[8],zero,xmm8[6,u,u,u],zero
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm15 = xmm7[u,7],zero,xmm7[5],zero,xmm7[u,u,u,8],zero,xmm7[6],zero,xmm7[u,u,u,9]
 ; AVX512BW-SLOW-NEXT:    vpor %xmm14, %xmm15, %xmm14
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm14, %zmm9, %zmm9
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3],xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm14, %zmm7, %zmm7
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm7 = zmm7[0,1,0,1,4,5,4,5]
 ; AVX512BW-SLOW-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
-; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm12, %zmm9 {%k2}
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm10 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm13[4],zero,xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero
-; AVX512BW-SLOW-NEXT:    vpor %xmm10, %xmm12, %xmm10
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3],xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
+; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm13, %zmm7 {%k1}
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm13 = xmm12[4],zero,xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero
+; AVX512BW-SLOW-NEXT:    vpor %xmm8, %xmm13, %xmm8
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm12[0],xmm11[0],xmm12[1],xmm11[1],xmm12[2],xmm11[2],xmm12[3],xmm11[3],xmm12[4],xmm11[4],xmm12[5],xmm11[5],xmm12[6],xmm11[6],xmm12[7],xmm11[7]
 ; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm10, %zmm11, %zmm10
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[0,1,0,1,4,5,4,5]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm7, %zmm7, %zmm11
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm8, %zmm11, %zmm8
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm9, %zmm11
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm12 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
 ; AVX512BW-SLOW-NEXT:    vpermw %zmm11, %zmm12, %zmm11
 ; AVX512BW-SLOW-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
-; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm11, %zmm10 {%k2}
+; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm11, %zmm8 {%k1}
 ; AVX512BW-SLOW-NEXT:    movabsq $8133997386832558192, %rcx # imm = 0x70E1C3870E1C3870
-; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm10, %zmm9 {%k2}
-; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm4 = ymm4[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[2,2,3,3,6,6,7,7]
-; AVX512BW-SLOW-NEXT:    movl $338170920, %ecx # imm = 0x14281428
-; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k2
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 {%k2} = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm4[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
-; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,3,3,4,6,7,7]
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm1, %ymm3 {%k1}
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm3[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    movl $101455920, %ecx # imm = 0x60C1830
-; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm2, %ymm1 {%k1}
-; AVX512BW-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
-; AVX512BW-SLOW-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vpermw %ymm7, %ymm2, %ymm2
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm5[27],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm6[27],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpor %ymm3, %ymm4, %ymm3
+; AVX512BW-SLOW-NEXT:    kmovq %rcx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm8, %zmm7 {%k1}
+; AVX512BW-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
+; AVX512BW-SLOW-NEXT:    # ymm8 = mem[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vpermw %ymm9, %ymm8, %ymm8
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm6[8],ymm5[8],ymm6[9],ymm5[9],ymm6[10],ymm5[10],ymm6[11],ymm5[11],ymm6[12],ymm5[12],ymm6[13],ymm5[13],ymm6[14],ymm5[14],ymm6[15],ymm5[15],ymm6[24],ymm5[24],ymm6[25],ymm5[25],ymm6[26],ymm5[26],ymm6[27],ymm5[27],ymm6[28],ymm5[28],ymm6[29],ymm5[29],ymm6[30],ymm5[30],ymm6[31],ymm5[31]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
 ; AVX512BW-SLOW-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
 ; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm8, %ymm5 {%k1}
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
+; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,3,3,4,6,7,7]
+; AVX512BW-SLOW-NEXT:    movl $676341840, %ecx # imm = 0x28502850
+; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k1
 ; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm2, %ymm3 {%k1}
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm3[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm0 = 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]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    movl $101455920, %ecx # imm = 0x60C1830
+; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm0, %ymm2 {%k1}
 ; AVX512BW-SLOW-NEXT:    movl $-507279602, %ecx # imm = 0xE1C3870E
 ; AVX512BW-SLOW-NEXT:    kmovd %ecx, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
-; AVX512BW-SLOW-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm9, (%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm8, 128(%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, 64(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm5, %ymm2 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa %ymm2, 192(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm7, (%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm10, 128(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm4, 64(%rax)
 ; AVX512BW-SLOW-NEXT:    vzeroupper
 ; AVX512BW-SLOW-NEXT:    retq
 ;
 ; AVX512BW-FAST-LABEL: store_i8_stride7_vf32:
 ; AVX512BW-FAST:       # %bb.0:
 ; AVX512BW-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX512BW-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %r10
-; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %ymm4
-; AVX512BW-FAST-NEXT:    vmovdqa (%rsi), %ymm3
-; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %ymm1
-; AVX512BW-FAST-NEXT:    vmovdqa (%rcx), %ymm2
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[0,1,0,1,14],zero,ymm1[14,15,0,1,14,15],zero,ymm1[13,14,15,16,17,16],zero,ymm1[30,31,30,31,16,17],zero,ymm1[31,28,29,30,31]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,zero,ymm2[14],zero,zero,zero,zero,zero,zero,ymm2[15],zero,zero,zero,zero,zero,zero,ymm2[16],zero,zero,zero,zero,zero,zero,ymm2[17],zero,zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vpor %ymm0, %ymm5, %ymm0
-; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %xmm8
-; AVX512BW-FAST-NEXT:    vmovdqa (%rsi), %xmm9
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm9[8],xmm8[8],xmm9[9],xmm8[9],xmm9[10],xmm8[10],xmm9[11],xmm8[11],xmm9[12],xmm8[12],xmm9[13],xmm8[13],xmm9[14],xmm8[14],xmm9[15],xmm8[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm5
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm4[0,1,14],zero,ymm4[12,13,0,1,14,15],zero,ymm4[3,12,13,2,3,16],zero,ymm4[30,31,28,29,16,17],zero,ymm4[31,18,19,28,29,18],zero
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero,zero,ymm3[18]
-; AVX512BW-FAST-NEXT:    vpor %ymm0, %ymm6, %ymm0
-; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %xmm12
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512BW-FAST-NEXT:    vmovdqa (%rsi), %ymm1
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %ymm2
+; AVX512BW-FAST-NEXT:    vmovdqa (%rcx), %ymm3
+; AVX512BW-FAST-NEXT:    vmovdqa (%r8), %ymm5
+; AVX512BW-FAST-NEXT:    vmovdqa (%r9), %ymm6
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm2[0,1,0,1,14],zero,ymm2[14,15,0,1,14,15],zero,ymm2[13,14,15,16,17,16],zero,ymm2[30,31,30,31,16,17],zero,ymm2[31,28,29,30,31]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = zero,zero,zero,zero,zero,ymm3[14],zero,zero,zero,zero,zero,zero,ymm3[15],zero,zero,zero,zero,zero,zero,ymm3[16],zero,zero,zero,zero,zero,zero,ymm3[17],zero,zero,zero,zero,zero
+; AVX512BW-FAST-NEXT:    vpor %ymm4, %ymm7, %ymm4
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %xmm7
+; AVX512BW-FAST-NEXT:    vmovdqa (%rsi), %xmm8
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm9 = xmm8[8],xmm7[8],xmm8[9],xmm7[9],xmm8[10],xmm7[10],xmm8[11],xmm7[11],xmm8[12],xmm7[12],xmm8[13],xmm7[13],xmm8[14],xmm7[14],xmm8[15],xmm7[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm9 = xmm9[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm9, %zmm9
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm0[0,1,14],zero,ymm0[12,13,0,1,14,15],zero,ymm0[3,12,13,2,3,16],zero,ymm0[30,31,28,29,16,17],zero,ymm0[31,18,19,28,29,18],zero
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,ymm1[14],zero,zero,zero,zero,zero,zero,ymm1[15],zero,zero,zero,zero,zero,zero,ymm1[16],zero,zero,zero,zero,zero,zero,ymm1[17],zero,zero,zero,zero,zero,zero,ymm1[18]
+; AVX512BW-FAST-NEXT:    vpor %ymm4, %ymm10, %ymm4
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %xmm13
 ; AVX512BW-FAST-NEXT:    vmovdqa (%rcx), %xmm14
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm14[8],xmm12[8],xmm14[9],xmm12[9],xmm14[10],xmm12[10],xmm14[11],xmm12[11],xmm14[12],xmm12[12],xmm14[13],xmm12[13],xmm14[14],xmm12[14],xmm14[15],xmm12[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm6, %zmm0
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm14[8],xmm13[8],xmm14[9],xmm13[9],xmm14[10],xmm13[10],xmm14[11],xmm13[11],xmm14[12],xmm13[12],xmm14[13],xmm13[13],xmm14[14],xmm13[14],xmm14[15],xmm13[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm10, %zmm4
 ; AVX512BW-FAST-NEXT:    movabsq $435749858791416001, %rcx # imm = 0x60C1830183060C1
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm5, %zmm0 {%k1}
-; AVX512BW-FAST-NEXT:    vbroadcasti64x4 {{.*#+}} zmm5 = mem[0,1,2,3,0,1,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[13],zero,zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero
-; AVX512BW-FAST-NEXT:    vbroadcasti64x4 {{.*#+}} zmm6 = mem[0,1,2,3,0,1,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vpor %ymm7, %ymm10, %ymm7
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm9, %zmm4 {%k1}
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[13],zero,zero,zero,zero,zero,zero,ymm6[14],zero,zero,zero,zero,zero,zero,ymm6[15],zero,zero,zero,zero,zero,zero,ymm6[16],zero,zero,zero,zero,zero,zero,ymm6[17],zero,zero,zero
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = zero,zero,zero,zero,zero,zero,ymm5[14],zero,zero,zero,zero,zero,zero,ymm5[15],zero,zero,zero,zero,zero,zero,ymm5[16],zero,zero,zero,zero,zero,zero,ymm5[17],zero,zero,zero,zero
+; AVX512BW-FAST-NEXT:    vpor %ymm9, %ymm10, %ymm9
 ; AVX512BW-FAST-NEXT:    vmovdqa (%r9), %xmm11
-; AVX512BW-FAST-NEXT:    vmovdqa (%r8), %xmm13
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm13[8],xmm11[8],xmm13[9],xmm11[9],xmm13[10],xmm11[10],xmm13[11],xmm11[11],xmm13[12],xmm11[12],xmm13[13],xmm11[13],xmm13[14],xmm11[14],xmm13[15],xmm11[15]
+; AVX512BW-FAST-NEXT:    vmovdqa (%r8), %xmm12
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm12[8],xmm11[8],xmm12[9],xmm11[9],xmm12[10],xmm11[10],xmm12[11],xmm11[11],xmm12[12],xmm11[12],xmm12[13],xmm11[13],xmm12[14],xmm11[14],xmm12[15],xmm11[15]
 ; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm10 = xmm10[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm10, %zmm10
-; AVX512BW-FAST-NEXT:    vbroadcasti64x4 {{.*#+}} zmm7 = mem[0,1,2,3,0,1,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm10, %zmm10
+; AVX512BW-FAST-NEXT:    vbroadcasti64x4 {{.*#+}} zmm9 = mem[0,1,2,3,0,1,2,3]
 ; AVX512BW-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm15 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6]
 ; AVX512BW-FAST-NEXT:    # ymm15 = mem[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vpermw %ymm7, %ymm15, %ymm15
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm16 = ymm7[12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX512BW-FAST-NEXT:    vpermw %ymm9, %ymm15, %ymm15
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm16 = ymm9[12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
 ; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm16, %zmm15, %zmm15
 ; AVX512BW-FAST-NEXT:    movabsq $2323999253380730912, %rcx # imm = 0x2040810204081020
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
 ; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm15, %zmm10 {%k1}
 ; AVX512BW-FAST-NEXT:    movabsq $4066998693416279096, %rcx # imm = 0x3870E1C3870E1C38
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm10, %zmm0 {%k1}
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm4, %zmm10
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm10 = zmm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,zmm10[19],zero,zmm10[21,20,21,22],zero,zmm10[20],zero,zmm10[22,23,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57],zero,zmm10[55],zero,zmm10[53,54,55,58],zero,zmm10[56],zero,zmm10[60,61,58,59]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm15
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm15 = zmm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm15[21],zero,zmm15[19],zero,zero,zero,zero,zmm15[22],zero,zmm15[20],zero,zero,zmm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zmm15[57],zero,zmm15[55],zero,zero,zero,zero,zmm15[58],zero,zmm15[56],zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm15 = zmm15[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm10, %zmm15, %zmm15
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm1, %zmm10
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm10 = zmm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm10[18,19,20,21],zero,zmm10[19],zero,zmm10[25,26,27,22],zero,zmm10[20],zero,zmm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm10[55],zero,zero,zero,zero,zmm10[58],zero,zmm10[56],zero,zero,zero,zero,zmm10[59],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm10 = zmm10[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm16
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm16 = zmm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm16[18],zero,zero,zero,zero,zmm16[21],zero,zmm16[19],zero,zero,zero,zero,zmm16[22],zero,zmm16[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm16[55],zero,zero,zero,zero,zmm16[58],zero,zmm16[56],zero,zero,zero,zero,zmm16[59],zero,zmm16[57]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm16 = zmm16[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm10, %zmm16, %zmm10
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm10, %zmm4 {%k1}
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm2[23],zero,ymm2[21,22,23,26],zero,ymm2[24],zero,ymm2[28,29,26,27]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpor %ymm10, %ymm15, %ymm10
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm15 = 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]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm15, %zmm15
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero,ymm1[27],zero
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm16 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero,ymm0[27],zero,ymm0[25]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vporq %ymm10, %ymm16, %ymm10
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm16 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[4],ymm3[4],ymm2[5],ymm3[5],ymm2[6],ymm3[6],ymm2[7],ymm3[7],ymm2[16],ymm3[16],ymm2[17],ymm3[17],ymm2[18],ymm3[18],ymm2[19],ymm3[19],ymm2[20],ymm3[20],ymm2[21],ymm3[21],ymm2[22],ymm3[22],ymm2[23],ymm3[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm16 = ymm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm16, %zmm10
 ; AVX512BW-FAST-NEXT:    movabsq $-9005497107459067808, %rcx # imm = 0x83060C180C183060
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
 ; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm15, %zmm10 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm15 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
-; AVX512BW-FAST-NEXT:    vpermw %zmm7, %zmm15, %zmm15
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm16 = zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm6[18],zero,zmm6[20,21,20,21],zero,zmm6[19],zero,zmm6[19,20,21,22],zero,zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm6[55],zero,zmm6[55,56,57,58],zero,zmm6[56],zero,zmm6[62,63]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm16 = zmm16[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm17 = zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm5[20],zero,zmm5[18],zero,zero,zero,zero,zmm5[21],zero,zmm5[19],zero,zero,zero,zero,zmm5[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm5[57],zero,zmm5[55],zero,zero,zero,zero,zmm5[58],zero,zmm5[56],zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm17 = zmm17[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm16, %zmm17, %zmm16
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm5[23],zero,ymm5[23,24,25,26],zero,ymm5[24],zero,ymm5[30,31]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm16 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vporq %ymm15, %ymm16, %ymm15
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm16 = ymm5[0],ymm6[0],ymm5[1],ymm6[1],ymm5[2],ymm6[2],ymm5[3],ymm6[3],ymm5[4],ymm6[4],ymm5[5],ymm6[5],ymm5[6],ymm6[6],ymm5[7],ymm6[7],ymm5[16],ymm6[16],ymm5[17],ymm6[17],ymm5[18],ymm6[18],ymm5[19],ymm6[19],ymm5[20],ymm6[20],ymm5[21],ymm6[21],ymm5[22],ymm6[22],ymm5[23],ymm6[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm16 = ymm16[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm16 = ymm16[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm15, %zmm16, %zmm15
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm16 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
+; AVX512BW-FAST-NEXT:    vpermw %zmm9, %zmm16, %zmm16
 ; AVX512BW-FAST-NEXT:    movabsq $1161999626690365456, %rcx # imm = 0x1020408102040810
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm15, %zmm16 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm16, %zmm15 {%k1}
 ; AVX512BW-FAST-NEXT:    movabsq $2033499346708139548, %rcx # imm = 0x1C3870E1C3870E1C
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm16, %zmm10 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm15, %zmm10 {%k1}
 ; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm15 = xmm14[u,u,u],zero,xmm14[7],zero,xmm14[5,u,u,u],zero,xmm14[8],zero,xmm14[6,u,u]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm16 = xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero,xmm12[u,u]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm16 = xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero,xmm13[u,u]
 ; AVX512BW-FAST-NEXT:    vporq %xmm15, %xmm16, %xmm15
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm12[0],xmm14[0],xmm12[1],xmm14[1],xmm12[2],xmm14[2],xmm12[3],xmm14[3],xmm12[4],xmm14[4],xmm12[5],xmm14[5],xmm12[6],xmm14[6],xmm12[7],xmm14[7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm15, %zmm12, %zmm12
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm12 = zmm12[0,1,0,1,4,5,4,5]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm14 = xmm9[u],zero,xmm9[7],zero,xmm9[5,u,u,u],zero,xmm9[8],zero,xmm9[6,u,u,u],zero
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm15 = xmm8[u,7],zero,xmm8[5],zero,xmm8[u,u,u,8],zero,xmm8[6],zero,xmm8[u,u,u,9]
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm13[0],xmm14[0],xmm13[1],xmm14[1],xmm13[2],xmm14[2],xmm13[3],xmm14[3],xmm13[4],xmm14[4],xmm13[5],xmm14[5],xmm13[6],xmm14[6],xmm13[7],xmm14[7]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm13 = xmm13[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm15, %zmm13, %zmm13
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm13 = zmm13[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm14 = xmm8[u],zero,xmm8[7],zero,xmm8[5,u,u,u],zero,xmm8[8],zero,xmm8[6,u,u,u],zero
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm15 = xmm7[u,7],zero,xmm7[5],zero,xmm7[u,u,u,8],zero,xmm7[6],zero,xmm7[u,u,u,9]
 ; AVX512BW-FAST-NEXT:    vpor %xmm14, %xmm15, %xmm14
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3],xmm8[4],xmm9[4],xmm8[5],xmm9[5],xmm8[6],xmm9[6],xmm8[7],xmm9[7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm14, %zmm8, %zmm8
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3],xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm14, %zmm7, %zmm7
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm7 = zmm7[0,1,0,1,4,5,4,5]
 ; AVX512BW-FAST-NEXT:    movabsq $871499720017774092, %rcx # imm = 0xC183060C183060C
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm12, %zmm8 {%k1}
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm9 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm12 = xmm13[4],zero,xmm13[u,u,u,7],zero,xmm13[5],zero,xmm13[u,u,u,8],zero,xmm13[6],zero
-; AVX512BW-FAST-NEXT:    vpor %xmm9, %xmm12, %xmm9
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm13[0],xmm11[0],xmm13[1],xmm11[1],xmm13[2],xmm11[2],xmm13[3],xmm11[3],xmm13[4],xmm11[4],xmm13[5],xmm11[5],xmm13[6],xmm11[6],xmm13[7],xmm11[7]
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm13, %zmm7 {%k1}
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm8 = zero,xmm11[4,u,u,u],zero,xmm11[7],zero,xmm11[5,u,u,u],zero,xmm11[8],zero,xmm11[6]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm13 = xmm12[4],zero,xmm12[u,u,u,7],zero,xmm12[5],zero,xmm12[u,u,u,8],zero,xmm12[6],zero
+; AVX512BW-FAST-NEXT:    vpor %xmm8, %xmm13, %xmm8
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm11 = xmm12[0],xmm11[0],xmm12[1],xmm11[1],xmm12[2],xmm11[2],xmm12[3],xmm11[3],xmm12[4],xmm11[4],xmm12[5],xmm11[5],xmm12[6],xmm11[6],xmm12[7],xmm11[7]
 ; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm9, %zmm11, %zmm9
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm9 = zmm9[0,1,0,1,4,5,4,5]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm7, %zmm11
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm8, %zmm11, %zmm8
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm8 = zmm8[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm9, %zmm11
 ; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm12 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
 ; AVX512BW-FAST-NEXT:    vpermw %zmm11, %zmm12, %zmm11
 ; AVX512BW-FAST-NEXT:    movabsq $4647998506761461824, %rcx # imm = 0x4081020408102040
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm11, %zmm9 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm11, %zmm8 {%k1}
 ; AVX512BW-FAST-NEXT:    movabsq $8133997386832558192, %rcx # imm = 0x70E1C3870E1C3870
 ; AVX512BW-FAST-NEXT:    kmovq %rcx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm9, %zmm8 {%k1}
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm4[28],zero,ymm4[30,31,30,31],zero,ymm4[29],zero,ymm4[31,28,29]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero,zero,zero
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm8, %zmm7 {%k1}
+; AVX512BW-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm8 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
+; AVX512BW-FAST-NEXT:    # ymm8 = mem[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vpermw %ymm9, %ymm8, %ymm8
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm6[8],ymm5[8],ymm6[9],ymm5[9],ymm6[10],ymm5[10],ymm6[11],ymm5[11],ymm6[12],ymm5[12],ymm6[13],ymm5[13],ymm6[14],ymm5[14],ymm6[15],ymm5[15],ymm6[24],ymm5[24],ymm6[25],ymm5[25],ymm6[26],ymm5[26],ymm6[27],ymm5[27],ymm6[28],ymm5[28],ymm6[29],ymm5[29],ymm6[30],ymm5[30],ymm6[31],ymm5[31]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX512BW-FAST-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
+; AVX512BW-FAST-NEXT:    kmovd %ecx, %k1
+; AVX512BW-FAST-NEXT:    vmovdqu8 %ymm8, %ymm5 {%k1}
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpor %ymm4, %ymm3, %ymm3
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpor %ymm2, %ymm1, %ymm1
+; AVX512BW-FAST-NEXT:    vpor %ymm3, %ymm2, %ymm2
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm0 = 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]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX512BW-FAST-NEXT:    movl $101455920, %ecx # imm = 0x60C1830
 ; AVX512BW-FAST-NEXT:    kmovd %ecx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
-; AVX512BW-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15]
-; AVX512BW-FAST-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vpermw %ymm7, %ymm2, %ymm2
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm5[27],zero,zero,zero,zero,ymm5[30],zero,ymm5[28],zero,zero,zero,zero,ymm5[31],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm6[27],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpor %ymm3, %ymm4, %ymm3
-; AVX512BW-FAST-NEXT:    movl $-2130574328, %ecx # imm = 0x81020408
-; AVX512BW-FAST-NEXT:    kmovd %ecx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %ymm2, %ymm3 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %ymm0, %ymm2 {%k1}
 ; AVX512BW-FAST-NEXT:    movl $-507279602, %ecx # imm = 0xE1C3870E
 ; AVX512BW-FAST-NEXT:    kmovd %ecx, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %ymm3, %ymm1 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqa %ymm1, 192(%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm8, (%rax)
+; AVX512BW-FAST-NEXT:    vmovdqu8 %ymm5, %ymm2 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqa %ymm2, 192(%rax)
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm7, (%rax)
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm10, 128(%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, 64(%rax)
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm4, 64(%rax)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq
   %in.vec0 = load <32 x i8>, ptr %in.vecptr0, align 64
   %in.vec1 = load <32 x i8>, ptr %in.vecptr1, align 64
   %in.vec2 = load <32 x i8>, ptr %in.vecptr2, align 64
   %in.vec3 = load <32 x i8>, ptr %in.vecptr3, align 64
   %in.vec4 = load <32 x i8>, ptr %in.vecptr4, align 64
   %in.vec5 = load <32 x i8>, ptr %in.vecptr5, align 64
; AVX512BW-FAST-NEXT:    retq
   %interleaved.vec = shufflevector <224 x i8> %8, <224 x i8> poison, <224 x i32> <i32 0, i32 32, i32 64, i32 96, i32 128, i32 160, i32 192, i32 1, i32 33, i32 65, i32 97, i32 129, i32 161, i32 193, i32 2, i32 34, i32 66, i32 98, i32 130, i32 162, i32 194, i32 3, i32 35, i32 67, i32 99, i32 131, i32 163, i32 195, i32 4, i32 36, i32 68, i32 100, i32 132, i32 164, i32 196, i32 5, i32 37, i32 69, i32 101, i32 133, i32 165, i32 197, i32 6, i32 38, i32 70, i32 102, i32 134, i32 166, i32 198, i32 7, i32 39, i32 71, i32 103, i32 135, i32 167, i32 199, i32 8, i32 40, i32 72, i32 104, i32 136, i32 168, i32 200, i32 9, i32 41, i32 73, i32 105, i32 137, i32 169, i32 201, i32 10, i32 42, i32 74, i32 106, i32 138, i32 170, i32 202, i32 11, i32 43, i32 75, i32 107, i32 139, i32 171, i32 203, i32 12, i32 44, i32 76, i32 108, i32 140, i32 172, i32 204, i32 13, i32 45, i32 77, i32 109, i32 141, i32 173, i32 205, i32 14, i32 46, i32 78, i32 110, i32 142, i32 174, i32 206, i32 15, i32 47, i32 79, i32 111, i32 143, i32 175, i32 207, i32 16, i32 48, i32 80, i32 112, i32 144, i32 176, i32 208, i32 17, i32 49, i32 81, i32 113, i32 145, i32 177, i32 209, i32 18, i32 50, i32 82, i32 114, i32 146, i32 178, i32 210, i32 19, i32 51, i32 83, i32 115, i32 147, i32 179, i32 211, i32 20, i32 52, i32 84, i32 116, i32 148, i32 180, i32 212, i32 21, i32 53, i32 85, i32 117, i32 149, i32 181, i32 213, i32 22, i32 54, i32 86, i32 118, i32 150, i32 182, i32 214, i32 23, i32 55, i32 87, i32 119, i32 151, i32 183, i32 215, i32 24, i32 56, i32 88, i32 120, i32 152, i32 184, i32 216, i32 25, i32 57, i32 89, i32 121, i32 153, i32 185, i32 217, i32 26, i32 58, i32 90, i32 122, i32 154, i32 186, i32 218, i32 27, i32 59, i32 91, i32 123, i32 155, i32 187, i32 219, i32 28, i32 60, i32 92, i32 124, i32 156, i32 188, i32 220, i32 29, i32 61, i32 93, i32 125, i32 157, i32 189, i32 221, i32 30, i32 62, i32 94, i32 126, i32 158, i32 190, i32 222, i32 31, i32 63, i32 95, i32 127, i32 159, i32 191, i32 223>
   store <224 x i8> %interleaved.vec, ptr %out.vec, align 64
   ret void
 }
 
 define void @store_i8_stride7_vf64(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.vecptr2, ptr %in.vecptr3, ptr %in.vecptr4, ptr %in.vecptr5, ptr %in.vecptr6, ptr %out.vec) nounwind {
 ; SSE-LABEL: store_i8_stride7_vf64:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    subq $648, %rsp # imm = 0x288
+; SSE-NEXT:    subq $472, %rsp # imm = 0x1D8
 ; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    movdqa 48(%rdi), %xmm1
+; SSE-NEXT:    movdqa 48(%rdi), %xmm12
 ; SSE-NEXT:    movdqa 48(%rsi), %xmm2
 ; SSE-NEXT:    movdqa 48(%rdx), %xmm3
-; SSE-NEXT:    movdqa 48(%rcx), %xmm9
-; SSE-NEXT:    movdqa 48(%r8), %xmm14
-; SSE-NEXT:    movdqa 48(%r9), %xmm8
-; SSE-NEXT:    movdqa 48(%rax), %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm12
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 48(%rcx), %xmm5
+; SSE-NEXT:    movdqa 48(%r8), %xmm4
+; SSE-NEXT:    movdqa 48(%r9), %xmm13
+; SSE-NEXT:    movdqa 48(%rax), %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm12[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm6, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm15, %xmm0
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm2, %xmm11
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm2, %xmm7
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[3,1,0,3]
-; SSE-NEXT:    movdqa %xmm6, %xmm2
+; SSE-NEXT:    movdqa %xmm15, %xmm2
 ; SSE-NEXT:    pandn %xmm1, %xmm2
 ; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm13, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm0, %xmm2
+; SSE-NEXT:    movdqa %xmm0, %xmm10
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm3, %xmm6
+; SSE-NEXT:    movdqa %xmm3, %xmm9
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm9[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,3,0,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm1, %xmm4
-; SSE-NEXT:    pandn %xmm3, %xmm4
-; SSE-NEXT:    por %xmm0, %xmm4
-; SSE-NEXT:    movdqa %xmm13, %xmm0
-; SSE-NEXT:    pandn %xmm4, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm14[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm1, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm5[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,0,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pandn %xmm1, %xmm3
 ; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
-; SSE-NEXT:    pand %xmm7, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm8[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm8, %xmm15
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm10, %xmm0
+; SSE-NEXT:    pandn %xmm3, %xmm0
+; SSE-NEXT:    por %xmm2, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pand %xmm14, %xmm0
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,2]
+; SSE-NEXT:    movdqa %xmm14, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm13[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm4
-; SSE-NEXT:    pandn %xmm0, %xmm4
-; SSE-NEXT:    por %xmm3, %xmm4
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
-; SSE-NEXT:    pand %xmm2, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm10[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
+; SSE-NEXT:    pand %xmm3, %xmm1
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm2, %xmm3
 ; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    por %xmm4, %xmm3
+; SSE-NEXT:    por %xmm1, %xmm3
 ; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa %xmm11, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm11[8],xmm0[9],xmm11[9],xmm0[10],xmm11[10],xmm0[11],xmm11[11],xmm0[12],xmm11[12],xmm0[13],xmm11[13],xmm0[14],xmm11[14],xmm0[15],xmm11[15]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm0[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE-NEXT:    movdqa %xmm12, %xmm0
+; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm7[8],xmm0[9],xmm7[9],xmm0[10],xmm7[10],xmm0[11],xmm7[11],xmm0[12],xmm7[12],xmm0[13],xmm7[13],xmm0[14],xmm7[14],xmm0[15],xmm7[15]
+; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pxor %xmm10, %xmm10
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm10[8],xmm0[9],xmm10[9],xmm0[10],xmm10[10],xmm0[11],xmm10[11],xmm0[12],xmm10[12],xmm0[13],xmm10[13],xmm0[14],xmm10[14],xmm0[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm0, %xmm3
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm6, %xmm3
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    movdqa %xmm5, %xmm1
+; SSE-NEXT:    movdqa %xmm5, %xmm11
+; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm9[8],xmm1[9],xmm9[9],xmm1[10],xmm9[10],xmm1[11],xmm9[11],xmm1[12],xmm9[12],xmm1[13],xmm9[13],xmm1[14],xmm9[14],xmm1[15],xmm9[15]
+; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    pandn %xmm3, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm12[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,2,2,2]
-; SSE-NEXT:    pand %xmm1, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm5
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm4, %xmm3
-; SSE-NEXT:    pandn %xmm5, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm6[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    movdqa %xmm6, %xmm12
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    movdqa %xmm9, %xmm5
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm9[8],xmm5[9],xmm9[9],xmm5[10],xmm9[10],xmm5[11],xmm9[11],xmm5[12],xmm9[12],xmm5[13],xmm9[13],xmm5[14],xmm9[14],xmm5[15],xmm9[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm5[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm8[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm7, %xmm8
-; SSE-NEXT:    por %xmm6, %xmm8
-; SSE-NEXT:    pand %xmm4, %xmm8
-; SSE-NEXT:    por %xmm3, %xmm8
-; SSE-NEXT:    movdqa %xmm14, %xmm4
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm14[0,1,2,3,5,6,6,7]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm10[8],xmm5[9],xmm10[9],xmm5[10],xmm10[10],xmm5[11],xmm10[11],xmm5[12],xmm10[12],xmm5[13],xmm10[13],xmm5[14],xmm10[14],xmm5[15],xmm10[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1],xmm1[2],xmm10[2],xmm1[3],xmm10[3],xmm1[4],xmm10[4],xmm1[5],xmm10[5],xmm1[6],xmm10[6],xmm1[7],xmm10[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[2,3],xmm5[0,1]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm5, %xmm1
+; SSE-NEXT:    pand %xmm6, %xmm1
+; SSE-NEXT:    por %xmm3, %xmm1
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm4[0,1,2,3,5,6,6,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    pandn %xmm3, %xmm6
-; SSE-NEXT:    pand %xmm1, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm6
-; SSE-NEXT:    movdqa %xmm15, %xmm3
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm15[8],xmm3[9],xmm15[9],xmm3[10],xmm15[10],xmm3[11],xmm15[11],xmm3[12],xmm15[12],xmm3[13],xmm15[13],xmm3[14],xmm15[14],xmm3[15],xmm15[15]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm3[1,1,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    pandn %xmm8, %xmm11
-; SSE-NEXT:    pand %xmm1, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm11
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm10[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm15, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
-; SSE-NEXT:    pand %xmm15, %xmm11
-; SSE-NEXT:    por %xmm11, %xmm8
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm5[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm12[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,1,3,2]
-; SSE-NEXT:    pand %xmm1, %xmm6
-; SSE-NEXT:    por %xmm8, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm9, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm14[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm11
-; SSE-NEXT:    pandn %xmm6, %xmm11
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm0[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm6
-; SSE-NEXT:    por %xmm11, %xmm6
-; SSE-NEXT:    pand %xmm9, %xmm6
-; SSE-NEXT:    por %xmm8, %xmm6
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm6, %xmm5
+; SSE-NEXT:    pandn %xmm3, %xmm5
+; SSE-NEXT:    pand %xmm6, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm5
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm4[0,1,2,3,4,4,6,5]
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm4[0,1,2,3,7,7,7,7]
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm13[0],xmm4[1],xmm13[1],xmm4[2],xmm13[2],xmm4[3],xmm13[3],xmm4[4],xmm13[4],xmm4[5],xmm13[5],xmm4[6],xmm13[6],xmm4[7],xmm13[7]
 ; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm4[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,1,3,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm11
-; SSE-NEXT:    pandn %xmm8, %xmm11
-; SSE-NEXT:    pand %xmm15, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm11
-; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm3[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm9, %xmm8
-; SSE-NEXT:    pandn %xmm6, %xmm8
-; SSE-NEXT:    pand %xmm9, %xmm11
-; SSE-NEXT:    por %xmm11, %xmm8
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm10[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm9, %xmm11
-; SSE-NEXT:    pandn %xmm6, %xmm11
-; SSE-NEXT:    pand %xmm9, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm11
-; SSE-NEXT:    movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm9, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm12[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,2,2,2]
-; SSE-NEXT:    pand %xmm9, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    pandn %xmm6, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]
+; SSE-NEXT:    movdqa %xmm13, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm13[8],xmm1[9],xmm13[9],xmm1[10],xmm13[10],xmm1[11],xmm13[11],xmm1[12],xmm13[12],xmm1[13],xmm13[13],xmm1[14],xmm13[14],xmm1[15],xmm13[15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,2,3]
+; SSE-NEXT:    movdqa %xmm15, %xmm2
 ; SSE-NEXT:    movdqa %xmm15, %xmm6
-; SSE-NEXT:    pandn %xmm0, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm14[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm6
-; SSE-NEXT:    pand %xmm1, %xmm6
+; SSE-NEXT:    pandn %xmm3, %xmm6
+; SSE-NEXT:    pand %xmm15, %xmm5
 ; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm8[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm5, %xmm4
+; SSE-NEXT:    pandn %xmm3, %xmm4
+; SSE-NEXT:    pand %xmm5, %xmm6
+; SSE-NEXT:    movdqa %xmm5, %xmm13
+; SSE-NEXT:    por %xmm6, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm9, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm11[8],xmm3[9],xmm11[9],xmm3[10],xmm11[10],xmm3[11],xmm11[11],xmm3[12],xmm11[12],xmm3[13],xmm11[13],xmm3[14],xmm11[14],xmm3[15],xmm11[15]
+; SSE-NEXT:    movdqa %xmm3, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm10[0],xmm5[1],xmm10[1],xmm5[2],xmm10[2],xmm5[3],xmm10[3],xmm5[4],xmm10[4],xmm5[5],xmm10[5],xmm5[6],xmm10[6],xmm5[7],xmm10[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm5[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm5
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm6, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm6
+; SSE-NEXT:    pandn %xmm5, %xmm6
 ; SSE-NEXT:    movdqa %xmm7, %xmm5
-; SSE-NEXT:    pandn %xmm0, %xmm5
-; SSE-NEXT:    pand %xmm7, %xmm6
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm12[8],xmm5[9],xmm12[9],xmm5[10],xmm12[10],xmm5[11],xmm12[11],xmm5[12],xmm12[12],xmm5[13],xmm12[13],xmm5[14],xmm12[14],xmm5[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm10[0],xmm5[1],xmm10[1],xmm5[2],xmm10[2],xmm5[3],xmm10[3],xmm5[4],xmm10[4],xmm5[5],xmm10[5],xmm5[6],xmm10[6],xmm5[7],xmm10[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm5[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm5[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm7, %xmm5
+; SSE-NEXT:    pand %xmm4, %xmm5
 ; SSE-NEXT:    por %xmm6, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm3[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
-; SSE-NEXT:    movdqa %xmm2, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    pand %xmm2, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm10[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    pshufd $246, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm6 = mem[2,1,3,3]
+; SSE-NEXT:    movdqa %xmm13, %xmm7
+; SSE-NEXT:    pandn %xmm6, %xmm7
+; SSE-NEXT:    pand %xmm13, %xmm5
+; SSE-NEXT:    por %xmm5, %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm1[1,2,2,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm15, %xmm6
+; SSE-NEXT:    pandn %xmm5, %xmm6
+; SSE-NEXT:    pand %xmm15, %xmm7
+; SSE-NEXT:    movdqa %xmm15, %xmm9
+; SSE-NEXT:    por %xmm7, %xmm6
+; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm8[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE-NEXT:    movdqa %xmm14, %xmm4
+; SSE-NEXT:    pandn %xmm5, %xmm4
+; SSE-NEXT:    pand %xmm14, %xmm6
+; SSE-NEXT:    por %xmm6, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm10[8],xmm3[9],xmm10[9],xmm3[10],xmm10[10],xmm3[11],xmm10[11],xmm3[12],xmm10[12],xmm3[13],xmm10[13],xmm3[14],xmm10[14],xmm3[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm3, %xmm5
+; SSE-NEXT:    psrldq {{.*#+}} xmm5 = xmm5[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm5, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
 ; SSE-NEXT:    movdqa %xmm4, %xmm5
-; SSE-NEXT:    pandn %xmm0, %xmm5
+; SSE-NEXT:    pandn %xmm3, %xmm5
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm0, %xmm3
 ; SSE-NEXT:    pand %xmm4, %xmm3
+; SSE-NEXT:    por %xmm5, %xmm3
+; SSE-NEXT:    pshufd $170, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm0 = mem[2,2,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa %xmm12, %xmm5
+; SSE-NEXT:    pandn %xmm0, %xmm5
+; SSE-NEXT:    pand %xmm12, %xmm3
+; SSE-NEXT:    movdqa %xmm12, %xmm15
 ; SSE-NEXT:    por %xmm3, %xmm5
-; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,5,6,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255]
+; SSE-NEXT:    movdqa %xmm3, %xmm1
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    pand %xmm3, %xmm5
+; SSE-NEXT:    movdqa %xmm3, %xmm6
+; SSE-NEXT:    por %xmm5, %xmm1
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,6,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    pandn %xmm0, %xmm4
+; SSE-NEXT:    pand %xmm3, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa (%rsi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,0,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm15, %xmm3
+; SSE-NEXT:    movdqa %xmm2, %xmm3
 ; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    movdqa (%rdi), %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm10[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa (%rdi), %xmm7
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm7[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    movdqa %xmm2, %xmm8
 ; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    movdqa (%rcx), %xmm6
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[2,1,2,3]
+; SSE-NEXT:    movdqa (%rcx), %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm10[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,0,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm4, %xmm5
+; SSE-NEXT:    movdqa %xmm9, %xmm5
 ; SSE-NEXT:    pandn %xmm0, %xmm5
 ; SSE-NEXT:    movdqa (%rdx), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    pand %xmm9, %xmm0
+; SSE-NEXT:    movdqa %xmm9, %xmm14
 ; SSE-NEXT:    por %xmm0, %xmm5
-; SSE-NEXT:    movdqa %xmm13, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    pandn %xmm5, %xmm0
-; SSE-NEXT:    pand %xmm13, %xmm3
+; SSE-NEXT:    pand %xmm1, %xmm3
 ; SSE-NEXT:    por %xmm3, %xmm0
-; SSE-NEXT:    movdqa (%r9), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,1,2,3]
+; SSE-NEXT:    movdqa (%r9), %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[2,1,2,3]
+; SSE-NEXT:    movdqa %xmm2, %xmm12
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm8
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    movdqa (%r8), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm15, %xmm5
+; SSE-NEXT:    pandn %xmm3, %xmm5
+; SSE-NEXT:    movdqa (%r8), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm7, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm8
-; SSE-NEXT:    movdqa (%rax), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pand %xmm15, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm5
+; SSE-NEXT:    movdqa (%rax), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm2[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm6, %xmm2
+; SSE-NEXT:    pandn %xmm3, %xmm6
+; SSE-NEXT:    pand %xmm2, %xmm5
 ; SSE-NEXT:    movdqa %xmm2, %xmm11
-; SSE-NEXT:    pandn %xmm3, %xmm11
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm11
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
-; SSE-NEXT:    movdqa %xmm9, %xmm3
-; SSE-NEXT:    pandn %xmm11, %xmm3
-; SSE-NEXT:    pand %xmm9, %xmm0
+; SSE-NEXT:    por %xmm5, %xmm6
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm6, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm0
 ; SSE-NEXT:    por %xmm0, %xmm3
 ; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa 16(%rsi), %xmm0
-; SSE-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,0,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm8, %xmm5
+; SSE-NEXT:    pandn %xmm0, %xmm5
 ; SSE-NEXT:    movdqa 16(%rdi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    pand %xmm8, %xmm0
+; SSE-NEXT:    movdqa %xmm8, %xmm6
+; SSE-NEXT:    por %xmm0, %xmm5
 ; SSE-NEXT:    movdqa 16(%rcx), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,0,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm4, %xmm8
+; SSE-NEXT:    movdqa %xmm9, %xmm8
 ; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    movdqa 16(%rdx), %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa 16(%rdx), %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    pand %xmm9, %xmm0
 ; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    movdqa %xmm13, %xmm0
+; SSE-NEXT:    movdqa %xmm1, %xmm0
 ; SSE-NEXT:    pandn %xmm8, %xmm0
-; SSE-NEXT:    pand %xmm13, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm0
-; SSE-NEXT:    movdqa 16(%r9), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,1,2,3]
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm8
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    movdqa 16(%r8), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm7, %xmm3
-; SSE-NEXT:    por %xmm3, %xmm8
-; SSE-NEXT:    movdqa 16(%rax), %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm3[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm2, %xmm11
-; SSE-NEXT:    pandn %xmm3, %xmm11
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm11
-; SSE-NEXT:    movdqa %xmm9, %xmm1
-; SSE-NEXT:    pandn %xmm11, %xmm1
-; SSE-NEXT:    pand %xmm9, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pand %xmm1, %xmm5
+; SSE-NEXT:    por %xmm5, %xmm0
+; SSE-NEXT:    movdqa 16(%r9), %xmm3
+; SSE-NEXT:    movdqa %xmm3, (%rsp) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,1,2,3]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,3,2,0]
+; SSE-NEXT:    movdqa %xmm15, %xmm9
+; SSE-NEXT:    pandn %xmm5, %xmm9
+; SSE-NEXT:    movdqa 16(%r8), %xmm3
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    pand %xmm15, %xmm5
+; SSE-NEXT:    por %xmm5, %xmm9
+; SSE-NEXT:    movdqa 16(%rax), %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,4,4,4]
+; SSE-NEXT:    movdqa %xmm11, %xmm13
+; SSE-NEXT:    pandn %xmm5, %xmm13
+; SSE-NEXT:    pand %xmm11, %xmm9
+; SSE-NEXT:    movdqa %xmm11, %xmm8
+; SSE-NEXT:    por %xmm9, %xmm13
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm13, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa 32(%rsi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[3,1,0,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm6, %xmm5
+; SSE-NEXT:    pandn %xmm0, %xmm5
 ; SSE-NEXT:    movdqa 32(%rdi), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    pand %xmm6, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm5
 ; SSE-NEXT:    movdqa 32(%rcx), %xmm0
 ; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,0,3]
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    movdqa %xmm4, %xmm11
-; SSE-NEXT:    movdqa %xmm4, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    movdqa 32(%rdx), %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm14, %xmm2
+; SSE-NEXT:    pandn %xmm0, %xmm14
+; SSE-NEXT:    movdqa 32(%rdx), %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    pand %xmm13, %xmm3
-; SSE-NEXT:    pandn %xmm8, %xmm13
-; SSE-NEXT:    por %xmm3, %xmm13
-; SSE-NEXT:    movdqa 32(%r9), %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,1,2,3]
-; SSE-NEXT:    movdqa %xmm1, %xmm14
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm14
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    pand %xmm1, %xmm5
+; SSE-NEXT:    pandn %xmm14, %xmm0
+; SSE-NEXT:    por %xmm5, %xmm0
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    movdqa 32(%r9), %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
 ; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    movdqa 32(%r8), %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm8[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm15, %xmm11
+; SSE-NEXT:    movdqa %xmm15, %xmm5
+; SSE-NEXT:    pandn %xmm0, %xmm5
+; SSE-NEXT:    movdqa 32(%r8), %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
-; SSE-NEXT:    pand %xmm2, %xmm3
-; SSE-NEXT:    movdqa 32(%rax), %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm12[3,3,3,3,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm5
+; SSE-NEXT:    movdqa %xmm8, %xmm2
+; SSE-NEXT:    pand %xmm8, %xmm5
+; SSE-NEXT:    movdqa 32(%rax), %xmm0
+; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,3,3,3,4,5,6,7]
 ; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,4,4]
 ; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    por %xmm3, %xmm2
-; SSE-NEXT:    pand %xmm9, %xmm13
-; SSE-NEXT:    pandn %xmm2, %xmm9
-; SSE-NEXT:    por %xmm13, %xmm9
-; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm2, %xmm1
-; SSE-NEXT:    pandn %xmm0, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm10[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm10, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm13[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    movdqa %xmm6, %xmm0
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm6[8],xmm0[9],xmm6[9],xmm0[10],xmm6[10],xmm0[11],xmm6[11],xmm0[12],xmm6[12],xmm0[13],xmm6[13],xmm0[14],xmm6[14],xmm0[15],xmm6[15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm3, %xmm1
-; SSE-NEXT:    pand %xmm10, %xmm1
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,1,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm0
-; SSE-NEXT:    movdqa %xmm15, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    pshufhw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm2 = mem[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
-; SSE-NEXT:    pand %xmm15, %xmm2
-; SSE-NEXT:    por %xmm3, %xmm2
-; SSE-NEXT:    pshufhw $164, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm9, %xmm11
-; SSE-NEXT:    pandn %xmm3, %xmm11
-; SSE-NEXT:    pand %xmm9, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm11
-; SSE-NEXT:    movdqa {{.*#+}} xmm9 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm9, %xmm2
-; SSE-NEXT:    pandn %xmm11, %xmm2
-; SSE-NEXT:    pand %xmm9, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa (%rsp), %xmm1 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm15, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    pshufhw $170, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    pand %xmm15, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm10, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm5[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    pand %xmm10, %xmm1
-; SSE-NEXT:    movdqa %xmm10, %xmm5
-; SSE-NEXT:    por %xmm3, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
-; SSE-NEXT:    movdqa %xmm0, %xmm10
-; SSE-NEXT:    movdqa %xmm0, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    pshufhw $233, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm2 = mem[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    por %xmm3, %xmm2
-; SSE-NEXT:    pshufhw $164, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm3 = mem[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm11
-; SSE-NEXT:    pandn %xmm3, %xmm11
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm11
-; SSE-NEXT:    movdqa %xmm9, %xmm2
-; SSE-NEXT:    pandn %xmm11, %xmm2
-; SSE-NEXT:    pand %xmm9, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    pshufhw $170, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; SSE-NEXT:    # xmm1 = mem[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    pand %xmm15, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm5, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm4[0,1,2,3,6,6,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,2,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,1,2,0,4,5,6,7]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pand %xmm7, %xmm1
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    pand %xmm5, %xmm1
-; SSE-NEXT:    por %xmm3, %xmm1
+; SSE-NEXT:    por %xmm5, %xmm2
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    pandn %xmm2, %xmm4
+; SSE-NEXT:    por %xmm1, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm10[0],xmm3[1],xmm10[1],xmm3[2],xmm10[2],xmm3[3],xmm10[3],xmm3[4],xmm10[4],xmm3[5],xmm10[5],xmm3[6],xmm10[6],xmm3[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm3, %xmm14
+; SSE-NEXT:    pxor %xmm8, %xmm8
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm14 = xmm14[0],xmm8[0],xmm14[1],xmm8[1],xmm14[2],xmm8[2],xmm14[3],xmm8[3],xmm14[4],xmm8[4],xmm14[5],xmm8[5],xmm14[6],xmm8[6],xmm14[7],xmm8[7]
 ; SSE-NEXT:    movdqa %xmm14, %xmm2
-; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm14[8],xmm2[9],xmm14[9],xmm2[10],xmm14[10],xmm2[11],xmm14[11],xmm2[12],xmm14[12],xmm2[13],xmm14[13],xmm2[14],xmm14[14],xmm2[15],xmm14[15]
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[1,1,2,3]
-; SSE-NEXT:    movdqa %xmm10, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm3
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm8[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
-; SSE-NEXT:    pand %xmm10, %xmm2
-; SSE-NEXT:    por %xmm3, %xmm2
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm12[0,1,2,3,4,5,6,6]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,3]
-; SSE-NEXT:    movdqa %xmm0, %xmm11
-; SSE-NEXT:    pandn %xmm3, %xmm11
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm0, %xmm14
-; SSE-NEXT:    por %xmm2, %xmm11
-; SSE-NEXT:    pand %xmm9, %xmm1
-; SSE-NEXT:    pandn %xmm11, %xmm9
-; SSE-NEXT:    por %xmm1, %xmm9
-; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm6[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm0, %xmm2
-; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm13[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
+; SSE-NEXT:    psrlq $16, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm14[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm5
 ; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    movdqa %xmm1, %xmm3
-; SSE-NEXT:    pandn %xmm2, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm15 = xmm15[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm15[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[0,1,1,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm4, %xmm1
-; SSE-NEXT:    pandn %xmm2, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm8[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    pand %xmm4, %xmm2
-; SSE-NEXT:    movdqa %xmm4, %xmm12
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    pand %xmm3, %xmm1
-; SSE-NEXT:    por %xmm11, %xmm1
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm5[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm11
-; SSE-NEXT:    pandn %xmm2, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm9[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm11
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm5, %xmm2
+; SSE-NEXT:    movdqa %xmm7, %xmm5
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm5 = xmm5[0],mem[0],xmm5[1],mem[1],xmm5[2],mem[2],xmm5[3],mem[3],xmm5[4],mem[4],xmm5[5],mem[5],xmm5[6],mem[6],xmm5[7],mem[7]
+; SSE-NEXT:    movdqa %xmm5, %xmm0
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm8[0],xmm0[1],xmm8[1],xmm0[2],xmm8[2],xmm0[3],xmm8[3],xmm0[4],xmm8[4],xmm0[5],xmm8[5],xmm0[6],xmm8[6],xmm0[7],xmm8[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm13 = xmm0[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm13 = xmm13[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm15 = xmm0[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm15 = xmm15[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm13, %xmm15
+; SSE-NEXT:    pand %xmm1, %xmm15
+; SSE-NEXT:    por %xmm2, %xmm15
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm10
-; SSE-NEXT:    pandn %xmm2, %xmm10
-; SSE-NEXT:    pand %xmm7, %xmm11
-; SSE-NEXT:    por %xmm11, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm2
-; SSE-NEXT:    pandn %xmm10, %xmm2
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    por %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm15[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm12[0],xmm4[1],xmm12[1],xmm4[2],xmm12[2],xmm4[3],xmm12[3],xmm4[4],xmm12[4],xmm4[5],xmm12[5],xmm4[6],xmm12[6],xmm4[7],xmm12[7]
+; SSE-NEXT:    movdqa %xmm4, %xmm13
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0],xmm8[0],xmm13[1],xmm8[1],xmm13[2],xmm8[2],xmm13[3],xmm8[3],xmm13[4],xmm8[4],xmm13[5],xmm8[5],xmm13[6],xmm8[6],xmm13[7],xmm8[7]
+; SSE-NEXT:    movdqa %xmm13, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm13, %xmm2
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm1[0,0,0,0]
+; SSE-NEXT:    movdqa %xmm11, %xmm1
+; SSE-NEXT:    pandn %xmm6, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,3]
+; SSE-NEXT:    pand %xmm11, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm6, %xmm2
 ; SSE-NEXT:    pandn %xmm1, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm8[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm1[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm10
-; SSE-NEXT:    por %xmm2, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm1, %xmm3
-; SSE-NEXT:    pandn %xmm10, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm6[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    movdqa %xmm12, %xmm1
-; SSE-NEXT:    pandn %xmm10, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm10 = xmm13[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm13, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm10[0,0,2,1]
-; SSE-NEXT:    pand %xmm1, %xmm10
-; SSE-NEXT:    por %xmm10, %xmm12
-; SSE-NEXT:    pand %xmm3, %xmm12
-; SSE-NEXT:    por %xmm2, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm5[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm14, %xmm10
-; SSE-NEXT:    pandn %xmm2, %xmm10
+; SSE-NEXT:    pand %xmm6, %xmm15
+; SSE-NEXT:    por %xmm15, %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm5 = xmm5[8],xmm8[8],xmm5[9],xmm8[9],xmm5[10],xmm8[10],xmm5[11],xmm8[11],xmm5[12],xmm8[12],xmm5[13],xmm8[13],xmm5[14],xmm8[14],xmm5[15],xmm8[15]
+; SSE-NEXT:    pxor %xmm2, %xmm2
+; SSE-NEXT:    psrlq $16, %xmm0
+; SSE-NEXT:    packuswb %xmm5, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm15, %xmm1
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm2[8],xmm3[9],xmm2[9],xmm3[10],xmm2[10],xmm3[11],xmm2[11],xmm3[12],xmm2[12],xmm3[13],xmm2[13],xmm3[14],xmm2[14],xmm3[15],xmm2[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm14[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm14 = xmm14[0,3],xmm3[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm14[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm13[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm13 = xmm13[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm13, %xmm1
 ; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm9[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm9, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    pand %xmm14, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm10
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm4, %xmm11
 ; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm13 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm13, %xmm14
-; SSE-NEXT:    pandn %xmm2, %xmm14
-; SSE-NEXT:    pand %xmm13, %xmm10
-; SSE-NEXT:    por %xmm10, %xmm14
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm8, %xmm6
+; SSE-NEXT:    pandn %xmm2, %xmm6
+; SSE-NEXT:    pand %xmm8, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm6
 ; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
 ; SSE-NEXT:    movdqa %xmm2, %xmm1
-; SSE-NEXT:    pandn %xmm14, %xmm1
-; SSE-NEXT:    pand %xmm2, %xmm12
-; SSE-NEXT:    por %xmm12, %xmm1
+; SSE-NEXT:    pandn %xmm6, %xmm1
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm1
 ; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm6[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm7, %xmm6
-; SSE-NEXT:    pandn %xmm2, %xmm6
-; SSE-NEXT:    movdqa %xmm0, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm0[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm2
-; SSE-NEXT:    por %xmm6, %xmm2
-; SSE-NEXT:    movdqa %xmm3, %xmm6
-; SSE-NEXT:    pandn %xmm2, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm15[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm13, %xmm10
-; SSE-NEXT:    pandn %xmm2, %xmm10
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm12[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,1,3]
-; SSE-NEXT:    pand %xmm13, %xmm2
-; SSE-NEXT:    movdqa %xmm13, %xmm14
-; SSE-NEXT:    por %xmm2, %xmm10
-; SSE-NEXT:    pand %xmm3, %xmm10
-; SSE-NEXT:    por %xmm6, %xmm10
+; SSE-NEXT:    movdqa %xmm3, %xmm0
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm3, %xmm0
+; SSE-NEXT:    movdqa %xmm15, %xmm1
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm5 = zero,zero,xmm5[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm0, %xmm5
+; SSE-NEXT:    pand %xmm15, %xmm5
+; SSE-NEXT:    por %xmm1, %xmm5
 ; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm2
 ; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    pandn %xmm10, %xmm2
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
 ; SSE-NEXT:    movdqa %xmm1, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm9[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,2,1]
-; SSE-NEXT:    pand %xmm1, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm11[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
+; SSE-NEXT:    pandn %xmm5, %xmm0
+; SSE-NEXT:    pxor %xmm13, %xmm13
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm13[8],xmm4[9],xmm13[9],xmm4[10],xmm13[10],xmm4[11],xmm13[11],xmm4[12],xmm13[12],xmm4[13],xmm13[13],xmm4[14],xmm13[14],xmm4[15],xmm13[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
 ; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm1
-; SSE-NEXT:    pandn %xmm5, %xmm1
-; SSE-NEXT:    pand %xmm3, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm3, %xmm5
-; SSE-NEXT:    pandn %xmm2, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm4[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    pandn %xmm1, %xmm4
 ; SSE-NEXT:    pand %xmm3, %xmm2
-; SSE-NEXT:    por %xmm5, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm1, %xmm5
-; SSE-NEXT:    pandn %xmm2, %xmm5
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm12[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm6
-; SSE-NEXT:    pandn %xmm2, %xmm6
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm9[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm2
-; SSE-NEXT:    por %xmm6, %xmm2
-; SSE-NEXT:    pand %xmm1, %xmm2
-; SSE-NEXT:    por %xmm5, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm13[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm14, %xmm6
-; SSE-NEXT:    pandn %xmm5, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm8[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[2,1,3,3]
-; SSE-NEXT:    pand %xmm14, %xmm5
-; SSE-NEXT:    por %xmm5, %xmm6
-; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm11[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm15, %xmm10
-; SSE-NEXT:    pandn %xmm5, %xmm10
-; SSE-NEXT:    pand %xmm15, %xmm6
-; SSE-NEXT:    por %xmm6, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm5, %xmm1
-; SSE-NEXT:    pandn %xmm10, %xmm1
-; SSE-NEXT:    pand %xmm5, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm1
-; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm9[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm14, %xmm10
-; SSE-NEXT:    pandn %xmm2, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm12[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,2,2,2]
-; SSE-NEXT:    pand %xmm14, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm10
-; SSE-NEXT:    pand %xmm1, %xmm10
-; SSE-NEXT:    por %xmm0, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm13[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm11[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm2, %xmm8
-; SSE-NEXT:    pand %xmm12, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
-; SSE-NEXT:    movdqa %xmm1, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm10
-; SSE-NEXT:    por %xmm10, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm9[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm14, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm14, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm2, %xmm8
-; SSE-NEXT:    movdqa (%rsp), %xmm13 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm13 = xmm13[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,1,0]
+; SSE-NEXT:    por %xmm2, %xmm4
+; SSE-NEXT:    pand %xmm6, %xmm4
+; SSE-NEXT:    por %xmm0, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
 ; SSE-NEXT:    movdqa %xmm12, %xmm0
-; SSE-NEXT:    pandn %xmm2, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    pand %xmm12, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm5[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    movdqa %xmm15, %xmm8
-; SSE-NEXT:    pandn %xmm2, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    pand %xmm15, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm3[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm10
-; SSE-NEXT:    pandn %xmm2, %xmm10
-; SSE-NEXT:    pand %xmm7, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm2
-; SSE-NEXT:    pandn %xmm10, %xmm2
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm2, (%rsp) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm13[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm10[8],xmm0[9],xmm10[9],xmm0[10],xmm10[10],xmm0[11],xmm10[11],xmm0[12],xmm10[12],xmm0[13],xmm10[13],xmm0[14],xmm10[14],xmm0[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm13[0],xmm1[1],xmm13[1],xmm1[2],xmm13[2],xmm1[3],xmm13[3],xmm1[4],xmm13[4],xmm1[5],xmm13[5],xmm1[6],xmm13[6],xmm1[7],xmm13[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
 ; SSE-NEXT:    movdqa %xmm4, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    movdqa %xmm12, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm12, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    pand %xmm4, %xmm8
-; SSE-NEXT:    movdqa %xmm4, %xmm12
-; SSE-NEXT:    por %xmm2, %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm11, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm7[8],xmm1[9],xmm7[9],xmm1[10],xmm7[10],xmm1[11],xmm7[11],xmm1[12],xmm7[12],xmm1[13],xmm7[13],xmm1[14],xmm7[14],xmm1[15],xmm7[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm13[0],xmm1[1],xmm13[1],xmm1[2],xmm13[2],xmm1[3],xmm13[3],xmm1[4],xmm13[4],xmm1[5],xmm13[5],xmm1[6],xmm13[6],xmm1[7],xmm13[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
 ; SSE-NEXT:    movdqa %xmm14, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm14, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm3[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm2 = xmm2[8],mem[8],xmm2[9],mem[9],xmm2[10],mem[10],xmm2[11],mem[11],xmm2[12],mem[12],xmm2[13],mem[13],xmm2[14],mem[14],xmm2[15],mem[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm13[0],xmm3[1],xmm13[1],xmm3[2],xmm13[2],xmm3[3],xmm13[3],xmm3[4],xmm13[4],xmm3[5],xmm13[5],xmm3[6],xmm13[6],xmm3[7],xmm13[7]
 ; SSE-NEXT:    movdqa %xmm3, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm1, %xmm10
-; SSE-NEXT:    pandn %xmm0, %xmm10
-; SSE-NEXT:    pand %xmm1, %xmm2
-; SSE-NEXT:    por %xmm2, %xmm10
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm2, %xmm0
-; SSE-NEXT:    pandn %xmm10, %xmm0
-; SSE-NEXT:    pand %xmm2, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm9[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm7, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    movdqa %xmm12, %xmm3
-; SSE-NEXT:    movdqa %xmm12, %xmm2
-; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm13[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm12[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    movdqa %xmm1, %xmm14
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    pand %xmm3, %xmm8
-; SSE-NEXT:    por %xmm2, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm5[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm1, %xmm8
-; SSE-NEXT:    pandn %xmm2, %xmm8
-; SSE-NEXT:    movdqa %xmm11, %xmm5
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm11[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,0,2,1]
-; SSE-NEXT:    pand %xmm1, %xmm2
-; SSE-NEXT:    movdqa %xmm1, %xmm11
-; SSE-NEXT:    por %xmm2, %xmm8
-; SSE-NEXT:    movdqa %xmm4, %xmm1
-; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm4[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm10 = xmm2[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm4, %xmm2
-; SSE-NEXT:    pandn %xmm10, %xmm2
-; SSE-NEXT:    pand %xmm4, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm2
-; SSE-NEXT:    pand %xmm3, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm4, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm4, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm12[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    movdqa %xmm12, %xmm13
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm10
-; SSE-NEXT:    pandn %xmm0, %xmm10
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm10, %xmm0
-; SSE-NEXT:    pand %xmm4, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm4[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm14, %xmm10
-; SSE-NEXT:    pandn %xmm8, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm5[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,1,3,3]
-; SSE-NEXT:    pand %xmm14, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm1[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm12
-; SSE-NEXT:    pandn %xmm8, %xmm12
-; SSE-NEXT:    pand %xmm15, %xmm10
-; SSE-NEXT:    por %xmm10, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm9
-; SSE-NEXT:    pandn %xmm12, %xmm9
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm9
-; SSE-NEXT:    movdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm3[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm8
+; SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,xmm4[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm5, %xmm4
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm9[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm8, %xmm6
+; SSE-NEXT:    pandn %xmm5, %xmm6
+; SSE-NEXT:    pand %xmm8, %xmm4
+; SSE-NEXT:    por %xmm4, %xmm6
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm5, %xmm4
+; SSE-NEXT:    pandn %xmm6, %xmm4
+; SSE-NEXT:    pand %xmm5, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm11[8],xmm7[9],xmm11[9],xmm7[10],xmm11[10],xmm7[11],xmm11[11],xmm7[12],xmm11[12],xmm7[13],xmm11[13],xmm7[14],xmm11[14],xmm7[15],xmm11[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm13[8],xmm7[9],xmm13[9],xmm7[10],xmm13[10],xmm7[11],xmm13[11],xmm7[12],xmm13[12],xmm7[13],xmm13[13],xmm7[14],xmm13[14],xmm7[15],xmm13[15]
+; SSE-NEXT:    movdqa %xmm7, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm7, %xmm4
+; SSE-NEXT:    packuswb %xmm1, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,3]
 ; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    movdqa %xmm6, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm2[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm10
-; SSE-NEXT:    pandn %xmm8, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm13[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,2,2,2]
-; SSE-NEXT:    pand %xmm3, %xmm8
-; SSE-NEXT:    movdqa %xmm3, %xmm2
-; SSE-NEXT:    por %xmm8, %xmm10
-; SSE-NEXT:    pand %xmm6, %xmm10
-; SSE-NEXT:    por %xmm0, %xmm10
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm5[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm8 = xmm1[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[2,1,3,2]
-; SSE-NEXT:    movdqa %xmm11, %xmm3
-; SSE-NEXT:    movdqa %xmm11, %xmm12
-; SSE-NEXT:    pandn %xmm8, %xmm12
-; SSE-NEXT:    pand %xmm11, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
-; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pandn %xmm12, %xmm1
-; SSE-NEXT:    pand %xmm0, %xmm10
-; SSE-NEXT:    por %xmm10, %xmm1
+; SSE-NEXT:    movdqa %xmm6, %xmm4
+; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm10 = xmm10[8],xmm12[8],xmm10[9],xmm12[9],xmm10[10],xmm12[10],xmm10[11],xmm12[11],xmm10[12],xmm12[12],xmm10[13],xmm12[13],xmm10[14],xmm12[14],xmm10[15],xmm12[15]
+; SSE-NEXT:    movdqa %xmm10, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm13[8],xmm1[9],xmm13[9],xmm1[10],xmm13[10],xmm1[11],xmm13[11],xmm1[12],xmm13[12],xmm1[13],xmm13[13],xmm1[14],xmm13[14],xmm1[15],xmm13[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm10 = xmm10[0],xmm13[0],xmm10[1],xmm13[1],xmm10[2],xmm13[2],xmm10[3],xmm13[3],xmm10[4],xmm13[4],xmm10[5],xmm13[5],xmm10[6],xmm13[6],xmm10[7],xmm13[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[3,1],xmm1[0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm5 = xmm10[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm5
+; SSE-NEXT:    pand %xmm6, %xmm5
+; SSE-NEXT:    por %xmm4, %xmm5
+; SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm13[8],xmm2[9],xmm13[9],xmm2[10],xmm13[10],xmm2[11],xmm13[11],xmm2[12],xmm13[12],xmm2[13],xmm13[13],xmm2[14],xmm13[14],xmm2[15],xmm13[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm9[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    pand %xmm4, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm3, %xmm1
+; SSE-NEXT:    pandn %xmm2, %xmm1
+; SSE-NEXT:    pand %xmm3, %xmm5
+; SSE-NEXT:    por %xmm5, %xmm1
 ; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm11 = xmm11[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm11[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm2, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    movdqa %xmm2, %xmm13
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm5, %xmm10
-; SSE-NEXT:    pandn %xmm8, %xmm10
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm13[8],xmm0[9],xmm13[9],xmm0[10],xmm13[10],xmm0[11],xmm13[11],xmm0[12],xmm13[12],xmm0[13],xmm13[13],xmm0[14],xmm13[14],xmm0[15],xmm13[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm6, %xmm1
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm7 = xmm7[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm7, %xmm0
+; SSE-NEXT:    pand %xmm6, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm0[0,1,1,0]
-; SSE-NEXT:    movdqa %xmm3, %xmm0
-; SSE-NEXT:    pandn %xmm8, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm14[8],xmm2[9],xmm14[9],xmm2[10],xmm14[10],xmm2[11],xmm14[11],xmm2[12],xmm14[12],xmm2[13],xmm14[13],xmm2[14],xmm14[14],xmm2[15],xmm14[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm13[8],xmm2[9],xmm13[9],xmm2[10],xmm13[10],xmm2[11],xmm13[11],xmm2[12],xmm13[12],xmm2[13],xmm13[13],xmm2[14],xmm13[14],xmm2[15],xmm13[15]
+; SSE-NEXT:    pxor %xmm10, %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm9[0,1,2,3,6,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
+; SSE-NEXT:    movdqa %xmm2, %xmm1
+; SSE-NEXT:    pandn %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm2, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm1
+; SSE-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm4[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,0,2,1]
-; SSE-NEXT:    pand %xmm3, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    pand %xmm5, %xmm0
-; SSE-NEXT:    por %xmm10, %xmm0
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm4 = xmm4[0],mem[0],xmm4[1],mem[1],xmm4[2],mem[2],xmm4[3],mem[3],xmm4[4],mem[4],xmm4[5],mem[5],xmm4[6],mem[6],xmm4[7],mem[7]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm2, %xmm1
+; SSE-NEXT:    psrlq $16, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm7, %xmm0
+; SSE-NEXT:    pandn %xmm3, %xmm0
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm12, %xmm1
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3],xmm1[4],mem[4],xmm1[5],mem[5],xmm1[6],mem[6],xmm1[7],mem[7]
+; SSE-NEXT:    movdqa %xmm1, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm10[0],xmm3[1],xmm10[1],xmm3[2],xmm10[2],xmm3[3],xmm10[3],xmm3[4],xmm10[4],xmm3[5],xmm10[5],xmm3[6],xmm10[6],xmm3[7],xmm10[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm5 = xmm3[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm5[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm3[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm5, %xmm6
+; SSE-NEXT:    pand %xmm7, %xmm6
+; SSE-NEXT:    por %xmm0, %xmm6
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; SSE-NEXT:    punpcklbw (%rsp), %xmm8 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm8 = xmm8[0],mem[0],xmm8[1],mem[1],xmm8[2],mem[2],xmm8[3],mem[3],xmm8[4],mem[4],xmm8[5],mem[5],xmm8[6],mem[6],xmm8[7],mem[7]
+; SSE-NEXT:    movdqa %xmm8, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0],xmm10[0],xmm5[1],xmm10[1],xmm5[2],xmm10[2],xmm5[3],xmm10[3],xmm5[4],xmm10[4],xmm5[5],xmm10[5],xmm5[6],xmm10[6],xmm5[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm5, %xmm7
+; SSE-NEXT:    pslldq {{.*#+}} xmm7 = zero,zero,xmm7[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm5, %xmm13
+; SSE-NEXT:    packuswb %xmm7, %xmm13
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm9[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,0,2,1]
-; SSE-NEXT:    movdqa %xmm15, %xmm10
-; SSE-NEXT:    pandn %xmm8, %xmm10
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm6[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,0,0,0]
-; SSE-NEXT:    pand %xmm15, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm10
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm5[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm12
-; SSE-NEXT:    pandn %xmm8, %xmm12
-; SSE-NEXT:    pand %xmm7, %xmm10
-; SSE-NEXT:    por %xmm10, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm10
-; SSE-NEXT:    pandn %xmm12, %xmm10
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm10
-; SSE-NEXT:    movdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm10, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm11[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    movdqa %xmm3, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    pand %xmm10, %xmm12
-; SSE-NEXT:    por %xmm8, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm9[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm13, %xmm8
-; SSE-NEXT:    pandn %xmm0, %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm14, %xmm13
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    pand %xmm13, %xmm8
-; SSE-NEXT:    por %xmm8, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm0, %xmm8
-; SSE-NEXT:    pandn %xmm14, %xmm8
-; SSE-NEXT:    pand %xmm0, %xmm12
-; SSE-NEXT:    por %xmm12, %xmm8
-; SSE-NEXT:    movdqa %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm11[2,2,3,3]
-; SSE-NEXT:    movdqa %xmm7, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm1[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm12, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm13, %xmm14
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm14
-; SSE-NEXT:    pand %xmm8, %xmm14
-; SSE-NEXT:    por %xmm12, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm2, %xmm12
-; SSE-NEXT:    pandn %xmm14, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm9[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
-; SSE-NEXT:    movdqa %xmm3, %xmm14
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm14
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm3, %xmm9
-; SSE-NEXT:    pandn %xmm0, %xmm9
-; SSE-NEXT:    pand %xmm3, %xmm14
-; SSE-NEXT:    por %xmm14, %xmm9
-; SSE-NEXT:    pand %xmm2, %xmm9
-; SSE-NEXT:    por %xmm12, %xmm9
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 # 16-byte Reload
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm11[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm3, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm9[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm0
+; SSE-NEXT:    pandn %xmm7, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm13[0,0,2,3]
+; SSE-NEXT:    pand %xmm11, %xmm7
+; SSE-NEXT:    por %xmm7, %xmm0
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm7
+; SSE-NEXT:    pandn %xmm0, %xmm7
+; SSE-NEXT:    pand %xmm11, %xmm6
+; SSE-NEXT:    por %xmm6, %xmm7
+; SSE-NEXT:    movdqa %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm10[8],xmm1[9],xmm10[9],xmm1[10],xmm10[10],xmm1[11],xmm10[11],xmm1[12],xmm10[12],xmm1[13],xmm10[13],xmm1[14],xmm10[14],xmm1[15],xmm10[15]
+; SSE-NEXT:    psrlq $16, %xmm3
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm3[0,1,2,2]
+; SSE-NEXT:    movdqa %xmm15, %xmm3
+; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm10[8],xmm4[9],xmm10[9],xmm4[10],xmm10[10],xmm4[11],xmm10[11],xmm4[12],xmm10[12],xmm4[13],xmm10[13],xmm4[14],xmm10[14],xmm4[15],xmm10[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm4[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm0
+; SSE-NEXT:    pand %xmm15, %xmm0
+; SSE-NEXT:    por %xmm3, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm5 = xmm5[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm5, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm9[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm6, %xmm5
+; SSE-NEXT:    pandn %xmm3, %xmm5
+; SSE-NEXT:    pand %xmm6, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm3, %xmm2
+; SSE-NEXT:    pandn %xmm5, %xmm2
 ; SSE-NEXT:    pand %xmm3, %xmm0
-; SSE-NEXT:    por %xmm12, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm4[0,1,2,3,5,5,5,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm14
-; SSE-NEXT:    pandn %xmm12, %xmm14
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm12 = xmm8[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm12
-; SSE-NEXT:    por %xmm14, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
-; SSE-NEXT:    pand %xmm2, %xmm12
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm4, %xmm0
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm4, %xmm0
+; SSE-NEXT:    movdqa %xmm15, %xmm2
 ; SSE-NEXT:    pandn %xmm0, %xmm2
-; SSE-NEXT:    por %xmm12, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm0, %xmm1
+; SSE-NEXT:    pand %xmm15, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm5, %xmm0
+; SSE-NEXT:    pandn %xmm1, %xmm0
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm10[8],xmm8[9],xmm10[9],xmm8[10],xmm10[10],xmm8[11],xmm10[11],xmm8[12],xmm10[12],xmm8[13],xmm10[13],xmm8[14],xmm10[14],xmm8[15],xmm10[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm8[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm8[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm9, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    pand %xmm5, %xmm3
+; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm10[1,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm13, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,4,4,6,5]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]
-; SSE-NEXT:    pand %xmm13, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm5[0,1,2,3,4,5,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm14
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    pand %xmm15, %xmm12
-; SSE-NEXT:    por %xmm12, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm2
-; SSE-NEXT:    pandn %xmm14, %xmm0
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm11[0,1,2,3,6,5,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; SSE-NEXT:    movdqa %xmm15, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm8[0,1,2,3,4,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,3,3]
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm14
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm4[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm14
-; SSE-NEXT:    pandn %xmm12, %xmm0
-; SSE-NEXT:    por %xmm14, %xmm0
+; SSE-NEXT:    movdqa %xmm10, %xmm0
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm13[8],xmm0[9],xmm13[9],xmm0[10],xmm13[10],xmm0[11],xmm13[11],xmm0[12],xmm13[12],xmm0[13],xmm13[13],xmm0[14],xmm13[14],xmm0[15],xmm13[15]
 ; SSE-NEXT:    movdqa %xmm0, %xmm1
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm6[0,1,2,3,7,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm10[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,2,2,3]
-; SSE-NEXT:    pand %xmm7, %xmm0
-; SSE-NEXT:    por %xmm12, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm5[0,1,2,3,6,7,7,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[2,1,3,2]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm14
-; SSE-NEXT:    pandn %xmm12, %xmm14
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm14
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    pandn %xmm14, %xmm0
+; SSE-NEXT:    pxor %xmm5, %xmm5
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1],xmm1[2],xmm5[2],xmm1[3],xmm5[3],xmm1[4],xmm5[4],xmm1[5],xmm5[5],xmm1[6],xmm5[6],xmm1[7],xmm5[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm14, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm12[8],xmm1[9],xmm12[9],xmm1[10],xmm12[10],xmm1[11],xmm12[11],xmm1[12],xmm12[12],xmm1[13],xmm12[13],xmm1[14],xmm12[14],xmm1[15],xmm12[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm5[0],xmm1[1],xmm5[1],xmm1[2],xmm5[2],xmm1[3],xmm5[3],xmm1[4],xmm5[4],xmm1[5],xmm5[5],xmm1[6],xmm5[6],xmm1[7],xmm5[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm3, %xmm1
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; SSE-NEXT:    movdqa (%rsp), %xmm9 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm9[8],xmm2[9],xmm9[9],xmm2[10],xmm9[10],xmm2[11],xmm9[11],xmm2[12],xmm9[12],xmm2[13],xmm9[13],xmm2[14],xmm9[14],xmm2[15],xmm9[15]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm3 = xmm3[0],xmm5[0],xmm3[1],xmm5[1],xmm3[2],xmm5[2],xmm3[3],xmm5[3],xmm3[4],xmm5[4],xmm3[5],xmm5[5],xmm3[6],xmm5[6],xmm3[7],xmm5[7]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,xmm4[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm3[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm6, %xmm4
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm11[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,2,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm8, %xmm7
+; SSE-NEXT:    pandn %xmm6, %xmm7
+; SSE-NEXT:    pand %xmm8, %xmm4
+; SSE-NEXT:    por %xmm4, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm6
+; SSE-NEXT:    pandn %xmm7, %xmm6
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm6
+; SSE-NEXT:    movdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm12, %xmm6
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm14[8],xmm6[9],xmm14[9],xmm6[10],xmm14[10],xmm6[11],xmm14[11],xmm6[12],xmm14[12],xmm6[13],xmm14[13],xmm6[14],xmm14[14],xmm6[15],xmm14[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm5[8],xmm6[9],xmm5[9],xmm6[10],xmm5[10],xmm6[11],xmm5[11],xmm6[12],xmm5[12],xmm6[13],xmm5[13],xmm6[14],xmm5[14],xmm6[15],xmm5[15]
+; SSE-NEXT:    movdqa %xmm6, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm6, %xmm4
+; SSE-NEXT:    movdqa %xmm6, %xmm7
+; SSE-NEXT:    packuswb %xmm1, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm8, %xmm4
+; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    movdqa %xmm13, %xmm6
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm10[8],xmm6[9],xmm10[9],xmm6[10],xmm10[10],xmm6[11],xmm10[11],xmm6[12],xmm10[12],xmm6[13],xmm10[13],xmm6[14],xmm10[14],xmm6[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm6, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm5[8],xmm1[9],xmm5[9],xmm1[10],xmm5[10],xmm1[11],xmm5[11],xmm1[12],xmm5[12],xmm1[13],xmm5[13],xmm1[14],xmm5[14],xmm1[15],xmm5[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3],xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[3,1],xmm1[0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm6 = xmm6[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm6
+; SSE-NEXT:    pand %xmm8, %xmm6
+; SSE-NEXT:    por %xmm4, %xmm6
+; SSE-NEXT:    psrldq {{.*#+}} xmm3 = xmm3[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm5[8],xmm2[9],xmm5[9],xmm2[10],xmm5[10],xmm2[11],xmm5[11],xmm2[12],xmm5[12],xmm2[13],xmm5[13],xmm2[14],xmm5[14],xmm2[15],xmm5[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm3
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm11[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm4, %xmm2
+; SSE-NEXT:    pandn %xmm1, %xmm2
+; SSE-NEXT:    pand %xmm4, %xmm3
+; SSE-NEXT:    por %xmm3, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm1, %xmm3
+; SSE-NEXT:    pandn %xmm2, %xmm3
+; SSE-NEXT:    pand %xmm1, %xmm6
+; SSE-NEXT:    por %xmm6, %xmm3
+; SSE-NEXT:    movdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm0 = xmm0[8],xmm5[8],xmm0[9],xmm5[9],xmm0[10],xmm5[10],xmm0[11],xmm5[11],xmm0[12],xmm5[12],xmm0[13],xmm5[13],xmm0[14],xmm5[14],xmm0[15],xmm5[15]
+; SSE-NEXT:    movdqa %xmm0, %xmm1
+; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm8, %xmm1
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm7 = xmm7[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm7, %xmm0
+; SSE-NEXT:    pand %xmm8, %xmm0
 ; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm9 = xmm9[8],mem[8],xmm9[9],mem[9],xmm9[10],mem[10],xmm9[11],mem[11],xmm9[12],mem[12],xmm9[13],mem[13],xmm9[14],mem[14],xmm9[15],mem[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm9 = xmm9[8],xmm5[8],xmm9[9],xmm5[9],xmm9[10],xmm5[10],xmm9[11],xmm5[11],xmm9[12],xmm5[12],xmm9[13],xmm5[13],xmm9[14],xmm5[14],xmm9[15],xmm5[15]
+; SSE-NEXT:    pxor %xmm10, %xmm10
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm9[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm9[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm11[0,1,2,3,6,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    movdqa %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm3, %xmm2
+; SSE-NEXT:    pand %xmm1, %xmm0
+; SSE-NEXT:    por %xmm0, %xmm2
+; SSE-NEXT:    movdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm5 = xmm5[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm5[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    movdqa %xmm8, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm6[0,2,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,0]
-; SSE-NEXT:    movdqa %xmm2, %xmm14
-; SSE-NEXT:    pandn %xmm0, %xmm14
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[0,0,2,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm14
+; SSE-NEXT:    movdqa %xmm5, %xmm1
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm1 = xmm1[0],mem[0],xmm1[1],mem[1],xmm1[2],mem[2],xmm1[3],mem[3],xmm1[4],mem[4],xmm1[5],mem[5],xmm1[6],mem[6],xmm1[7],mem[7]
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    psrlq $16, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm3, %xmm4
 ; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm14
-; SSE-NEXT:    pandn %xmm12, %xmm0
-; SSE-NEXT:    por %xmm14, %xmm0
-; SSE-NEXT:    movdqa %xmm0, %xmm11
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm12 = xmm1[0,1,1,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm12 = xmm12[0,0,2,1]
-; SSE-NEXT:    pandn %xmm12, %xmm15
-; SSE-NEXT:    por %xmm0, %xmm15
+; SSE-NEXT:    movdqa %xmm0, %xmm3
+; SSE-NEXT:    pandn %xmm4, %xmm3
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm7 = xmm7[0],mem[0],xmm7[1],mem[1],xmm7[2],mem[2],xmm7[3],mem[3],xmm7[4],mem[4],xmm7[5],mem[5],xmm7[6],mem[6],xmm7[7],mem[7]
+; SSE-NEXT:    movdqa %xmm7, %xmm4
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm10[0],xmm4[1],xmm10[1],xmm4[2],xmm10[2],xmm4[3],xmm10[3],xmm4[4],xmm10[4],xmm4[5],xmm10[5],xmm4[6],xmm10[6],xmm4[7],xmm10[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm6 = xmm4[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm6 = xmm6[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm4[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm6, %xmm8
+; SSE-NEXT:    pand %xmm0, %xmm8
+; SSE-NEXT:    por %xmm3, %xmm8
 ; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm14[0,0,0,0,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa %xmm7, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pand %xmm7, %xmm15
-; SSE-NEXT:    por %xmm15, %xmm12
-; SSE-NEXT:    movdqa {{.*#+}} xmm10 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; SSE-NEXT:    pand %xmm10, %xmm11
-; SSE-NEXT:    pandn %xmm12, %xmm10
-; SSE-NEXT:    por %xmm11, %xmm10
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm6[0,1,2,2]
-; SSE-NEXT:    movdqa %xmm7, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[1,1,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm0
+; SSE-NEXT:    movdqa %xmm14, %xmm12
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm12 = xmm12[0],mem[0],xmm12[1],mem[1],xmm12[2],mem[2],xmm12[3],mem[3],xmm12[4],mem[4],xmm12[5],mem[5],xmm12[6],mem[6],xmm12[7],mem[7]
+; SSE-NEXT:    movdqa %xmm12, %xmm6
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm6 = xmm6[0],xmm10[0],xmm6[1],xmm10[1],xmm6[2],xmm10[2],xmm6[3],xmm10[3],xmm6[4],xmm10[4],xmm6[5],xmm10[5],xmm6[6],xmm10[6],xmm6[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm6, %xmm3
+; SSE-NEXT:    pslldq {{.*#+}} xmm3 = zero,zero,xmm3[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm6, %xmm13
+; SSE-NEXT:    packuswb %xmm3, %xmm13
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm9[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    movdqa %xmm11, %xmm0
+; SSE-NEXT:    pandn %xmm3, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm13[0,0,2,3]
+; SSE-NEXT:    pand %xmm11, %xmm3
 ; SSE-NEXT:    por %xmm3, %xmm0
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    movdqa %xmm11, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[1,1,2,1]
-; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,4]
-; SSE-NEXT:    movdqa {{.*#+}} xmm15 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; SSE-NEXT:    movdqa %xmm15, %xmm12
-; SSE-NEXT:    pandn %xmm0, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[1,1,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm3, %xmm11
+; SSE-NEXT:    pandn %xmm0, %xmm11
+; SSE-NEXT:    pand %xmm3, %xmm8
+; SSE-NEXT:    por %xmm8, %xmm11
+; SSE-NEXT:    movdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm7 = xmm7[8],xmm10[8],xmm7[9],xmm10[9],xmm7[10],xmm10[10],xmm7[11],xmm10[11],xmm7[12],xmm10[12],xmm7[13],xmm10[13],xmm7[14],xmm10[14],xmm7[15],xmm10[15]
+; SSE-NEXT:    psrlq $16, %xmm4
+; SSE-NEXT:    packuswb %xmm7, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm4[0,1,2,2]
+; SSE-NEXT:    movdqa %xmm15, %xmm4
+; SSE-NEXT:    pandn %xmm0, %xmm4
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm10[8],xmm1[9],xmm10[9],xmm1[10],xmm10[10],xmm1[11],xmm10[11],xmm1[12],xmm10[12],xmm1[13],xmm10[13],xmm1[14],xmm10[14],xmm1[15],xmm10[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm0 = xmm2[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,3],xmm1[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm0
 ; SSE-NEXT:    pand %xmm15, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm12
-; SSE-NEXT:    pand %xmm11, %xmm12
-; SSE-NEXT:    por %xmm3, %xmm12
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm1[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,1,1,3]
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    movdqa %xmm8, %xmm15
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm14[1,1,1,1,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    movdqa {{.*#+}} xmm8 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
-; SSE-NEXT:    movdqa %xmm8, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm3
-; SSE-NEXT:    pand %xmm8, %xmm15
-; SSE-NEXT:    por %xmm15, %xmm3
-; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; SSE-NEXT:    pand %xmm11, %xmm12
-; SSE-NEXT:    pandn %xmm3, %xmm11
-; SSE-NEXT:    por %xmm12, %xmm11
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm5[2,2,3,3]
-; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm4[2,1,3,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,0,2,1]
-; SSE-NEXT:    pand %xmm7, %xmm3
-; SSE-NEXT:    pandn %xmm0, %xmm7
-; SSE-NEXT:    por %xmm3, %xmm7
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm13[0,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,1,3]
-; SSE-NEXT:    pand %xmm8, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm6[0,1,2,3,5,6,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,2,2,2]
-; SSE-NEXT:    pandn %xmm3, %xmm8
-; SSE-NEXT:    por %xmm0, %xmm8
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm8
-; SSE-NEXT:    pandn %xmm7, %xmm0
-; SSE-NEXT:    por %xmm8, %xmm0
-; SSE-NEXT:    movdqa %xmm0, %xmm4
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm2[2,2,2,3,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,2,1]
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    por %xmm4, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm6[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm6 = xmm6[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm6, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm9[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    movdqa %xmm9, %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm3, %xmm8
+; SSE-NEXT:    pandn %xmm4, %xmm8
+; SSE-NEXT:    pand %xmm3, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm8
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm2, %xmm3
+; SSE-NEXT:    pandn %xmm8, %xmm3
 ; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    pshufhw {{.*#+}} xmm3 = xmm1[0,1,2,3,4,6,5,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[2,1,3,2]
-; SSE-NEXT:    pandn %xmm3, %xmm2
-; SSE-NEXT:    por %xmm0, %xmm2
-; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm3, (%rsp) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm1, %xmm0
+; SSE-NEXT:    pslldq {{.*#+}} xmm0 = zero,zero,zero,zero,zero,zero,xmm0[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm15, %xmm1
+; SSE-NEXT:    pandn %xmm0, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm7[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm7 = zero,zero,xmm7[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm0, %xmm7
+; SSE-NEXT:    pand %xmm15, %xmm7
+; SSE-NEXT:    por %xmm1, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    movdqa %xmm3, %xmm1
+; SSE-NEXT:    pandn %xmm7, %xmm1
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm10[8],xmm12[9],xmm10[9],xmm12[10],xmm10[10],xmm12[11],xmm10[11],xmm12[12],xmm10[12],xmm12[13],xmm10[13],xmm12[14],xmm10[14],xmm12[15],xmm10[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm12[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm12[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm0, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm11[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[0,0,0,0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm6 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    movdqa %xmm6, %xmm0
+; SSE-NEXT:    pandn %xmm4, %xmm0
+; SSE-NEXT:    pand %xmm6, %xmm2
+; SSE-NEXT:    por %xmm2, %xmm0
+; SSE-NEXT:    pand %xmm3, %xmm0
+; SSE-NEXT:    por %xmm1, %xmm0
+; SSE-NEXT:    movdqa %xmm5, %xmm8
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm9[8],xmm8[9],xmm9[9],xmm8[10],xmm9[10],xmm8[11],xmm9[11],xmm8[12],xmm9[12],xmm8[13],xmm9[13],xmm8[14],xmm9[14],xmm8[15],xmm9[15]
+; SSE-NEXT:    movdqa %xmm8, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1],xmm1[2],xmm10[2],xmm1[3],xmm10[3],xmm1[4],xmm10[4],xmm1[5],xmm10[5],xmm1[6],xmm10[6],xmm1[7],xmm10[7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    psrlq $16, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm2, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm5, %xmm2
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm6[8],xmm2[9],xmm6[9],xmm2[10],xmm6[10],xmm2[11],xmm6[11],xmm2[12],xmm6[12],xmm2[13],xmm6[13],xmm2[14],xmm6[14],xmm2[15],xmm6[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm4, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0]
 ; SSE-NEXT:    pand %xmm3, %xmm2
-; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm14[2,2,2,2,4,5,6,7]
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT:    pandn %xmm0, %xmm3
+; SSE-NEXT:    pandn %xmm1, %xmm3
 ; SSE-NEXT:    por %xmm2, %xmm3
-; SSE-NEXT:    movdqa {{.*#+}} xmm0 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
-; SSE-NEXT:    pand %xmm0, %xmm3
-; SSE-NEXT:    pandn %xmm4, %xmm0
-; SSE-NEXT:    por %xmm0, %xmm3
+; SSE-NEXT:    movdqa %xmm14, %xmm1
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm13[8],xmm1[9],xmm13[9],xmm1[10],xmm13[10],xmm1[11],xmm13[11],xmm1[12],xmm13[12],xmm1[13],xmm13[13],xmm1[14],xmm13[14],xmm1[15],xmm13[15]
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm2, %xmm4
+; SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,xmm4[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm2[2,2,3,3]
+; SSE-NEXT:    packuswb %xmm7, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm12 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
+; SSE-NEXT:    pand %xmm12, %xmm4
+; SSE-NEXT:    pshufhw {{.*#+}} xmm7 = xmm11[0,1,2,3,4,5,5,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,2,2,3]
+; SSE-NEXT:    pandn %xmm7, %xmm12
+; SSE-NEXT:    por %xmm4, %xmm12
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm3
+; SSE-NEXT:    pandn %xmm12, %xmm4
+; SSE-NEXT:    por %xmm3, %xmm4
+; SSE-NEXT:    movdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; SSE-NEXT:    movdqa %xmm6, %xmm3
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm5[8],xmm3[9],xmm5[9],xmm3[10],xmm5[10],xmm3[11],xmm5[11],xmm3[12],xmm5[12],xmm3[13],xmm5[13],xmm3[14],xmm5[14],xmm3[15],xmm5[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm3 = xmm3[8],xmm10[8],xmm3[9],xmm10[9],xmm3[10],xmm10[10],xmm3[11],xmm10[11],xmm3[12],xmm10[12],xmm3[13],xmm10[13],xmm3[14],xmm10[14],xmm3[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm3, %xmm4
+; SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,xmm4[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm3, %xmm7
+; SSE-NEXT:    movdqa %xmm3, %xmm6
+; SSE-NEXT:    packuswb %xmm4, %xmm7
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm7[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255]
+; SSE-NEXT:    movdqa %xmm5, %xmm7
+; SSE-NEXT:    pandn %xmm4, %xmm7
+; SSE-NEXT:    punpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm9 = xmm9[8],mem[8],xmm9[9],mem[9],xmm9[10],mem[10],xmm9[11],mem[11],xmm9[12],mem[12],xmm9[13],mem[13],xmm9[14],mem[14],xmm9[15],mem[15]
+; SSE-NEXT:    movdqa %xmm9, %xmm4
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm4 = xmm4[8],xmm10[8],xmm4[9],xmm10[9],xmm4[10],xmm10[10],xmm4[11],xmm10[11],xmm4[12],xmm10[12],xmm4[13],xmm10[13],xmm4[14],xmm10[14],xmm4[15],xmm10[15]
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm9 = xmm9[0],xmm10[0],xmm9[1],xmm10[1],xmm9[2],xmm10[2],xmm9[3],xmm10[3],xmm9[4],xmm10[4],xmm9[5],xmm10[5],xmm9[6],xmm10[6],xmm9[7],xmm10[7]
+; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[3,1],xmm4[0,3]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm12 = xmm9[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm4, %xmm12
+; SSE-NEXT:    pand %xmm5, %xmm12
+; SSE-NEXT:    movdqa %xmm5, %xmm4
+; SSE-NEXT:    por %xmm7, %xmm12
+; SSE-NEXT:    psrldq {{.*#+}} xmm2 = xmm2[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm1 = xmm1[8],xmm10[8],xmm1[9],xmm10[9],xmm1[10],xmm10[10],xmm1[11],xmm10[11],xmm1[12],xmm10[12],xmm1[13],xmm10[13],xmm1[14],xmm10[14],xmm1[15],xmm10[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm11[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,3]
+; SSE-NEXT:    pandn %xmm1, %xmm3
+; SSE-NEXT:    por %xmm2, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm5 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm5, %xmm12
+; SSE-NEXT:    pandn %xmm3, %xmm5
+; SSE-NEXT:    por %xmm12, %xmm5
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm8 = xmm8[8],xmm10[8],xmm8[9],xmm10[9],xmm8[10],xmm10[10],xmm8[11],xmm10[11],xmm8[12],xmm10[12],xmm8[13],xmm10[13],xmm8[14],xmm10[14],xmm8[15],xmm10[15]
+; SSE-NEXT:    movdqa %xmm8, %xmm1
+; SSE-NEXT:    psrldq {{.*#+}} xmm1 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm1, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm6[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm6 = xmm6[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm6, %xmm1
+; SSE-NEXT:    pand %xmm4, %xmm1
+; SSE-NEXT:    pandn %xmm8, %xmm4
+; SSE-NEXT:    por %xmm1, %xmm4
+; SSE-NEXT:    movdqa %xmm13, %xmm2
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm14[8],xmm2[9],xmm14[9],xmm2[10],xmm14[10],xmm2[11],xmm14[11],xmm2[12],xmm14[12],xmm2[13],xmm14[13],xmm2[14],xmm14[14],xmm2[15],xmm14[15]
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm2 = xmm2[8],xmm10[8],xmm2[9],xmm10[9],xmm2[10],xmm10[10],xmm2[11],xmm10[11],xmm2[12],xmm10[12],xmm2[13],xmm10[13],xmm2[14],xmm10[14],xmm2[15],xmm10[15]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,4,5,5,4]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; SSE-NEXT:    pand %xmm7, %xmm2
+; SSE-NEXT:    pshufhw {{.*#+}} xmm1 = xmm11[0,1,2,3,6,7,7,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[2,1,3,2]
+; SSE-NEXT:    pandn %xmm1, %xmm7
+; SSE-NEXT:    por %xmm2, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm11 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
+; SSE-NEXT:    pand %xmm11, %xmm4
+; SSE-NEXT:    pandn %xmm7, %xmm11
+; SSE-NEXT:    por %xmm4, %xmm11
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm13 = xmm13[0],mem[0],xmm13[1],mem[1],xmm13[2],mem[2],xmm13[3],mem[3],xmm13[4],mem[4],xmm13[5],mem[5],xmm13[6],mem[6],xmm13[7],mem[7]
+; SSE-NEXT:    movdqa %xmm13, %xmm1
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0],xmm10[0],xmm1[1],xmm10[1],xmm1[2],xmm10[2],xmm1[3],xmm10[3],xmm1[4],xmm10[4],xmm1[5],xmm10[5],xmm1[6],xmm10[6],xmm1[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm1, %xmm2
+; SSE-NEXT:    psrlq $16, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[0,0,1,1]
+; SSE-NEXT:    packuswb %xmm2, %xmm4
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Reload
+; SSE-NEXT:    punpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
+; SSE-NEXT:    # xmm12 = xmm12[0],mem[0],xmm12[1],mem[1],xmm12[2],mem[2],xmm12[3],mem[3],xmm12[4],mem[4],xmm12[5],mem[5],xmm12[6],mem[6],xmm12[7],mem[7]
+; SSE-NEXT:    movdqa %xmm12, %xmm2
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm10[0],xmm2[1],xmm10[1],xmm2[2],xmm10[2],xmm2[3],xmm10[3],xmm2[4],xmm10[4],xmm2[5],xmm10[5],xmm2[6],xmm10[6],xmm2[7],xmm10[7]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm7 = xmm2[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm7[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm2[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm7, %xmm8
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm3, %xmm8
+; SSE-NEXT:    pandn %xmm4, %xmm3
+; SSE-NEXT:    por %xmm8, %xmm3
+; SSE-NEXT:    movdqa %xmm3, %xmm9
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm6, %xmm4
+; SSE-NEXT:    punpcklbw {{.*#+}} xmm4 = xmm4[0],xmm10[0],xmm4[1],xmm10[1],xmm4[2],xmm10[2],xmm4[3],xmm10[3],xmm4[4],xmm10[4],xmm4[5],xmm10[5],xmm4[6],xmm10[6],xmm4[7],xmm10[7]
+; SSE-NEXT:    movdqa %xmm4, %xmm7
+; SSE-NEXT:    pslldq {{.*#+}} xmm7 = zero,zero,xmm7[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    movdqa %xmm4, %xmm8
+; SSE-NEXT:    packuswb %xmm7, %xmm8
+; SSE-NEXT:    pshufd {{.*#+}} xmm7 = xmm8[0,0,2,3]
+; SSE-NEXT:    movdqa {{.*#+}} xmm14 = [255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255]
+; SSE-NEXT:    pand %xmm14, %xmm7
+; SSE-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; SSE-NEXT:    pshuflw {{.*#+}} xmm8 = xmm3[0,0,0,0,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm8 = xmm8[0,0,0,0]
+; SSE-NEXT:    pandn %xmm8, %xmm14
+; SSE-NEXT:    por %xmm7, %xmm14
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; SSE-NEXT:    pand %xmm7, %xmm9
+; SSE-NEXT:    pandn %xmm14, %xmm7
+; SSE-NEXT:    por %xmm9, %xmm7
+; SSE-NEXT:    movdqa %xmm7, %xmm9
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm12 = xmm12[8],xmm10[8],xmm12[9],xmm10[9],xmm12[10],xmm10[10],xmm12[11],xmm10[11],xmm12[12],xmm10[12],xmm12[13],xmm10[13],xmm12[14],xmm10[14],xmm12[15],xmm10[15]
+; SSE-NEXT:    psrlq $16, %xmm2
+; SSE-NEXT:    packuswb %xmm12, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,2]
+; SSE-NEXT:    movdqa %xmm15, %xmm7
+; SSE-NEXT:    pandn %xmm2, %xmm7
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm13 = xmm13[8],xmm10[8],xmm13[9],xmm10[9],xmm13[10],xmm10[10],xmm13[11],xmm10[11],xmm13[12],xmm10[12],xmm13[13],xmm10[13],xmm13[14],xmm10[14],xmm13[15],xmm10[15]
+; SSE-NEXT:    pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,4,5,6,6]
+; SSE-NEXT:    shufps {{.*#+}} xmm1 = xmm1[0,3],xmm13[0,0]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[2,1,2,3]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    pand %xmm15, %xmm2
+; SSE-NEXT:    por %xmm7, %xmm2
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[2,2,3,3]
+; SSE-NEXT:    psrldq {{.*#+}} xmm4 = xmm4[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; SSE-NEXT:    packuswb %xmm4, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm7, %xmm1
+; SSE-NEXT:    movdqa %xmm3, %xmm8
+; SSE-NEXT:    pshuflw {{.*#+}} xmm4 = xmm3[1,1,1,1,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm4[0,0,0,0]
+; SSE-NEXT:    pandn %xmm4, %xmm7
+; SSE-NEXT:    por %xmm1, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; SSE-NEXT:    pand %xmm1, %xmm2
+; SSE-NEXT:    pandn %xmm7, %xmm1
+; SSE-NEXT:    por %xmm2, %xmm1
+; SSE-NEXT:    movdqa %xmm1, %xmm3
+; SSE-NEXT:    movdqa %xmm13, %xmm1
+; SSE-NEXT:    pslldq {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,xmm1[0,1,2,3,4,5,6,7,8,9]
+; SSE-NEXT:    packuswb %xmm13, %xmm1
+; SSE-NEXT:    movdqa %xmm12, %xmm4
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm12[2,2,3,3]
+; SSE-NEXT:    pslldq {{.*#+}} xmm4 = zero,zero,xmm4[0,1,2,3,4,5,6,7,8,9,10,11,12,13]
+; SSE-NEXT:    packuswb %xmm2, %xmm4
+; SSE-NEXT:    pand %xmm15, %xmm4
+; SSE-NEXT:    pandn %xmm1, %xmm15
+; SSE-NEXT:    por %xmm4, %xmm15
+; SSE-NEXT:    punpckhbw {{.*#+}} xmm6 = xmm6[8],xmm10[8],xmm6[9],xmm10[9],xmm6[10],xmm10[10],xmm6[11],xmm10[11],xmm6[12],xmm10[12],xmm6[13],xmm10[13],xmm6[14],xmm10[14],xmm6[15],xmm10[15]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm6[3,3,3,3,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,2]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm6[0,1,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,1,2,1]
+; SSE-NEXT:    packuswb %xmm1, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; SSE-NEXT:    pand %xmm4, %xmm2
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm8[2,2,2,2,4,5,6,7]
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; SSE-NEXT:    pandn %xmm1, %xmm4
+; SSE-NEXT:    por %xmm2, %xmm4
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [255,255,255,0,0,0,0,255,255,255,0,0,0,0,255,255]
+; SSE-NEXT:    pand %xmm1, %xmm4
+; SSE-NEXT:    pandn %xmm15, %xmm1
+; SSE-NEXT:    por %xmm1, %xmm4
 ; SSE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; SSE-NEXT:    movdqa %xmm3, 368(%rax)
-; SSE-NEXT:    movdqa %xmm11, 352(%rax)
-; SSE-NEXT:    movdqa %xmm10, 336(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 320(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 288(%rax)
-; SSE-NEXT:    movdqa %xmm9, 256(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movdqa %xmm4, 368(%rax)
+; SSE-NEXT:    movdqa %xmm3, 352(%rax)
+; SSE-NEXT:    movdqa %xmm9, 336(%rax)
+; SSE-NEXT:    movdqa %xmm11, 320(%rax)
+; SSE-NEXT:    movdqa %xmm5, 304(%rax)
+; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; SSE-NEXT:    movaps %xmm1, 288(%rax)
+; SSE-NEXT:    movdqa %xmm0, 256(%rax)
+; SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 240(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 224(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 208(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movaps %xmm0, 192(%rax)
+; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 176(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 144(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 128(%rax)
-; SSE-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 112(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 96(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; SSE-NEXT:    movaps %xmm0, 80(%rax)
+; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 64(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 32(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 16(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, (%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 304(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 192(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; SSE-NEXT:    movaps %xmm0, 80(%rax)
-; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 272(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 160(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 48(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 432(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 400(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 416(%rax)
 ; SSE-NEXT:    movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; SSE-NEXT:    movaps %xmm0, 384(%rax)
-; SSE-NEXT:    addq $648, %rsp # imm = 0x288
+; SSE-NEXT:    addq $472, %rsp # imm = 0x1D8
 ; SSE-NEXT:    retq
 ;
 ; AVX1-ONLY-LABEL: store_i8_stride7_vf64:
 ; AVX1-ONLY:       # %bb.0:
 ; AVX1-ONLY-NEXT:    subq $632, %rsp # imm = 0x278
 ; AVX1-ONLY-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX1-ONLY-NEXT:    vmovdqa 16(%rax), %xmm6
 ; AVX1-ONLY-NEXT:    vpshufb {{.*#+}} xmm0 = zero,xmm6[13,u,u,u,u],zero,zero,xmm6[14,u,u,u,u],zero,zero,xmm6[15]

 ; AVX1-ONLY-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
 ; AVX1-ONLY-NEXT:    vmovaps %xmm0, 400(%rax)
 ; AVX1-ONLY-NEXT:    addq $632, %rsp # imm = 0x278
 ; AVX1-ONLY-NEXT:    vzeroupper
 ; AVX1-ONLY-NEXT:    retq
 ;
 ; AVX2-SLOW-LABEL: store_i8_stride7_vf64:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    subq $808, %rsp # imm = 0x328
+; AVX2-SLOW-NEXT:    subq $744, %rsp # imm = 0x2E8
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm6
-; AVX2-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm7
-; AVX2-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%r8), %ymm5
-; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%r9), %ymm4
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm2
+; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm4
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa 32(%rax), %ymm3
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 32(%r8), %ymm5
+; AVX2-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 32(%r9), %ymm6
+; AVX2-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rax), %ymm7
+; AVX2-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[23],zero,ymm0[27,20,21,26],zero,ymm0[24],zero,ymm0[26,27,26,27],zero,ymm0[25]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero,ymm1[27],zero
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero,zero
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero,ymm6[27]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm4[25],zero,ymm4[23],zero,zero,zero,zero,ymm4[26],zero,ymm4[24],zero,zero,zero,zero,ymm4[27]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,24,25,26,27,24,25,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,25,24,23,u,u,u,u,u,u,u,u,u]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,25,24,23,u,u,u,u,u,u,u,u,u]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[27],zero,ymm0[27,28,29,30],zero,ymm0[28],zero,ymm0[26,27,30,31],zero,ymm0[29]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm1
+; AVX2-SLOW-NEXT:    vmovdqa (%r8), %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm1[27],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqa (%rax), %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-SLOW-NEXT:    vmovdqa (%r9), %ymm2
+; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15],ymm2[24],ymm1[24],ymm2[25],ymm1[25],ymm2[26],ymm1[26],ymm2[27],ymm1[27],ymm2[28],ymm1[28],ymm2[29],ymm1[29],ymm2[30],ymm1[30],ymm2[31],ymm1[31]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm1 = ymm1[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,2,3,3,4,6,7,7]
 ; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %ymm2
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
 ; AVX2-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [255,0,0,0,0,255,0,255,0,0,0,0,255,0,255,0,255,0,0,0,0,255,0,255,0,0,0,0,255,0,255,0]
 ; AVX2-SLOW-NEXT:    # ymm3 = mem[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm2, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm2
-; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm2[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[2,2,3,3,6,6,7,7]
-; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm3
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %ymm3
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,14,u,12,u,u,u,u,15,u,13,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
-; AVX2-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [0,0,255,0,255,0,0,0,0,255,0,255,0,0,0,0,0,0,255,0,255,0,0,0,0,255,0,255,0,0,0,0]
-; AVX2-SLOW-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-SLOW-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %ymm2
+; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15],ymm3[24],ymm2[24],ymm3[25],ymm2[25],ymm3[26],ymm2[26],ymm3[27],ymm2[27],ymm3[28],ymm2[28],ymm3[29],ymm2[29],ymm3[30],ymm2[30],ymm3[31],ymm2[31]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm12
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm0
 ; AVX2-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill

 ; AVX2-SLOW-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm11
 ; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm5
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX2-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
-; AVX2-SLOW-NEXT:    vmovdqa %xmm5, (%rsp) # 16-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %xmm3
 ; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm6
 ; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3],xmm6[4],xmm3[4],xmm6[5],xmm3[5],xmm6[6],xmm3[6],xmm6[7],xmm3[7]
 ; AVX2-SLOW-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-SLOW-NEXT:    vpshufb %xmm8, %xmm1, %xmm1
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm8
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%r9), %xmm1
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%r8), %xmm0
-; AVX2-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa %xmm0, (%rsp) # 16-byte Spill
 ; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; AVX2-SLOW-NEXT:    vmovdqa %xmm1, %xmm2
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm9 = <u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u>
 ; AVX2-SLOW-NEXT:    vpshufb %xmm9, %xmm0, %xmm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vmovdqa 32(%rax), %xmm1
 ; AVX2-SLOW-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,1,0,0,4,5,6,7]

 ; AVX2-SLOW-NEXT:    vpor %xmm0, %xmm11, %xmm0
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm15, %ymm10, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm10 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
 ; AVX2-SLOW-NEXT:    vmovdqa %xmm2, %xmm5
 ; AVX2-SLOW-NEXT:    vpshufb %xmm10, %xmm2, %xmm11
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm12 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
-; AVX2-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; AVX2-SLOW-NEXT:    vmovdqa (%rsp), %xmm2 # 16-byte Reload
 ; AVX2-SLOW-NEXT:    vpshufb %xmm12, %xmm2, %xmm13
 ; AVX2-SLOW-NEXT:    vpor %xmm11, %xmm13, %xmm11
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm13 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
 ; AVX2-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
 ; AVX2-SLOW-NEXT:    vpshufb %xmm13, %xmm3, %xmm15
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,0,1,0]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u>

 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm11 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
 ; AVX2-SLOW-NEXT:    vpshufb %xmm11, %xmm8, %xmm8
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm0, %ymm8, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
 ; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm1[8],xmm6[8],xmm1[9],xmm6[9],xmm1[10],xmm6[10],xmm1[11],xmm6[11],xmm1[12],xmm6[12],xmm1[13],xmm6[13],xmm1[14],xmm6[14],xmm1[15],xmm6[15]
 ; AVX2-SLOW-NEXT:    vpshufb %xmm10, %xmm6, %xmm6
-; AVX2-SLOW-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX2-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
 ; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm7 = xmm1[8],xmm7[8],xmm1[9],xmm7[9],xmm1[10],xmm7[10],xmm1[11],xmm7[11],xmm1[12],xmm7[12],xmm1[13],xmm7[13],xmm1[14],xmm7[14],xmm1[15],xmm7[15]
 ; AVX2-SLOW-NEXT:    vpshufb %xmm11, %xmm7, %xmm7
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm6, %ymm7, %ymm6
 ; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm5[8],xmm2[9],xmm5[9],xmm2[10],xmm5[10],xmm2[11],xmm5[11],xmm2[12],xmm5[12],xmm2[13],xmm5[13],xmm2[14],xmm5[14],xmm2[15],xmm5[15]
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} xmm8 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
 ; AVX2-SLOW-NEXT:    vpshufb %xmm8, %xmm4, %xmm4

 ; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm1, %ymm2, %ymm1
 ; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm0, %ymm3, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm6, %ymm1, %ymm0
 ; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22>
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm0, %ymm1, %ymm1
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm2, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm1, %ymm3, %ymm1
-; AVX2-SLOW-NEXT:    vpshuflw $150, {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm3 = mem[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,1,1,3,4,5,5,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,3,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm0, %ymm14, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm10[23],zero,ymm10[27,20,21,26],zero,ymm10[24],zero,ymm10[26,27,26,27],zero,ymm10[25]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm2, %ymm8, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm11[23],zero,zero,zero,zero,ymm11[26],zero,ymm11[24],zero,zero,zero,zero,ymm11[27],zero
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vpor %ymm0, %ymm2, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm15[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,1,3,4,5,5,7]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,3,2]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
-; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm11[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,0,1,1,4,4,5,5]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,5,u,3,u,u,u,u,6,u,4,u,u,u,u,u,u,5,u,3,u,u,u,u,6,u,4,u,u>
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm13, %ymm4
-; AVX2-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [0,0,0,255,0,255,0,0,0,0,255,0,255,0,0,0,0,0,0,255,0,255,0,0,0,0,255,0,255,0,0,0]
-; AVX2-SLOW-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm2, %ymm4, %ymm2
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm7, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm4 = ymm6[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm4 = ymm4[0,0,1,1,4,4,5,5]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm4, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20>
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm4, %ymm10, %ymm5
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm12[18],zero,zero,zero,zero,ymm12[21],zero,ymm12[19],zero,zero,zero,zero,ymm12[22],zero,ymm12[20],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm5, %ymm9, %ymm5
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm12[25],zero,ymm12[23],zero,zero,zero,zero,ymm12[26],zero,ymm12[24],zero,zero,zero,zero
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm5, %ymm2, %ymm2
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm4, %ymm1, %ymm4
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[18],zero,zero,zero,zero,ymm0[21],zero,ymm0[19],zero,zero,zero,zero,ymm0[22],zero,ymm0[20],zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm9[25],zero,ymm9[23],zero,zero,zero,zero,ymm9[26],zero,ymm9[24],zero,zero,zero,zero,ymm9[27]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm4, %ymm3, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm5, %ymm2
-; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm3, %ymm2, %ymm2
-; AVX2-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm6[23],zero,ymm6[27,20,21,26],zero,ymm6[24],zero,ymm6[26,27,26,27],zero,ymm6[25]
+; AVX2-SLOW-NEXT:    vpor %ymm2, %ymm3, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm14[25],zero,ymm14[23],zero,zero,zero,zero,ymm14[26],zero,ymm14[24],zero,zero
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero,zero,ymm7[27],zero
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm15[25],zero,ymm15[23],zero,zero,zero,zero,ymm15[26],zero,ymm15[24],zero,zero,zero
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
 ; AVX2-SLOW-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm1[25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero
-; AVX2-SLOW-NEXT:    vmovdqa %ymm1, %ymm6
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[25],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero,ymm0[27]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm3, %ymm4, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm2, %ymm3, %ymm2
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm14[25],zero,ymm14[23],zero,zero,zero,zero,ymm14[26],zero,ymm14[24],zero,zero
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm8[25],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero,zero
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpor %ymm3, %ymm4, %ymm3
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm1
-; AVX2-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
-; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm13, %ymm4
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,128,14,128,128,128,128,1,128,15,128,128,128,128,2,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128,18,128]
-; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm11, %ymm8
-; AVX2-SLOW-NEXT:    vpor %ymm4, %ymm8, %ymm4
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm10, %ymm9
-; AVX2-SLOW-NEXT:    vmovdqa %ymm10, %ymm15
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,128,0,128,14,128,128,128,128,1,128,15,128,128,128,128,128,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128]
-; AVX2-SLOW-NEXT:    vpshufb %ymm10, %ymm12, %ymm11
-; AVX2-SLOW-NEXT:    vmovdqa %ymm12, %ymm14
-; AVX2-SLOW-NEXT:    vpor %ymm9, %ymm11, %ymm9
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm4, %ymm9, %ymm4
-; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm7, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm1, %ymm5
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm4, %ymm2, %ymm3, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb %ymm2, %ymm0, %ymm3
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,128,14,128,128,128,128,1,128,15,128,128,128,128,2,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128,18,128]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb %ymm4, %ymm0, %ymm5
 ; AVX2-SLOW-NEXT:    vpor %ymm3, %ymm5, %ymm3
-; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm6, %ymm5
-; AVX2-SLOW-NEXT:    vpshufb %ymm10, %ymm0, %ymm8
-; AVX2-SLOW-NEXT:    vpor %ymm5, %ymm8, %ymm5
-; AVX2-SLOW-NEXT:    vpblendvb %ymm11, %ymm3, %ymm5, %ymm3
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [128,1,2,3,0,128,14,128,0,1,0,1,128,15,128,15,128,17,18,19,16,128,30,128,16,17,16,17,128,31,128,31]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm0, %ymm8
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
+; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm0, %ymm6
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [128,128,0,128,14,128,128,128,128,1,128,15,128,128,128,128,128,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128]
 ; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm9, %ymm1, %ymm10
-; AVX2-SLOW-NEXT:    vpor %ymm8, %ymm10, %ymm8
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm10, %ymm2, %ymm11
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm8, %ymm11, %ymm8
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm6, %ymm5
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm9, %ymm6, %ymm9
-; AVX2-SLOW-NEXT:    vpor %ymm5, %ymm9, %ymm5
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpshufb %ymm10, %ymm6, %ymm9
-; AVX2-SLOW-NEXT:    vpblendvb %ymm12, %ymm5, %ymm9, %ymm5
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm4, %ymm8, %ymm4
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm3, %ymm5, %ymm3
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm15[11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
-; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm8 = ymm14[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm8 = ymm8[0,2,3,3,4,6,7,7]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm5, %ymm8, %ymm5
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm13[u,u,u,14,u,12,u,u,u,u,15,u,13,u,u,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u,u,u]
-; AVX2-SLOW-NEXT:    vpshufhw $190, {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Folded Reload
-; AVX2-SLOW-NEXT:    # ymm9 = mem[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[2,2,3,3,6,6,7,7]
-; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm9, %ymm8, %ymm8
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm5, %ymm8, %ymm5
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,26,27,30,31,30,31,28,29,28,29,28,29,28,29]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u>
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm5, %ymm8, %ymm5
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,28,27,u,u,u,31,30,u,u,u,u,u,u,u,u]
-; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,2,2]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u>
+; AVX2-SLOW-NEXT:    vpshufb %ymm7, %ymm1, %ymm8
+; AVX2-SLOW-NEXT:    vpor %ymm6, %ymm8, %ymm6
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm3, %ymm6, %ymm3
+; AVX2-SLOW-NEXT:    vpshufb %ymm2, %ymm11, %ymm2
+; AVX2-SLOW-NEXT:    vpshufb %ymm4, %ymm10, %ymm4
+; AVX2-SLOW-NEXT:    vpor %ymm2, %ymm4, %ymm2
+; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm12, %ymm4
+; AVX2-SLOW-NEXT:    vpshufb %ymm7, %ymm9, %ymm5
+; AVX2-SLOW-NEXT:    vpor %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpblendvb %ymm8, %ymm2, %ymm4, %ymm4
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,1,2,3,0,128,14,128,0,1,0,1,128,15,128,15,128,17,18,19,16,128,30,128,16,17,16,17,128,31,128,31]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb %ymm2, %ymm10, %ymm5
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb %ymm6, %ymm11, %ymm7
+; AVX2-SLOW-NEXT:    vpor %ymm5, %ymm7, %ymm5
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpshufb %ymm7, %ymm12, %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm5, %ymm8, %ymm5
-; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm8 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-SLOW-NEXT:    vpshufb %ymm2, %ymm15, %ymm2
+; AVX2-SLOW-NEXT:    vpshufb %ymm6, %ymm14, %ymm6
+; AVX2-SLOW-NEXT:    vpor %ymm2, %ymm6, %ymm2
+; AVX2-SLOW-NEXT:    vpshufb %ymm7, %ymm13, %ymm6
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm2, %ymm6, %ymm6
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm3, %ymm5, %ymm2
+; AVX2-SLOW-NEXT:    vpblendvb %ymm7, %ymm4, %ymm6, %ymm3
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm0[11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u,29,u]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm1, %ymm7
+; AVX2-SLOW-NEXT:    vpshufhw {{.*#+}} ymm5 = ymm1[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm5 = ymm5[0,2,3,3,4,6,7,7]
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm5 = ymm8[8],ymm9[8],ymm8[9],ymm9[9],ymm8[10],ymm9[10],ymm8[11],ymm9[11],ymm8[12],ymm9[12],ymm8[13],ymm9[13],ymm8[14],ymm9[14],ymm8[15],ymm9[15],ymm8[24],ymm9[24],ymm8[25],ymm9[25],ymm8[26],ymm9[26],ymm8[27],ymm9[27],ymm8[28],ymm9[28],ymm8[29],ymm9[29],ymm8[30],ymm9[30],ymm8[31],ymm9[31]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm1, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,26,27,30,31,30,31,28,29,28,29,28,29,28,29]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,28,27,u,u,u,31,30,u,u,u,u,u,u,u,u]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm6, %ymm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm7[0],ymm0[0],ymm7[1],ymm0[1],ymm7[2],ymm0[2],ymm7[3],ymm0[3],ymm7[4],ymm0[4],ymm7[5],ymm0[5],ymm7[6],ymm0[6],ymm7[7],ymm0[7],ymm7[16],ymm0[16],ymm7[17],ymm0[17],ymm7[18],ymm0[18],ymm7[19],ymm0[19],ymm7[20],ymm0[20],ymm7[21],ymm0[21],ymm7[22],ymm0[22],ymm7[23],ymm0[23]
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm6 = ymm8[0],ymm9[0],ymm8[1],ymm9[1],ymm8[2],ymm9[2],ymm8[3],ymm9[3],ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[16],ymm9[16],ymm8[17],ymm9[17],ymm8[18],ymm9[18],ymm8[19],ymm9[19],ymm8[20],ymm9[20],ymm8[21],ymm9[21],ymm8[22],ymm9[22],ymm8[23],ymm9[23]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = <4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9>
+; AVX2-SLOW-NEXT:    vpshufb %ymm7, %ymm5, %ymm5
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u>
+; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm5, %ymm6, %ymm5
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm6 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm6 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[2],mem[2],ymm0[3],mem[3],ymm0[4],mem[4],ymm0[5],mem[5],ymm0[6],mem[6],ymm0[7],mem[7],ymm0[16],mem[16],ymm0[17],mem[17],ymm0[18],mem[18],ymm0[19],mem[19],ymm0[20],mem[20],ymm0[21],mem[21],ymm0[22],mem[22],ymm0[23],mem[23]
+; AVX2-SLOW-NEXT:    vpshufb %ymm7, %ymm6, %ymm6
+; AVX2-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vpunpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %ymm0, %ymm7 # 32-byte Folded Reload
+; AVX2-SLOW-NEXT:    # ymm7 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[2],mem[2],ymm0[3],mem[3],ymm0[4],mem[4],ymm0[5],mem[5],ymm0[6],mem[6],ymm0[7],mem[7],ymm0[16],mem[16],ymm0[17],mem[17],ymm0[18],mem[18],ymm0[19],mem[19],ymm0[20],mem[20],ymm0[21],mem[21],ymm0[22],mem[22],ymm0[23],mem[23]
+; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm7, %ymm7
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm6, %ymm7, %ymm6
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm7 = ymm10[0],ymm11[0],ymm10[1],ymm11[1],ymm10[2],ymm11[2],ymm10[3],ymm11[3],ymm10[4],ymm11[4],ymm10[5],ymm11[5],ymm10[6],ymm11[6],ymm10[7],ymm11[7],ymm10[16],ymm11[16],ymm10[17],ymm11[17],ymm10[18],ymm11[18],ymm10[19],ymm11[19],ymm10[20],ymm11[20],ymm10[21],ymm11[21],ymm10[22],ymm11[22],ymm10[23],ymm11[23]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13>
+; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm7, %ymm7
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm12[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,3,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm7, %ymm9, %ymm7
+; AVX2-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm9 = ymm15[0],ymm14[0],ymm15[1],ymm14[1],ymm15[2],ymm14[2],ymm15[3],ymm14[3],ymm15[4],ymm14[4],ymm15[5],ymm14[5],ymm15[6],ymm14[6],ymm15[7],ymm14[7],ymm15[16],ymm14[16],ymm15[17],ymm14[17],ymm15[18],ymm14[18],ymm15[19],ymm14[19],ymm15[20],ymm14[20],ymm15[21],ymm14[21],ymm15[22],ymm14[22],ymm15[23],ymm14[23]
+; AVX2-SLOW-NEXT:    vpshufb %ymm8, %ymm9, %ymm8
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm5, %ymm8, %ymm5
+; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} ymm9 = ymm13[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} ymm9 = ymm9[0,1,1,3,4,5,5,7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,2,3,2]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm5, %ymm7, %ymm5
+; AVX2-SLOW-NEXT:    vpblendvb %ymm9, %ymm6, %ymm8, %ymm6
 ; AVX2-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-SLOW-NEXT:    vmovdqa %ymm3, 96(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm4, 320(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, 320(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm6, 128(%rax)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm5, 352(%rax)
+; AVX2-SLOW-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 160(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm0, 128(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 224(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-SLOW-NEXT:    vmovaps %ymm0, 352(%rax)
-; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 288(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 256(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 192(%rax)
 ; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, 384(%rax)
-; AVX2-SLOW-NEXT:    vmovdqa %ymm5, 416(%rax)
-; AVX2-SLOW-NEXT:    addq $808, %rsp # imm = 0x328
+; AVX2-SLOW-NEXT:    vmovdqa %ymm4, 416(%rax)
+; AVX2-SLOW-NEXT:    addq $744, %rsp # imm = 0x2E8
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: store_i8_stride7_vf64:
 ; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    subq $648, %rsp # imm = 0x288
+; AVX2-FAST-NEXT:    subq $600, %rsp # imm = 0x258
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %ymm6
-; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %ymm7
-; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %ymm3
-; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %ymm4
-; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %ymm5
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[23],zero,ymm1[27,20,21,26],zero,ymm1[24],zero,ymm1[26,27,26,27],zero,ymm1[25]
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm0
+; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero
-; AVX2-FAST-NEXT:    vmovdqa %ymm2, %ymm9
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %ymm4
+; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %ymm2
 ; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %ymm3
+; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %ymm6
+; AVX2-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %ymm7
+; AVX2-FAST-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[23],zero,ymm0[27,20,21,26],zero,ymm0[24],zero,ymm0[26,27,26,27],zero,ymm0[25]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero,ymm1[27],zero
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero,ymm6[27]
-; AVX2-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm4[25],zero,ymm4[23],zero,zero,zero,zero,ymm4[26],zero,ymm4[24],zero,zero,zero,zero,ymm4[27]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,24,25,26,27,24,25,30,31]
-; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,25,24,23,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,25,24,23,u,u,u,u,u,u,u,u,u]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm6[17,18,19,30],zero,ymm6[28],zero,ymm6[28,29,30,31],zero,ymm6[29],zero,ymm6[31]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm7[30],zero,ymm7[28],zero,zero,zero,zero,ymm7[31],zero,ymm7[29],zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm9[30],zero,ymm9[28],zero,zero,zero,zero,ymm9[31],zero,ymm9[29],zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm8[30],zero,ymm8[28],zero,zero,zero,zero,ymm8[31],zero,ymm8[29],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,26,27,30,31,30,31,28,29,28,29,28,29,28,29]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,28,27,u,u,u,31,30,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %xmm13
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm13[0],xmm0[1],xmm13[1],xmm0[2],xmm13[2],xmm0[3],xmm13[3],xmm0[4],xmm13[4],xmm0[5],xmm13[5],xmm0[6],xmm13[6],xmm0[7],xmm13[7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm13, (%rsp) # 16-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
 ; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm3
+; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm2
+; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %xmm4
-; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm4, %xmm7
-; AVX2-FAST-NEXT:    vmovdqa %xmm3, %xmm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
-; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm4[0],xmm2[0],xmm4[1],xmm2[1],xmm4[2],xmm2[2],xmm4[3],xmm2[3],xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
+; AVX2-FAST-NEXT:    vmovdqa %xmm4, %xmm6
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm5 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
+; AVX2-FAST-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm0, %ymm2, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm2
 ; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
 ; AVX2-FAST-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm1
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm3
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm9
-; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm9[0],xmm1[0],xmm9[1],xmm1[1],xmm9[2],xmm1[2],xmm9[3],xmm1[3],xmm9[4],xmm1[4],xmm9[5],xmm1[5],xmm9[6],xmm1[6],xmm9[7],xmm1[7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; AVX2-FAST-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-NEXT:    vpshufb %xmm5, %xmm1, %xmm1
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm10
+; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm0, %ymm1, %ymm10
 ; AVX2-FAST-NEXT:    vmovdqa 32(%rax), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,1,0,0,4,5,6,7]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,0,1,2,0,0,1]
 ; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %xmm6
+; AVX2-FAST-NEXT:    vmovdqa 32(%r9), %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa 32(%r8), %xmm14
-; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm14[0],xmm6[0],xmm14[1],xmm6[1],xmm14[2],xmm6[2],xmm14[3],xmm6[3],xmm14[4],xmm6[4],xmm14[5],xmm6[5],xmm14[6],xmm6[6],xmm14[7],xmm6[7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm14[0],xmm1[0],xmm14[1],xmm1[1],xmm14[2],xmm1[2],xmm14[3],xmm1[3],xmm14[4],xmm1[4],xmm14[5],xmm1[5],xmm14[6],xmm1[6],xmm14[7],xmm1[7]
+; AVX2-FAST-NEXT:    vmovdqa %xmm1, %xmm7
+; AVX2-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm11 = <u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm4, %xmm4
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,1]
+; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm9, %xmm9
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm4, %ymm0, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm9, %ymm0, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa (%rax), %xmm0
 ; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm0[1,1,0,0,4,5,6,7]
 ; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm2, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa (%r9), %xmm2
 ; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm3
-; AVX2-FAST-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm15 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3],xmm3[4],xmm2[4],xmm3[5],xmm2[5],xmm3[6],xmm2[6],xmm3[7],xmm2[7]
+; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm9
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm15 = xmm9[0],xmm2[0],xmm9[1],xmm2[1],xmm9[2],xmm2[2],xmm9[3],xmm2[3],xmm9[4],xmm2[4],xmm9[5],xmm2[5],xmm9[6],xmm2[6],xmm9[7],xmm2[7]
 ; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm15, %xmm11
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm11, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm11, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm1 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm11, %ymm10, %ymm0, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
 ; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm13, %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm10 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm2, %xmm11
 ; AVX2-FAST-NEXT:    vpor %xmm1, %xmm11, %xmm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm11 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
-; AVX2-FAST-NEXT:    vmovdqa %xmm8, %xmm5
-; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm8, %xmm12
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm5, %xmm12
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm15 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
-; AVX2-FAST-NEXT:    vmovdqa %xmm7, %xmm4
-; AVX2-FAST-NEXT:    vpshufb %xmm15, %xmm7, %xmm13
+; AVX2-FAST-NEXT:    vmovdqa %xmm6, %xmm4
+; AVX2-FAST-NEXT:    vpshufb %xmm15, %xmm6, %xmm13
 ; AVX2-FAST-NEXT:    vpor %xmm12, %xmm13, %xmm12
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm1, %ymm12, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm8, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm7, %xmm10
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm6, %xmm10
 ; AVX2-FAST-NEXT:    vpor %xmm0, %xmm10, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm3, %xmm10
-; AVX2-FAST-NEXT:    vpshufb %xmm15, %xmm9, %xmm11
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %xmm15, %xmm3, %xmm11
 ; AVX2-FAST-NEXT:    vpor %xmm10, %xmm11, %xmm10
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm0, %ymm10, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm10 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
-; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm6, %xmm11
+; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm7, %xmm11
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm12 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
 ; AVX2-FAST-NEXT:    vpshufb %xmm12, %xmm14, %xmm13
 ; AVX2-FAST-NEXT:    vpor %xmm11, %xmm13, %xmm11
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm13 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm13, %xmm2, %xmm15
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[0,0,1,0]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm11, %ymm15, %ymm11
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm6, %xmm10
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm7, %xmm10
 ; AVX2-FAST-NEXT:    vpshufb %xmm12, %xmm9, %xmm12
 ; AVX2-FAST-NEXT:    vpor %xmm10, %xmm12, %xmm10
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpshufb %xmm13, %xmm15, %xmm12
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,1,0]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm1, %ymm10, %ymm12, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm10, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
 ; AVX2-FAST-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm5[8],xmm4[8],xmm5[9],xmm4[9],xmm5[10],xmm4[10],xmm5[11],xmm4[11],xmm5[12],xmm4[12],xmm5[13],xmm4[13],xmm5[14],xmm4[14],xmm5[15],xmm4[15]
-; AVX2-FAST-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
 ; AVX2-FAST-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm1 # 16-byte Folded Reload
 ; AVX2-FAST-NEXT:    # xmm1 = xmm1[8],mem[8],xmm1[9],mem[9],xmm1[10],mem[10],xmm1[11],mem[11],xmm1[12],mem[12],xmm1[13],mem[13],xmm1[14],mem[14],xmm1[15],mem[15]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm10 = <6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7>
 ; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm11 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
 ; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm1, %xmm1
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm3, %xmm1 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm1 = xmm3[8],mem[8],xmm3[9],mem[9],xmm3[10],mem[10],xmm3[11],mem[11],xmm3[12],mem[12],xmm3[13],mem[13],xmm3[14],mem[14],xmm3[15],mem[15]
+; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8],xmm3[8],xmm1[9],xmm3[9],xmm1[10],xmm3[10],xmm1[11],xmm3[11],xmm1[12],xmm3[12],xmm1[13],xmm3[13],xmm1[14],xmm3[14],xmm1[15],xmm3[15]
 ; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm1, %xmm1
-; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm8[8],xmm7[8],xmm8[9],xmm7[9],xmm8[10],xmm7[10],xmm8[11],xmm7[11],xmm8[12],xmm7[12],xmm8[13],xmm7[13],xmm8[14],xmm7[14],xmm8[15],xmm7[15]
-; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm8, %xmm8
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm8[8],xmm6[8],xmm8[9],xmm6[9],xmm8[10],xmm6[10],xmm8[11],xmm6[11],xmm8[12],xmm6[12],xmm8[13],xmm6[13],xmm8[14],xmm6[14],xmm8[15],xmm6[15]
+; AVX2-FAST-NEXT:    vpshufb %xmm11, %xmm6, %xmm6
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm1, %ymm8, %ymm1
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm1, %ymm6, %ymm1
 ; AVX2-FAST-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm14, %xmm3 # 16-byte Folded Reload
 ; AVX2-FAST-NEXT:    # xmm3 = xmm14[8],mem[8],xmm14[9],mem[9],xmm14[10],mem[10],xmm14[11],mem[11],xmm14[12],mem[12],xmm14[13],mem[13],xmm14[14],mem[14],xmm14[15],mem[15]
 ; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} xmm5 = xmm2[0,1,2,3,4,5,5,6]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [2,2,3,3,2,2,3,3]
-; AVX2-FAST-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm7, %ymm5
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
-; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm3, %xmm3
+; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [2,2,3,3,2,2,3,3]
+; AVX2-FAST-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm6, %ymm5
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm10 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
+; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm3, %xmm3
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm3, %ymm5, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm3, %ymm5, %ymm3
 ; AVX2-FAST-NEXT:    vpshufhw {{.*#+}} xmm5 = xmm15[0,1,2,3,4,5,5,6]
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm7, %ymm5
-; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm9[8],xmm6[8],xmm9[9],xmm6[9],xmm9[10],xmm6[10],xmm9[11],xmm6[11],xmm9[12],xmm6[12],xmm9[13],xmm6[13],xmm9[14],xmm6[14],xmm9[15],xmm6[15]
-; AVX2-FAST-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
+; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm6, %ymm5
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm9[8],xmm7[8],xmm9[9],xmm7[9],xmm9[10],xmm7[10],xmm9[11],xmm7[11],xmm9[12],xmm7[12],xmm9[13],xmm7[13],xmm9[14],xmm7[14],xmm9[15],xmm7[15]
+; AVX2-FAST-NEXT:    vpshufb %xmm10, %xmm2, %xmm2
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm2, %ymm5, %ymm2
+; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm2, %ymm5, %ymm2
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm0, %ymm3, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm2, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa (%rax), %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm4
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm6
+; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm5
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm5[8],ymm6[8],ymm5[9],ymm6[9],ymm5[10],ymm6[10],ymm5[11],ymm6[11],ymm5[12],ymm6[12],ymm5[13],ymm6[13],ymm5[14],ymm6[14],ymm5[15],ymm6[15],ymm5[24],ymm6[24],ymm5[25],ymm6[25],ymm5[26],ymm6[26],ymm5[27],ymm6[27],ymm5[28],ymm6[28],ymm5[29],ymm6[29],ymm5[30],ymm6[30],ymm5[31],ymm6[31]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0>
+; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm7
 ; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm1
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[17,18,19,30],zero,ymm1[28],zero,ymm1[28,29,30,31],zero,ymm1[29],zero,ymm1[31]
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, %ymm10
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm3
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero
-; AVX2-FAST-NEXT:    vmovdqa %ymm3, %ymm10
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm2
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero
+; AVX2-FAST-NEXT:    vmovdqa %ymm2, %ymm11
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm13
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm13[30],zero,ymm13[28],zero,zero,zero,zero,ymm13[31],zero,ymm13[29],zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm11
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm11[30],zero,ymm11[28],zero,zero,zero,zero,ymm11[31],zero,ymm11[29],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm4, %ymm6, %ymm4
+; AVX2-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm3
+; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm2
+; AVX2-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm9 = ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15],ymm3[24],ymm2[24],ymm3[25],ymm2[25],ymm3[26],ymm2[26],ymm3[27],ymm2[27],ymm3[28],ymm2[28],ymm3[29],ymm2[29],ymm3[30],ymm2[30],ymm3[31],ymm2[31]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm9[u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm3, %ymm4, %ymm6
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[27],zero,ymm0[27,28,29,30],zero,ymm0[28],zero,ymm0[26,27,30,31],zero,ymm0[29]
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, %ymm1
+; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
+; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm8, %ymm7, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm4
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm4[27],zero,zero,zero,zero,ymm4[30],zero,ymm4[28],zero,zero,zero,zero,ymm4[31],zero
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm3[23],zero,ymm3[27,20,21,26],zero,ymm3[24],zero,ymm3[26,27,26,27],zero,ymm3[25]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm3, %ymm7, %ymm7
-; AVX2-FAST-NEXT:    vmovdqa (%rax), %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, %ymm3
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm7
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm6, %ymm7, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm11[23],zero,ymm11[27,20,21,26],zero,ymm11[24],zero,ymm11[26,27,26,27],zero,ymm11[25]
-; AVX2-FAST-NEXT:    vmovdqa %ymm11, %ymm15
+; AVX2-FAST-NEXT:    vpor %ymm7, %ymm8, %ymm7
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm11[25],zero,ymm11[23],zero,zero,zero,zero,ymm11[26],zero,ymm11[24],zero,zero,zero,zero
+; AVX2-FAST-NEXT:    vmovdqa %ymm11, %ymm14
 ; AVX2-FAST-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm13[23],zero,zero,zero,zero,ymm13[26],zero,ymm13[24],zero,zero,zero,zero,ymm13[27],zero
-; AVX2-FAST-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm10[25],zero,ymm10[23],zero,zero,zero,zero,ymm10[26],zero,ymm10[24],zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vmovdqa %ymm10, %ymm5
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm10[25],zero,ymm10[23],zero,zero,zero,zero,ymm10[26],zero,ymm10[24],zero,zero,zero,zero,ymm10[27]
+; AVX2-FAST-NEXT:    vmovdqa %ymm10, %ymm13
 ; AVX2-FAST-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27]
-; AVX2-FAST-NEXT:    vmovdqa %ymm2, %ymm7
-; AVX2-FAST-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpor %ymm8, %ymm9, %ymm8
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
-; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm4[25],zero,ymm4[23],zero,zero,zero,zero,ymm4[26],zero,ymm4[24],zero,zero
+; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm7, %ymm8, %ymm7
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm5[25],zero,ymm5[23],zero,zero,zero,zero,ymm5[26],zero,ymm5[24],zero,zero
+; AVX2-FAST-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm1[25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpor %ymm8, %ymm9, %ymm8
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm9 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-FAST-NEXT:    vmovdqa %ymm4, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm6, %ymm8, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm0, %ymm8
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128]
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm0, %ymm11
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm8, %ymm11, %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm2, %ymm12
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm14
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm12, %ymm14, %ymm12
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm8, %ymm12, %ymm3
-; AVX2-FAST-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %ymm9, %ymm5, %ymm9
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm7, %ymm10
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm7 = ymm4[0],ymm0[0],ymm4[1],ymm0[1],ymm4[2],ymm0[2],ymm4[3],ymm0[3],ymm4[4],ymm0[4],ymm4[5],ymm0[5],ymm4[6],ymm0[6],ymm4[7],ymm0[7],ymm4[16],ymm0[16],ymm4[17],ymm0[17],ymm4[18],ymm0[18],ymm4[19],ymm0[19],ymm4[20],ymm0[20],ymm4[21],ymm0[21],ymm4[22],ymm0[22],ymm4[23],ymm0[23]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9>
+; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm7, %ymm7
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm10 = ymm9[0],ymm15[0],ymm9[1],ymm15[1],ymm9[2],ymm15[2],ymm9[3],ymm15[3],ymm9[4],ymm15[4],ymm9[5],ymm15[5],ymm9[6],ymm15[6],ymm9[7],ymm15[7],ymm9[16],ymm15[16],ymm9[17],ymm15[17],ymm9[18],ymm15[18],ymm9[19],ymm15[19],ymm9[20],ymm15[20],ymm9[21],ymm15[21],ymm9[22],ymm15[22],ymm9[23],ymm15[23]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u>
+; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm10, %ymm10
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm9, %ymm10, %ymm9
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm13, %ymm10
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm7, %ymm10, %ymm4
+; AVX2-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm10 = ymm13[0],ymm14[0],ymm13[1],ymm14[1],ymm13[2],ymm14[2],ymm13[3],ymm14[3],ymm13[4],ymm14[4],ymm13[5],ymm14[5],ymm13[6],ymm14[6],ymm13[7],ymm14[7],ymm13[16],ymm14[16],ymm13[17],ymm14[17],ymm13[18],ymm14[18],ymm13[19],ymm14[19],ymm13[20],ymm14[20],ymm13[21],ymm14[21],ymm13[22],ymm14[22],ymm13[23],ymm14[23]
+; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm10, %ymm8
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm10 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[4],ymm2[4],ymm3[5],ymm2[5],ymm3[6],ymm2[6],ymm3[7],ymm2[7],ymm3[16],ymm2[16],ymm3[17],ymm2[17],ymm3[18],ymm2[18],ymm3[19],ymm2[19],ymm3[20],ymm2[20],ymm3[21],ymm2[21],ymm3[22],ymm2[22],ymm3[23],ymm2[23]
+; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm10, %ymm10
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
 ; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm15, %ymm11
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm10, %ymm11, %ymm10
-; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm9, %ymm10, %ymm9
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm8, %ymm10, %ymm10
+; AVX2-FAST-NEXT:    vpshuflw $150, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm8 = mem[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [4,5,4,5,5,7,4,5]
+; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm11, %ymm8
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm4[0],ymm7[0],ymm4[1],ymm7[1],ymm4[2],ymm7[2],ymm4[3],ymm7[3],ymm4[4],ymm7[4],ymm4[5],ymm7[5],ymm4[6],ymm7[6],ymm4[7],ymm7[7],ymm4[16],ymm7[16],ymm4[17],ymm7[17],ymm4[18],ymm7[18],ymm4[19],ymm7[19],ymm4[20],ymm7[20],ymm4[21],ymm7[21],ymm4[22],ymm7[22],ymm4[23],ymm7[23]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13>
+; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm12, %ymm12
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm12, %ymm8, %ymm8
+; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm1[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm11, %ymm11
+; AVX2-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm6[0],ymm5[0],ymm6[1],ymm5[1],ymm6[2],ymm5[2],ymm6[3],ymm5[3],ymm6[4],ymm5[4],ymm6[5],ymm5[5],ymm6[6],ymm5[6],ymm6[7],ymm5[7],ymm6[16],ymm5[16],ymm6[17],ymm5[17],ymm6[18],ymm5[18],ymm6[19],ymm5[19],ymm6[20],ymm5[20],ymm6[21],ymm5[21],ymm6[22],ymm5[22],ymm6[23],ymm5[23]
+; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm12, %ymm12
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm12, %ymm11, %ymm11
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8, %ymm1 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm10, %ymm11, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
 ; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm15, %ymm11
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm5, %ymm13
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [0,128,14,128,128,128,128,1,128,15,128,128,128,128,2,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128,18,128]
+; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm9, %ymm13
 ; AVX2-FAST-NEXT:    vpor %ymm11, %ymm13, %ymm11
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm13 = ymm8[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [4,5,4,5,5,7,4,5]
-; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm14, %ymm13
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm11, %ymm13, %ymm11
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm4, %ymm10
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm3, %ymm12
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-NEXT:    vpor %ymm10, %ymm12, %ymm10
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} ymm12 = ymm6[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX2-FAST-NEXT:    vpermd %ymm12, %ymm14, %ymm12
-; AVX2-FAST-NEXT:    vpblendvb %ymm0, %ymm10, %ymm12, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm7 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm9, %ymm0, %ymm9
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm10
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,128,14,128,128,128,128,1,128,15,128,128,128,128,2,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128,18,128]
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm1, %ymm12
-; AVX2-FAST-NEXT:    vpor %ymm10, %ymm12, %ymm10
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm13, %ymm13
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [128,128,0,128,14,128,128,128,128,1,128,15,128,128,128,128,128,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128]
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm14, %ymm1, %ymm7
-; AVX2-FAST-NEXT:    vpor %ymm7, %ymm13, %ymm7
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm10, %ymm7, %ymm7
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
+; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm0, %ymm14
+; AVX2-FAST-NEXT:    vmovdqa %ymm0, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = [128,128,0,128,14,128,128,128,128,1,128,15,128,128,128,128,128,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128]
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm5, %ymm9
+; AVX2-FAST-NEXT:    vpor %ymm14, %ymm9, %ymm9
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm11, %ymm9, %ymm9
+; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm2, %ymm0
+; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm3, %ymm1
 ; AVX2-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vmovdqu (%rsp), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm14, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpshufb %ymm13, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm2, %ymm2
 ; AVX2-FAST-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpblendvb %ymm13, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpblendvb %ymm14, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,1,2,3,0,128,14,128,0,1,0,1,128,15,128,15,128,17,18,19,16,128,30,128,16,17,16,17,128,31,128,31]
-; AVX2-FAST-NEXT:    vpshufb %ymm1, %ymm5, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
-; AVX2-FAST-NEXT:    vpshufb %ymm5, %ymm15, %ymm10
+; AVX2-FAST-NEXT:    vpshufb %ymm1, %ymm4, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm7, %ymm10
 ; AVX2-FAST-NEXT:    vpor %ymm2, %ymm10, %ymm2
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm8, %ymm11
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm13, %ymm11
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm2, %ymm11, %ymm2
-; AVX2-FAST-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vpshufb %ymm5, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vpor %ymm1, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm6, %ymm3
+; AVX2-FAST-NEXT:    vpshufb %ymm1, %ymm6, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm6, %ymm3
+; AVX2-FAST-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpshufb %ymm10, %ymm3, %ymm3
 ; AVX2-FAST-NEXT:    vpblendvb %ymm12, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm7, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm9, %ymm2, %ymm2
 ; AVX2-FAST-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm5[17,18,19,30],zero,ymm5[28],zero,ymm5[28,29,30,31],zero,ymm5[29],zero,ymm5[31]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm8[30],zero,ymm8[28],zero,zero,zero,zero,ymm8[31],zero,ymm8[29],zero
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX2-FAST-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %ymm3, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-NEXT:    # ymm3 = ymm3[8],mem[8],ymm3[9],mem[9],ymm3[10],mem[10],ymm3[11],mem[11],ymm3[12],mem[12],ymm3[13],mem[13],ymm3[14],mem[14],ymm3[15],mem[15],ymm3[24],mem[24],ymm3[25],mem[25],ymm3[26],mem[26],ymm3[27],mem[27],ymm3[28],mem[28],ymm3[29],mem[29],ymm3[30],mem[30],ymm3[31],mem[31]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,26,27,30,31,30,31,28,29,28,29,28,29,28,29]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,28,27,u,u,u,31,30,u,u,u,u,u,u,u,u]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u>
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; AVX2-FAST-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-NEXT:    vmovdqa %ymm0, 96(%rax)
 ; AVX2-FAST-NEXT:    vmovdqa %ymm2, 320(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm9, 128(%rax)
+; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT:    vmovaps %ymm0, 128(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 352(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 160(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 192(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 224(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 288(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 256(%rax)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm0, 416(%rax)
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, 416(%rax)
 ; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-NEXT:    vmovaps %ymm0, 384(%rax)
-; AVX2-FAST-NEXT:    addq $648, %rsp # imm = 0x288
+; AVX2-FAST-NEXT:    addq $600, %rsp # imm = 0x258
 ; AVX2-FAST-NEXT:    vzeroupper
 ; AVX2-FAST-NEXT:    retq
 ;
 ; AVX2-FAST-PERLANE-LABEL: store_i8_stride7_vf64:
 ; AVX2-FAST-PERLANE:       # %bb.0:
-; AVX2-FAST-PERLANE-NEXT:    subq $648, %rsp # imm = 0x288
+; AVX2-FAST-PERLANE-NEXT:    subq $616, %rsp # imm = 0x268
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdx), %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rcx), %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rax), %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[23],zero,ymm1[27,20,21,26],zero,ymm1[24],zero,ymm1[26,27,26,27],zero,ymm1[25]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdx), %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rcx), %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rax), %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[23],zero,ymm0[27,20,21,26],zero,ymm0[24],zero,ymm0[26,27,26,27],zero,ymm0[25]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero,ymm1[27],zero
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero,zero,zero,ymm6[27]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm3[25],zero,ymm3[23],zero,zero,zero,zero,ymm3[26],zero,ymm3[24],zero,zero,zero,zero,ymm3[27]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm2, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,24,25,26,27,24,25,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,22,23,22,23,24,25,26,27,24,25,30,31]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,25,24,23,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,25,24,23,u,u,u,u,u,u,u,u,u]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255>
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255]
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm6[17,18,19,30],zero,ymm6[28],zero,ymm6[28,29,30,31],zero,ymm6[29],zero,ymm6[31]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm7[30],zero,ymm7[28],zero,zero,zero,zero,ymm7[31],zero,ymm7[29],zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm9[30],zero,ymm9[28],zero,zero,zero,zero,ymm9[31],zero,ymm9[29],zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm8[30],zero,ymm8[28],zero,zero,zero,zero,ymm8[31],zero,ymm8[29],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,26,27,30,31,30,31,28,29,28,29,28,29,28,29]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,28,27,u,u,u,31,30,u,u,u,u,u,u,u,u]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %xmm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rsi), %xmm12
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm13[0],xmm0[1],xmm13[1],xmm0[2],xmm13[2],xmm0[3],xmm13[3],xmm0[4],xmm13[4],xmm0[5],xmm13[5],xmm0[6],xmm13[6],xmm0[7],xmm13[7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm12[0],xmm0[1],xmm12[1],xmm0[2],xmm12[2],xmm0[3],xmm12[3],xmm0[4],xmm12[4],xmm0[5],xmm12[5],xmm0[6],xmm12[6],xmm0[7],xmm12[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm1 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rcx), %xmm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rcx), %xmm6
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdx), %xmm2
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm3 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm2[0],xmm6[0],xmm2[1],xmm6[1],xmm2[2],xmm6[2],xmm2[3],xmm6[3],xmm2[4],xmm6[4],xmm2[5],xmm6[5],xmm2[6],xmm6[6],xmm2[7],xmm6[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm5 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm2, %xmm2
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm0, %ymm2, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm5
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm5[0],xmm2[0],xmm5[1],xmm2[1],xmm5[2],xmm2[2],xmm5[3],xmm2[3],xmm5[4],xmm2[4],xmm5[5],xmm2[5],xmm5[6],xmm2[6],xmm5[7],xmm2[7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, (%rsp) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm2, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm8[0],xmm0[1],xmm8[1],xmm0[2],xmm8[2],xmm0[3],xmm8[3],xmm0[4],xmm8[4],xmm0[5],xmm8[5],xmm0[6],xmm8[6],xmm0[7],xmm8[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm8, (%rsp) # 16-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm6
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm6[0],xmm7[0],xmm6[1],xmm7[1],xmm6[2],xmm7[2],xmm6[3],xmm7[3],xmm6[4],xmm7[4],xmm6[5],xmm7[5],xmm6[6],xmm7[6],xmm6[7],xmm7[7]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3],xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm1, %xmm1
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm0, %ymm1, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rax), %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm11 = [2,3,2,3,0,1,0,1,8,9,10,11,2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm10 = [2,3,2,3,0,1,0,1,8,9,10,11,2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm0, %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %xmm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %xmm15
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm15[0],xmm1[0],xmm15[1],xmm1[1],xmm15[2],xmm1[2],xmm15[3],xmm1[3],xmm15[4],xmm1[4],xmm15[5],xmm1[5],xmm15[6],xmm1[6],xmm15[7],xmm1[7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, %xmm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm9 = <u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm9, %xmm3, %xmm3
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm3, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %xmm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm14 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3],xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm9, %xmm14, %xmm14
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %xmm9
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm9, %xmm11
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,1,0]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm14, %ymm11, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm12, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r9), %xmm15
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%r8), %xmm14
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm14[0],xmm15[0],xmm14[1],xmm15[1],xmm14[2],xmm15[2],xmm14[3],xmm15[3],xmm14[4],xmm15[4],xmm14[5],xmm15[5],xmm14[6],xmm15[6],xmm14[7],xmm15[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm1, %xmm9, %xmm9
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = <u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm9, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm9
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} xmm13 = xmm9[0],xmm2[0],xmm9[1],xmm2[1],xmm9[2],xmm2[2],xmm9[3],xmm2[3],xmm9[4],xmm2[4],xmm9[5],xmm2[5],xmm9[6],xmm2[6],xmm9[7],xmm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm1, %xmm13, %xmm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm7, %xmm10
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm13, %ymm10, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm11, %ymm0, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm8, %ymm10, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm3, %ymm5, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm12 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm12, %xmm13, %xmm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm10 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm11 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm12, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm0, %xmm11
-; AVX2-FAST-PERLANE-NEXT:    vpor %xmm8, %xmm11, %xmm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm11 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm3, %xmm14
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm0, %xmm10
+; AVX2-FAST-PERLANE-NEXT:    vpor %xmm3, %xmm10, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm10 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm6, %xmm13
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm1, %xmm13
-; AVX2-FAST-PERLANE-NEXT:    vpor %xmm14, %xmm13, %xmm13
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm8, %ymm13, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm12, %xmm2, %xmm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm5, %xmm10
-; AVX2-FAST-PERLANE-NEXT:    vpor %xmm12, %xmm10, %xmm10
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm7, %xmm11
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm6, %xmm0
-; AVX2-FAST-PERLANE-NEXT:    vpor %xmm0, %xmm11, %xmm0
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm10, %ymm0, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm10 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm4, %xmm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm12 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm12, %xmm15, %xmm13
-; AVX2-FAST-PERLANE-NEXT:    vpor %xmm11, %xmm13, %xmm11
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm13 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm13, %xmm2, %xmm14
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,0,1,0]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm11, %ymm14, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm14, %xmm10
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm1, %xmm12
+; AVX2-FAST-PERLANE-NEXT:    vpor %xmm13, %xmm12, %xmm12
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm3, %ymm12, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm8, %xmm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm8, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpor %xmm5, %xmm11, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm4, %xmm10
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm12, %xmm4, %xmm12
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpor %xmm0, %xmm10, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm5, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm5 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm15, %xmm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm11 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm14, %xmm12
 ; AVX2-FAST-PERLANE-NEXT:    vpor %xmm10, %xmm12, %xmm10
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm13, %xmm9, %xmm12
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,1,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm10, %ymm12, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11, %ymm11 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm0, %ymm8, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm12 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm12, %xmm2, %xmm13
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = <255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm10, %ymm13, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm13, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm9, %xmm11
+; AVX2-FAST-PERLANE-NEXT:    vpor %xmm5, %xmm11, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm12, %xmm7, %xmm11
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm5, %ymm11, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm10 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm0, %ymm3, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm3[8],xmm1[8],xmm3[9],xmm1[9],xmm3[10],xmm1[10],xmm3[11],xmm1[11],xmm3[12],xmm1[12],xmm3[13],xmm1[13],xmm3[14],xmm1[14],xmm3[15],xmm1[15]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm6[8],xmm1[8],xmm6[9],xmm1[9],xmm6[10],xmm1[10],xmm6[11],xmm1[11],xmm6[12],xmm1[12],xmm6[13],xmm1[13],xmm6[14],xmm1[14],xmm6[15],xmm1[15]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm8 # 16-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # xmm8 = xmm1[8],mem[8],xmm1[9],mem[9],xmm1[10],mem[10],xmm1[11],mem[11],xmm1[12],mem[12],xmm1[13],mem[13],xmm1[14],mem[14],xmm1[15],mem[15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm10 = <6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm3 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm3 = xmm1[8],mem[8],xmm1[9],mem[9],xmm1[10],mem[10],xmm1[11],mem[11],xmm1[12],mem[12],xmm1[13],mem[13],xmm1[14],mem[14],xmm1[15],mem[15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm5 = <6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm0, %xmm0
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm11 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm8, %xmm8
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm0, %ymm8, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm7[8],xmm6[8],xmm7[9],xmm6[9],xmm7[10],xmm6[10],xmm7[11],xmm6[11],xmm7[12],xmm6[12],xmm7[13],xmm6[13],xmm7[14],xmm6[14],xmm7[15],xmm6[15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm6, %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm10 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm3, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm0, %ymm3, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm4[8],xmm1[9],xmm4[9],xmm1[10],xmm4[10],xmm1[11],xmm4[11],xmm1[12],xmm4[12],xmm1[13],xmm4[13],xmm1[14],xmm4[14],xmm1[15],xmm4[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm5, %xmm3, %xmm3
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsp), %xmm1 # 16-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm7 = xmm1[8],xmm5[8],xmm1[9],xmm5[9],xmm1[10],xmm5[10],xmm1[11],xmm5[11],xmm1[12],xmm5[12],xmm1[13],xmm5[13],xmm1[14],xmm5[14],xmm1[15],xmm5[15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm11, %xmm7, %xmm7
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm6, %ymm7, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm15, %xmm1 # 16-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    # xmm1 = xmm15[8],mem[8],xmm15[9],mem[9],xmm15[10],mem[10],xmm15[11],mem[11],xmm15[12],mem[12],xmm15[13],mem[13],xmm15[14],mem[14],xmm15[15],mem[15]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,9,10,11,8,9,10,11,10,11,12,13,10,11,12,13]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm2, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm1[8],xmm8[8],xmm1[9],xmm8[9],xmm1[10],xmm8[10],xmm1[11],xmm8[11],xmm1[12],xmm8[12],xmm1[13],xmm8[13],xmm1[14],xmm8[14],xmm1[15],xmm8[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm10, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm3, %ymm5, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm14[8],xmm15[8],xmm14[9],xmm15[9],xmm14[10],xmm15[10],xmm14[11],xmm15[11],xmm14[12],xmm15[12],xmm14[13],xmm15[13],xmm14[14],xmm15[14],xmm14[15],xmm15[15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm4 = [8,9,10,11,8,9,10,11,10,11,12,13,10,11,12,13]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm2, %xmm5
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm7 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm7, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm8 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm8, %xmm1, %xmm1
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm1, %ymm5, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm9, %xmm2
-; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm4[8],xmm14[8],xmm4[9],xmm14[9],xmm4[10],xmm14[10],xmm4[11],xmm14[11],xmm4[12],xmm14[12],xmm4[13],xmm14[13],xmm4[14],xmm14[14],xmm4[15],xmm14[15]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm7, %xmm3, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm1, %ymm5, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm7, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm9[8],xmm13[8],xmm9[9],xmm13[9],xmm9[10],xmm13[10],xmm9[11],xmm13[11],xmm9[12],xmm13[12],xmm9[13],xmm13[13],xmm9[14],xmm13[14],xmm9[15],xmm13[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm8, %xmm4, %xmm4
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[0,1,0,1]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm3, %ymm2, %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm6, %ymm4, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm6, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm3, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm12[17,18,19,30],zero,ymm12[28],zero,ymm12[28,29,30,31],zero,ymm12[29],zero,ymm12[31]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm3
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm4[8],ymm5[8],ymm4[9],ymm5[9],ymm4[10],ymm5[10],ymm4[11],ymm5[11],ymm4[12],ymm5[12],ymm4[13],ymm5[13],ymm4[14],ymm5[14],ymm4[15],ymm5[15],ymm4[24],ymm5[24],ymm4[25],ymm5[25],ymm4[26],ymm5[26],ymm4[27],ymm5[27],ymm4[28],ymm5[28],ymm4[29],ymm5[29],ymm4[30],ymm5[30],ymm4[31],ymm5[31]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[17,18,19,30],zero,ymm1[28],zero,ymm1[28,29,30,31],zero,ymm1[29],zero,ymm1[31]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm0, %ymm1, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm15
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm15[30],zero,ymm15[28],zero,zero,zero,zero,ymm15[31],zero,ymm15[29],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm4, %ymm6, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm0, %ymm1, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{.*#+}} ymm9 = 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]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm9[u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm3, %ymm4, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[27],zero,ymm0[27,28,29,30],zero,ymm0[28],zero,ymm0[26,27,30,31],zero,ymm0[29]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm8, %ymm7, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm0[27],zero,zero,zero,zero,ymm0[30],zero,ymm0[28],zero,zero,zero,zero,ymm0[31],zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm3, %ymm7, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rax), %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, (%rsp) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = [255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm6, %ymm7, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm15[23],zero,ymm15[27,20,21,26],zero,ymm15[24],zero,ymm15[26,27,26,27],zero,ymm15[25]
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm5[23],zero,zero,zero,zero,ymm5[26],zero,ymm5[24],zero,zero,zero,zero,ymm5[27],zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm5, %ymm14
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm6, %ymm7, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm10[25],zero,ymm10[23],zero,zero,zero,zero,ymm10[26],zero,ymm10[24],zero,zero,zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, %ymm5
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[23],zero,ymm1[27,20,21,26],zero,ymm1[24],zero,ymm1[26,27,26,27],zero,ymm1[25]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm12[25],zero,ymm12[23],zero,zero,zero,zero,ymm12[26],zero,ymm12[24],zero,zero,zero,zero,ymm12[27]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm12, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero,ymm2[27],zero
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm6, %ymm7, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm4[25],zero,ymm4[23],zero,zero,zero,zero,ymm4[26],zero,ymm4[24],zero,zero
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm1[25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm11[25],zero,ymm11[23],zero,zero,zero,zero,ymm11[26],zero,ymm11[24],zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm11, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm11, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm10[25],zero,ymm10[23],zero,zero,zero,zero,ymm10[26],zero,ymm10[24],zero,zero,zero,zero,ymm10[27]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm7, %ymm8, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm4[25],zero,ymm4[23],zero,zero,zero,zero,ymm4[26],zero,ymm4[24],zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm6, %ymm7, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm5[25],zero,ymm5[23],zero,zero,zero,zero,ymm5[26],zero,ymm5[24],zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm9 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm8, %ymm9, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = [0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm7, %ymm8, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm8, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm7 = ymm3[0],ymm0[0],ymm3[1],ymm0[1],ymm3[2],ymm0[2],ymm3[3],ymm0[3],ymm3[4],ymm0[4],ymm3[5],ymm0[5],ymm3[6],ymm0[6],ymm3[7],ymm0[7],ymm3[16],ymm0[16],ymm3[17],ymm0[17],ymm3[18],ymm0[18],ymm3[19],ymm0[19],ymm3[20],ymm0[20],ymm3[21],ymm0[21],ymm3[22],ymm0[22],ymm3[23],ymm0[23]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm8, %ymm7, %ymm7
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm9, %ymm6, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm10 = ymm15[0],ymm9[0],ymm15[1],ymm9[1],ymm15[2],ymm9[2],ymm15[3],ymm9[3],ymm15[4],ymm9[4],ymm15[5],ymm9[5],ymm15[6],ymm9[6],ymm15[7],ymm9[7],ymm15[16],ymm9[16],ymm15[17],ymm9[17],ymm15[18],ymm9[18],ymm15[19],ymm9[19],ymm15[20],ymm9[20],ymm15[21],ymm9[21],ymm15[22],ymm9[22],ymm15[23],ymm9[23]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm10, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm7, %ymm10, %ymm7
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128>
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm2, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,21,128,19,128,128,128,128,22,128,20,128,128,128]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm1, %ymm13
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm11, %ymm13, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm7, %ymm11, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm8, %ymm5, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm7, %ymm10, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm10 = ymm13[0],ymm14[0],ymm13[1],ymm14[1],ymm13[2],ymm14[2],ymm13[3],ymm14[3],ymm13[4],ymm14[4],ymm13[5],ymm14[5],ymm13[6],ymm14[6],ymm13[7],ymm14[7],ymm13[16],ymm14[16],ymm13[17],ymm14[17],ymm13[18],ymm14[18],ymm13[19],ymm14[19],ymm13[20],ymm14[20],ymm13[21],ymm14[21],ymm13[22],ymm14[22],ymm13[23],ymm14[23]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm8, %ymm10, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm10 = ymm6[0],ymm2[0],ymm6[1],ymm2[1],ymm6[2],ymm2[2],ymm6[3],ymm2[3],ymm6[4],ymm2[4],ymm6[5],ymm2[5],ymm6[6],ymm2[6],ymm6[7],ymm2[7],ymm6[16],ymm2[16],ymm6[17],ymm2[17],ymm6[18],ymm2[18],ymm6[19],ymm2[19],ymm6[20],ymm2[20],ymm6[21],ymm2[21],ymm6[22],ymm2[22],ymm6[23],ymm2[23]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm10, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm9, %ymm12, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm8, %ymm9, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm14, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm15, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm9, %ymm10, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm8, %ymm9, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22>
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm9, %ymm14, %ymm10
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22,128]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm5, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm10, %ymm12, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,18,19,18,19,20,21,18,19,20,21,28,29,30,31>
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm12, %ymm4, %ymm13
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,2,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm10, %ymm13, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm9, %ymm13, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm8, %ymm10, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,18,19,18,19,20,21,18,19,20,21,28,29,30,31>
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm3, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm9, %ymm11, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu (%rsp), %ymm7 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm12, %ymm7, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm8, %ymm3, %ymm11
 ; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,2,3,2]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm0, %ymm9, %ymm11, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, {{[-0-9]+}}(%r{{[sb]}}p), %ymm10, %ymm10 # 32-byte Folded Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm9, %ymm8, %ymm0, %ymm8
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm0 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm2, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [0,128,14,128,128,128,128,1,128,15,128,128,128,128,2,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128,18,128]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm1, %ymm11
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm9, %ymm11, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm1, %ymm12
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,0,128,14,128,128,128,128,1,128,15,128,128,128,128,128,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm6, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm6, %ymm12, %ymm6
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm9, %ymm6, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm3[0],ymm6[0],ymm3[1],ymm6[1],ymm3[2],ymm6[2],ymm3[3],ymm6[3],ymm3[4],ymm6[4],ymm3[5],ymm6[5],ymm3[6],ymm6[6],ymm3[7],ymm6[7],ymm3[16],ymm6[16],ymm3[17],ymm6[17],ymm3[18],ymm6[18],ymm3[19],ymm6[19],ymm3[20],ymm6[20],ymm3[21],ymm6[21],ymm3[22],ymm6[22],ymm3[23],ymm6[23]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = <8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm12, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm14 = <u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm12, %ymm11, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklbw {{.*#+}} ymm12 = ymm5[0],ymm4[0],ymm5[1],ymm4[1],ymm5[2],ymm4[2],ymm5[3],ymm4[3],ymm5[4],ymm4[4],ymm5[5],ymm4[5],ymm5[6],ymm4[6],ymm5[7],ymm4[7],ymm5[16],ymm4[16],ymm5[17],ymm4[17],ymm5[18],ymm4[18],ymm5[19],ymm4[19],ymm5[20],ymm4[20],ymm5[21],ymm4[21],ymm5[22],ymm4[22],ymm5[23],ymm4[23]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm12, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm8, %ymm1, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,2,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm14, %ymm12, %ymm8, %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = [255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm7, %ymm11, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm13, %ymm10, %ymm12, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, (%rsp) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm9, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = [0,128,14,128,128,128,128,1,128,15,128,128,128,128,2,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128,18,128]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm12, %ymm15, %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm15, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm11, %ymm13, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm13 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm0, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = [128,128,0,128,14,128,128,128,128,1,128,15,128,128,128,128,128,128,16,128,30,128,128,128,128,17,128,31,128,128,128,128]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm4, %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm14, %ymm15, %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255,0,0,u,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm11, %ymm14, %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm2, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm10, %ymm15, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm11, %ymm1, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm9, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm13, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm2, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm15, %ymm0, %ymm1, %ymm0
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,1,2,3,0,128,14,128,0,1,0,1,128,15,128,15,128,17,18,19,16,128,30,128,16,17,16,17,128,31,128,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm5, %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm5, %ymm14, %ymm9
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm2, %ymm9, %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm9 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm9, %ymm4, %ymm10
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm11 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm2, %ymm10, %ymm2
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm3, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm5, %ymm13, %ymm4
-; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm4, %ymm1
-; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm9, %ymm7, %ymm3
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm11, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm3, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm3, %ymm6, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm2, %ymm7, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm13, %ymm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm12 = <255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u,u,u,255,255,0,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm2, %ymm10, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm3, %ymm5, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm7, %ymm3, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm12, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255,255,255,0,0,0,255,255]
-; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm6, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm11, %ymm2, %ymm2
 ; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm3, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm4[17,18,19,30],zero,ymm4[28],zero,ymm4[28,29,30,31],zero,ymm4[29],zero,ymm4[31]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm8[30],zero,ymm8[28],zero,zero,zero,zero,ymm8[31],zero,ymm8[29],zero
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %ymm9, %ymm3 # 32-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # ymm3 = ymm9[8],mem[8],ymm9[9],mem[9],ymm9[10],mem[10],ymm9[11],mem[11],ymm9[12],mem[12],ymm9[13],mem[13],ymm9[14],mem[14],ymm9[15],mem[15],ymm9[24],mem[24],ymm9[25],mem[25],ymm9[26],mem[26],ymm9[27],mem[27],ymm9[28],mem[28],ymm9[29],mem[29],ymm9[30],mem[30],ymm9[31],mem[31]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,28,29,24,25,u,u,u,30,31,26,27,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u,0,0,255,255,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,26,27,30,31,30,31,28,29,28,29,28,29,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u,255,255,255,255,0,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29,28,27,u,u,u,31,30,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u,255,255,255,255,255,0,u>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm4, %ymm1, %ymm3, %ymm1
 ; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 96(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, 320(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm8, 128(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 128(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 352(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 160(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 192(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 64(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 32(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, (%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 224(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 288(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 256(%rax)
-; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 416(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 416(%rax)
 ; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
 ; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 384(%rax)
-; AVX2-FAST-PERLANE-NEXT:    addq $648, %rsp # imm = 0x288
+; AVX2-FAST-PERLANE-NEXT:    addq $616, %rsp # imm = 0x268
 ; AVX2-FAST-PERLANE-NEXT:    vzeroupper
 ; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-SLOW-LABEL: store_i8_stride7_vf64:
 ; AVX512F-SLOW:       # %bb.0:
-; AVX512F-SLOW-NEXT:    subq $1368, %rsp # imm = 0x558
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm13
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm13, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm13, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm16
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm15
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm15, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm3, %ymm18
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm15, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    subq $968, %rsp # imm = 0x3C8
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm5
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm5, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm25
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm7
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm7, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm28
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
 ; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
-; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm23
-; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %ymm1
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <13,u,u,u,u,u,128,14,u,u,u,u,u,128,15,u,u,u,u,u,128,16,u,u,u,u,u,128,17,u,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm3, %ymm29
-; AVX512F-SLOW-NEXT:    vporq %ymm0, %ymm1, %ymm24
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %ymm12
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm2[27],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29]
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm2, %ymm10
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm12[25],zero,ymm12[23],zero,zero,zero,zero,ymm12[26],zero,ymm12[24],zero,zero
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm5, %ymm1, %ymm1
+; AVX512F-SLOW-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm1, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm1, %ymm3
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %ymm7
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <13,u,u,u,u,u,128,14,u,u,u,u,u,128,15,u,u,u,u,u,128,16,u,u,u,u,u,128,17,u,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm4, %ymm7, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm7, %ymm8
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm7, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vporq %ymm0, %ymm1, %ymm21
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %xmm7
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm7[0],xmm1[0],xmm7[1],xmm1[1],xmm7[2],xmm1[2],xmm7[3],xmm1[3],xmm7[4],xmm1[4],xmm7[5],xmm1[5],xmm7[6],xmm1[6],xmm7[7],xmm1[7]
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm7, %xmm13
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm1, %xmm14
+; AVX512F-SLOW-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm1 = ymm8[8],ymm3[8],ymm8[9],ymm3[9],ymm8[10],ymm3[10],ymm8[11],ymm3[11],ymm8[12],ymm3[12],ymm8[13],ymm3[13],ymm8[14],ymm3[14],ymm8[15],ymm3[15],ymm8[24],ymm3[24],ymm8[25],ymm3[25],ymm8[26],ymm3[26],ymm8[27],ymm3[27],ymm8[28],ymm3[28],ymm8[29],ymm3[29],ymm8[30],ymm3[30],ymm8[31],ymm3[31]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm1[2,3,2,3],zmm0[0,1,0,1]
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm8
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm9
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm9[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm9[30],zero,ymm9[28],zero,zero,zero,zero,ymm9[31],zero,ymm9[29],zero,zero
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm9, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,128,25,128,23,128,128,128,128,26,128,24,128,128,128,128>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm8, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm10
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm3, %xmm1, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm3, %xmm18
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm1, %xmm24
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm10, %xmm3
+; AVX512F-SLOW-NEXT:    vpor %xmm0, %xmm3, %xmm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm11
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm11[30],zero,ymm11[28],zero,zero,zero,zero,ymm11[31],zero,ymm11[29],zero,zero,zero
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm15
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm11
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm3, %xmm11, %xmm7
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm15, %xmm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm15, %xmm22
+; AVX512F-SLOW-NEXT:    vpor %xmm7, %xmm8, %xmm7
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm15 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm15, %xmm14, %xmm7
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm14 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm14, %xmm13, %xmm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm14, %xmm23
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm13, %xmm17
+; AVX512F-SLOW-NEXT:    vpor %xmm7, %xmm8, %xmm7
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm7
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm7, %ymm6
+; AVX512F-SLOW-NEXT:    vmovdqa %ymm7, %ymm13
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm7
+; AVX512F-SLOW-NEXT:    vpshufb %ymm5, %ymm7, %ymm5
+; AVX512F-SLOW-NEXT:    vpor %ymm6, %ymm5, %ymm5
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm5
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm5, %ymm5
 ; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %ymm6
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm4, %ymm6, %ymm3
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm3, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm6, %ymm2
+; AVX512F-SLOW-NEXT:    vpor %ymm5, %ymm2, %ymm2
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %ymm5
+; AVX512F-SLOW-NEXT:    vpshufb %ymm9, %ymm5, %ymm2
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm5, %ymm29
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %ymm5
 ; AVX512F-SLOW-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm5, %ymm1
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm5[18],zero,zero,zero,zero,ymm5[21],zero,ymm5[19],zero,zero,zero,zero,ymm5[22],zero,ymm5[20]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm7[23],zero,ymm7[21,22,23,26],zero,ymm7[24],zero,ymm7[28,29,26,27]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,ymm7[18,19,20,21],zero,ymm7[19],zero,ymm7[25,26,27,22],zero,ymm7[20],zero
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm7, %ymm26
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm4, %ymm15, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm13[21],zero,ymm13[19],zero,zero,zero,zero,ymm13[22],zero,ymm13[20],zero,zero
+; AVX512F-SLOW-NEXT:    vpshufb %ymm4, %ymm5, %ymm4
+; AVX512F-SLOW-NEXT:    vporq %ymm2, %ymm4, %ymm31
+; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm12
+; AVX512F-SLOW-NEXT:    vpshufb %xmm3, %xmm12, %xmm2
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm3
+; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm3, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm3, %xmm19
+; AVX512F-SLOW-NEXT:    vporq %xmm2, %xmm0, %xmm20
+; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm4
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm18, %xmm0
+; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm9
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm9, %xmm1
+; AVX512F-SLOW-NEXT:    vporq %xmm0, %xmm1, %xmm30
+; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm14
+; AVX512F-SLOW-NEXT:    vpshufb %xmm15, %xmm14, %xmm0
+; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm8
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm23, %xmm1
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm8, %xmm1
+; AVX512F-SLOW-NEXT:    vporq %xmm0, %xmm1, %xmm25
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm7[30],zero,ymm7[28],zero,zero,zero,zero,ymm7[31],zero,ymm7[29],zero,zero
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm7, %ymm28
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm13[25],zero,ymm13[23],zero,zero,zero,zero,ymm13[26],zero,ymm13[24],zero,zero,zero,zero
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm13, %ymm18
 ; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm14 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm14, %xmm30
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm14 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm14, %xmm2, %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm14, %xmm21
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm2, %xmm27
-; AVX512F-SLOW-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %ymm2
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm15 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm15, %zmm0
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm0, %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm3, %xmm3
-; AVX512F-SLOW-NEXT:    vporq %xmm2, %xmm3, %xmm22
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r9), %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%r8), %xmm4
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm14 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm14, %xmm3, %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm3, %xmm19
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm5 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm5, %xmm4, %xmm3
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm4, %xmm17
-; AVX512F-SLOW-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm25, %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm8, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm28, %ymm3
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm9, %ymm3
-; AVX512F-SLOW-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm16, %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm11, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm11, %ymm15
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm18, %ymm3
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm6, %ymm3
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm6, %ymm13
-; AVX512F-SLOW-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm23, %ymm2
-; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm10, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm10, %ymm25
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm29, %ymm3
-; AVX512F-SLOW-NEXT:    vpshufb %ymm3, %ymm12, %ymm3
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm12, %ymm29
-; AVX512F-SLOW-NEXT:    vpor %ymm2, %ymm3, %ymm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rsi), %xmm2
-; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm2, %xmm16
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdi), %xmm7
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm7, %xmm0
-; AVX512F-SLOW-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%rcx), %xmm11
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm30, %xmm0
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm11, %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa (%rdx), %xmm8
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm21, %xmm0
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm8, %xmm4
-; AVX512F-SLOW-NEXT:    vpor %xmm1, %xmm4, %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa (%r9), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %xmm14, %xmm0, %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqa (%r8), %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %xmm5, %xmm2, %xmm5
-; AVX512F-SLOW-NEXT:    vpor %xmm1, %xmm5, %xmm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm24, %zmm0, %zmm1
-; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = 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]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm5, %xmm5
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm0, %xmm20
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm5[0,1,0,1],zmm1[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm21, %zmm0, %zmm1
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm15 = xmm8[8],xmm14[8],xmm8[9],xmm14[9],xmm8[10],xmm14[10],xmm8[11],xmm14[11],xmm8[12],xmm14[12],xmm8[13],xmm14[13],xmm8[14],xmm14[14],xmm8[15],xmm14[15]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm7 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm7, %xmm15, %xmm15
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm15[0,1,0,1],zmm1[4,5,6,7]
 ; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
 ; AVX512F-SLOW-NEXT:    vmovdqa (%rax), %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,5,6]
 ; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm9 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; AVX512F-SLOW-NEXT:    vpandn %ymm1, %ymm9, %ymm9
-; AVX512F-SLOW-NEXT:    vmovdqa (%rax), %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm14 = zero,ymm0[13],zero,zero,zero,zero,zero,zero,ymm0[14],zero,zero,zero,zero,zero,zero,ymm0[15],zero,zero,zero,zero,zero,zero,ymm0[16],zero,zero,zero,zero,zero,zero,ymm0[17],zero,zero
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm9, %zmm21
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm26, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm0[30],zero,ymm0[28],zero,zero,zero,zero,ymm0[31],zero,ymm0[29],zero,zero
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm30
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u>
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm6, %ymm12
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm28
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; AVX512F-SLOW-NEXT:    vpandn %ymm1, %ymm15, %ymm15
+; AVX512F-SLOW-NEXT:    vmovdqa (%rax), %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm1, %ymm13
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm15, %zmm21
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm24, %xmm1
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm10[0],xmm1[0],xmm10[1],xmm1[1],xmm10[2],xmm1[2],xmm10[3],xmm1[3],xmm10[4],xmm1[4],xmm10[5],xmm1[5],xmm10[6],xmm1[6],xmm10[7],xmm1[7]
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm13 = xmm1[8],xmm10[8],xmm1[9],xmm10[9],xmm1[10],xmm10[10],xmm1[11],xmm10[11],xmm1[12],xmm10[12],xmm1[13],xmm10[13],xmm1[14],xmm10[14],xmm1[15],xmm10[15]
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm15 = xmm4[8],xmm9[8],xmm4[9],xmm9[9],xmm4[10],xmm9[10],xmm4[11],xmm9[11],xmm4[12],xmm9[12],xmm4[13],xmm9[13],xmm4[14],xmm9[14],xmm4[15],xmm9[15]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = <6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm15, %xmm3
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm13, %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm1, %zmm3, %zmm27
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm22, %xmm1
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3],xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm11 = xmm11[8],xmm1[8],xmm11[9],xmm1[9],xmm11[10],xmm1[10],xmm11[11],xmm1[11],xmm11[12],xmm1[12],xmm11[13],xmm1[13],xmm11[14],xmm1[14],xmm11[15],xmm1[15]
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm19, %xmm15
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm13 = xmm12[8],xmm15[8],xmm12[9],xmm15[9],xmm12[10],xmm15[10],xmm12[11],xmm15[11],xmm12[12],xmm15[12],xmm12[13],xmm15[13],xmm12[14],xmm15[14],xmm12[15],xmm15[15]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm13, %xmm10
+; AVX512F-SLOW-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %xmm1, %xmm11, %xmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm10 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm1, %zmm10, %zmm26
+; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm17, %xmm10
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm10, %xmm11 # 16-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # xmm11 = xmm10[8],mem[8],xmm10[9],mem[9],xmm10[10],mem[10],xmm10[11],mem[11],xmm10[12],mem[12],xmm10[13],mem[13],xmm10[14],mem[14],xmm10[15],mem[15]
+; AVX512F-SLOW-NEXT:    vpshufb %xmm7, %xmm11, %xmm7
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm1 = zmm1[0,1,0,1],zmm7[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm29, %ymm1
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm7 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm7 = ymm1[0],mem[0],ymm1[1],mem[1],ymm1[2],mem[2],ymm1[3],mem[3],ymm1[4],mem[4],ymm1[5],mem[5],ymm1[6],mem[6],ymm1[7],mem[7],ymm1[16],mem[16],ymm1[17],mem[17],ymm1[18],mem[18],ymm1[19],mem[19],ymm1[20],mem[20],ymm1[21],mem[21],ymm1[22],mem[22],ymm1[23],mem[23]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = <8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm7, %ymm7
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm7 = zmm31[0,1,2,3],zmm7[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm2, %ymm6
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm2 = ymm2[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,3,2]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; AVX512F-SLOW-NEXT:    vpandn %ymm2, %ymm7, %ymm2
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm6, %zmm16
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm15[0],xmm12[0],xmm15[1],xmm12[1],xmm15[2],xmm12[2],xmm15[3],xmm12[3],xmm15[4],xmm12[4],xmm15[5],xmm12[5],xmm15[6],xmm12[6],xmm15[7],xmm12[7]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm6 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm6, %xmm3, %xmm3
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm3, %ymm24
+; AVX512F-SLOW-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
+; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm20, %zmm2, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm9[0],xmm4[0],xmm9[1],xmm4[1],xmm9[2],xmm4[2],xmm9[3],xmm4[3],xmm9[4],xmm4[4],xmm9[5],xmm4[5],xmm9[6],xmm4[6],xmm9[7],xmm4[7]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm3 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
+; AVX512F-SLOW-NEXT:    vpshufb %xmm3, %xmm5, %xmm4
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm4, %ymm23
+; AVX512F-SLOW-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm30, %zmm2, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm8[0],xmm14[0],xmm8[1],xmm14[1],xmm8[2],xmm14[2],xmm8[3],xmm14[3],xmm8[4],xmm14[4],xmm8[5],xmm14[5],xmm8[6],xmm14[6],xmm8[7],xmm14[7]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm31 = zmm2[0,1,0,1],zmm25[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm0[1,1,0,0,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm3[0,1,2,0]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm5, %zmm30
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm11[0],ymm8[0],ymm11[1],ymm8[1],ymm11[2],ymm8[2],ymm11[3],ymm8[3],ymm11[4],ymm8[4],ymm11[5],ymm8[5],ymm11[6],ymm8[6],ymm11[7],ymm8[7],ymm11[16],ymm8[16],ymm11[17],ymm8[17],ymm11[18],ymm8[18],ymm11[19],ymm8[19],ymm11[20],ymm8[20],ymm11[21],ymm8[21],ymm11[22],ymm8[22],ymm11[23],ymm8[23]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm5, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm22
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm9[0],ymm12[0],ymm9[1],ymm12[1],ymm9[2],ymm12[2],ymm9[3],ymm12[3],ymm9[4],ymm12[4],ymm9[5],ymm12[5],ymm9[6],ymm12[6],ymm9[7],ymm12[7],ymm9[16],ymm12[16],ymm9[17],ymm12[17],ymm9[18],ymm12[18],ymm9[19],ymm12[19],ymm9[20],ymm12[20],ymm9[21],ymm12[21],ymm9[22],ymm12[22],ymm9[23],ymm12[23]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm5, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm17
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm18, %ymm7
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm28, %ymm15
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm15[0],ymm7[0],ymm15[1],ymm7[1],ymm15[2],ymm7[2],ymm15[3],ymm7[3],ymm15[4],ymm7[4],ymm15[5],ymm7[5],ymm15[6],ymm7[6],ymm15[7],ymm7[7],ymm15[16],ymm7[16],ymm15[17],ymm7[17],ymm15[18],ymm7[18],ymm15[19],ymm7[19],ymm15[20],ymm7[20],ymm15[21],ymm7[21],ymm15[22],ymm7[22],ymm15[23],ymm7[23]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm5, %ymm14
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
 ; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm4 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm4[27],zero,zero,zero,zero,ymm4[30],zero,ymm4[28],zero,zero,zero,zero,ymm4[31],zero,ymm4[29]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm12, %ymm10
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm19, %xmm9
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm17, %xmm1
-; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm1[0],xmm9[0],xmm1[1],xmm9[1],xmm1[2],xmm9[2],xmm1[3],xmm9[3],xmm1[4],xmm9[4],xmm1[5],xmm9[5],xmm1[6],xmm9[6],xmm1[7],xmm9[7]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm10[0,1,2,3],zmm12[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm10 = ymm13[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm13, %ymm26
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm10 = ymm10[2,2,3,3,6,6,7,7]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm12 = <9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u,9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u>
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm15, %ymm3
-; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm15, %ymm15
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm15, %zmm23
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm13 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm13[30],zero,ymm13[28],zero,zero,zero,zero,ymm13[31],zero,ymm13[29],zero,zero,zero
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm18
-; AVX512F-SLOW-NEXT:    vpshufb %ymm12, %ymm13, %ymm12
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm13 = ymm10[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm13 = ymm13[0,0,1,1,4,4,5,5]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm13, %zmm24
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm27, %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm14 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm13 = 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]
-; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm15 = xmm11[8],xmm8[8],xmm11[9],xmm8[9],xmm11[10],xmm8[10],xmm11[11],xmm8[11],xmm11[12],xmm8[12],xmm11[13],xmm8[13],xmm11[14],xmm8[14],xmm11[15],xmm8[15]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = <6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm15, %xmm2
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm13, %xmm0
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm0, %zmm2, %zmm27
-; AVX512F-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX512F-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
-; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm15 = 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]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm16, %xmm2
-; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm13 = xmm2[8],xmm7[8],xmm2[9],xmm7[9],xmm2[10],xmm7[10],xmm2[11],xmm7[11],xmm2[12],xmm7[12],xmm2[13],xmm7[13],xmm2[14],xmm7[14],xmm2[15],xmm7[15]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm0 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm13, %xmm12
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm12, %ymm31
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm15, %xmm0
-; AVX512F-SLOW-NEXT:    vinserti32x4 $2, %xmm0, %zmm22, %zmm22
-; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm15 = xmm1[8],xmm9[8],xmm1[9],xmm9[9],xmm1[10],xmm9[10],xmm1[11],xmm9[11],xmm1[12],xmm9[12],xmm1[13],xmm9[13],xmm1[14],xmm9[14],xmm1[15],xmm9[15]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %xmm20, %xmm0
-; AVX512F-SLOW-NEXT:    vpshufb %xmm0, %xmm15, %xmm15
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,0,1],zmm15[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u>
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm0, %ymm1
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm4[0],ymm2[0],ymm4[1],ymm2[1],ymm4[2],ymm2[2],ymm4[3],ymm2[3],ymm4[4],ymm2[4],ymm4[5],ymm2[5],ymm4[6],ymm2[6],ymm4[7],ymm2[7],ymm4[16],ymm2[16],ymm4[17],ymm2[17],ymm4[18],ymm2[18],ymm4[19],ymm2[19],ymm4[20],ymm2[20],ymm4[21],ymm2[21],ymm4[22],ymm2[22],ymm4[23],ymm2[23]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm5, %ymm0
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm25
 ; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm0, %ymm13
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[18],zero,zero,zero,zero,ymm0[21],zero,ymm0[19],zero,zero,zero,zero,ymm0[22],zero,ymm0[20]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm6 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm0[25],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm28 = ymm6[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm10[0],ymm0[0],ymm10[1],ymm0[1],ymm10[2],ymm0[2],ymm10[3],ymm0[3],ymm10[4],ymm0[4],ymm10[5],ymm0[5],ymm10[6],ymm0[6],ymm10[7],ymm0[7],ymm10[16],ymm0[16],ymm10[17],ymm0[17],ymm10[18],ymm0[18],ymm10[19],ymm0[19],ymm10[20],ymm0[20],ymm10[21],ymm0[21],ymm10[22],ymm0[22],ymm10[23],ymm0[23]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm10[23],zero,ymm10[23,24,25,26],zero,ymm10[24],zero,ymm10[30,31]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vporq %ymm28, %ymm3, %ymm1
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm1
+; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm28 = zmm0[2,3,2,3],zmm1[4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm2, %ymm5
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm7, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm20
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm0[23],zero,ymm0[21,22,23,26],zero,ymm0[24],zero,ymm0[28,29,26,27]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,ymm0[18,19,20,21],zero,ymm0[19],zero,ymm0[25,26,27,22],zero,ymm0[20],zero
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm4[30],zero,ymm4[28],zero,zero,zero,zero,ymm4[31],zero,ymm4[29],zero,zero
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm4, %ymm1
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm1, %ymm18
+; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm15, %ymm0
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm19
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm3[21],zero,ymm3[19],zero,zero,zero,zero,ymm3[22],zero,ymm3[20],zero,zero
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm0, %ymm17
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm29, %ymm15
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm28, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb %ymm0, %ymm15, %ymm9
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm6, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[25],zero,ymm6[23],zero,zero,zero,zero,ymm6[26],zero,ymm6[24],zero,zero
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm15, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25,128,23,128,23,24,25,26,128,24,128,30,31>
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm25, %ymm0
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm0, %ymm3
-; AVX512F-SLOW-NEXT:    vmovdqa %ymm4, %ymm15
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm4, %ymm4
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm4, %ymm29
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128>
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm15, %ymm4
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm4, %ymm28
-; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm0, %ymm0
-; AVX512F-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm6 = zero,ymm0[13],zero,zero,zero,zero,zero,zero,ymm0[14],zero,zero,zero,zero,zero,zero,ymm0[15],zero,zero,zero,zero,zero,zero,ymm0[16],zero,zero,zero,zero,zero,zero,ymm0[17],zero,zero
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm15 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm15 = ymm15[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,2,3,2]
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm25 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; AVX512F-SLOW-NEXT:    vpandnq %ymm15, %ymm25, %ymm15
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm15, %zmm6, %zmm6
-; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3],xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm15 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm15, %xmm5, %xmm5
-; AVX512F-SLOW-NEXT:    vpshufb %xmm15, %xmm7, %xmm7
-; AVX512F-SLOW-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm7, %zmm25 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm1[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm15 = ymm30[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm8[0],xmm11[0],xmm8[1],xmm11[1],xmm8[2],xmm11[2],xmm8[3],xmm11[3],xmm8[4],xmm11[4],xmm8[5],xmm11[5],xmm8[6],xmm11[6],xmm8[7],xmm11[7]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm8 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
-; AVX512F-SLOW-NEXT:    vpshufb %xmm8, %xmm14, %xmm14
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm18 = ymm18[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} ymm7 = ymm10[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[2,2,3,3,6,6,7,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm9[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm3[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm13[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm12[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm20[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm19[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm17[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm26, %ymm1
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm1[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,1,1,4,4,5,5]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
-; AVX512F-SLOW-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm4, %ymm4
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm4 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vporq %zmm4, %zmm0, %zmm0
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [18374966859431673855,18446463693966278655,18374966859431673855,18446463693966278655]
-; AVX512F-SLOW-NEXT:    vpand %ymm4, %ymm11, %ymm8
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm8, %zmm9, %zmm8
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm9 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vporq %zmm9, %zmm8, %zmm8
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm9 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm11 = zmm23[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm11
-; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm11
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm0, %zmm8, %zmm11
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm0 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm9 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vporq %zmm0, %zmm9, %zmm0
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm9 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm17 = zmm24[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm17
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm0, %zmm8, %zmm17
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm27[0,1,0,1,4,5,4,5]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm9 = zmm22[0,1,0,1,4,5,4,5]
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm0, %zmm8, %zmm9
-; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm4, %ymm16, %ymm15
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm1[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $236, %ymm4, %ymm10, %ymm0
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm15, %zmm1
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm7[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpternlogq $236, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm18, %ymm4
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm4, %zmm4
-; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm1, %zmm5, %zmm4
-; AVX512F-SLOW-NEXT:    vpor %ymm13, %ymm12, %ymm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm3, %zmm1
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
-; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm1, %zmm5, %zmm0
-; AVX512F-SLOW-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm20 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm31[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm5 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm7 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm0 = ymm9[8],ymm12[8],ymm9[9],ymm12[9],ymm9[10],ymm12[10],ymm9[11],ymm12[11],ymm9[12],ymm12[12],ymm9[13],ymm12[13],ymm9[14],ymm12[14],ymm9[15],ymm12[15],ymm9[24],ymm12[24],ymm9[25],ymm12[25],ymm9[26],ymm12[26],ymm9[27],ymm12[27],ymm9[28],ymm12[28],ymm9[29],ymm12[29],ymm9[30],ymm12[30],ymm9[31],ymm12[31]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u,9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm12, %ymm4
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm9, %ymm3
+; AVX512F-SLOW-NEXT:    vpshufb %ymm6, %ymm8, %ymm6
+; AVX512F-SLOW-NEXT:    vpshufb %ymm2, %ymm11, %ymm10
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm9 = ymm11[8],ymm8[8],ymm11[9],ymm8[9],ymm11[10],ymm8[10],ymm11[11],ymm8[11],ymm11[12],ymm8[12],ymm11[13],ymm8[13],ymm11[14],ymm8[14],ymm11[15],ymm8[15],ymm11[24],ymm8[24],ymm11[25],ymm8[25],ymm11[26],ymm8[26],ymm11[27],ymm8[27],ymm11[28],ymm8[28],ymm11[29],ymm8[29],ymm11[30],ymm8[30],ymm11[31],ymm8[31]
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u>
+; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm0[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm24[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm5[2,3,2,3]
 ; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; AVX512F-SLOW-NEXT:    vpunpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm8, %xmm8 # 16-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # xmm8 = xmm8[0],mem[0],xmm8[1],mem[1],xmm8[2],mem[2],xmm8[3],mem[3],xmm8[4],mem[4],xmm8[5],mem[5],xmm8[6],mem[6],xmm8[7],mem[7]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %xmm10
-; AVX512F-SLOW-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm8, %zmm8 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm8 = zmm8[0,1,0,1],mem[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm12 = xmm1[1,1,0,0,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm12 = xmm12[0,1,2,0]
-; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} xmm13 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm13, %xmm1, %xmm14
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm12, %zmm12
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm14 = xmm10[1,1,0,0,4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm14 = xmm14[0,1,2,0]
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm15 = mem[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm16 = ymm29[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm18 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm18 = mem[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm19 = ymm28[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm3 = ymm3[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,1,1,3,4,5,5,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpshufb %xmm13, %xmm10, %xmm13
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[0,0,1,0]
-; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,4,5,5,6]
-; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm10 = xmm10[2,2,3,3]
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm22 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm22 = mem[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %ymm24 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # ymm24 = mem[2,3,2,3]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2, %zmm2 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5, %zmm5 # 32-byte Folded Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm5
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm23[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm22[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm17[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpshufb %ymm15, %ymm9, %ymm9
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [18374966859431673855,18446463693966278655,18374966859431673855,18446463693966278655]
+; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm2, %ymm8, %ymm7
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm7, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm7 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm0, %zmm7, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm14, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm8, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm0, %zmm7, %zmm5
+; AVX512F-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = [18374967954648269055,71777218572844800,18374967954648269055,71777218572844800]
+; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm0, %ymm4, %ymm3
+; AVX512F-SLOW-NEXT:    vpternlogq $248, %ymm0, %ymm6, %ymm10
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm27[0,1,0,1,4,5,4,5]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm4 = zmm26[0,1,0,1,4,5,4,5]
+; AVX512F-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm17 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm0, %zmm17, %zmm4
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm12, %zmm0
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm18[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpor %ymm3, %ymm13, %ymm3
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm25[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm7, %zmm3
+; AVX512F-SLOW-NEXT:    vpternlogq $226, %zmm0, %zmm17, %zmm3
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm20[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpand %ymm2, %ymm0, %ymm0
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm19[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
+; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # zmm2 = mem[2,3,2,3,6,7,6,7]
+; AVX512F-SLOW-NEXT:    vporq %zmm2, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm9[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm10, %zmm2
+; AVX512F-SLOW-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm19 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm19 = mem[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # ymm8 = mem[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm18 = ymm9[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm7 # 32-byte Reload
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm10 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero
+; AVX512F-SLOW-NEXT:    vmovdqa64 %ymm29, %ymm6
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm12 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm6[23],zero,ymm6[23,24,25,26],zero,ymm6[24],zero,ymm6[30,31]
+; AVX512F-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm13 = ymm7[8],ymm6[8],ymm7[9],ymm6[9],ymm7[10],ymm6[10],ymm7[11],ymm6[11],ymm7[12],ymm6[12],ymm7[13],ymm6[13],ymm7[14],ymm6[14],ymm7[15],ymm6[15],ymm7[24],ymm6[24],ymm7[25],ymm6[25],ymm7[26],ymm6[26],ymm7[27],ymm6[27],ymm7[28],ymm6[28],ymm7[29],ymm6[29],ymm7[30],ymm6[30],ymm7[31],ymm6[31]
+; AVX512F-SLOW-NEXT:    vmovdqa 32(%rax), %xmm14
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm13 = ymm13[10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm14[1,1,0,0,4,5,6,7]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm15 = xmm15[0,1,2,0]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} xmm6 = xmm14[4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,0,1,0]
+; AVX512F-SLOW-NEXT:    vpshufhw {{.*#+}} xmm14 = xmm14[0,1,2,3,4,5,5,6]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} xmm14 = xmm14[2,2,3,3]
+; AVX512F-SLOW-NEXT:    vpshufb {{.*#+}} ymm9 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512F-SLOW-NEXT:    vpshuflw {{.*#+}} ymm7 = ymm11[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512F-SLOW-NEXT:    vpshufd {{.*#+}} ymm7 = ymm7[0,1,1,3,4,5,5,7]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
+; AVX512F-SLOW-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm2
+; AVX512F-SLOW-NEXT:    vpor %ymm10, %ymm12, %ymm0
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm0, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $184, %zmm2, %zmm17, %zmm0
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm19, %zmm2 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm8, %zmm8 # 32-byte Folded Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm8
 ; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
 ; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm21
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm21
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm14[0,0,1,0]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm7, %zmm2
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm2
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm2
-; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
-; AVX512F-SLOW-NEXT:    # zmm4 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm11, %zmm4
-; AVX512F-SLOW-NEXT:    vporq %ymm15, %ymm16, %ymm5
-; AVX512F-SLOW-NEXT:    vporq %ymm18, %ymm19, %ymm7
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm5
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm5 = zmm7[0,1,2,3],zmm5[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,3,2]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm3, %zmm1
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm1
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm17, %zmm1
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm10[0,1,0,1]
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm13, %zmm3
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm3
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm3
-; AVX512F-SLOW-NEXT:    vporq %ymm22, %ymm24, %ymm5
-; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm5
-; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm7 # 64-byte Reload
-; AVX512F-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm5 = zmm7[0,1,2,3],zmm5[4,5,6,7]
-; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm6
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm6
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm25[0,1,0,1,4,5,4,5]
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm5 = zmm20[0,1,0,1,4,5,4,5]
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm5
-; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm12[0,0,1,0,4,4,5,4]
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm0
-; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm0
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm21
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm15[0,0,1,0]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm18, %zmm2
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm8 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm8, %zmm2
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm2
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm14[0,1,0,1]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm6, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm6 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm1
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm1
+; AVX512F-SLOW-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-SLOW-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm16
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm16
+; AVX512F-SLOW-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # zmm4 = mem[0,1,0,1,4,5,4,5]
+; AVX512F-SLOW-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # zmm5 = mem[0,1,0,1,4,5,4,5]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm5
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} zmm4 = zmm30[0,0,1,0,4,4,5,4]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm31, %zmm4
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm4
+; AVX512F-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm7[2,2,3,2]
+; AVX512F-SLOW-NEXT:    vinserti64x4 $1, %ymm9, %zmm5, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm28, %zmm5
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm5
+; AVX512F-SLOW-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Folded Reload
+; AVX512F-SLOW-NEXT:    # zmm3 = mem[2,3,2,3,6,7,6,7]
+; AVX512F-SLOW-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm3
 ; AVX512F-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm0, (%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm6, 320(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm3, 256(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm1, 128(%rax)
-; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm4, 384(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm5, 128(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm4, (%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm16, 320(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm1, 256(%rax)
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm2, 192(%rax)
+; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm3, 384(%rax)
 ; AVX512F-SLOW-NEXT:    vmovdqa64 %zmm21, 64(%rax)
-; AVX512F-SLOW-NEXT:    addq $1368, %rsp # imm = 0x558
+; AVX512F-SLOW-NEXT:    addq $968, %rsp # imm = 0x3C8
 ; AVX512F-SLOW-NEXT:    vzeroupper
 ; AVX512F-SLOW-NEXT:    retq
 ;
 ; AVX512F-FAST-LABEL: store_i8_stride7_vf64:
 ; AVX512F-FAST:       # %bb.0:
-; AVX512F-FAST-NEXT:    subq $1432, %rsp # imm = 0x598
-; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %ymm2
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %ymm1
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm19
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero,zero,zero
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm17
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %ymm3
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %ymm8
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm3[30],zero,ymm3[28],zero,zero,zero,zero,ymm3[31],zero,ymm3[29],zero,zero,zero
-; AVX512F-FAST-NEXT:    vmovdqa %ymm3, %ymm10
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero,zero,zero,ymm8[27],zero,ymm8[25]
-; AVX512F-FAST-NEXT:    vmovdqu %ymm8, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %ymm2
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %ymm1
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[27],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29]
-; AVX512F-FAST-NEXT:    vmovdqa %ymm1, %ymm15
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[25],zero,ymm2[23],zero,zero,zero,zero,ymm2[26],zero,ymm2[24],zero,zero
-; AVX512F-FAST-NEXT:    vmovdqa %ymm2, %ymm6
-; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    subq $1384, %rsp # imm = 0x568
 ; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,ymm0[14],zero,zero,zero,zero,zero,zero,ymm0[15],zero,zero,zero,zero,zero,zero,ymm0[16],zero,zero,zero,zero,zero,zero,ymm0[17],zero,zero,zero,zero,zero,zero,ymm0[18]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm0, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,14],zero,ymm1[12,13,0,1,14,15],zero,ymm1[3,12,13,2,3,16],zero,ymm1[30,31,28,29,16,17],zero,ymm1[31,18,19,28,29,18],zero
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
 ; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm1, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm24
+; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm27
 ; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %ymm1
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm3, %ymm16
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
+; AVX512F-FAST-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm28
 ; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm9, %ymm1, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm25
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[13,u,u,u,u,u],zero,ymm1[14,u,u,u,u,u],zero,ymm1[15,u,u,u,u,u],zero,ymm1[16,u,u,u,u,u],zero,ymm1[17,u,u,u]
-; AVX512F-FAST-NEXT:    vporq %ymm0, %ymm1, %ymm22
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %xmm3
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm1, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm2, %xmm23
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm1, %xmm30
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm3, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm27
-; AVX512F-FAST-NEXT:    vpor %xmm0, %xmm1, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa (%r8), %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm2, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa %ymm2, %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm16
+; AVX512F-FAST-NEXT:    vmovdqa (%r9), %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <13,u,u,u,u,u,128,14,u,u,u,u,u,128,15,u,u,u,u,u,128,16,u,u,u,u,u,128,17,u,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm2, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm3, %ymm17
+; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vporq %ymm0, %ymm1, %ymm19
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %xmm14
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %xmm1
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm14[0],xmm1[1],xmm14[1],xmm1[2],xmm14[2],xmm1[3],xmm14[3],xmm1[4],xmm14[4],xmm1[5],xmm14[5],xmm1[6],xmm14[6],xmm1[7],xmm14[7]
+; AVX512F-FAST-NEXT:    vmovdqa %xmm1, %xmm13
+; AVX512F-FAST-NEXT:    vmovdqa %xmm14, %xmm7
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm1 = 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]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm1[2,3,2,3],zmm0[0,1,0,1]
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %xmm5
-; AVX512F-FAST-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %xmm8
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %xmm14
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm14, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm8, %xmm3
+; AVX512F-FAST-NEXT:    vpor %xmm2, %xmm3, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %xmm11
 ; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %xmm3
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm3, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm20
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm5, %xmm3
-; AVX512F-FAST-NEXT:    vporq %xmm0, %xmm3, %xmm21
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %xmm12
-; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %xmm11
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm7 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
-; AVX512F-FAST-NEXT:    vpshufb %xmm7, %xmm12, %xmm3
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm12, %xmm29
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
-; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm11, %xmm5
-; AVX512F-FAST-NEXT:    vpor %xmm3, %xmm5, %xmm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm3
-; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm3, %ymm3
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm16, %ymm5
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm19, %ymm12
-; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm12, %ymm5
-; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm5, %ymm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = zero,zero,zero,ymm10[14],zero,zero,zero,zero,zero,zero,ymm10[15],zero,zero,zero,zero,zero,zero,ymm10[16],zero,zero,zero,zero,zero,zero,ymm10[17],zero,zero,zero,zero,zero,zero,ymm10[18]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm10, %ymm18
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm8[0,1,14],zero,ymm8[12,13,0,1,14,15],zero,ymm8[3,12,13,2,3,16],zero,ymm8[30,31,28,29,16,17],zero,ymm8[31,18,19,28,29,18],zero
-; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm5, %ymm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %ymm9, %ymm15, %ymm3
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm6[13,u,u,u,u,u],zero,ymm6[14,u,u,u,u,u],zero,ymm6[15,u,u,u,u,u],zero,ymm6[16,u,u,u,u,u],zero,ymm6[17,u,u,u]
-; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm5, %ymm3
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm14
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm14, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %xmm3
-; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm3, %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm16
-; AVX512F-FAST-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %xmm10
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm23, %xmm1
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm10, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %xmm9
-; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm9, %xmm4
-; AVX512F-FAST-NEXT:    vpor %xmm1, %xmm4, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r9), %xmm2
-; AVX512F-FAST-NEXT:    vpshufb %xmm7, %xmm2, %xmm1
-; AVX512F-FAST-NEXT:    vmovdqa %xmm2, %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa (%r8), %xmm2
-; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm2, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa %xmm2, %xmm5
-; AVX512F-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vpor %xmm1, %xmm0, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm3, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm30
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
+; AVX512F-FAST-NEXT:    vpshufb %xmm5, %xmm11, %xmm3
+; AVX512F-FAST-NEXT:    vpor %xmm2, %xmm3, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm7, %xmm6
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm7, %xmm24
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm13, %xmm7
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm13, %xmm29
+; AVX512F-FAST-NEXT:    vporq %xmm6, %xmm7, %xmm26
+; AVX512F-FAST-NEXT:    vmovdqa (%rsi), %xmm9
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm9, %xmm4
+; AVX512F-FAST-NEXT:    vmovdqa (%rdi), %xmm6
+; AVX512F-FAST-NEXT:    vpshufb %xmm5, %xmm6, %xmm5
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm6, %xmm25
+; AVX512F-FAST-NEXT:    vporq %xmm4, %xmm5, %xmm20
+; AVX512F-FAST-NEXT:    vmovdqa (%rcx), %xmm15
+; AVX512F-FAST-NEXT:    vpshufb %xmm0, %xmm15, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm15, %xmm31
+; AVX512F-FAST-NEXT:    vmovdqa (%rdx), %xmm4
+; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm4, %xmm1
+; AVX512F-FAST-NEXT:    vporq %xmm0, %xmm1, %xmm22
+; AVX512F-FAST-NEXT:    vmovdqa (%r9), %xmm15
+; AVX512F-FAST-NEXT:    vpshufb %xmm2, %xmm15, %xmm0
+; AVX512F-FAST-NEXT:    vmovdqa %xmm15, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa (%r8), %xmm5
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm5, %xmm1
+; AVX512F-FAST-NEXT:    vporq %xmm0, %xmm1, %xmm21
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rcx), %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm27, %ymm0
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa %ymm1, %ymm7
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdx), %ymm15
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm28, %ymm1
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm15, %ymm1
+; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero,zero,zero,ymm0[27],zero,ymm0[25]
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %ymm23 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm23, %ymm1
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22],zero,ymm1[20],zero,zero
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rsi), %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm10, %ymm0, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %ymm12, %ymm1, %ymm1
+; AVX512F-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm24, %ymm7
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero,zero
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm7[18],zero,zero,zero,zero,ymm7[21],zero,ymm7[19],zero,zero,zero,zero,ymm7[22],zero,ymm7[20]
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r8), %ymm1
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm16, %ymm0
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa 32(%r9), %ymm6
+; AVX512F-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm1
+; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm6, %ymm1
+; AVX512F-FAST-NEXT:    vporq %ymm0, %ymm1, %ymm18
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm15[30],zero,ymm15[28],zero,zero,zero,zero,ymm15[31],zero,ymm15[29],zero,zero
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm7[25],zero,ymm7[23],zero,zero,zero,zero,ymm7[26],zero,ymm7[24],zero,zero,zero,zero
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm7, %ymm17
 ; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27>
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm2, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,ymm2[18,19,20,21],zero,ymm2[19],zero,ymm2[25,26,27,22],zero,ymm2[20],zero
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %ymm1
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm28
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm12, %zmm0
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm22, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm5[8],xmm4[8],xmm5[9],xmm4[9],xmm5[10],xmm4[10],xmm5[11],xmm4[11],xmm5[12],xmm4[12],xmm5[13],xmm4[13],xmm5[14],xmm4[14],xmm5[15],xmm4[15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm12 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
-; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm1, %xmm1
-; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm1[0,1,0,1],zmm0[4,5,6,7]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm19, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa %xmm2, %xmm6
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm5[8],xmm2[8],xmm5[9],xmm2[9],xmm5[10],xmm2[10],xmm5[11],xmm2[11],xmm5[12],xmm2[12],xmm5[13],xmm2[13],xmm5[14],xmm2[14],xmm5[15],xmm2[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm5, %xmm23
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm13 = <u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10>
+; AVX512F-FAST-NEXT:    vpshufb %xmm13, %xmm12, %xmm12
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm12[0,1,0,1],zmm0[4,5,6,7]
 ; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [2,2,3,3,2,2,3,3]
-; AVX512F-FAST-NEXT:    # ymm6 = mem[0,1,0,1]
-; AVX512F-FAST-NEXT:    vmovdqa (%rax), %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} xmm1 = xmm0[0,1,2,3,4,5,5,6]
-; AVX512F-FAST-NEXT:    vpermd %ymm1, %ymm6, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
-; AVX512F-FAST-NEXT:    vpandn %ymm1, %ymm4, %ymm4
+; AVX512F-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm12 = [2,2,3,3,2,2,3,3]
+; AVX512F-FAST-NEXT:    # ymm12 = mem[0,1,0,1]
+; AVX512F-FAST-NEXT:    vmovdqa (%rax), %xmm1
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,5,6]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm12, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm16 = [255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255]
+; AVX512F-FAST-NEXT:    vpandnq %ymm0, %ymm16, %ymm16
 ; AVX512F-FAST-NEXT:    vmovdqa (%rax), %ymm0
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [128,13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128]
-; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm0, %ymm8
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm26
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm4, %zmm22
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm31
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm0[13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,29,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm25, %ymm1
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[27],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm24 = [18374967954648269055,71777218572844800,18374967954648269055,71777218572844800]
-; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm24, %ymm8, %ymm15
-; AVX512F-FAST-NEXT:    vmovdqa %xmm11, %xmm4
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm29, %xmm2
-; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm8 = xmm11[0],xmm2[0],xmm11[1],xmm2[1],xmm11[2],xmm2[2],xmm11[3],xmm2[3],xmm11[4],xmm2[4],xmm11[5],xmm2[5],xmm11[6],xmm2[6],xmm11[7],xmm2[7]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm29 = zmm15[0,1,2,3],zmm8[0,1,0,1]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm27, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm30, %xmm1
-; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm3, %xmm28
-; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm1 = 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]
-; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm10[8],xmm9[8],xmm10[9],xmm9[9],xmm10[10],xmm9[10],xmm10[11],xmm9[11],xmm10[12],xmm9[12],xmm10[13],xmm9[13],xmm10[14],xmm9[14],xmm10[15],xmm9[15]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm9, %xmm27
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm10, %xmm25
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = zero,ymm0[13],zero,zero,zero,zero,zero,zero,ymm0[14],zero,zero,zero,zero,zero,zero,ymm0[15],zero,zero,zero,zero,zero,zero,ymm0[16],zero,zero,zero,zero,zero,zero,ymm0[17],zero,zero
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm10, %zmm16, %zmm19
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm7 = xmm8[0],xmm14[0],xmm8[1],xmm14[1],xmm8[2],xmm14[2],xmm8[3],xmm14[3],xmm8[4],xmm14[4],xmm8[5],xmm14[5],xmm8[6],xmm14[6],xmm8[7],xmm14[7]
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm14[8],xmm8[8],xmm14[9],xmm8[9],xmm14[10],xmm8[10],xmm14[11],xmm8[11],xmm14[12],xmm8[12],xmm14[13],xmm8[13],xmm14[14],xmm8[14],xmm14[15],xmm8[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm31, %xmm14
+; AVX512F-FAST-NEXT:    vmovdqa %xmm4, %xmm5
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm14[8],xmm4[8],xmm14[9],xmm4[9],xmm14[10],xmm4[10],xmm14[11],xmm4[11],xmm14[12],xmm4[12],xmm14[13],xmm4[13],xmm14[14],xmm4[14],xmm14[15],xmm4[15]
 ; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = <6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7>
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm1, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm20, %xmm11
-; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm9 = xmm0[0],xmm11[0],xmm0[1],xmm11[1],xmm0[2],xmm11[2],xmm0[3],xmm11[3],xmm0[4],xmm11[4],xmm0[5],xmm11[5],xmm0[6],xmm11[6],xmm0[7],xmm11[7]
-; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm11[8],xmm0[8],xmm11[9],xmm0[9],xmm11[10],xmm0[10],xmm11[11],xmm0[11],xmm11[12],xmm0[12],xmm11[13],xmm0[13],xmm11[14],xmm0[14],xmm11[15],xmm0[15]
-; AVX512F-FAST-NEXT:    vmovdqa %xmm14, %xmm8
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm16, %xmm5
-; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm14[8],xmm5[8],xmm14[9],xmm5[9],xmm14[10],xmm5[10],xmm14[11],xmm5[11],xmm14[12],xmm5[12],xmm14[13],xmm5[13],xmm14[14],xmm5[14],xmm14[15],xmm5[15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm3, %xmm0
-; AVX512F-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm21, %zmm30
-; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm0 = xmm4[8],xmm2[8],xmm4[9],xmm2[9],xmm4[10],xmm2[10],xmm4[11],xmm2[11],xmm4[12],xmm2[12],xmm4[13],xmm2[13],xmm4[14],xmm2[14],xmm4[15],xmm2[15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u>
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm7, %ymm11
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm1
-; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm3
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm21 = ymm3[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm19, %ymm3
-; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm3, %ymm10
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm20 = ymm10[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm2[0,1,0,1],zmm0[0,1,0,1]
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %xmm10
-; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm10[0,1,2,3,4,5,5,6]
-; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm6, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm12 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[18],zero,zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22],zero,ymm1[20]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm18, %ymm0
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm0[9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u,25,u,23,u,u,u,u,26,u,24,u,u,u,u,27,u]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm19 = ymm7[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,ymm3[18,19,20,21],zero,ymm3[19],zero,ymm3[25,26,27,22],zero,ymm3[20],zero
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm16
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm0[21],zero,ymm0[19],zero,zero,zero,zero,ymm0[22],zero,ymm0[20],zero,zero
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm26, %ymm1
-; AVX512F-FAST-NEXT:    vpshufb %ymm1, %ymm0, %ymm7
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,5,4,5,5,7,4,5]
-; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm26 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
-; AVX512F-FAST-NEXT:    vpandnq %ymm0, %ymm26, %ymm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm7, %zmm7
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30,128,28,128,30,31,30,31,128,29,128,31,28,29>
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm15, %ymm13
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm17 = ymm13[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm14 = ymm0[13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,29,u,27,u,u,u,u,30,u,28,u,u,u,u,31,u]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm18 = ymm14[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm5[0],xmm8[0],xmm5[1],xmm8[1],xmm5[2],xmm8[2],xmm5[3],xmm8[3],xmm5[4],xmm8[4],xmm5[5],xmm8[5],xmm5[6],xmm8[6],xmm5[7],xmm8[7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm9, %xmm9
-; AVX512F-FAST-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 16-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm0[23],zero,ymm0[23,24,25,26],zero,ymm0[24],zero,ymm0[30,31]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm26
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm1[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm23, %ymm2
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm2[30],zero,ymm2[28],zero,zero,zero,zero,ymm2[31],zero,ymm2[29],zero,zero,zero
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm0[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm1, %ymm4
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm9[0,1,0,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm31 = ymm31[2,3,2,3]
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm27, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm25, %xmm5
-; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3],xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
-; AVX512F-FAST-NEXT:    vmovdqa64 %xmm28, %xmm5
-; AVX512F-FAST-NEXT:    vpshufb %xmm8, %xmm5, %xmm14
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[0,1,0,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm16[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqu %ymm2, (%rsp) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %xmm3, %xmm10, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti32x4 $2, %xmm2, %zmm3, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm30, %xmm3
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm11[0],xmm3[0],xmm11[1],xmm3[1],xmm11[2],xmm3[2],xmm11[3],xmm3[3],xmm11[4],xmm3[4],xmm11[5],xmm3[5],xmm11[6],xmm3[6],xmm11[7],xmm3[7]
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm3[8],xmm11[8],xmm3[9],xmm11[9],xmm3[10],xmm11[10],xmm3[11],xmm11[11],xmm3[12],xmm11[12],xmm3[13],xmm11[13],xmm3[14],xmm11[14],xmm3[15],xmm11[15]
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm25, %xmm8
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm10 = xmm9[8],xmm8[8],xmm9[9],xmm8[9],xmm9[10],xmm8[10],xmm9[11],xmm8[11],xmm9[12],xmm8[12],xmm9[13],xmm8[13],xmm9[14],xmm8[14],xmm9[15],xmm8[15]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = <2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u>
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm10, %xmm10
+; AVX512F-FAST-NEXT:    vmovdqu %ymm10, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm3, %xmm3
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti32x4 $2, %xmm3, %zmm4, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm24, %xmm3
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm29, %xmm4
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm4[8],xmm3[8],xmm4[9],xmm3[9],xmm4[10],xmm3[10],xmm4[11],xmm3[11],xmm4[12],xmm3[12],xmm4[13],xmm3[13],xmm4[14],xmm3[14],xmm4[15],xmm3[15]
+; AVX512F-FAST-NEXT:    vpshufb %xmm13, %xmm3, %xmm3
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm3 = zmm26[0,1,0,1],zmm3[0,1,0,1]
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa 32(%rax), %xmm3
+; AVX512F-FAST-NEXT:    vpshufhw {{.*#+}} xmm4 = xmm3[0,1,2,3,4,5,5,6]
+; AVX512F-FAST-NEXT:    vpermd %ymm4, %ymm12, %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [0,1,0,1,2,0,0,1]
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm10 = xmm3[1,1,0,0,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm10, %ymm11, %ymm4
+; AVX512F-FAST-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm12 = xmm8[0],xmm9[0],xmm8[1],xmm9[1],xmm8[2],xmm9[2],xmm8[3],xmm9[3],xmm8[4],xmm9[4],xmm8[5],xmm9[5],xmm8[6],xmm9[6],xmm8[7],xmm9[7]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm2, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm31
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm12, %xmm2
+; AVX512F-FAST-NEXT:    vinserti32x4 $2, %xmm20, %zmm2, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm5[0],xmm14[0],xmm5[1],xmm14[1],xmm5[2],xmm14[2],xmm5[3],xmm14[3],xmm5[4],xmm14[4],xmm5[5],xmm14[5],xmm5[6],xmm14[6],xmm5[7],xmm14[7]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm12 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
+; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm7, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm30
+; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm4, %xmm4
+; AVX512F-FAST-NEXT:    vinserti32x4 $2, %xmm22, %zmm4, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %xmm23, %xmm2
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm2[0],xmm6[0],xmm2[1],xmm6[1],xmm2[2],xmm6[2],xmm2[3],xmm6[3],xmm2[4],xmm6[4],xmm2[5],xmm6[5],xmm2[6],xmm6[6],xmm2[7],xmm6[7]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm4[0,1,0,1],zmm21[0,1,0,1]
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm1[1,1,0,0,4,5,6,7]
+; AVX512F-FAST-NEXT:    vpermd %ymm4, %ymm11, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm4 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm3, %xmm2
+; AVX512F-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vpshufb %xmm4, %xmm1, %xmm1
+; AVX512F-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm8[25],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm2[23],zero,ymm2[23,24,25,26],zero,ymm2[24],zero,ymm2[30,31]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpshufb %xmm8, %xmm0, %xmm0
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,128,19,128,21,20,21,22,128,20,128,22,23>
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm15, %ymm15
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
-; AVX512F-FAST-NEXT:    vinserti32x4 $2, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm23 # 16-byte Folded Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm2[9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u,25,u,23,u,u,u,u,26,u,24,u,u,u,u,27,u]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpshufb %ymm8, %ymm1, %ymm2
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpor %ymm6, %ymm4, %ymm1
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm9, %zmm1, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [18374966859431673855,18446463693966278655,18374966859431673855,18446463693966278655]
-; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm4, %ymm11, %ymm31
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm31, %zmm9
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm11 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm1, %zmm11, %zmm9
-; AVX512F-FAST-NEXT:    vpor %ymm5, %ymm12, %ymm1
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm5, %zmm1
-; AVX512F-FAST-NEXT:    vpor %ymm3, %ymm15, %ymm3
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm5, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm1, %zmm11, %zmm3
-; AVX512F-FAST-NEXT:    vpandq %ymm4, %ymm21, %ymm1
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm20, %zmm1
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm4 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm4, %zmm1, %zmm1
-; AVX512F-FAST-NEXT:    vpandq %ymm24, %ymm19, %ymm4
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm17, %zmm4, %zmm4
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm5 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm5, %zmm4, %zmm4
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm4
-; AVX512F-FAST-NEXT:    vpandq %ymm24, %ymm18, %ymm1
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm13, %zmm1
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm5 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm5, %zmm1, %zmm1
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
-; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm4, %zmm5, %zmm1
-; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm4 = mem[0,1,0,1,4,5,4,5]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm11 = zmm30[0,1,0,1,4,5,4,5]
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm4, %zmm5, %zmm11
-; AVX512F-FAST-NEXT:    vpandq %ymm24, %ymm0, %ymm0
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm2 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm2, %zmm0, %zmm0
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm2 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm2 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm4 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vporq %zmm2, %zmm4, %zmm2
-; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm5, %zmm2
-; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX512F-FAST-NEXT:    vpunpcklbw {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 16-byte Folded Reload
-; AVX512F-FAST-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1],xmm0[2],mem[2],xmm0[3],mem[3],xmm0[4],mem[4],xmm0[5],mem[5],xmm0[6],mem[6],xmm0[7],mem[7]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512F-FAST-NEXT:    vshufi64x2 $0, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm16 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm16 = zmm0[0,1,0,1],mem[0,1,0,1]
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm10[1,1,0,0,4,5,6,7]
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,1,0,1,2,0,0,1]
-; AVX512F-FAST-NEXT:    vpermd %ymm4, %ymm5, %ymm18
-; AVX512F-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} xmm12 = xmm0[1,1,0,0,4,5,6,7]
-; AVX512F-FAST-NEXT:    vpermd %ymm12, %ymm5, %ymm17
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} xmm12 = [4,5,4,5,4,5,8,9,6,7,6,7,6,7,6,7]
-; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm10, %xmm10
-; AVX512F-FAST-NEXT:    vpshufb %xmm12, %xmm0, %xmm12
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm13 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,20,128,18,128,128,128,128,21,128,19,128,128,128,128,22>
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm0, %ymm14
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm15 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm0[25],zero,ymm0[23],zero,zero,zero,zero,ymm0[26],zero,ymm0[24],zero,zero
-; AVX512F-FAST-NEXT:    vpshufb %ymm13, %ymm0, %ymm13
-; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,128,18,128,20,21,20,21,128,19,128,19,20,21,22,128>
-; AVX512F-FAST-NEXT:    vmovdqa64 %ymm26, %ymm4
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm4, %ymm5
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm2[0],ymm8[0],ymm2[1],ymm8[1],ymm2[2],ymm8[2],ymm2[3],ymm8[3],ymm2[4],ymm8[4],ymm2[5],ymm8[5],ymm2[6],ymm8[6],ymm2[7],ymm8[7],ymm2[16],ymm8[16],ymm2[17],ymm8[17],ymm2[18],ymm8[18],ymm2[19],ymm8[19],ymm2[20],ymm8[20],ymm2[21],ymm8[21],ymm2[22],ymm8[22],ymm2[23],ymm8[23]
+; AVX512F-FAST-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm1
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13>
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm4, %ymm3
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm16 = zmm3[2,3,2,3],zmm1[4,5,6,7]
 ; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm6[23],zero,ymm6[23,24,25,26],zero,ymm6[24],zero,ymm6[30,31]
-; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm8
-; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm6 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
-; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm19 = [4,5,4,5,5,7,4,5]
-; AVX512F-FAST-NEXT:    vpermd %ymm6, %ymm19, %ymm19
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
-; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
-; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm20 # 64-byte Folded Reload
-; AVX512F-FAST-NEXT:    # zmm20 = mem[2,3,2,3,6,7,6,7]
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm20
-; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm1 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm21 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    # ymm21 = mem[0,1,0,1]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[0,0,1,0]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,3,2,3]
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm8 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm6[0],ymm8[0],ymm6[1],ymm8[1],ymm6[2],ymm8[2],ymm6[3],ymm8[3],ymm6[4],ymm8[4],ymm6[5],ymm8[5],ymm6[6],ymm8[6],ymm6[7],ymm8[7],ymm6[16],ymm8[16],ymm6[17],ymm8[17],ymm6[18],ymm8[18],ymm6[19],ymm8[19],ymm6[20],ymm8[20],ymm6[21],ymm8[21],ymm6[22],ymm8[22],ymm6[23],ymm8[23]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9>
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm1, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm29
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm17, %ymm11
+; AVX512F-FAST-NEXT:    vmovdqa %ymm15, %ymm13
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm15[0],ymm11[0],ymm15[1],ymm11[1],ymm15[2],ymm11[2],ymm15[3],ymm11[3],ymm15[4],ymm11[4],ymm15[5],ymm11[5],ymm15[6],ymm11[6],ymm15[7],ymm11[7],ymm15[16],ymm11[16],ymm15[17],ymm11[17],ymm15[18],ymm11[18],ymm15[19],ymm11[19],ymm15[20],ymm11[20],ymm15[21],ymm11[21],ymm15[22],ymm11[22],ymm15[23],ymm11[23]
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm1, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm24
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm3 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm10 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm3[0],ymm10[0],ymm3[1],ymm10[1],ymm3[2],ymm10[2],ymm3[3],ymm10[3],ymm3[4],ymm10[4],ymm3[5],ymm10[5],ymm3[6],ymm10[6],ymm3[7],ymm10[7],ymm3[16],ymm10[16],ymm3[17],ymm10[17],ymm3[18],ymm10[18],ymm3[19],ymm10[19],ymm3[20],ymm10[20],ymm3[21],ymm10[21],ymm3[22],ymm10[22],ymm3[23],ymm10[23]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm2, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm25
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm9 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm7 = ymm12[0],ymm9[0],ymm12[1],ymm9[1],ymm12[2],ymm9[2],ymm12[3],ymm9[3],ymm12[4],ymm9[4],ymm12[5],ymm9[5],ymm12[6],ymm9[6],ymm12[7],ymm9[7],ymm12[16],ymm9[16],ymm12[17],ymm9[17],ymm12[18],ymm9[18],ymm12[19],ymm9[19],ymm12[20],ymm9[20],ymm12[21],ymm9[21],ymm12[22],ymm9[22],ymm12[23],ymm9[23]
+; AVX512F-FAST-NEXT:    vpshufb %ymm4, %ymm7, %ymm1
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm1, %ymm23
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm15 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm14 # 32-byte Reload
+; AVX512F-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm15[0],ymm14[0],ymm15[1],ymm14[1],ymm15[2],ymm14[2],ymm15[3],ymm14[3],ymm15[4],ymm14[4],ymm15[5],ymm14[5],ymm15[6],ymm14[6],ymm15[7],ymm14[7],ymm15[16],ymm14[16],ymm15[17],ymm14[17],ymm15[18],ymm14[18],ymm15[19],ymm14[19],ymm15[20],ymm14[20],ymm15[21],ymm14[21],ymm15[22],ymm14[22],ymm15[23],ymm14[23]
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm4, %ymm1
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,26,27,28,29,30,31,30,31,28,29,28,29,30,31]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm27
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,26,27,24,25,22,23,24,25,26,27,26,27,24,25]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm26
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm0 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,5,4,5,5,7,4,5]
+; AVX512F-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm18[0,1,2,3],zmm1[2,3,2,3]
+; AVX512F-FAST-NEXT:    vmovdqu64 %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm28, %ymm0
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = zero,ymm0[13],zero,zero,zero,zero,zero,zero,ymm0[14],zero,zero,zero,zero,zero,zero,ymm0[15],zero,zero,zero,zero,zero,zero,ymm0[16],zero,zero,zero,zero,zero,zero,ymm0[17],zero,zero
+; AVX512F-FAST-NEXT:    vpshuflw {{.*#+}} ymm5 = ymm0[2,1,1,2,4,5,6,7,10,9,9,10,12,13,14,15]
+; AVX512F-FAST-NEXT:    vpermd %ymm5, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255,0,255,255,255,255,255,255]
+; AVX512F-FAST-NEXT:    vpandn %ymm2, %ymm5, %ymm2
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm8[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm8[25],zero,ymm8[23],zero,zero,zero,zero,ymm8[26],zero,ymm8[24],zero,zero,zero,zero
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm22
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm8, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm17
+; AVX512F-FAST-NEXT:    vpshufb %ymm5, %ymm11, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm28
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29],zero,zero
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,128,23,128,21,22,23,26,128,24,128,28,29,26,27>
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm2
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm2, %ymm18
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm13, %ymm0
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm0, %ymm21
+; AVX512F-FAST-NEXT:    vmovdqa %ymm3, %ymm4
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm3[8],ymm10[8],ymm3[9],ymm10[9],ymm3[10],ymm10[10],ymm3[11],ymm10[11],ymm3[12],ymm10[12],ymm3[13],ymm10[13],ymm3[14],ymm10[14],ymm3[15],ymm10[15],ymm3[24],ymm10[24],ymm3[25],ymm10[25],ymm3[26],ymm10[26],ymm3[27],ymm10[27],ymm3[28],ymm10[28],ymm3[29],ymm10[29],ymm3[30],ymm10[30],ymm3[31],ymm10[31]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u,9,u,7,u,u,u,u,10,u,8,u,u,u,u,11,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm10, %ymm5
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25>
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm4, %ymm4
+; AVX512F-FAST-NEXT:    vpshufb %ymm3, %ymm9, %ymm10
+; AVX512F-FAST-NEXT:    vpshufb %ymm0, %ymm12, %ymm6
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm0 = ymm12[8],ymm9[8],ymm12[9],ymm9[9],ymm12[10],ymm9[10],ymm12[11],ymm9[11],ymm12[12],ymm9[12],ymm12[13],ymm9[13],ymm12[14],ymm9[14],ymm12[15],ymm9[15],ymm12[24],ymm9[24],ymm12[25],ymm9[25],ymm12[26],ymm9[26],ymm12[27],ymm9[27],ymm12[28],ymm9[28],ymm12[29],ymm9[29],ymm12[30],ymm9[30],ymm12[31],ymm9[31]
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u>
+; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm2, %ymm2
+; AVX512F-FAST-NEXT:    vpshufb %ymm7, %ymm0, %ymm11
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm13 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm14[25],zero,ymm14[23],zero,zero,zero,zero,ymm14[26],zero,ymm14[24],zero,zero
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm15[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,ymm15[23],zero,ymm15[23,24,25,26],zero,ymm15[24],zero,ymm15[30,31]
+; AVX512F-FAST-NEXT:    vmovdqa64 %ymm7, %ymm20
+; AVX512F-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm7 = ymm14[8],ymm15[8],ymm14[9],ymm15[9],ymm14[10],ymm15[10],ymm14[11],ymm15[11],ymm14[12],ymm15[12],ymm14[13],ymm15[13],ymm14[14],ymm15[14],ymm14[15],ymm15[15],ymm14[24],ymm15[24],ymm14[25],ymm15[25],ymm14[26],ymm15[26],ymm14[27],ymm15[27],ymm14[28],ymm15[28],ymm14[29],ymm15[29],ymm14[30],ymm15[30],ymm14[31],ymm15[31]
+; AVX512F-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11,26,23,22,u,u,u,29,28,25,24,u,u,u,31,30,27]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm2[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm31[0,1,0,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm9 = ymm17[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm30[0,1,0,1]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm12 = ymm12[0,0,1,0]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm15[2,3,2,3]
 ; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm13 = ymm13[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[2,3,2,3]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm1, %zmm1 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm21, %zmm21 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm21
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm1 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm22
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm21, %zmm22
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm18, %zmm6, %zmm1
-; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm29
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm29
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm10, %zmm1 # 32-byte Folded Reload
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm6 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm1
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm11, %zmm1
-; AVX512F-FAST-NEXT:    vpor %ymm5, %ymm14, %ymm5
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm0, %zmm5
-; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm6 # 64-byte Reload
-; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm5 = zmm6[0,1,2,3],zmm5[4,5,6,7]
-; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm7
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm7
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm24[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm23[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm10[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [18374966859431673855,18446463693966278655,18374966859431673855,18446463693966278655]
+; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm3, %ymm9, %ymm8
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm2, %zmm8, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm8, %zmm2
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm15, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm9 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm9, %zmm9
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm8, %zmm9
+; AVX512F-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [18374967954648269055,71777218572844800,18374967954648269055,71777218572844800]
+; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm0, %ymm5, %ymm4
+; AVX512F-FAST-NEXT:    vpternlogq $248, %ymm0, %ymm10, %ymm6
+; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    # zmm0 = mem[0,1,0,1,4,5,4,5]
+; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    # zmm5 = mem[0,1,0,1,4,5,4,5]
+; AVX512F-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm8 = [255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255,255,255,255,255,0,0,255]
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm8, %zmm5
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm25[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm22[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm18[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm10, %ymm4
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm29[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm10, %zmm4
+; AVX512F-FAST-NEXT:    vpternlogq $226, %zmm0, %zmm8, %zmm4
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm28[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm21[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    # zmm3 = mem[2,3,2,3,6,7,6,7]
+; AVX512F-FAST-NEXT:    vporq %zmm3, %zmm0, %zmm0
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm11[2,3,2,3]
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm6, %zmm3
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm3
+; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm0 = mem[0,1,0,1]
+; AVX512F-FAST-NEXT:    vpermq $68, (%rsp), %ymm6 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm10 = ymm27[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm11 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm11 = mem[0,0,1,0]
+; AVX512F-FAST-NEXT:    vpermq $16, {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    # ymm12 = mem[0,0,1,0]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm26[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm15 = ymm13[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm17 = ymm20[2,3,2,3]
+; AVX512F-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[2,3,2,3]
+; AVX512F-FAST-NEXT:    vporq %ymm15, %ymm17, %ymm15
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm15, %zmm7
+; AVX512F-FAST-NEXT:    vpternlogq $184, %zmm3, %zmm8, %zmm7
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm0, %zmm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm6, %zmm3 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm19
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm19
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm10, %zmm0 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm0
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm11, %zmm2 # 32-byte Folded Reload
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm2
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm2
 ; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Folded Reload
 ; AVX512F-FAST-NEXT:    # zmm3 = mem[0,1,0,1,4,5,4,5]
-; AVX512F-FAST-NEXT:    vpermq {{.*#+}} zmm5 = zmm23[0,1,0,1,4,5,4,5]
+; AVX512F-FAST-NEXT:    vpermq $68, {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    # zmm5 = mem[0,1,0,1,4,5,4,5]
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm3, %zmm5
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm17, %zmm3
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm16, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm12, %zmm3, %zmm3
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm6 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm6, %zmm3
 ; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm3
-; AVX512F-FAST-NEXT:    vpor %ymm4, %ymm15, %ymm4
-; AVX512F-FAST-NEXT:    vpor %ymm13, %ymm8, %ymm5
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm4
-; AVX512F-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm4 = zmm5[0,1,2,3],zmm4[4,5,6,7]
-; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm19, %zmm0
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm0
-; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm5 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm5, %zmm5
+; AVX512F-FAST-NEXT:    vpternlogq $228, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm5, %zmm16
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm16
+; AVX512F-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Reload
+; AVX512F-FAST-NEXT:    vpternlogq $248, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm4, %zmm1
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm9, %zmm1
+; AVX512F-FAST-NEXT:    vpermq $238, {{[-0-9]+}}(%r{{[sb]}}p), %zmm4 # 64-byte Folded Reload
+; AVX512F-FAST-NEXT:    # zmm4 = mem[2,3,2,3,6,7,6,7]
+; AVX512F-FAST-NEXT:    vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm7, %zmm4
 ; AVX512F-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm0, 128(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm1, 320(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm16, 128(%rax)
 ; AVX512F-FAST-NEXT:    vmovdqa64 %zmm3, (%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm7, 320(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm1, 256(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm29, 192(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm22, 64(%rax)
-; AVX512F-FAST-NEXT:    vmovdqa64 %zmm20, 384(%rax)
-; AVX512F-FAST-NEXT:    addq $1432, %rsp # imm = 0x598
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm2, 256(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm0, 192(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm19, 64(%rax)
+; AVX512F-FAST-NEXT:    vmovdqa64 %zmm4, 384(%rax)
+; AVX512F-FAST-NEXT:    addq $1384, %rsp # imm = 0x568
 ; AVX512F-FAST-NEXT:    vzeroupper
 ; AVX512F-FAST-NEXT:    retq
 ;
 ; AVX512BW-SLOW-LABEL: store_i8_stride7_vf64:
 ; AVX512BW-SLOW:       # %bb.0:
 ; AVX512BW-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512BW-SLOW-NEXT:    vmovdqa (%rax), %ymm13
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rax), %ymm9
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm1, %ymm13, %ymm0
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rax), %ymm3
+; AVX512BW-SLOW-NEXT:    vmovdqu %ymm3, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rax), %ymm10
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm0, %ymm3, %ymm1
 ; AVX512BW-SLOW-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6]
 ; AVX512BW-SLOW-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vpermw %ymm13, %ymm2, %ymm2
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
-; AVX512BW-SLOW-NEXT:    vmovdqa (%r9), %ymm10
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm2, %ymm10, %ymm3
-; AVX512BW-SLOW-NEXT:    vmovdqa (%r8), %ymm12
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm4, %ymm12, %ymm5
-; AVX512BW-SLOW-NEXT:    vpor %ymm3, %ymm5, %ymm3
+; AVX512BW-SLOW-NEXT:    vpermw %ymm3, %ymm2, %ymm2
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm2, %zmm3
+; AVX512BW-SLOW-NEXT:    vmovdqa (%r9), %ymm2
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm1 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm1, %ymm2, %ymm4
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %ymm2, %ymm28
+; AVX512BW-SLOW-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512BW-SLOW-NEXT:    vmovdqa64 (%r8), %ymm21
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm2 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm2, %ymm21, %ymm5
+; AVX512BW-SLOW-NEXT:    vpor %ymm4, %ymm5, %ymm4
 ; AVX512BW-SLOW-NEXT:    vmovdqa (%r9), %xmm6
 ; AVX512BW-SLOW-NEXT:    vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX512BW-SLOW-NEXT:    vmovdqa (%r8), %xmm5
 ; AVX512BW-SLOW-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm5[8],xmm6[8],xmm5[9],xmm6[9],xmm5[10],xmm6[10],xmm5[11],xmm6[11],xmm5[12],xmm6[12],xmm5[13],xmm6[13],xmm5[14],xmm6[14],xmm5[15],xmm6[15]
 ; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm5, %zmm11
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
 ; AVX512BW-SLOW-NEXT:    movabsq $2323999253380730912, %r10 # imm = 0x2040810204081020
 ; AVX512BW-SLOW-NEXT:    kmovq %r10, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm11 {%k1}
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdx), %ymm17
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm0, %ymm17, %ymm5
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rcx), %ymm18
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm3, %ymm18, %ymm7
-; AVX512BW-SLOW-NEXT:    vpor %ymm5, %ymm7, %ymm5
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdx), %xmm29
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rcx), %xmm23
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm7 = xmm23[8],xmm29[8],xmm23[9],xmm29[9],xmm23[10],xmm29[10],xmm23[11],xmm29[11],xmm23[12],xmm29[12],xmm23[13],xmm29[13],xmm23[14],xmm29[14],xmm23[15],xmm29[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm7 = ymm7[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm7, %zmm19
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %ymm20
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm3, %zmm4 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdx), %ymm15
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm3, %ymm15, %ymm5
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rcx), %ymm13
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm14 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm14, %ymm13, %ymm8
+; AVX512BW-SLOW-NEXT:    vpor %ymm5, %ymm8, %ymm5
+; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdx), %xmm20
+; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rcx), %xmm18
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm18[8],xmm20[8],xmm18[9],xmm20[9],xmm18[10],xmm20[10],xmm18[11],xmm20[11],xmm18[12],xmm20[12],xmm18[13],xmm20[13],xmm18[14],xmm20[14],xmm18[15],xmm20[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm8, %zmm19
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rdi), %ymm12
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm24 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm24, %ymm20, %ymm5
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rsi), %ymm21
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm26 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm26, %ymm21, %ymm7
-; AVX512BW-SLOW-NEXT:    vpor %ymm5, %ymm7, %ymm5
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %xmm22
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm14
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rsi), %xmm16
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm15
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm25 = xmm16[8],xmm22[8],xmm16[9],xmm22[9],xmm16[10],xmm22[10],xmm16[11],xmm22[11],xmm16[12],xmm22[12],xmm16[13],xmm22[13],xmm16[14],xmm22[14],xmm16[15],xmm22[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm25 = xmm25[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm25 = ymm25[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm25, %zmm8
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm24, %ymm12, %ymm5
+; AVX512BW-SLOW-NEXT:    vmovdqa (%rsi), %ymm11
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm27 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm27, %ymm11, %ymm16
+; AVX512BW-SLOW-NEXT:    vporq %ymm5, %ymm16, %ymm5
+; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %xmm16
+; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%rdi), %xmm22
+; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rsi), %xmm17
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm23 = xmm17[8],xmm16[8],xmm17[9],xmm16[9],xmm17[10],xmm16[10],xmm17[11],xmm16[11],xmm17[12],xmm16[12],xmm17[13],xmm16[13],xmm17[14],xmm16[14],xmm17[15],xmm16[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm23 = xmm23[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm23, %zmm9
 ; AVX512BW-SLOW-NEXT:    movabsq $435749860008887046, %r10 # imm = 0x60C183060C18306
 ; AVX512BW-SLOW-NEXT:    kmovq %r10, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm19, %zmm8 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm19, %zmm9 {%k1}
 ; AVX512BW-SLOW-NEXT:    movabsq $4066998693416279096, %r10 # imm = 0x3870E1C3870E1C38
 ; AVX512BW-SLOW-NEXT:    kmovq %r10, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm11, %zmm8 {%k1}
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm11 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10]
-; AVX512BW-SLOW-NEXT:    vpermw %ymm9, %ymm11, %ymm11
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm1, %ymm9, %ymm1
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm11, %zmm1, %zmm1
-; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%r9), %ymm28
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm2, %ymm28, %ymm2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm4, %zmm9 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm4 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10]
+; AVX512BW-SLOW-NEXT:    vpermw %ymm10, %ymm4, %ymm4
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm0, %ymm10, %ymm0
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%r9), %ymm30
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm1, %ymm30, %ymm1
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%r8), %ymm31
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm4, %ymm31, %ymm4
-; AVX512BW-SLOW-NEXT:    vpor %ymm2, %ymm4, %ymm2
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm31[20],zero,ymm31[18],zero,ymm31[20,21,20,21],zero,ymm31[19],zero,ymm31[19,20,21,22],zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm11 = ymm28[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm28[20],zero,ymm28[18],zero,zero,zero,zero,ymm28[21],zero,ymm28[19],zero,zero,zero,zero,ymm28[22]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpor %ymm4, %ymm11, %ymm4
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm2, %zmm4
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm2, %ymm31, %ymm2
+; AVX512BW-SLOW-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm31[0],ymm30[0],ymm31[1],ymm30[1],ymm31[2],ymm30[2],ymm31[3],ymm30[3],ymm31[4],ymm30[4],ymm31[5],ymm30[5],ymm31[6],ymm30[6],ymm31[7],ymm30[7],ymm31[16],ymm30[16],ymm31[17],ymm30[17],ymm31[18],ymm30[18],ymm31[19],ymm30[19],ymm31[20],ymm30[20],ymm31[21],ymm30[21],ymm31[22],ymm30[22],ymm31[23],ymm30[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm1, %zmm1
 ; AVX512BW-SLOW-NEXT:    movabsq $145249953336295682, %r10 # imm = 0x204081020408102
 ; AVX512BW-SLOW-NEXT:    kmovq %r10, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm4 {%k1}
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rsi), %ymm2
-; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%rdi), %ymm25
-; AVX512BW-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm25[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,1,1,4,4,5,5]
-; AVX512BW-SLOW-NEXT:    movl $676341840, %esi # imm = 0x28502850
-; AVX512BW-SLOW-NEXT:    kmovd %esi, %k1
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 {%k1} = ymm2[u,u,u,u,5,u,3,u,u,u,u,6,u,4,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm24, %ymm25, %ymm11
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm26, %ymm2, %ymm19
-; AVX512BW-SLOW-NEXT:    vporq %ymm11, %ymm19, %ymm11
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm11, %zmm11
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rdx), %ymm5
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm0, %ymm5, %ymm19
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rcx), %ymm1
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm3, %ymm1, %ymm3
-; AVX512BW-SLOW-NEXT:    vporq %ymm19, %ymm3, %ymm3
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm19 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm5[18],zero,ymm5[18,19,20,21],zero,ymm5[19],zero,ymm5[25,26,27,22],zero,ymm5[20],zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm19 = ymm19[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm24 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[18],zero,zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22],zero,ymm1[20]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm24 = ymm24[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vporq %ymm19, %ymm24, %ymm19
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm19, %zmm3, %zmm3
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%rdx), %ymm25
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm3, %ymm25, %ymm0
+; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%rcx), %ymm26
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm14, %ymm26, %ymm2
+; AVX512BW-SLOW-NEXT:    vpor %ymm0, %ymm2, %ymm0
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm2 = ymm25[0],ymm26[0],ymm25[1],ymm26[1],ymm25[2],ymm26[2],ymm25[3],ymm26[3],ymm25[4],ymm26[4],ymm25[5],ymm26[5],ymm25[6],ymm26[6],ymm25[7],ymm26[7],ymm25[16],ymm26[16],ymm25[17],ymm26[17],ymm25[18],ymm26[18],ymm25[19],ymm26[19],ymm25[20],ymm26[20],ymm25[21],ymm26[21],ymm25[22],ymm26[22],ymm25[23],ymm26[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%rdi), %ymm23
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm24, %ymm23, %ymm2
+; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%rsi), %ymm24
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm27, %ymm24, %ymm3
+; AVX512BW-SLOW-NEXT:    vpor %ymm2, %ymm3, %ymm2
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm3 = ymm23[0],ymm24[0],ymm23[1],ymm24[1],ymm23[2],ymm24[2],ymm23[3],ymm24[3],ymm23[4],ymm24[4],ymm23[5],ymm24[5],ymm23[6],ymm24[6],ymm23[7],ymm24[7],ymm23[16],ymm24[16],ymm23[17],ymm24[17],ymm23[18],ymm24[18],ymm23[19],ymm24[19],ymm23[20],ymm24[20],ymm23[21],ymm24[21],ymm23[22],ymm24[22],ymm23[23],ymm24[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm3, %zmm2, %zmm14
+; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm7
 ; AVX512BW-SLOW-NEXT:    movabsq $3485998880071096368, %rsi # imm = 0x3060C183060C1830
-; AVX512BW-SLOW-NEXT:    kmovq %rsi, %k3
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm3, %zmm11 {%k3}
-; AVX512BW-SLOW-NEXT:    movabsq $-4357498600088870461, %rsi # imm = 0xC3870E1C3870E1C3
 ; AVX512BW-SLOW-NEXT:    kmovq %rsi, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm4, %zmm11 {%k2}
-; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm20[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[2,2,3,3,6,6,7,7]
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,14,u,12,u,u,u,u,15,u,13,u,u,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,u,u>
-; AVX512BW-SLOW-NEXT:    movl $338170920, %esi # imm = 0x14281428
-; AVX512BW-SLOW-NEXT:    kmovd %esi, %k2
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm6, %ymm21, %ymm3 {%k2}
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm4 = xmm14[0],xmm15[0],xmm14[1],xmm15[1],xmm14[2],xmm15[2],xmm14[3],xmm15[3],xmm14[4],xmm15[4],xmm14[5],xmm15[5],xmm14[6],xmm15[6],xmm14[7],xmm15[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm3, %zmm26
-; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm3 = ymm17[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[0,2,3,3,4,6,7,7]
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm7 = <11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u>
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm7, %ymm18, %ymm4
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm3, %ymm4 {%k1}
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm19 = ymm4[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm4
-; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm3
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm27 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm27 = xmm27[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm27, %zmm19, %zmm19
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm26, %zmm19 {%k3}
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm26 = ymm10[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm10[27],zero,zero,zero,zero,ymm10[30],zero,ymm10[28],zero,zero,zero,zero,ymm10[31],zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm26 = ymm26[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm27 = ymm12[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm12[27],zero,zero,zero,zero,ymm12[30],zero,ymm12[28],zero,zero,zero,zero,ymm12[31],zero,ymm12[29]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vporq %ymm26, %ymm27, %ymm30
-; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%r9), %xmm26
-; AVX512BW-SLOW-NEXT:    vmovdqa64 32(%r8), %xmm27
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm27[0],xmm26[0],xmm27[1],xmm26[1],xmm27[2],xmm26[2],xmm27[3],xmm26[3],xmm27[4],xmm26[4],xmm27[5],xmm26[5],xmm27[6],xmm26[6],xmm27[7],xmm26[7]
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm14 {%k2}
+; AVX512BW-SLOW-NEXT:    movabsq $-4357498600088870461, %rsi # imm = 0xC3870E1C3870E1C3
+; AVX512BW-SLOW-NEXT:    kmovq %rsi, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm14 {%k1}
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm0 = ymm12[8],ymm11[8],ymm12[9],ymm11[9],ymm12[10],ymm11[10],ymm12[11],ymm11[11],ymm12[12],ymm11[12],ymm12[13],ymm11[13],ymm12[14],ymm11[14],ymm12[15],ymm11[15],ymm12[24],ymm11[24],ymm12[25],ymm11[25],ymm12[26],ymm11[26],ymm12[27],ymm11[27],ymm12[28],ymm11[28],ymm12[29],ymm11[29],ymm12[30],ymm11[30],ymm12[31],ymm11[31]
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm27 = <u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u>
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm27, %ymm0, %ymm0
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm22[0],xmm7[0],xmm22[1],xmm7[1],xmm22[2],xmm7[2],xmm22[3],xmm7[3],xmm22[4],xmm7[4],xmm22[5],xmm7[5],xmm22[6],xmm7[6],xmm22[7],xmm7[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm2
+; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm15[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,3,3,4,6,7,7]
+; AVX512BW-SLOW-NEXT:    movl $676341840, %esi # imm = 0x28502850
+; AVX512BW-SLOW-NEXT:    kmovd %esi, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u,11,u,u,u,u,14,u,12,u,u,u,u,15,u,13,u>
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm6, %ymm13, %ymm1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm0, %ymm1 {%k1}
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm3 = ymm1[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm5
+; AVX512BW-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm4
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm19 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3],xmm5[4],xmm4[4],xmm5[5],xmm4[5],xmm5[6],xmm4[6],xmm5[7],xmm4[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm19 = xmm19[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm19 = ymm19[0,1,0,1]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm19, %zmm3, %zmm19
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm19 {%k2}
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm28[8],ymm21[8],ymm28[9],ymm21[9],ymm28[10],ymm21[10],ymm28[11],ymm21[11],ymm28[12],ymm21[12],ymm28[13],ymm21[13],ymm28[14],ymm21[14],ymm28[15],ymm21[15],ymm28[24],ymm21[24],ymm28[25],ymm21[25],ymm28[26],ymm21[26],ymm28[27],ymm21[27],ymm28[28],ymm21[28],ymm28[29],ymm21[29],ymm28[30],ymm21[30],ymm28[31],ymm21[31]
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm8 = <10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11,10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11>
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm8, %ymm2, %ymm2
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm29 = ymm2[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqa 32(%r9), %xmm3
+; AVX512BW-SLOW-NEXT:    vmovdqa 32(%r8), %xmm2
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3],xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
 ; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm30, %zmm0
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rax), %zmm30
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm24 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15,17,17,16,16,17,17,16,16,20,21,17,17,17,17,16,16]
-; AVX512BW-SLOW-NEXT:    vpermw %zmm30, %zmm24, %zmm24
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm29, %zmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rax), %zmm29
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm28 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15,17,17,16,16,17,17,16,16,20,21,17,17,17,17,16,16]
+; AVX512BW-SLOW-NEXT:    vpermw %zmm29, %zmm28, %zmm28
 ; AVX512BW-SLOW-NEXT:    movabsq $580999813345182728, %rax # imm = 0x810204081020408
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k3
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm24, %zmm0 {%k3}
-; AVX512BW-SLOW-NEXT:    movabsq $1016749673354069774, %rax # imm = 0xE1C3870E1C3870E
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k3
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm19 {%k3}
-; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm0 = ymm25[0,1,2,3,6,7,7,6,8,9,10,11,14,15,15,14]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,2,3,3,6,6,7,7]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm6, %ymm2, %ymm0 {%k2}
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128>
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm6, %ymm2, %ymm2
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm24 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm24, %ymm25, %ymm25
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm25 = ymm25[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vporq %ymm2, %ymm25, %ymm2
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm25
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm5[24,25],zero,ymm5[23],zero,ymm5[21,22,23,26],zero,ymm5[24],zero,ymm5[28,29,26,27]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm1[25],zero,ymm1[23],zero,zero,zero,zero,ymm1[26],zero,ymm1[24],zero,zero,zero,zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpor %ymm0, %ymm2, %ymm0
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm7, %ymm1, %ymm1
-; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm2 = ymm5[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[0,2,3,3,4,6,7,7]
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm2, %ymm1 {%k1}
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm0
-; AVX512BW-SLOW-NEXT:    movabsq $1742999440035548184, %rax # imm = 0x183060C183060C18
 ; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm25 {%k2}
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%r9), %zmm0
-; AVX512BW-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm1 = zmm31[0,1,2,3],zmm0[4,5,6,7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm1 = zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,zmm1[23],zero,zmm1[23,24,25,26],zero,zmm1[24],zero,zmm1[30,31,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,61],zero,zmm1[59],zero,zero,zero,zero,zmm1[62],zero,zmm1[60],zero,zero,zero,zero,zmm1[63],zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%r8), %zmm2
-; AVX512BW-SLOW-NEXT:    vshufi64x2 {{.*#+}} zmm5 = zmm28[0,1,2,3],zmm2[4,5,6,7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm5 = zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm5[25],zero,zmm5[23],zero,zero,zero,zero,zmm5[26],zero,zmm5[24],zero,zero,zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm5[59],zero,zero,zero,zero,zmm5[62],zero,zmm5[60],zero,zero,zero,zero,zmm5[63],zero,zmm5[61]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm5 = zmm5[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vporq %zmm1, %zmm5, %zmm1
-; AVX512BW-SLOW-NEXT:    movabsq $6971997760142192736, %rax # imm = 0x60C183060C183060
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm28, %zmm0 {%k2}
+; AVX512BW-SLOW-NEXT:    movabsq $1016749673354069774, %rax # imm = 0xE1C3870E1C3870E
 ; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm25 {%k2}
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm1 = [11,13,12,11,12,13,13,12,11,13,12,11,12,13,13,12,62,61,62,63,63,62,62,63,62,61,62,63,63,62,62,63]
-; AVX512BW-SLOW-NEXT:    vpermi2w %zmm30, %zmm9, %zmm1
-; AVX512BW-SLOW-NEXT:    movabsq $-9150747060186627967, %rax # imm = 0x8102040810204081
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k3
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm25 {%k3}
-; AVX512BW-SLOW-NEXT:    vpshuflw {{.*#+}} ymm1 = ymm20[1,2,2,3,4,5,6,7,9,10,10,11,12,13,14,15]
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,1,1,4,4,5,5]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm1 {%k1} = ymm21[u,u,u,u,5,u,3,u,u,u,u,6,u,4,u,u,u,u,u,u,21,u,19,u,u,u,u,22,u,20,u,u]
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm6, %ymm21, %ymm5
-; AVX512BW-SLOW-NEXT:    vpshufb %ymm24, %ymm20, %ymm6
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm19 {%k2}
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm25[24,25],zero,ymm25[23],zero,ymm25[21,22,23,26],zero,ymm25[24],zero,ymm25[28,29,26,27]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm28 = ymm26[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm26[25],zero,ymm26[23],zero,zero,zero,zero,ymm26[26],zero,ymm26[24],zero,zero,zero,zero
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm28 = ymm28[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vporq %ymm0, %ymm28, %ymm0
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm6, %ymm26, %ymm6
+; AVX512BW-SLOW-NEXT:    vpshufhw {{.*#+}} ymm25 = ymm25[0,1,2,3,7,6,6,7,8,9,10,11,15,14,14,15]
+; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} ymm25 = ymm25[0,2,3,3,4,6,7,7]
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %ymm25, %ymm6 {%k1}
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vpor %ymm5, %ymm6, %ymm5
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm5, %zmm1, %zmm1
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdx), %zmm5
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm17, %zmm5, %zmm5
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rcx), %zmm6
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm18, %zmm6, %zmm6
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm5 = zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm5[18,19,20,21],zero,zmm5[19],zero,zmm5[25,26,27,22],zero,zmm5[20],zero,zmm5[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57],zero,zmm5[55],zero,zmm5[53,54,55,58],zero,zmm5[56],zero,zmm5[60,61,58,59]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm5 = zmm5[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm6 = zmm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm6[18],zero,zero,zero,zero,zmm6[21],zero,zmm6[19],zero,zero,zero,zero,zmm6[22],zero,zmm6[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zmm6[57],zero,zmm6[55],zero,zero,zero,zero,zmm6[58],zero,zmm6[56],zero,zero,zero,zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm6 = zmm6[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vporq %zmm5, %zmm6, %zmm17
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm17 {%k2}
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm12, %zmm2, %zmm1
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm1 = zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm1[18],zero,zmm1[20,21,20,21],zero,zmm1[19],zero,zmm1[19,20,21,22],zero,zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm1[55],zero,zmm1[55,56,57,58],zero,zmm1[56],zero,zmm1[62,63]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm10, %zmm0, %zmm0
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm0[20],zero,zmm0[18],zero,zero,zero,zero,zmm0[21],zero,zmm0[19],zero,zero,zero,zero,zmm0[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm0[57],zero,zmm0[55],zero,zero,zero,zero,zmm0[58],zero,zmm0[56],zero,zero
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512BW-SLOW-NEXT:    vporq %zmm1, %zmm0, %zmm1
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm13, %zmm30, %zmm0
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
-; AVX512BW-SLOW-NEXT:    vpermw %zmm0, %zmm2, %zmm2
-; AVX512BW-SLOW-NEXT:    movabsq $1161999626690365456, %rax # imm = 0x1020408102040810
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm26 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128>
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm26, %ymm24, %ymm6
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm25 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm25, %ymm23, %ymm28
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm28 = ymm28[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vporq %ymm6, %ymm28, %ymm6
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm23 = ymm23[8],ymm24[8],ymm23[9],ymm24[9],ymm23[10],ymm24[10],ymm23[11],ymm24[11],ymm23[12],ymm24[12],ymm23[13],ymm24[13],ymm23[14],ymm24[14],ymm23[15],ymm24[15],ymm23[24],ymm24[24],ymm23[25],ymm24[25],ymm23[26],ymm24[26],ymm23[27],ymm24[27],ymm23[28],ymm24[28],ymm23[29],ymm24[29],ymm23[30],ymm24[30],ymm23[31],ymm24[31]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm27, %ymm23, %ymm23
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm23, %zmm6, %zmm24
+; AVX512BW-SLOW-NEXT:    movabsq $1742999440035548184, %rax # imm = 0x183060C183060C18
 ; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k1}
-; AVX512BW-SLOW-NEXT:    movabsq $2033499346708139548, %rax # imm = 0x1C3870E1C3870E1C
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm24 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm27 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,24,25,24,25,128,23,128,23,24,25,26,128,24,128,30,31]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm27, %ymm31, %ymm0
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} ymm23 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128>
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm23, %ymm30, %ymm6
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpor %ymm0, %ymm6, %ymm0
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} ymm6 = ymm30[8],ymm31[8],ymm30[9],ymm31[9],ymm30[10],ymm31[10],ymm30[11],ymm31[11],ymm30[12],ymm31[12],ymm30[13],ymm31[13],ymm30[14],ymm31[14],ymm30[15],ymm31[15],ymm30[24],ymm31[24],ymm30[25],ymm31[25],ymm30[26],ymm31[26],ymm30[27],ymm31[27],ymm30[28],ymm31[28],ymm30[29],ymm31[29],ymm30[30],ymm31[30],ymm30[31],ymm31[31]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm6, %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    movabsq $6971997760142192736, %rax # imm = 0x60C183060C183060
 ; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm17 {%k1}
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm1, %xmm3, %xmm2
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm5, %xmm4, %xmm6
-; AVX512BW-SLOW-NEXT:    vpor %xmm2, %xmm6, %xmm2
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm3, %zmm2, %zmm2
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm3 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm3, %xmm15, %xmm4
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm6, %xmm14, %xmm7
-; AVX512BW-SLOW-NEXT:    vpor %xmm4, %xmm7, %xmm4
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm7 = xmm15[8],xmm14[8],xmm15[9],xmm14[9],xmm15[10],xmm14[10],xmm15[11],xmm14[11],xmm15[12],xmm14[12],xmm15[13],xmm14[13],xmm15[14],xmm14[14],xmm15[15],xmm14[15]
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm24 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [11,13,12,11,12,13,13,12,11,13,12,11,12,13,13,12,62,61,62,63,63,62,62,63,62,61,62,63,63,62,62,63]
+; AVX512BW-SLOW-NEXT:    vpermi2w %zmm29, %zmm10, %zmm0
+; AVX512BW-SLOW-NEXT:    movabsq $-9150747060186627967, %rax # imm = 0x8102040810204081
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm24 {%k2}
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm6 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm6, %xmm4, %xmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm8 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm8, %xmm5, %xmm28
+; AVX512BW-SLOW-NEXT:    vporq %xmm0, %xmm28, %xmm0
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm4[8],xmm5[8],xmm4[9],xmm5[9],xmm4[10],xmm5[10],xmm4[11],xmm5[11],xmm4[12],xmm5[12],xmm4[13],xmm5[13],xmm4[14],xmm5[14],xmm4[15],xmm5[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm1, %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm1 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm1, %xmm7, %xmm4
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} xmm28 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm28, %xmm22, %xmm30
+; AVX512BW-SLOW-NEXT:    vporq %xmm4, %xmm30, %xmm4
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm7[8],xmm22[8],xmm7[9],xmm22[9],xmm7[10],xmm22[10],xmm7[11],xmm22[11],xmm7[12],xmm22[12],xmm7[13],xmm22[13],xmm7[14],xmm22[14],xmm7[15],xmm22[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm5, %zmm4, %zmm4
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm5 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm0 = zmm4[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm5, %zmm0 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm4 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm4, %xmm3, %xmm5
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} xmm22 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm22, %xmm2, %xmm30
+; AVX512BW-SLOW-NEXT:    vporq %xmm5, %xmm30, %xmm5
+; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm2 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm2, %zmm5, %zmm2
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [2,2,2,4,2,2,2,4,3,3,3,3,2,2,2,4,52,53,52,53,53,54,53,54,52,53,52,53,53,54,53,54]
+; AVX512BW-SLOW-NEXT:    vpermi2w %zmm29, %zmm10, %zmm3
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm2 = zmm2[0,1,0,1,4,5,4,5]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm7, %zmm4, %zmm4
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm4 = zmm4[0,1,0,1,4,5,4,5]
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm4 {%k2}
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm2 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm2, %xmm26, %xmm7
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm10 = [2,2,2,4,2,2,2,4,3,3,3,3,2,2,2,4,52,53,52,53,53,54,53,54,52,53,52,53,53,54,53,54]
-; AVX512BW-SLOW-NEXT:    vpermi2w %zmm30, %zmm9, %zmm10
-; AVX512BW-SLOW-NEXT:    vmovdqa {{.*#+}} xmm9 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm9, %xmm27, %xmm12
-; AVX512BW-SLOW-NEXT:    vpor %xmm7, %xmm12, %xmm7
-; AVX512BW-SLOW-NEXT:    vpunpckhbw {{.*#+}} xmm12 = xmm27[8],xmm26[8],xmm27[9],xmm26[9],xmm27[10],xmm26[10],xmm27[11],xmm26[11],xmm27[12],xmm26[12],xmm27[13],xmm26[13],xmm27[14],xmm26[14],xmm27[15],xmm26[15]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm12 = xmm12[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm12, %zmm7, %zmm7
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm7 = zmm7[0,1,0,1,4,5,4,5]
 ; AVX512BW-SLOW-NEXT:    movabsq $290499906672591364, %rax # imm = 0x408102040810204
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm10, %zmm7 {%k1}
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k2}
 ; AVX512BW-SLOW-NEXT:    movabsq $-8714997200177740921, %rax # imm = 0x870E1C3870E1C387
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm7, %zmm4 {%k1}
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm1, %xmm23, %xmm1
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm5, %xmm29, %xmm5
-; AVX512BW-SLOW-NEXT:    vpor %xmm1, %xmm5, %xmm1
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm29[0],xmm23[0],xmm29[1],xmm23[1],xmm29[2],xmm23[2],xmm29[3],xmm23[3],xmm29[4],xmm23[4],xmm29[5],xmm23[5],xmm29[6],xmm23[6],xmm29[7],xmm23[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm1, %zmm5, %zmm1
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm3, %xmm16, %xmm3
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm6, %xmm22, %xmm5
-; AVX512BW-SLOW-NEXT:    vpor %xmm3, %xmm5, %xmm3
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm5 = xmm22[0],xmm16[0],xmm22[1],xmm16[1],xmm22[2],xmm16[2],xmm22[3],xmm16[3],xmm22[4],xmm16[4],xmm22[5],xmm16[5],xmm22[6],xmm16[6],xmm22[7],xmm16[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm3, %zmm5, %zmm3
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm6, %xmm18, %xmm2
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm8, %xmm20, %xmm3
+; AVX512BW-SLOW-NEXT:    vpor %xmm2, %xmm3, %xmm2
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm20[0],xmm18[0],xmm20[1],xmm18[1],xmm20[2],xmm18[2],xmm20[3],xmm18[3],xmm20[4],xmm18[4],xmm20[5],xmm18[5],xmm20[6],xmm18[6],xmm20[7],xmm18[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm2, %zmm3, %zmm2
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm1, %xmm17, %xmm1
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm28, %xmm16, %xmm3
+; AVX512BW-SLOW-NEXT:    vpor %xmm1, %xmm3, %xmm1
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm16[0],xmm17[0],xmm16[1],xmm17[1],xmm16[2],xmm17[2],xmm16[3],xmm17[3],xmm16[4],xmm17[4],xmm16[5],xmm17[5],xmm16[6],xmm17[6],xmm16[7],xmm17[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm1, %zmm3, %zmm1
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm2 = zmm2[0,1,0,1,4,5,4,5]
 ; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[0,1,0,1,4,5,4,5]
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm3 = zmm3[0,1,0,1,4,5,4,5]
 ; AVX512BW-SLOW-NEXT:    movabsq $871499720017774092, %rax # imm = 0xC183060C183060C
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm3 {%k1}
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k2}
 ; AVX512BW-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm2, %xmm5, %xmm1
-; AVX512BW-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
-; AVX512BW-SLOW-NEXT:    vpshufb %xmm9, %xmm6, %xmm2
-; AVX512BW-SLOW-NEXT:    vpor %xmm1, %xmm2, %xmm1
-; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3],xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
-; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm1, %zmm2, %zmm1
-; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
-; AVX512BW-SLOW-NEXT:    vpermw %zmm0, %zmm2, %zmm0
-; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm4, %xmm5, %xmm2
+; AVX512BW-SLOW-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX512BW-SLOW-NEXT:    vpshufb %xmm22, %xmm4, %xmm3
+; AVX512BW-SLOW-NEXT:    vpor %xmm2, %xmm3, %xmm2
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3],xmm4[4],xmm5[4],xmm4[5],xmm5[5],xmm4[6],xmm5[6],xmm4[7],xmm5[7]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u]
+; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm2, %zmm3, %zmm2
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} zmm2 = zmm2[0,1,0,1,4,5,4,5]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm29, %zmm3 # 32-byte Folded Reload
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
+; AVX512BW-SLOW-NEXT:    vpermw %zmm3, %zmm4, %zmm4
 ; AVX512BW-SLOW-NEXT:    movabsq $4647998506761461824, %rax # imm = 0x4081020408102040
-; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm0, %zmm1 {%k1}
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm4, %zmm2 {%k2}
 ; AVX512BW-SLOW-NEXT:    movabsq $8133997386832558192, %rax # imm = 0x70E1C3870E1C3870
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k2
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k2}
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm26, %ymm11, %ymm2
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm25, %ymm12, %ymm4
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpor %ymm2, %ymm4, %ymm2
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm12[0],ymm11[0],ymm12[1],ymm11[1],ymm12[2],ymm11[2],ymm12[3],ymm11[3],ymm12[4],ymm11[4],ymm12[5],ymm11[5],ymm12[6],ymm11[6],ymm12[7],ymm11[7],ymm12[16],ymm11[16],ymm12[17],ymm11[17],ymm12[18],ymm11[18],ymm12[19],ymm11[19],ymm12[20],ymm11[20],ymm12[21],ymm11[21],ymm12[22],ymm11[22],ymm12[23],ymm11[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm4, %zmm2
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm4 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm15[24,25],zero,ymm15[23],zero,ymm15[21,22,23,26],zero,ymm15[24],zero,ymm15[28,29,26,27]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm13[25],zero,ymm13[23],zero,zero,zero,zero,ymm13[26],zero,ymm13[24],zero,zero,zero,zero
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpor %ymm4, %ymm5, %ymm4
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm15[0],ymm13[0],ymm15[1],ymm13[1],ymm15[2],ymm13[2],ymm15[3],ymm13[3],ymm15[4],ymm13[4],ymm15[5],ymm13[5],ymm15[6],ymm13[6],ymm15[7],ymm13[7],ymm15[16],ymm13[16],ymm15[17],ymm13[17],ymm15[18],ymm13[18],ymm15[19],ymm13[19],ymm15[20],ymm13[20],ymm15[21],ymm13[21],ymm15[22],ymm13[22],ymm15[23],ymm13[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm4
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm4 {%k1}
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm27, %ymm21, %ymm2
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm6 # 32-byte Reload
+; AVX512BW-SLOW-NEXT:    vpshufb %ymm23, %ymm6, %ymm5
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vpor %ymm2, %ymm5, %ymm2
+; AVX512BW-SLOW-NEXT:    vpunpcklbw {{.*#+}} ymm5 = ymm21[0],ymm6[0],ymm21[1],ymm6[1],ymm21[2],ymm6[2],ymm21[3],ymm6[3],ymm21[4],ymm6[4],ymm21[5],ymm6[5],ymm21[6],ymm6[6],ymm21[7],ymm6[7],ymm21[16],ymm6[16],ymm21[17],ymm6[17],ymm21[18],ymm6[18],ymm21[19],ymm6[19],ymm21[20],ymm6[20],ymm21[21],ymm6[21],ymm21[22],ymm6[22],ymm21[23],ymm6[23]
+; AVX512BW-SLOW-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX512BW-SLOW-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm2, %zmm5, %zmm2
+; AVX512BW-SLOW-NEXT:    vmovdqa64 {{.*#+}} zmm5 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
+; AVX512BW-SLOW-NEXT:    vpermw %zmm3, %zmm5, %zmm3
+; AVX512BW-SLOW-NEXT:    movabsq $1161999626690365456, %rax # imm = 0x1020408102040810
 ; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
-; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm1, %zmm3 {%k1}
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm3, %zmm2 {%k1}
+; AVX512BW-SLOW-NEXT:    movabsq $2033499346708139548, %rax # imm = 0x1C3870E1C3870E1C
+; AVX512BW-SLOW-NEXT:    kmovq %rax, %k1
+; AVX512BW-SLOW-NEXT:    vmovdqu8 %zmm2, %zmm4 {%k1}
 ; AVX512BW-SLOW-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm3, (%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm11, 320(%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm4, 256(%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm17, 128(%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm8, 64(%rax)
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm25, 384(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm14, 320(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm4, 128(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm1, (%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, 256(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm9, 64(%rax)
+; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm24, 384(%rax)
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm19, 192(%rax)
 ; AVX512BW-SLOW-NEXT:    vzeroupper
 ; AVX512BW-SLOW-NEXT:    retq
 ;
 ; AVX512BW-FAST-LABEL: store_i8_stride7_vf64:
 ; AVX512BW-FAST:       # %bb.0:
-; AVX512BW-FAST-NEXT:    subq $200, %rsp
 ; AVX512BW-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm5
-; AVX512BW-FAST-NEXT:    vmovdqu64 %zmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rsi), %zmm3
-; AVX512BW-FAST-NEXT:    vmovdqu64 %zmm3, (%rsp) # 64-byte Spill
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdx), %zmm2
-; AVX512BW-FAST-NEXT:    vmovdqu64 %zmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
-; AVX512BW-FAST-NEXT:    vmovdqa (%rax), %ymm6
-; AVX512BW-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX512BW-FAST-NEXT:    vmovdqa (%rax), %ymm2
+; AVX512BW-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
 ; AVX512BW-FAST-NEXT:    vmovdqa 32(%rax), %ymm12
-; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm0, %ymm6, %ymm1
-; AVX512BW-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6]
-; AVX512BW-FAST-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vpermw %ymm6, %ymm4, %ymm4
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm4, %zmm6
-; AVX512BW-FAST-NEXT:    vmovdqa (%r9), %ymm14
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm18 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm18, %ymm14, %ymm7
-; AVX512BW-FAST-NEXT:    vmovdqa (%r8), %ymm1
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm20 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm20, %ymm1, %ymm8
-; AVX512BW-FAST-NEXT:    vmovdqa64 %ymm1, %ymm24
-; AVX512BW-FAST-NEXT:    vpor %ymm7, %ymm8, %ymm7
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%r9), %xmm25
-; AVX512BW-FAST-NEXT:    vmovdqa (%r8), %xmm10
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm10[8],xmm25[8],xmm10[9],xmm25[9],xmm10[10],xmm25[10],xmm10[11],xmm25[11],xmm10[12],xmm25[12],xmm10[13],xmm25[13],xmm10[14],xmm25[14],xmm10[15],xmm25[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm8, %zmm22
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = [12,13,2,3,12,13,0,1,14,15,2,3,0,1,14,15,28,29,18,19,28,29,16,17,30,31,18,19,16,17,30,31]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm11, %ymm2, %ymm0
+; AVX512BW-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,5,4,5,5,6,5,6,4,5,4,5,5,6,5,6]
+; AVX512BW-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vpermw %ymm2, %ymm1, %ymm1
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm3
+; AVX512BW-FAST-NEXT:    vmovdqa (%r9), %ymm9
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [13,128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm0, %ymm9, %ymm4
+; AVX512BW-FAST-NEXT:    vmovdqa64 (%r8), %ymm21
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} ymm14 = [128,128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm14, %ymm21, %ymm5
+; AVX512BW-FAST-NEXT:    vpor %ymm4, %ymm5, %ymm4
+; AVX512BW-FAST-NEXT:    vmovdqa (%r9), %xmm1
+; AVX512BW-FAST-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512BW-FAST-NEXT:    vmovdqa (%r8), %xmm2
+; AVX512BW-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm5 = xmm2[8],xmm1[8],xmm2[9],xmm1[9],xmm2[10],xmm1[10],xmm2[11],xmm1[11],xmm2[12],xmm1[12],xmm2[13],xmm1[13],xmm2[14],xmm1[14],xmm2[15],xmm1[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm4, %zmm5, %zmm18
 ; AVX512BW-FAST-NEXT:    movabsq $2323999253380730912, %r10 # imm = 0x2040810204081020
 ; AVX512BW-FAST-NEXT:    kmovq %r10, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm6, %zmm22 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm18 {%k1}
 ; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %ymm1
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm21 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm21, %ymm1, %ymm8
-; AVX512BW-FAST-NEXT:    vmovdqa %ymm1, %ymm6
-; AVX512BW-FAST-NEXT:    vmovdqa (%rcx), %ymm1
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm23 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm23, %ymm1, %ymm11
-; AVX512BW-FAST-NEXT:    vmovdqa %ymm1, %ymm7
-; AVX512BW-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX512BW-FAST-NEXT:    vpor %ymm8, %ymm11, %ymm8
-; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %xmm15
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rcx), %xmm16
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm11 = xmm16[8],xmm15[8],xmm16[9],xmm15[9],xmm16[10],xmm15[10],xmm16[11],xmm15[11],xmm16[12],xmm15[12],xmm16[13],xmm15[13],xmm16[14],xmm15[14],xmm16[15],xmm15[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm11 = xmm11[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm11 = ymm11[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm11, %zmm26
-; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %ymm11
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm28 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm28, %ymm11, %ymm8
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm19 = [0,1,0,1,14,128,14,15,0,1,14,15,128,13,14,15,16,17,16,128,30,31,30,31,16,17,128,31,28,29,30,31]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm19, %ymm1, %ymm5
+; AVX512BW-FAST-NEXT:    vmovdqa64 (%rcx), %ymm22
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm20 = [128,128,128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm20, %ymm22, %ymm8
+; AVX512BW-FAST-NEXT:    vpor %ymm5, %ymm8, %ymm5
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdx), %xmm3
+; AVX512BW-FAST-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512BW-FAST-NEXT:    vmovdqa (%rcx), %xmm2
+; AVX512BW-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm8 = xmm2[8],xmm3[8],xmm2[9],xmm3[9],xmm2[10],xmm3[10],xmm2[11],xmm3[11],xmm2[12],xmm3[12],xmm2[13],xmm3[13],xmm2[14],xmm3[14],xmm2[15],xmm3[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm8 = xmm8[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm8, %zmm23
+; AVX512BW-FAST-NEXT:    vmovdqa (%rdi), %ymm15
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm24 = [0,1,14,128,12,13,0,1,14,15,128,3,12,13,2,3,16,128,30,31,28,29,16,17,128,31,18,19,28,29,18,128]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm24, %ymm15, %ymm5
 ; AVX512BW-FAST-NEXT:    vmovdqa (%rsi), %ymm13
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm29 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
-; AVX512BW-FAST-NEXT:    vpshufb %ymm29, %ymm13, %ymm17
-; AVX512BW-FAST-NEXT:    vporq %ymm8, %ymm17, %ymm8
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %xmm17
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rsi), %xmm19
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm27 = xmm19[8],xmm17[8],xmm19[9],xmm17[9],xmm19[10],xmm17[10],xmm19[11],xmm17[11],xmm19[12],xmm17[12],xmm19[13],xmm17[13],xmm19[14],xmm17[14],xmm19[15],xmm17[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm27 = xmm27[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm8, %zmm27, %zmm8
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm25 = [128,128,128,14,128,128,128,128,128,128,15,128,128,128,128,128,128,16,128,128,128,128,128,128,17,128,128,128,128,128,128,18]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm25, %ymm13, %ymm16
+; AVX512BW-FAST-NEXT:    vporq %ymm5, %ymm16, %ymm5
+; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %xmm16
+; AVX512BW-FAST-NEXT:    vmovdqa64 (%rsi), %xmm17
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm26 = xmm17[8],xmm16[8],xmm17[9],xmm16[9],xmm17[10],xmm16[10],xmm17[11],xmm16[11],xmm17[12],xmm16[12],xmm17[13],xmm16[13],xmm17[14],xmm16[14],xmm17[15],xmm16[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm26 = xmm26[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm26 = ymm26[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm5, %zmm26, %zmm10
 ; AVX512BW-FAST-NEXT:    movabsq $435749860008887046, %r10 # imm = 0x60C183060C18306
 ; AVX512BW-FAST-NEXT:    kmovq %r10, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm26, %zmm8 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm23, %zmm10 {%k1}
 ; AVX512BW-FAST-NEXT:    movabsq $4066998693416279096, %r10 # imm = 0x3870E1C3870E1C38
 ; AVX512BW-FAST-NEXT:    kmovq %r10, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm22, %zmm8 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm22 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10]
-; AVX512BW-FAST-NEXT:    vpermw %ymm12, %ymm22, %ymm22
-; AVX512BW-FAST-NEXT:    vpshufb %ymm0, %ymm12, %ymm0
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm22, %zmm0, %zmm22
-; AVX512BW-FAST-NEXT:    vmovdqa 32(%r9), %ymm1
-; AVX512BW-FAST-NEXT:    vpshufb %ymm18, %ymm1, %ymm18
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%r8), %ymm27
-; AVX512BW-FAST-NEXT:    vpshufb %ymm20, %ymm27, %ymm20
-; AVX512BW-FAST-NEXT:    vporq %ymm18, %ymm20, %ymm18
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm20 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm27[20],zero,ymm27[18],zero,ymm27[20,21,20,21],zero,ymm27[19],zero,ymm27[19,20,21,22],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm20 = ymm20[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm26 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm1[20],zero,ymm1[18],zero,zero,zero,zero,ymm1[21],zero,ymm1[19],zero,zero,zero,zero,ymm1[22]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm26 = ymm26[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vporq %ymm20, %ymm26, %ymm20
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm18, %zmm26
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm18, %zmm10 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm18 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10]
+; AVX512BW-FAST-NEXT:    vpermw %ymm12, %ymm18, %ymm18
+; AVX512BW-FAST-NEXT:    vpshufb %ymm11, %ymm12, %ymm11
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm18, %zmm11, %zmm18
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%r9), %ymm27
+; AVX512BW-FAST-NEXT:    vpshufb %ymm0, %ymm27, %ymm0
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%r8), %ymm29
+; AVX512BW-FAST-NEXT:    vpshufb %ymm14, %ymm29, %ymm11
+; AVX512BW-FAST-NEXT:    vpor %ymm0, %ymm11, %ymm0
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm14 = ymm29[0],ymm27[0],ymm29[1],ymm27[1],ymm29[2],ymm27[2],ymm29[3],ymm27[3],ymm29[4],ymm27[4],ymm29[5],ymm27[5],ymm29[6],ymm27[6],ymm29[7],ymm27[7],ymm29[16],ymm27[16],ymm29[17],ymm27[17],ymm29[18],ymm27[18],ymm29[19],ymm27[19],ymm29[20],ymm27[20],ymm29[21],ymm27[21],ymm29[22],ymm27[22],ymm29[23],ymm27[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm14 = ymm14[8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm14 = ymm14[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm0, %zmm0
 ; AVX512BW-FAST-NEXT:    movabsq $145249953336295682, %r10 # imm = 0x204081020408102
 ; AVX512BW-FAST-NEXT:    kmovq %r10, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm22, %zmm26 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rdx), %ymm22
-; AVX512BW-FAST-NEXT:    vpshufb %ymm21, %ymm22, %ymm18
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rcx), %ymm30
-; AVX512BW-FAST-NEXT:    vpshufb %ymm23, %ymm30, %ymm20
-; AVX512BW-FAST-NEXT:    vporq %ymm18, %ymm20, %ymm18
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm20 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm22[18],zero,ymm22[18,19,20,21],zero,ymm22[19],zero,ymm22[25,26,27,22],zero,ymm22[20],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm20 = ymm20[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm21 = ymm30[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm30[18],zero,zero,zero,zero,ymm30[21],zero,ymm30[19],zero,zero,zero,zero,ymm30[22],zero,ymm30[20]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm21 = ymm21[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vporq %ymm20, %ymm21, %ymm20
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm18, %zmm21
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rdi), %ymm31
-; AVX512BW-FAST-NEXT:    vpshufb %ymm28, %ymm31, %ymm18
-; AVX512BW-FAST-NEXT:    vmovdqa 32(%rsi), %ymm0
-; AVX512BW-FAST-NEXT:    vpshufb %ymm29, %ymm0, %ymm20
-; AVX512BW-FAST-NEXT:    vporq %ymm18, %ymm20, %ymm18
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm20 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm31[18,19,20,21],zero,ymm31[19],zero,ymm31[21,20,21,22],zero,ymm31[20],zero,ymm31[22,23]
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm18, %zmm0 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rdx), %ymm23
+; AVX512BW-FAST-NEXT:    vpshufb %ymm19, %ymm23, %ymm14
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rcx), %ymm19
+; AVX512BW-FAST-NEXT:    vpshufb %ymm20, %ymm19, %ymm18
+; AVX512BW-FAST-NEXT:    vporq %ymm14, %ymm18, %ymm14
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm20 = ymm23[0],ymm19[0],ymm23[1],ymm19[1],ymm23[2],ymm19[2],ymm23[3],ymm19[3],ymm23[4],ymm19[4],ymm23[5],ymm19[5],ymm23[6],ymm19[6],ymm23[7],ymm19[7],ymm23[16],ymm19[16],ymm23[17],ymm19[17],ymm23[18],ymm19[18],ymm23[19],ymm19[19],ymm23[20],ymm19[20],ymm23[21],ymm19[21],ymm23[22],ymm19[22],ymm23[23],ymm19[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm20 = ymm20[4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm20 = ymm20[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm23 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm0[21],zero,ymm0[19],zero,zero,zero,zero,ymm0[22],zero,ymm0[20],zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vporq %ymm20, %ymm23, %ymm20
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rcx), %zmm23
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm18, %zmm18
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%r8), %zmm9
-; AVX512BW-FAST-NEXT:    vmovdqu64 %zmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm14, %zmm26
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rdi), %ymm28
+; AVX512BW-FAST-NEXT:    vpshufb %ymm24, %ymm28, %ymm14
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rsi), %ymm24
+; AVX512BW-FAST-NEXT:    vpshufb %ymm25, %ymm24, %ymm20
+; AVX512BW-FAST-NEXT:    vporq %ymm14, %ymm20, %ymm14
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm25 = ymm28[0],ymm24[0],ymm28[1],ymm24[1],ymm28[2],ymm24[2],ymm28[3],ymm24[3],ymm28[4],ymm24[4],ymm28[5],ymm24[5],ymm28[6],ymm24[6],ymm28[7],ymm24[7],ymm28[16],ymm24[16],ymm28[17],ymm24[17],ymm28[18],ymm24[18],ymm28[19],ymm24[19],ymm28[20],ymm24[20],ymm28[21],ymm24[21],ymm28[22],ymm24[22],ymm28[23],ymm24[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm25 = ymm25[u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm25 = ymm25[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm25, %zmm14, %zmm14
 ; AVX512BW-FAST-NEXT:    movabsq $3485998880071096368, %r10 # imm = 0x3060C183060C1830
 ; AVX512BW-FAST-NEXT:    kmovq %r10, %k2
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm21, %zmm18 {%k2}
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%r9), %zmm21
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm26, %zmm14 {%k2}
 ; AVX512BW-FAST-NEXT:    movabsq $-4357498600088870461, %r10 # imm = 0xC3870E1C3870E1C3
 ; AVX512BW-FAST-NEXT:    kmovq %r10, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm26, %zmm18 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rax), %zmm26
-; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm22 = zmm22[0,1,2,3],zmm23[4,5,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm22 = zmm22[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,zmm22[23],zero,zmm22[21,22,23,26],zero,zmm22[24],zero,zmm22[28,29,26,27,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,59],zero,zero,zero,zero,zmm22[62],zero,zmm22[60],zero,zero,zero,zero,zmm22[63],zero,zmm22[61],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm22 = zmm22[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm28 = zmm30[0,1,2,3],zmm2[4,5,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm28 = zmm28[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zmm28[25],zero,zmm28[23],zero,zero,zero,zero,zmm28[26],zero,zmm28[24],zero,zero,zero,zero,zmm28[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm28[62],zero,zmm28[60],zero,zero,zero,zero,zmm28[63],zero,zmm28[61],zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm28 = zmm28[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm22, %zmm28, %zmm29
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rdi), %xmm28
-; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm5[4,5,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25],zero,zmm0[23],zero,zero,zero,zero,zmm0[26],zero,zmm0[24],zero,zero,zero,zero,zmm0[27],zero,zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,60,61,62],zero,zmm0[60],zero,zmm0[62,63,62,63],zero,zmm0[61],zero,zmm0[63,60,61]
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm14 {%k1}
+; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm23[0,1,2,3],mem[4,5,6,7]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,zmm0[23],zero,zmm0[21,22,23,26],zero,zmm0[24],zero,zmm0[28,29,26,27,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,59],zero,zero,zero,zero,zmm0[62],zero,zmm0[60],zero,zero,zero,zero,zmm0[63],zero,zmm0[61],zero
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm22 = zmm31[0,1,2,3],zmm3[4,5,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm22 = zmm22[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm22[23],zero,zero,zero,zero,zmm22[26],zero,zmm22[24],zero,zero,zero,zero,zmm22[27],zero,zmm22[25,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zmm22[62],zero,zmm22[60],zero,zero,zero,zero,zmm22[63],zero,zmm22[61],zero,zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm22 = zmm22[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm0, %zmm22, %zmm22
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rsi), %xmm30
-; AVX512BW-FAST-NEXT:    movabsq $1742999440035548184, %rax # imm = 0x183060C183060C18
-; AVX512BW-FAST-NEXT:    kmovq %rax, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm29, %zmm22 {%k1}
-; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm27[0,1,2,3],zmm21[4,5,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,24,25],zero,zmm0[23],zero,zmm0[23,24,25,26],zero,zmm0[24],zero,zmm0[30,31,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,61],zero,zmm0[59],zero,zero,zero,zero,zmm0[62],zero,zmm0[60],zero,zero,zero,zero,zmm0[63],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm1 = zmm1[0,1,2,3],zmm9[4,5,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm1 = zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm1[25],zero,zmm1[23],zero,zero,zero,zero,zmm1[26],zero,zmm1[24],zero,zero,zmm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm1[59],zero,zero,zero,zero,zmm1[62],zero,zmm1[60],zero,zero,zero,zero,zmm1[63],zero,zmm1[61]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm0, %zmm1, %zmm0
+; AVX512BW-FAST-NEXT:    vshufi64x2 {{.*#+}} zmm19 = zmm19[0,1,2,3],mem[4,5,6,7]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm19 = zmm19[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zmm19[25],zero,zmm19[23],zero,zero,zero,zero,zmm19[26],zero,zmm19[24],zero,zero,zero,zero,zmm19[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm19[62],zero,zmm19[60],zero,zero,zero,zero,zmm19[63],zero,zmm19[61],zero,zero
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm19 = zmm19[2,3,2,3,6,7,6,7]
+; AVX512BW-FAST-NEXT:    vporq %zmm0, %zmm19, %zmm0
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm25 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,25,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128>
+; AVX512BW-FAST-NEXT:    vpshufb %ymm25, %ymm24, %ymm19
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm19 = ymm19[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm26 = [128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,128,23,128,128,128,128,26,128,24,128,128,128,128,27,128,25]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm26, %ymm28, %ymm23
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vporq %ymm19, %ymm23, %ymm19
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm23 = ymm28[8],ymm24[8],ymm28[9],ymm24[9],ymm28[10],ymm24[10],ymm28[11],ymm24[11],ymm28[12],ymm24[12],ymm28[13],ymm24[13],ymm28[14],ymm24[14],ymm28[15],ymm24[15],ymm28[24],ymm24[24],ymm28[25],ymm24[25],ymm28[26],ymm24[26],ymm28[27],ymm24[27],ymm28[28],ymm24[28],ymm28[29],ymm24[29],ymm28[30],ymm24[30],ymm28[31],ymm24[31]
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm30 = <u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u,u,u,12,13,8,9,u,u,u,14,15,10,11,u,u,u>
+; AVX512BW-FAST-NEXT:    vpshufb %ymm30, %ymm23, %ymm23
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm23 = ymm23[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm23, %zmm19, %zmm19
+; AVX512BW-FAST-NEXT:    movabsq $1742999440035548184, %r10 # imm = 0x183060C183060C18
+; AVX512BW-FAST-NEXT:    kmovq %r10, %k1
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm19 {%k1}
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm29[24,25,24,25],zero,ymm29[23],zero,ymm29[23,24,25,26],zero,ymm29[24],zero,ymm29[30,31]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} ymm24 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,128,128,128,128,25,128,23,128,128,128,128,26,128,24,128,128>
+; AVX512BW-FAST-NEXT:    vpshufb %ymm24, %ymm27, %ymm28
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm28 = ymm28[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vporq %ymm0, %ymm28, %ymm0
+; AVX512BW-FAST-NEXT:    vmovdqa64 (%rax), %zmm28
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm27 = ymm27[8],ymm29[8],ymm27[9],ymm29[9],ymm27[10],ymm29[10],ymm27[11],ymm29[11],ymm27[12],ymm29[12],ymm27[13],ymm29[13],ymm27[14],ymm29[14],ymm27[15],ymm29[15],ymm27[24],ymm29[24],ymm27[25],ymm29[25],ymm27[26],ymm29[26],ymm27[27],ymm29[27],ymm27[28],ymm29[28],ymm27[29],ymm29[29],ymm27[30],ymm29[30],ymm27[31],ymm29[31]
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = <10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11,10,7,6,u,u,u,13,12,9,8,u,u,u,15,14,11>
+; AVX512BW-FAST-NEXT:    vpshufb %ymm7, %ymm27, %ymm27
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm27, %zmm0, %zmm27
+; AVX512BW-FAST-NEXT:    vmovdqa 32(%rdi), %xmm6
 ; AVX512BW-FAST-NEXT:    movabsq $6971997760142192736, %rax # imm = 0x60C183060C183060
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm22 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [11,13,12,11,12,13,13,12,11,13,12,11,12,13,13,12,62,61,62,63,63,62,62,63,62,61,62,63,63,62,62,63]
-; AVX512BW-FAST-NEXT:    vpermi2w %zmm26, %zmm12, %zmm0
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm27, %zmm19 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm27 = [11,13,12,11,12,13,13,12,11,13,12,11,12,13,13,12,62,61,62,63,63,62,62,63,62,61,62,63,63,62,62,63]
+; AVX512BW-FAST-NEXT:    vpermi2w %zmm28, %zmm12, %zmm27
 ; AVX512BW-FAST-NEXT:    movabsq $-9150747060186627967, %rax # imm = 0x8102040810204081
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k3
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm22 {%k3}
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,28,29,30],zero,ymm11[28],zero,ymm11[30,31,30,31],zero,ymm11[29],zero,ymm11[31,28,29]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm13[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,ymm13[30],zero,ymm13[28],zero,zero,zero,zero,ymm13[31],zero,ymm13[29],zero,zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm28[0],xmm30[0],xmm28[1],xmm30[1],xmm28[2],xmm30[2],xmm28[3],xmm30[3],xmm28[4],xmm30[4],xmm28[5],xmm30[5],xmm28[6],xmm30[6],xmm28[7],xmm30[7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm0, %zmm2
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm7[30],zero,ymm7[28],zero,zero,zero,zero,ymm7[31],zero,ymm7[29],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm27 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm6[30],zero,ymm6[28],zero,zero,zero,zero,ymm6[31],zero,ymm6[29],zero,zero
-; AVX512BW-FAST-NEXT:    vmovdqa64 %ymm6, %ymm20
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm27, %zmm19 {%k3}
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm27 = ymm15[8],ymm13[8],ymm15[9],ymm13[9],ymm15[10],ymm13[10],ymm15[11],ymm13[11],ymm15[12],ymm13[12],ymm15[13],ymm13[13],ymm15[14],ymm13[14],ymm15[15],ymm13[15],ymm15[24],ymm13[24],ymm15[25],ymm13[25],ymm15[26],ymm13[26],ymm15[27],ymm13[27],ymm15[28],ymm13[28],ymm15[29],ymm13[29],ymm15[30],ymm13[30],ymm15[31],ymm13[31]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm30, %ymm27, %ymm27
+; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rsi), %xmm30
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm31 = xmm6[0],xmm30[0],xmm6[1],xmm30[1],xmm6[2],xmm30[2],xmm6[3],xmm30[3],xmm6[4],xmm30[4],xmm6[5],xmm30[5],xmm6[6],xmm30[6],xmm6[7],xmm30[7]
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} xmm29 = <0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5>
+; AVX512BW-FAST-NEXT:    vpshufb %xmm29, %xmm31, %xmm31
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm31 = ymm31[0,1,0,1]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm31, %zmm27, %zmm3
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm27 = ymm22[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,27],zero,zero,zero,zero,ymm22[30],zero,ymm22[28],zero,zero,zero,zero,ymm22[31],zero,ymm22[29],zero
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm27 = ymm27[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vporq %ymm0, %ymm27, %ymm27
-; AVX512BW-FAST-NEXT:    vmovdqa64 32(%rdx), %xmm31
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm31 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,ymm1[30],zero,ymm1[28],zero,zero,zero,zero,ymm1[31],zero,ymm1[29],zero,zero
+; AVX512BW-FAST-NEXT:    vmovdqa64 %ymm1, %ymm23
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm31 = ymm31[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vporq %ymm27, %ymm31, %ymm27
+; AVX512BW-FAST-NEXT:    vmovdqa 32(%rdx), %xmm2
 ; AVX512BW-FAST-NEXT:    vmovdqa 32(%rcx), %xmm1
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm31[0],xmm1[0],xmm31[1],xmm1[1],xmm31[2],xmm1[2],xmm31[3],xmm1[3],xmm31[4],xmm1[4],xmm31[5],xmm1[5],xmm31[6],xmm1[6],xmm31[7],xmm1[7]
-; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm6 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm6, %xmm0, %xmm0
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3],xmm2[4],xmm1[4],xmm2[5],xmm1[5],xmm2[6],xmm1[6],xmm2[7],xmm1[7]
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} xmm31 = <4,5,0,1,u,u,u,6,7,2,3,u,u,u,8,9>
+; AVX512BW-FAST-NEXT:    vpshufb %xmm31, %xmm0, %xmm0
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm27, %zmm27
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm2, %zmm27 {%k2}
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm14[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,29],zero,ymm14[27],zero,zero,zero,zero,ymm14[30],zero,ymm14[28],zero,zero,zero,zero,ymm14[31],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm24[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,ymm24[27],zero,zero,zero,zero,ymm24[30],zero,ymm24[28],zero,zero,zero,zero,ymm24[31],zero,ymm24[29]
-; AVX512BW-FAST-NEXT:    vmovdqa64 %ymm24, %ymm7
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,3,2,3]
-; AVX512BW-FAST-NEXT:    vpor %ymm0, %ymm2, %ymm2
-; AVX512BW-FAST-NEXT:    vmovdqa 32(%r9), %xmm4
-; AVX512BW-FAST-NEXT:    vmovdqa 32(%r8), %xmm3
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm3[0],xmm4[0],xmm3[1],xmm4[1],xmm3[2],xmm4[2],xmm3[3],xmm4[3],xmm3[4],xmm4[4],xmm3[5],xmm4[5],xmm3[6],xmm4[6],xmm3[7],xmm4[7]
-; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm9 = <u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm9, %xmm0, %xmm0
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm27 {%k2}
+; AVX512BW-FAST-NEXT:    vmovdqa64 %ymm21, %ymm8
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} ymm0 = ymm9[8],ymm21[8],ymm9[9],ymm21[9],ymm9[10],ymm21[10],ymm9[11],ymm21[11],ymm9[12],ymm21[12],ymm9[13],ymm21[13],ymm9[14],ymm21[14],ymm9[15],ymm21[15],ymm9[24],ymm21[24],ymm9[25],ymm21[25],ymm9[26],ymm21[26],ymm9[27],ymm21[27],ymm9[28],ymm21[28],ymm9[29],ymm21[29],ymm9[30],ymm21[30],ymm9[31],ymm21[31]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm7, %ymm0, %ymm0
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm0[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vmovdqa 32(%r9), %xmm5
+; AVX512BW-FAST-NEXT:    vmovdqa 32(%r8), %xmm4
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3],xmm4[4],xmm5[4],xmm4[5],xmm5[5],xmm4[6],xmm5[6],xmm4[7],xmm5[7]
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm11 = <u,u,4,5,0,1,u,u,u,6,7,2,3,u,u,u>
+; AVX512BW-FAST-NEXT:    vpshufb %xmm11, %xmm0, %xmm0
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm2, %zmm0
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15,17,17,16,16,17,17,16,16,20,21,17,17,17,17,16,16]
-; AVX512BW-FAST-NEXT:    vpermw %zmm26, %zmm2, %zmm2
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm3, %zmm0
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [14,13,14,15,15,14,14,15,14,13,14,15,15,14,14,15,17,17,16,16,17,17,16,16,20,21,17,17,17,17,16,16]
+; AVX512BW-FAST-NEXT:    vpermw %zmm28, %zmm3, %zmm3
 ; AVX512BW-FAST-NEXT:    movabsq $580999813345182728, %rax # imm = 0x810204081020408
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm2, %zmm0 {%k2}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k2}
 ; AVX512BW-FAST-NEXT:    movabsq $1016749673354069774, %rax # imm = 0xE1C3870E1C3870E
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
 ; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm27 {%k2}
-; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm5 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm5, %xmm1, %xmm2
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm7 = <u,u,u,128,7,128,5,u,u,u,128,8,128,6,u,u>
+; AVX512BW-FAST-NEXT:    vpshufb %xmm7, %xmm1, %xmm3
 ; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = <u,u,u,7,128,5,128,u,u,u,8,128,6,128,u,u>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm0, %xmm31, %xmm24
-; AVX512BW-FAST-NEXT:    vporq %xmm2, %xmm24, %xmm2
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8],xmm31[8],xmm1[9],xmm31[9],xmm1[10],xmm31[10],xmm1[11],xmm31[11],xmm1[12],xmm31[12],xmm1[13],xmm31[13],xmm1[14],xmm31[14],xmm1[15],xmm31[15]
+; AVX512BW-FAST-NEXT:    vpshufb %xmm0, %xmm2, %xmm18
+; AVX512BW-FAST-NEXT:    vporq %xmm3, %xmm18, %xmm3
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm1[8],xmm2[8],xmm1[9],xmm2[9],xmm1[10],xmm2[10],xmm1[11],xmm2[11],xmm1[12],xmm2[12],xmm1[13],xmm2[13],xmm1[14],xmm2[14],xmm1[15],xmm2[15]
 ; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[6,3,2,u,u,u,9,8,5,4,u,u,u,11,10,7]
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm1, %zmm2, %zmm1
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm1, %zmm3, %zmm1
 ; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,128,7,128,5,u,u,u,128,8,128,6,u,u,u,128>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm2, %xmm30, %xmm24
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} xmm31 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm31, %xmm28, %xmm29
-; AVX512BW-FAST-NEXT:    vporq %xmm24, %xmm29, %xmm24
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm28 = xmm30[8],xmm28[8],xmm30[9],xmm28[9],xmm30[10],xmm28[10],xmm30[11],xmm28[11],xmm30[12],xmm28[12],xmm30[13],xmm28[13],xmm30[14],xmm28[14],xmm30[15],xmm28[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm28 = xmm28[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm28, %zmm24, %zmm24
+; AVX512BW-FAST-NEXT:    vpshufb %xmm2, %xmm30, %xmm3
+; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} xmm18 = <u,7,128,5,128,u,u,u,8,128,6,128,u,u,u,9>
+; AVX512BW-FAST-NEXT:    vpshufb %xmm18, %xmm6, %xmm20
+; AVX512BW-FAST-NEXT:    vporq %xmm3, %xmm20, %xmm3
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm6 = xmm30[8],xmm6[8],xmm30[9],xmm6[9],xmm30[10],xmm6[10],xmm30[11],xmm6[11],xmm30[12],xmm6[12],xmm30[13],xmm6[13],xmm30[14],xmm6[14],xmm30[15],xmm6[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[2,u,u,u,9,8,5,4,u,u,u,11,10,7,6,u]
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm6, %zmm3, %zmm3
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[0,1,0,1,4,5,4,5]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm24 = zmm24[0,1,0,1,4,5,4,5]
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm1, %zmm24 {%k1}
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm21 = zmm3[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm1, %zmm21 {%k1}
 ; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = <128,4,u,u,u,128,7,128,5,u,u,u,128,8,128,6>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm1, %xmm4, %xmm28
-; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} xmm29 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
-; AVX512BW-FAST-NEXT:    vpshufb %xmm29, %xmm3, %xmm30
-; AVX512BW-FAST-NEXT:    vporq %xmm28, %xmm30, %xmm28
-; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm3 = xmm3[8],xmm4[8],xmm3[9],xmm4[9],xmm3[10],xmm4[10],xmm3[11],xmm4[11],xmm3[12],xmm4[12],xmm3[13],xmm4[13],xmm3[14],xmm4[14],xmm3[15],xmm4[15]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm3, %zmm28, %zmm3
+; AVX512BW-FAST-NEXT:    vpshufb %xmm1, %xmm5, %xmm3
+; AVX512BW-FAST-NEXT:    vmovdqa {{.*#+}} xmm6 = <4,128,u,u,u,7,128,5,128,u,u,u,8,128,6,128>
+; AVX512BW-FAST-NEXT:    vpshufb %xmm6, %xmm4, %xmm20
+; AVX512BW-FAST-NEXT:    vporq %xmm3, %xmm20, %xmm3
+; AVX512BW-FAST-NEXT:    vpunpckhbw {{.*#+}} xmm4 = xmm4[8],xmm5[8],xmm4[9],xmm5[9],xmm4[10],xmm5[10],xmm4[11],xmm5[11],xmm4[12],xmm5[12],xmm4[13],xmm5[13],xmm4[14],xmm5[14],xmm4[15],xmm5[15]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[u,6,7,2,3,u,u,u,8,9,4,5,u,u,u,10]
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm4, %zmm3, %zmm3
 ; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [2,2,2,4,2,2,2,4,3,3,3,3,2,2,2,4,52,53,52,53,53,54,53,54,52,53,52,53,53,54,53,54]
-; AVX512BW-FAST-NEXT:    vpermi2w %zmm26, %zmm12, %zmm4
+; AVX512BW-FAST-NEXT:    vpermi2w %zmm28, %zmm12, %zmm4
 ; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm3 = zmm3[0,1,0,1,4,5,4,5]
 ; AVX512BW-FAST-NEXT:    movabsq $290499906672591364, %rax # imm = 0x408102040810204
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
 ; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm4, %zmm3 {%k2}
 ; AVX512BW-FAST-NEXT:    movabsq $-8714997200177740921, %rax # imm = 0x870E1C3870E1C387
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm24 {%k2}
-; AVX512BW-FAST-NEXT:    vpshufb %xmm5, %xmm16, %xmm3
-; AVX512BW-FAST-NEXT:    vpshufb %xmm0, %xmm15, %xmm0
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm21 {%k2}
+; AVX512BW-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; AVX512BW-FAST-NEXT:    vpshufb %xmm7, %xmm5, %xmm3
+; AVX512BW-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX512BW-FAST-NEXT:    vpshufb %xmm0, %xmm4, %xmm0
 ; AVX512BW-FAST-NEXT:    vpor %xmm3, %xmm0, %xmm0
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm15[0],xmm16[0],xmm15[1],xmm16[1],xmm15[2],xmm16[2],xmm15[3],xmm16[3],xmm15[4],xmm16[4],xmm15[5],xmm16[5],xmm15[6],xmm16[6],xmm15[7],xmm16[7]
-; AVX512BW-FAST-NEXT:    vpshufb %xmm6, %xmm3, %xmm3
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm4[0],xmm5[0],xmm4[1],xmm5[1],xmm4[2],xmm5[2],xmm4[3],xmm5[3],xmm4[4],xmm5[4],xmm4[5],xmm5[5],xmm4[6],xmm5[6],xmm4[7],xmm5[7]
+; AVX512BW-FAST-NEXT:    vpshufb %xmm31, %xmm3, %xmm3
 ; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm3, %zmm0
-; AVX512BW-FAST-NEXT:    vpshufb %xmm2, %xmm19, %xmm2
-; AVX512BW-FAST-NEXT:    vpshufb %xmm31, %xmm17, %xmm3
+; AVX512BW-FAST-NEXT:    vpshufb %xmm2, %xmm17, %xmm2
+; AVX512BW-FAST-NEXT:    vpshufb %xmm18, %xmm16, %xmm3
 ; AVX512BW-FAST-NEXT:    vpor %xmm2, %xmm3, %xmm2
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm17[0],xmm19[0],xmm17[1],xmm19[1],xmm17[2],xmm19[2],xmm17[3],xmm19[3],xmm17[4],xmm19[4],xmm17[5],xmm19[5],xmm17[6],xmm19[6],xmm17[7],xmm19[7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[0,1,u,u,u,6,7,2,3,u,u,u,8,9,4,5]
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm3 = xmm16[0],xmm17[0],xmm16[1],xmm17[1],xmm16[2],xmm17[2],xmm16[3],xmm17[3],xmm16[4],xmm17[4],xmm16[5],xmm17[5],xmm16[6],xmm17[6],xmm16[7],xmm17[7]
+; AVX512BW-FAST-NEXT:    vpshufb %xmm29, %xmm3, %xmm3
 ; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm2, %zmm3, %zmm2
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[0,1,0,1,4,5,4,5]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm2 = zmm2[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm3 = zmm0[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm2[0,1,0,1,4,5,4,5]
 ; AVX512BW-FAST-NEXT:    movabsq $871499720017774092, %rax # imm = 0xC183060C183060C
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm2 {%k2}
-; AVX512BW-FAST-NEXT:    vpshufb %xmm1, %xmm25, %xmm0
-; AVX512BW-FAST-NEXT:    vpshufb %xmm29, %xmm10, %xmm1
-; AVX512BW-FAST-NEXT:    vpor %xmm0, %xmm1, %xmm0
-; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm10[0],xmm25[0],xmm10[1],xmm25[1],xmm10[2],xmm25[2],xmm10[3],xmm25[3],xmm10[4],xmm25[4],xmm10[5],xmm25[5],xmm10[6],xmm25[6],xmm10[7],xmm25[7]
-; AVX512BW-FAST-NEXT:    vpshufb %xmm9, %xmm1, %xmm1
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[0,1,0,1,4,5,4,5]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm26, %zmm1 # 32-byte Folded Reload
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k2}
+; AVX512BW-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Reload
+; AVX512BW-FAST-NEXT:    vpshufb %xmm1, %xmm3, %xmm1
+; AVX512BW-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
+; AVX512BW-FAST-NEXT:    vpshufb %xmm6, %xmm4, %xmm2
+; AVX512BW-FAST-NEXT:    vpor %xmm1, %xmm2, %xmm1
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} xmm2 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3],xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX512BW-FAST-NEXT:    vpshufb %xmm11, %xmm2, %xmm2
+; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm1, %zmm2, %zmm1
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[0,1,0,1,4,5,4,5]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm28, %zmm2 # 32-byte Folded Reload
 ; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm3 = [1,1,0,0,1,1,0,0,4,5,1,1,1,1,0,0,18,18,18,20,18,18,18,20,19,19,19,19,18,18,18,20]
-; AVX512BW-FAST-NEXT:    vpermw %zmm1, %zmm3, %zmm3
+; AVX512BW-FAST-NEXT:    vpermw %zmm2, %zmm3, %zmm3
 ; AVX512BW-FAST-NEXT:    movabsq $4647998506761461824, %rax # imm = 0x4081020408102040
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm0 {%k2}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm3, %zmm1 {%k2}
 ; AVX512BW-FAST-NEXT:    movabsq $8133997386832558192, %rax # imm = 0x70E1C3870E1C3870
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k2
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm2 {%k2}
-; AVX512BW-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm13, %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vmovdqu64 (%rsp), %zmm3 # 64-byte Reload
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm11, %zmm3, %zmm3
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21],zero,zmm0[19],zero,zmm0[21,20,21,22],zero,zmm0[20],zero,zmm0[22,23,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,57],zero,zmm0[55],zero,zero,zero,zero,zmm0[58],zero,zmm0[56],zero,zero,zero,zero,zmm0[59],zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm3 = zmm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm3[21],zero,zmm3[19],zero,zero,zero,zero,zmm3[22],zero,zmm3[20],zero,zero,zmm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm3[55],zero,zero,zero,zero,zmm3[58],zero,zmm3[56],zero,zero,zero,zero,zmm3[59],zero,zmm3[57]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm3 = zmm3[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm0, %zmm3, %zmm0
-; AVX512BW-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm3 # 64-byte Reload
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm20, %zmm3, %zmm3
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, {{[-0-9]+}}(%r{{[sb]}}p), %zmm23, %zmm4 # 32-byte Folded Reload
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm3 = zmm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18],zero,zmm3[18,19,20,21],zero,zmm3[19],zero,zmm3[25,26,27,22],zero,zmm3[20],zero,zmm3[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57],zero,zmm3[55],zero,zmm3[53,54,55,58],zero,zmm3[56],zero,zmm3[60,61,58,59]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm3 = zmm3[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm4 = zmm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm4[18],zero,zero,zero,zero,zmm4[21],zero,zmm4[19],zero,zero,zero,zero,zmm4[22],zero,zmm4[20,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zmm4[57],zero,zmm4[55],zero,zero,zero,zero,zmm4[58],zero,zmm4[56],zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm4 = zmm4[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm3, %zmm4, %zmm3
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm3 {%k1}
-; AVX512BW-FAST-NEXT:    vmovdqu64 {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm7, %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm14, %zmm21, %zmm4
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20],zero,zmm0[18],zero,zmm0[20,21,20,21],zero,zmm0[19],zero,zmm0[19,20,21,22],zero,zmm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,56,57,56,57],zero,zmm0[55],zero,zmm0[55,56,57,58],zero,zmm0[56],zero,zmm0[62,63]
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} zmm4 = zmm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zmm4[20],zero,zmm4[18],zero,zero,zero,zero,zmm4[21],zero,zmm4[19],zero,zero,zero,zero,zmm4[22,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,zero,zero,zmm4[57],zero,zmm4[55],zero,zero,zero,zero,zmm4[58],zero,zmm4[56],zero,zero
-; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} zmm4 = zmm4[2,3,2,3,6,7,6,7]
-; AVX512BW-FAST-NEXT:    vporq %zmm0, %zmm4, %zmm0
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k2}
+; AVX512BW-FAST-NEXT:    vpshufb %ymm25, %ymm13, %ymm1
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm26, %ymm15, %ymm3
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpor %ymm1, %ymm3, %ymm1
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm3 = ymm15[0],ymm13[0],ymm15[1],ymm13[1],ymm15[2],ymm13[2],ymm15[3],ymm13[3],ymm15[4],ymm13[4],ymm15[5],ymm13[5],ymm15[6],ymm13[6],ymm15[7],ymm13[7],ymm15[16],ymm13[16],ymm15[17],ymm13[17],ymm15[18],ymm13[18],ymm15[19],ymm13[19],ymm15[20],ymm13[20],ymm15[21],ymm13[21],ymm15[22],ymm13[22],ymm15[23],ymm13[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,10,11,6,7,u,u,u,12,13,8,9,u,u,u,u,u,26,27,22,23,u,u,u,28,29,24,25,u,u]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm3, %zmm1
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm23[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25],zero,ymm23[23],zero,ymm23[21,22,23,26],zero,ymm23[24],zero,ymm23[28,29,26,27]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm22[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u],zero,zero,ymm22[25],zero,ymm22[23],zero,zero,zero,zero,ymm22[26],zero,ymm22[24],zero,zero,zero,zero
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpor %ymm3, %ymm4, %ymm3
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm23[0],ymm22[0],ymm23[1],ymm22[1],ymm23[2],ymm22[2],ymm23[3],ymm22[3],ymm23[4],ymm22[4],ymm23[5],ymm22[5],ymm23[6],ymm22[6],ymm23[7],ymm22[7],ymm23[16],ymm22[16],ymm23[17],ymm22[17],ymm23[18],ymm22[18],ymm23[19],ymm22[19],ymm23[20],ymm22[20],ymm23[21],ymm22[21],ymm23[22],ymm22[22],ymm23[23],ymm22[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[4,5,u,u,u,10,11,6,7,u,u,u,12,13,8,9,20,21,u,u,u,26,27,22,23,u,u,u,28,29,24,25]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm3, %zmm4, %zmm3
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm1, %zmm3 {%k1}
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm8[24,25,24,25],zero,ymm8[23],zero,ymm8[23,24,25,26],zero,ymm8[24],zero,ymm8[30,31]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpshufb %ymm24, %ymm9, %ymm4
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vpor %ymm1, %ymm4, %ymm1
+; AVX512BW-FAST-NEXT:    vpunpcklbw {{.*#+}} ymm4 = ymm8[0],ymm9[0],ymm8[1],ymm9[1],ymm8[2],ymm9[2],ymm8[3],ymm9[3],ymm8[4],ymm9[4],ymm8[5],ymm9[5],ymm8[6],ymm9[6],ymm8[7],ymm9[7],ymm8[16],ymm9[16],ymm8[17],ymm9[17],ymm8[18],ymm9[18],ymm8[19],ymm9[19],ymm8[20],ymm9[20],ymm8[21],ymm9[21],ymm8[22],ymm9[22],ymm8[23],ymm9[23]
+; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[8,9,4,5,u,u,u,10,11,6,7,u,u,u,12,13,24,25,20,21,u,u,u,26,27,22,23,u,u,u,28,29]
+; AVX512BW-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,2,3]
+; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm1, %zmm4, %zmm1
 ; AVX512BW-FAST-NEXT:    vmovdqa64 {{.*#+}} zmm4 = [10,9,9,10,10,9,9,10,9,10,14,15,10,9,9,10,27,29,28,27,28,29,29,28,27,29,28,27,28,29,29,28]
-; AVX512BW-FAST-NEXT:    vpermw %zmm1, %zmm4, %zmm1
+; AVX512BW-FAST-NEXT:    vpermw %zmm2, %zmm4, %zmm2
 ; AVX512BW-FAST-NEXT:    movabsq $1161999626690365456, %rax # imm = 0x1020408102040810
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm1, %zmm0 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm2, %zmm1 {%k1}
 ; AVX512BW-FAST-NEXT:    movabsq $2033499346708139548, %rax # imm = 0x1C3870E1C3870E1C
 ; AVX512BW-FAST-NEXT:    kmovq %rax, %k1
-; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm0, %zmm3 {%k1}
+; AVX512BW-FAST-NEXT:    vmovdqu8 %zmm1, %zmm3 {%k1}
 ; AVX512BW-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm14, 320(%rax)
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm3, 128(%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm18, 320(%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm2, (%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm24, 256(%rax)
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rax)
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm21, 256(%rax)
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm27, 192(%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm8, 64(%rax)
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm22, 384(%rax)
-; AVX512BW-FAST-NEXT:    addq $200, %rsp
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm10, 64(%rax)
+; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm19, 384(%rax)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq
   %in.vec0 = load <64 x i8>, ptr %in.vecptr0, align 64
   %in.vec1 = load <64 x i8>, ptr %in.vecptr1, align 64
   %in.vec2 = load <64 x i8>, ptr %in.vecptr2, align 64
   %in.vec3 = load <64 x i8>, ptr %in.vecptr3, align 64
   %in.vec4 = load <64 x i8>, ptr %in.vecptr4, align 64
   %in.vec5 = load <64 x i8>, ptr %in.vecptr5, align 64