DAGCombiner: Extend reduceBuildVecToTrunc to handle non-zero offset

[zvi]

Summary

Adding support for combining power2-strided build_vector's where the
first build_vectori's operand is extracted from a non-zero index.

Example:

v4i32 build_vector((extract_elt V, 1),

(extract_elt V, 3),
(extract_elt V, 5),
(extract_elt V, 7))

-->
v4i32 truncate (bitcast (shuffle<1,u,3,u,5,u,7,u> V, u) to v4i64)

[RKSimon]

include/llvm/Target/TargetLowering.h

2782

It would possibly make sense to change this to

isDesirableToCombineBuildVectorToShuffleTruncate(ArrayRef<int> ShuffleMask, EVT SrcVT, EVT TruncVT)

checking with a general shuffle mask - I'd expect any non-lane crossing shuffle mask to be valid on AVX512.

[guyblank]

test/CodeGen/X86/shuffle-strided-with-offset-256.ll

39

should this be

vpmovdb %zmm0, (%rsi)

same for AVX512VL

[guyblank]

include/llvm/Target/TargetLowering.h

2779

shuffle mask should be <1,u,3,u,5,u,7,u>

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

14380–14381

update comment

14390

I think you should add
Offset != 0
to the condition

14416

you can use TLI

lib/Target/X86/X86ISelLowering.cpp

35835

what about VPERMW for 16-bit elements?

[zvi]

include/llvm/Target/TargetLowering.h

2779

right

2782

Thanks. Will look into this suggestion.

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

14380–14381

ok

14390

Makes sense. To remain consistent with current flow.

14416

ok

lib/Target/X86/X86ISelLowering.cpp

35835

We need to improve lowerBuildVector to make use of VPERMW. Until then we should go with this combine. Will add a TODO.

test/CodeGen/X86/shuffle-strided-with-offset-256.ll

39

Look like a follow-up work needed here.

[zvi]

test/CodeGen/X86/shuffle-strided-with-offset-256.ll

39

Ah you were referring to the regalloc issue that's been reported in pr32719

[zvi]

Addressing Guy's comments

[zvi]

Follow Simon's suggestion to generalize isDesirableToCombineBuildVectorToShuffleTruncate() to allow it to handle more cases.
The patch still only handles monotonically increasing indices - added TODO's to be revisited in follow-up work.

[guyblank]

I'm ok with this.

@RKSimon ?

[RKSimon]

A few minor comments, but looks almost ready.

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

14428

Construct the truncate.

lib/Target/X86/X86ISelLowering.cpp

35833

assert for isShuffleMaskLegal to double check?

35844

You could probably replace all of this with:

if(!is128BitLaneCrossingShuffleMask(SrcVT, ShuffleMask))
  return false;

35847

Better to do this at the top to early-out if we don't support it. Then just return true at the end.

[zvi]

Implemented Simon's suggestions.

[RKSimon]

lib/Target/X86/X86ISelLowering.cpp

35810

No, this should be an early-out:

if (SrcVT.getScalarSizeInBits() == 32 || !Subtarget.hasAVX2())
  return false;

[zvi]

lib/Target/X86/X86ISelLowering.cpp

35810

Whoops this is not what i intended to upload.

[zvi]

The previous revision was a mishap caused by hitting save on a stale copy. Sorry.

[RKSimon]

LGTM with one minor comment

lib/Target/X86/X86ISelLowering.cpp

35813

Do we need the isSimple() test? It's handled by isShuffleMaskLegal

[zvi]

lib/Target/X86/X86ISelLowering.cpp

35813

Then in that case i will drop it. Thanks.

[zvi]

Drop the isSimple() test.

include/llvm/Target/TargetLowering.h

public:
   //                    (extract_elt V, 4),
   //                    (extract_elt V, 6))
   //  -->
   // v4i32 truncate (bitcast V to v4i64)
   virtual bool isDesirableToCombineBuildVectorToTruncate() const {
     return false;
   }
 
+  // Return true if it is profitable to combine a BUILD_VECTOR with a stride-pattern
+  // to a shuffle and a truncate.
+  // Example of such a combine:
+  // v4i32 build_vector((extract_elt V, 1),
+  //                    (extract_elt V, 3),
+  //                    (extract_elt V, 5),
+  //                    (extract_elt V, 7))
+  //  -->
+  // v4i32 truncate (bitcast (shuffle<1,u,3,u,5,u,7,u> V, u) to v4i64)
+  virtual bool isDesirableToCombineBuildVectorToShuffleTruncate(
+      ArrayRef<int> ShuffleMask, EVT SrcVT, EVT TruncVT) const {
+    return false;
+  }
+
   /// Return true if the target has native support for the specified value type
   /// and it is 'desirable' to use the type for the given node type. e.g. On x86
   /// i16 is legal, but undesirable since i16 instruction encodings are longer
   /// and some i16 instructions are slow.
   virtual bool isTypeDesirableForOp(unsigned /*Opc*/, EVT VT) const {
     // By default, assume all legal types are desirable.
     return isTypeLegal(VT);
   }

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

for (unsigned In = 0, Len = CurSize / 2; In < Len; ++In) {
           DAG.getVectorShuffle(VT, DL, Shuffles[Left], Shuffles[Right], Mask);
     }
   }
 
   return Shuffles[0];
 }
 
 // Check to see if this is a BUILD_VECTOR of a bunch of EXTRACT_VECTOR_ELT
-// operations which can be matched to a truncate.
+// operations which can be matched to a truncate or to a shuffle-truncate.
 SDValue DAGCombiner::reduceBuildVecToTrunc(SDNode *N) {
   // TODO: Add support for big-endian.
   if (DAG.getDataLayout().isBigEndian())
     return SDValue();
   if (N->getNumOperands() < 2)
     return SDValue();
   SDLoc DL(N);
   EVT VT = N->getValueType(0);
if (any_of(N->ops(), [VT](SDValue Op) {
                Op.getValueType() != VT.getVectorElementType();
       }))
     return SDValue();
 
   // Helper for obtaining an EXTRACT_VECTOR_ELT's constant index
   auto GetExtractIdx = [](SDValue Extract) {
     return cast<ConstantSDNode>(Extract.getOperand(1))->getSExtValue();
   };
 
-  // The first BUILD_VECTOR operand must be an an extract from index zero
-  // (assuming no undef and little-endian).
-  if (GetExtractIdx(N->getOperand(0)) != 0)
-    return SDValue();
+  // The offset is defined to be the BUILD_VECTOR's first operand (assuming no
+  // undef and little-endian).
+  int Offset = GetExtractIdx(N->getOperand(0));
 
-  // Compute the stride from the first index.
-  int Stride = GetExtractIdx(N->getOperand(1));
+  // Compute the stride from the next operand.
+  int Stride = GetExtractIdx(N->getOperand(1)) - Offset;
   SDValue ExtractedFromVec = N->getOperand(0).getOperand(0);
 
   // Proceed only if the stride and the types can be matched to a truncate.
   if ((Stride == 1 || !isPowerOf2_32(Stride)) ||
       (ExtractedFromVec.getValueType().getVectorNumElements() !=
        Stride * NumElems) ||
       (VT.getScalarSizeInBits() * Stride > 64))
     return SDValue();
 
   // Check remaining operands are consistent with the computed stride.
   for (unsigned i = 1; i != NumElems; ++i) {
     SDValue Op = N->getOperand(i);
 
     if ((Op.getOperand(0) != ExtractedFromVec) ||
-        (GetExtractIdx(Op) != Stride * i))
+        (GetExtractIdx(Op) != Stride * i + Offset))
       return SDValue();
   }
 
-  // All checks were ok, construct the truncate.
+  SDValue Res = ExtractedFromVec;
+  EVT TruncVT =
+      VT.isFloatingPoint() ? VT.changeVectorElementTypeToInteger() : VT;
+  if (Offset) {
+    // If the first index is non-zero, need to shuffle elements of interest to
+    // lower parts of the vector's elements the truncate will act upon.
+    // TODO: Generalize to compute the permute-shuffle that will prepare any
+    // element permutation for the truncate, and let the target decide if
+    // profitable.
+    EVT ExtractedVT = ExtractedFromVec.getValueType();
+    SmallVector<int, 64> Mask;
+    for (unsigned i = 0; i != NumElems; ++i) {
+      Mask.push_back(Offset + i * Stride);
+      // Pad the elements that will be lost after the truncate with undefs.
+      Mask.append(Stride - 1, -1);
+    }
+    if (!TLI.isShuffleMaskLegal(Mask, ExtractedVT) ||
+        !TLI.isDesirableToCombineBuildVectorToShuffleTruncate(Mask, ExtractedVT,
+                                                              TruncVT))
+      return SDValue();
+    Res = DAG.getVectorShuffle(ExtractedVT, SDLoc(N), Res,
+                               DAG.getUNDEF(ExtractedVT), Mask);
+  }
+  // Construct the truncate.
   LLVMContext &Ctx = *DAG.getContext();
   EVT NewVT = VT.getVectorVT(
       Ctx, EVT::getIntegerVT(Ctx, VT.getScalarSizeInBits() * Stride), NumElems);
-  EVT TruncVT =
-      VT.isFloatingPoint() ? VT.changeVectorElementTypeToInteger() : VT;
 
-  SDValue Res = DAG.getBitcast(NewVT, ExtractedFromVec);
+  Res = DAG.getBitcast(NewVT, Res);
   Res = DAG.getNode(ISD::TRUNCATE, SDLoc(N), TruncVT, Res);
   return DAG.getBitcast(VT, Res);
 }
 
 SDValue DAGCombiner::visitBUILD_VECTOR(SDNode *N) {
   EVT VT = N->getValueType(0);
 
   // A vector built entirely of undefs is undef.

lib/Target/X86/X86ISelLowering.h

public:
     //                    (extract_elt V, 4),
     //                    (extract_elt V, 6))
     //  -->
     // v4i32 truncate (bitcast V to v4i64)
     bool isDesirableToCombineBuildVectorToTruncate() const override {
       return true;
     }
 
+    // Return true if it is profitable to combine a BUILD_VECTOR with a
+    // stride-pattern to a shuffle and a truncate.
+    // Example of such a combine:
+    // v4i32 build_vector((extract_elt V, 1),
+    //                    (extract_elt V, 3),
+    //                    (extract_elt V, 5),
+    //                    (extract_elt V, 7))
+    //  -->
+    // v4i32 truncate (bitcast (shuffle<1,u,3,u,4,u,5,u,6,u,7,u> V, u) to
+    // v4i64)
+    bool isDesirableToCombineBuildVectorToShuffleTruncate(
+        ArrayRef<int> ShuffleMask, EVT SrcVT, EVT TruncVT) const override;
+
     /// Return true if the target has native support for
     /// the specified value type and it is 'desirable' to use the type for the
     /// given node type. e.g. On x86 i16 is legal, but undesirable since i16
     /// instruction encodings are longer and some i16 instructions are slow.
     bool isTypeDesirableForOp(unsigned Opc, EVT VT) const override;
 
     /// Return true if the target has native support for the
     /// specified value type and it is 'desirable' to use the type. e.g. On x86

lib/Target/X86/X86ISelLowering.cpp

     Promote = true;
   }
   }
 
   PVT = MVT::i32;
   return Promote;
 }
 
+bool X86TargetLowering::
+    isDesirableToCombineBuildVectorToShuffleTruncate(
+        ArrayRef<int> ShuffleMask, EVT SrcVT, EVT TruncVT) const {
+
+  assert(SrcVT.getVectorNumElements() == ShuffleMask.size() &&
+         "Element count mismatch");
+  assert(
+      Subtarget.getTargetLowering()->isShuffleMaskLegal(ShuffleMask, SrcVT) &&
+      "Shuffle Mask expected to be legal");
+
+  // For 32-bit elements VPERMD is better than shuffle+truncate.
+  // TODO: After we improve lowerBuildVector, add execption for VPERMW.
+  return (SrcVT.getScalarSizeInBits() != 32 && Subtarget.hasAVX2());
+
+  if (!SrcVT.isSimple() ||
+      is128BitLaneCrossingShuffleMask(SrcVT.getSimpleVT(), ShuffleMask))
+    return false;
+
+  return true;
+}
+
 //===----------------------------------------------------------------------===//
 //                           X86 Inline Assembly Support
 //===----------------------------------------------------------------------===//
 
 // Helper to match a string separated by whitespace.
 static bool matchAsm(StringRef S, ArrayRef<const char *> Pieces) {
   S = S.substr(S.find_first_not_of(" \t")); // Skip leading whitespace.
 
   for (StringRef Piece : Pieces) {
     if (!S.startswith(Piece)) // Check if the piece matches.
       return false;
 
     S = S.substr(Piece.size());
     StringRef::size_type Pos = S.find_first_not_of(" \t");
     if (Pos == 0) // We matched a prefix.
       return false;
 
     S = S.substr(Pos);
   }
 
   return S.empty();
 }
 
 static bool clobbersFlagRegisters(const SmallVector<StringRef, 4> &AsmPieces) {
 
   if (AsmPieces.size() == 3 || AsmPieces.size() == 4) {
     if (std::count(AsmPieces.begin(), AsmPieces.end(), "~{cc}") &&
         std::count(AsmPieces.begin(), AsmPieces.end(), "~{flags}") &&
         std::count(AsmPieces.begin(), AsmPieces.end(), "~{fpsr}")) {
 
       if (AsmPieces.size() == 3)
         return true;
       else if (std::count(AsmPieces.begin(), AsmPieces.end(), "~{dirflag}"))
         return true;
     }
   }
   return false;

test/CodeGen/X86/shuffle-strided-with-offset-256.ll

 ; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v16i8_1:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovsxwd %ymm0, %zmm0
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512F-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v16i8_1:
 ; AVX512VL:       # BB#0:
 ; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u>
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovsxwd %ymm0, %zmm0
+; AVX512VL-NEXT:    vpmovdb %zmm0, %xmm0
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v16i8_1:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512BW-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v16i8_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = <1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u>
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlw $8, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovwb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
   store <16 x i8> %strided.vec, <16 x i8>* %S
   ret void
 }
 
 define void @shuffle_v16i16_to_v8i16_1(<16 x i16>* %L, <8 x i16>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v16i16_to_v8i16_1:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; AVX1-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_to_v8i16_1:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,6,7,10,11,14,15,10,11,14,15,14,15],zero,zero,ymm0[18,19,22,23,26,27,30,31,26,27,30,31,30,31],zero,zero
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
 ; AVX2-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v8i16_1:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512F-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512F-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_to_v8i16_1:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovdw %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v16i16_to_v8i16_1:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512BW-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
 ; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v16i16_to_v8i16_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovdw %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <16 x i16>, <16 x i16>* %L
   %strided.vec = shufflevector <16 x i16> %vec, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   store <8 x i16> %strided.vec, <8 x i16>* %S
   ret void
 }
 

 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; AVX1-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v8i8_1:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[1,u,5,u,9,u,13,u,u,u,u,u,u,u,u,u,17,u,21,u,25,u,29,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v8i8_1:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v8i8_1:
 ; AVX512VL:       # BB#0:
 ; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v8i8_1:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512BW-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v8i8_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [1,1,5,5,9,9,13,13,13,13,5,5,12,12,13,13]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlw $8, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v8i8_2(<32 x i8>* %L, <8 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v8i8_2:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; AVX1-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v8i8_2:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,30,31,26,27,28,29,30,31]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v8i8_2:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512F-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v8i8_2:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v8i8_2:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,6,10,14,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512BW-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v8i8_2:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v8i8_3(<32 x i8>* %L, <8 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v8i8_3:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,7,11,15,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; AVX1-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v8i8_3:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,7,11,15,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[3,u,7,u,11,u,15,u,u,u,u,u,u,u,u,u,19,u,23,u,27,u,31,u,u,u,u,u,u,u,u,u]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v8i8_3:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,7,11,15,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512F-NEXT:    vpsrld $24, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v8i8_3:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,7,11,15,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrld $24, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovdb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v8i8_3:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,7,11,15,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512BW-NEXT:    vpsrld $24, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v8i8_3:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [3,3,7,7,11,11,15,15,7,7,15,15,6,6,7,7]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $24, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovdb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; AVX1-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_to_v4i16_1:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX2-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16_1:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512F-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512F-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512F-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX512F-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_to_v4i16_1:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,1,3,3,4,5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,1,3,3,4,5,6,7]
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdw %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqw %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v16i16_to_v4i16_1:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,3,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512BW-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v16i16_to_v4i16_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,1,3,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,1,3,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdw %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <16 x i16>, <16 x i16>* %L
   %strided.vec = shufflevector <16 x i16> %vec, <16 x i16> undef, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
   store <4 x i16> %strided.vec, <4 x i16>* %S
   ret void
 }
 
 define void @shuffle_v16i16_to_v4i16_2(<16 x i16>* %L, <4 x i16>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
 ; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; AVX1-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm0 = [1,3,5,7,5,7,7,7]
+; AVX2-NEXT:    vpermd (%rdi), %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX2-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512F-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512F-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512F-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX512F-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
-; AVX512VL-NEXT:    vpmovdw %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512VL-NEXT:    vpmovqw %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,0,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v16i16_to_v4i16_2:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
-; AVX512BWVL-NEXT:    vpmovdw %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <16 x i16>, <16 x i16>* %L
   %strided.vec = shufflevector <16 x i16> %vec, <16 x i16> undef, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
   store <4 x i16> %strided.vec, <4 x i16>* %S
   ret void
 }
 

 ; AVX1-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
 ; AVX1-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v16i16_to_v4i16_3:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX2-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX2-NEXT:    vpsrlq $48, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX2-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v16i16_to_v4i16_3:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512F-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512F-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512F-NEXT:    vpsrlq $48, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX512F-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_to_v4i16_3:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdw %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrlq $48, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqw %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v16i16_to_v4i16_3:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512BW-NEXT:    vpsrlq $48, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v16i16_to_v4i16_3:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdw %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $48, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqw %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <16 x i16>, <16 x i16>* %L
   %strided.vec = shufflevector <16 x i16> %vec, <16 x i16> undef, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
   store <4 x i16> %strided.vec, <4 x i16>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_1(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_1:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_1:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_1:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_1:
 ; AVX512VL:       # BB#0:
 ; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,0,1,1,8,8,9,9,8,8,9,9,10,10,11,11]
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_1:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [0,0,1,1,8,8,9,9,8,8,9,9,10,10,11,11]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpsrld $16, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlw $8, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 1, i32 9, i32 17, i32 25>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_2(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_2:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,10,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_2:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,10,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_2:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,10,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_2:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,1,3,3,4,5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,1,3,3,4,5,6,7]
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_2:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <2,10,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpsrld $16, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_2:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[1,1,3,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,1,3,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 2, i32 10, i32 18, i32 26>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_3(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_3:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,11,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_3:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,11,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsrld $24, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_3:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,11,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpsrld $24, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_3:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [10,10,11,11,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrld $24, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_3:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <3,11,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpsrld $24, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_3:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [10,10,11,11,2,2,3,3,8,8,9,9,10,10,11,11]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $24, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 3, i32 11, i32 19, i32 27>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_4(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_4:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <4,12,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_4:
 ; AVX2:       # BB#0:
-; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <4,12,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vmovdqa {{.*#+}} ymm0 = [1,3,5,7,5,7,7,7]
+; AVX2-NEXT:    vpermd (%rdi), %ymm0, %ymm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_4:
 ; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <4,12,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_4:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_4:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <4,12,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_4:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpshufd {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 4, i32 12, i32 20, i32 28>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_5(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_5:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <5,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_5:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <5,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsrlq $40, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_5:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <5,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpsrlq $40, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_5:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrlq $40, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_5:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <5,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpsrlq $40, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_5:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BWVL-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $40, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 5, i32 13, i32 21, i32 29>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_6(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_6:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <6,14,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_6:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <6,14,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsrlq $48, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_6:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <6,14,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpsrlq $48, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_6:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrlq $48, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_6:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <6,14,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpsrlq $48, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_6:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BWVL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[3,1,1,3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $48, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 6, i32 14, i32 22, i32 30>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 
 define void @shuffle_v32i8_to_v4i8_7(<32 x i8>* %L, <4 x i8>* %S) nounwind {
 ; AVX1-LABEL: shuffle_v32i8_to_v4i8_7:
 ; AVX1:       # BB#0:
 ; AVX1-NEXT:    vmovdqa (%rdi), %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = <7,15,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
 ; AVX1-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: shuffle_v32i8_to_v4i8_7:
 ; AVX2:       # BB#0:
 ; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <7,15,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX2-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-NEXT:    vpsrlq $56, %ymm0, %ymm0
+; AVX2-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,2,2,3,4,6,6,7]
+; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX2-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: shuffle_v32i8_to_v4i8_7:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512F-NEXT:    vmovdqa {{.*#+}} xmm2 = <7,15,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512F-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512F-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512F-NEXT:    vpsrlq $56, %ymm0, %ymm0
+; AVX512F-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512F-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512F-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v32i8_to_v4i8_7:
 ; AVX512VL:       # BB#0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,14,14,15,15,14,14,15,15,4,4,5,5,6,6]
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512VL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512VL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512VL-NEXT:    vpsrlq $56, (%rdi), %ymm0
+; AVX512VL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i8_to_v4i8_7:
 ; AVX512BW:       # BB#0:
 ; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BW-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = <7,15,u,u,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX512BW-NEXT:    vpsrlq $56, %ymm0, %ymm0
+; AVX512BW-NEXT:    vpmovqd %zmm0, %ymm0
+; AVX512BW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u]
 ; AVX512BW-NEXT:    vmovd %xmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i8_to_v4i8_7:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} xmm2 = [7,7,14,14,15,15,14,14,15,15,4,4,5,5,6,6]
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
-; AVX512BWVL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX512BWVL-NEXT:    vpmovdb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $56, (%rdi), %ymm0
+; AVX512BWVL-NEXT:    vpmovqb %ymm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i8>, <32 x i8>* %L
   %strided.vec = shufflevector <32 x i8> %vec, <32 x i8> undef, <4 x i32> <i32 7, i32 15, i32 23, i32 31>
   store <4 x i8> %strided.vec, <4 x i8>* %S
   ret void
 }
 

test/CodeGen/X86/shuffle-strided-with-offset-512.ll

 ; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX512VL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX512VL-NEXT:    vmovdqa %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v32i8_1:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u,17,19,21,23,25,27,29,31]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u,17,19,21,23,25,27,29,31,u,u,u,u,u,u,u,u]
-; AVX512BW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX512BW-NEXT:    vmovdqa %ymm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlw $8, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v32i8_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u,17,19,21,23,25,27,29,31]
-; AVX512BWVL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u,17,19,21,23,25,27,29,31,u,u,u,u,u,u,u,u]
-; AVX512BWVL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX512BWVL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX512BWVL-NEXT:    vmovdqu %ymm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlw $8, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovwb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <32 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 33, i32 35, i32 37, i32 39, i32 41, i32 43, i32 45, i32 47, i32 49, i32 51, i32 53, i32 55, i32 57, i32 59, i32 61, i32 63>
   store <32 x i8> %strided.vec, <32 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
 ; AVX512VL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX512VL-NEXT:    vmovdqa %ymm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i16_to_v16i16_1:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[6,7,2,3,4,5,6,7,2,3,6,7,10,11,14,15,22,23,18,19,20,21,22,23,18,19,22,23,26,27,30,31]
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,30,31,26,27,28,29,30,31]
-; AVX512BW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX512BW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX512BW-NEXT:    vmovdqa %ymm0, (%rsi)
+; AVX512BW-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovdw %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i16_to_v16i16_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BWVL-NEXT:    vmovdqu {{.*#+}} ymm2 = [1,3,5,7,17,19,21,23,9,11,13,15,25,27,29,31]
-; AVX512BWVL-NEXT:    vpermi2w %ymm1, %ymm0, %ymm2
-; AVX512BWVL-NEXT:    vpermq {{.*#+}} ymm0 = ymm2[0,2,1,3]
-; AVX512BWVL-NEXT:    vmovdqu %ymm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovdw %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i16>, <32 x i16>* %L
   %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>
   store <16 x i16> %strided.vec, <16 x i16>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v16i8_1:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX512BW-NEXT:    vpextrb $1, %xmm0, %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm1
-; AVX512BW-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
-; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlw $8, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovdb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v16i8_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlw $8, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovdb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>
   store <16 x i8> %strided.vec, <16 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v16i8_2:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX512BW-NEXT:    vpextrb $2, %xmm0, %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm1
-; AVX512BW-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
-; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovdb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v16i8_2:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovdb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>
   store <16 x i8> %strided.vec, <16 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v16i8_3:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX512BW-NEXT:    vpextrb $3, %xmm0, %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm1
-; AVX512BW-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
-; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrld $24, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovdb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v16i8_3:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $1, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $3, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $5, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $7, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $9, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $11, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $13, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $24, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovdb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>
   store <16 x i8> %strided.vec, <16 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i16_to_v8i16_1:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BW-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX512BW-NEXT:    vpextrw $1, %xmm0, %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm1
-; AVX512BW-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrw $1, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $2, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrw $5, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrw $1, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3,4],xmm2[5],xmm1[6,7]
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $1, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm0
-; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqw %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i16_to_v8i16_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrw $1, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrw $1, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrw $5, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrw $1, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3,4],xmm2[5],xmm1[6,7]
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrw $1, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrw $5, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqw %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i16>, <32 x i16>* %L
   %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>
   store <8 x i16> %strided.vec, <8 x i16>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i16_to_v8i16_2:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BW-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX512BW-NEXT:    vpextrw $2, %xmm0, %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm1
-; AVX512BW-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2],xmm1[3,4,5,6,7]
-; AVX512BW-NEXT:    vpextrw $6, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrw $2, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrw $6, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $5, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $2, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm0
-; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpshufd {{.*#+}} zmm0 = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; AVX512BW-NEXT:    vpmovqw %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i16_to_v8i16_2:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrw $2, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2],xmm1[3,4,5,6,7]
-; AVX512BWVL-NEXT:    vpextrw $6, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrw $2, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrw $6, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $5, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrw $2, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrw $6, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpshufd {{.*#+}} zmm0 = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; AVX512BWVL-NEXT:    vpmovqw %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i16>, <32 x i16>* %L
   %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>
   store <8 x i16> %strided.vec, <8 x i16>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; AVX512VL-NEXT:    vmovdqa %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v32i16_to_v8i16_3:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BW-NEXT:    vpextrw $7, %xmm0, %eax
-; AVX512BW-NEXT:    vpextrw $3, %xmm0, %ecx
-; AVX512BW-NEXT:    vmovd %ecx, %xmm1
-; AVX512BW-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrw $3, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $2, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrw $7, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vpextrw $3, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpextrw $7, %xmm2, %eax
-; AVX512BW-NEXT:    vpinsrw $5, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vpextrw $3, %xmm0, %eax
-; AVX512BW-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,6],xmm0[7]
-; AVX512BW-NEXT:    vmovdqa %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlq $48, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqw %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v32i16_to_v8i16_3:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu16 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrw $7, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrw $3, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrw $3, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrw $7, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrw $3, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrw $7, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrw $5, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrw $3, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3,4,5,6],xmm0[7]
-; AVX512BWVL-NEXT:    vmovdqu %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $48, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqw %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <32 x i16>, <32 x i16>* %L
   %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>
   store <8 x i16> %strided.vec, <8 x i16>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_1:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $1, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $1, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $9, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $1, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $9, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $1, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlw $8, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_1:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $1, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $9, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlw $8, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 1, i32 9, i32 17, i32 25, i32 33, i32 41, i32 49, i32 57>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_2:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $2, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $2, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $10, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $2, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $10, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $2, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_2:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $2, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $10, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $16, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 2, i32 10, i32 18, i32 26, i32 34, i32 42, i32 50, i32 58>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_3:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $3, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $3, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $11, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $3, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $11, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $3, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrld $24, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_3:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $3, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $11, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrld $24, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 3, i32 11, i32 19, i32 27, i32 35, i32 43, i32 51, i32 59>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_4:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $12, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $4, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $12, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $4, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $12, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $4, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $12, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $4, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpshufd {{.*#+}} zmm0 = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_4:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $4, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $4, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $12, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $4, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $12, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $4, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $12, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpshufd {{.*#+}} zmm0 = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 4, i32 12, i32 20, i32 28, i32 36, i32 44, i32 52, i32 60>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_5:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $5, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $5, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $13, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $5, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $13, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $5, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlq $40, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_5:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $5, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $13, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $40, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 5, i32 13, i32 21, i32 29, i32 37, i32 45, i32 53, i32 61>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_6:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $6, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $6, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $14, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $6, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $14, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $6, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlq $48, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_6:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $6, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $14, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $48, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 6, i32 14, i32 22, i32 30, i32 38, i32 46, i32 54, i32 62>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

 ; AVX512VL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
 ; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3]
 ; AVX512VL-NEXT:    vmovq %xmm0, (%rsi)
 ; AVX512VL-NEXT:    vzeroupper
 ; AVX512VL-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v8i8_7:
 ; AVX512BW:       # BB#0:
-; AVX512BW-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BW-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm1, %r8d
-; AVX512BW-NEXT:    vpextrb $7, %xmm1, %r9d
-; AVX512BW-NEXT:    vextracti32x4 $2, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm1, %r10d
-; AVX512BW-NEXT:    vpextrb $7, %xmm1, %r11d
-; AVX512BW-NEXT:    vextracti32x4 $1, %zmm0, %xmm1
-; AVX512BW-NEXT:    vpextrb $15, %xmm1, %eax
-; AVX512BW-NEXT:    vpextrb $7, %xmm1, %ecx
-; AVX512BW-NEXT:    vpextrb $15, %xmm0, %edx
-; AVX512BW-NEXT:    vpextrb $7, %xmm0, %edi
-; AVX512BW-NEXT:    vmovd %edi, %xmm0
-; AVX512BW-NEXT:    vpinsrb $1, %edx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $2, %ecx, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $3, %eax, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $4, %r11d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $5, %r10d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $6, %r9d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vpinsrb $7, %r8d, %xmm0, %xmm0
-; AVX512BW-NEXT:    vmovq %xmm0, (%rsi)
+; AVX512BW-NEXT:    vpsrlq $56, (%rdi), %zmm0
+; AVX512BW-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512BWVL-LABEL: shuffle_v64i8_to_v8i8_7:
 ; AVX512BWVL:       # BB#0:
-; AVX512BWVL-NEXT:    vmovdqu8 (%rdi), %zmm0
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm0, %ecx
-; AVX512BWVL-NEXT:    vmovd %ecx, %xmm1
-; AVX512BWVL-NEXT:    vpinsrb $2, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $4, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $6, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $8, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm2, %eax
-; AVX512BWVL-NEXT:    vpinsrb $10, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; AVX512BWVL-NEXT:    vpextrb $7, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $12, %eax, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpextrb $15, %xmm0, %eax
-; AVX512BWVL-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm0
-; AVX512BWVL-NEXT:    vpmovwb %xmm0, (%rsi)
+; AVX512BWVL-NEXT:    vpsrlq $56, (%rdi), %zmm0
+; AVX512BWVL-NEXT:    vpmovqb %zmm0, (%rsi)
 ; AVX512BWVL-NEXT:    vzeroupper
 ; AVX512BWVL-NEXT:    retq
   %vec = load <64 x i8>, <64 x i8>* %L
   %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 7, i32 15, i32 23, i32 31, i32 39, i32 47, i32 55, i32 63>
   store <8 x i8> %strided.vec, <8 x i8>* %S
   ret void
 }
 

test/CodeGen/X86/vector-shuffle-512-v32.ll

 ; KNL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
 ; KNL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[1,3,2,3,4,5,6,7]
 ; KNL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
 ; KNL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: pr32967:
 ; SKX:       ## BB#0:
-; SKX-NEXT:    vpextrw $5, %xmm0, %eax
-; SKX-NEXT:    vpextrw $1, %xmm0, %ecx
-; SKX-NEXT:    vmovd %ecx, %xmm1
-; SKX-NEXT:    vpinsrw $1, %eax, %xmm1, %xmm1
-; SKX-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
-; SKX-NEXT:    vpextrw $1, %xmm2, %eax
-; SKX-NEXT:    vpinsrw $2, %eax, %xmm1, %xmm1
-; SKX-NEXT:    vpextrw $5, %xmm2, %eax
-; SKX-NEXT:    vpinsrw $3, %eax, %xmm1, %xmm1
-; SKX-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
-; SKX-NEXT:    vpextrw $1, %xmm2, %eax
-; SKX-NEXT:    vpinsrw $4, %eax, %xmm1, %xmm1
-; SKX-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3,4],xmm2[5],xmm1[6,7]
-; SKX-NEXT:    vextracti32x4 $3, %zmm0, %xmm0
-; SKX-NEXT:    vpextrw $1, %xmm0, %eax
-; SKX-NEXT:    vpinsrw $6, %eax, %xmm1, %xmm1
-; SKX-NEXT:    vpextrw $5, %xmm0, %eax
-; SKX-NEXT:    vpinsrw $7, %eax, %xmm1, %xmm0
+; SKX-NEXT:    vpsrld $16, %zmm0, %zmm0
+; SKX-NEXT:    vpmovqw %zmm0, %xmm0
 ; SKX-NEXT:    vzeroupper
 ; SKX-NEXT:    retq
  %shuffle = shufflevector <32 x i16> %v, <32 x i16> undef, <8 x i32> <i32 1,i32 5,i32 9,i32 13,i32 17,i32 21,i32 25,i32 29>
  ret <8 x i16> %shuffle
 }