diff --git a/llvm/lib/Target/X86/X86.td b/llvm/lib/Target/X86/X86.td
--- a/llvm/lib/Target/X86/X86.td
+++ b/llvm/lib/Target/X86/X86.td
@@ -323,10 +323,15 @@
                                       "platform configuration instruction">;
 // On recent X86 (port bound) processors, its preferable to combine to a single shuffle
 // using a variable mask over multiple fixed shuffles.
-def FeatureFastVariableShuffle
-    : SubtargetFeature<"fast-variable-shuffle",
-                       "HasFastVariableShuffle",
-                       "true", "Shuffles with variable masks are fast">;
+def FeatureFastVariableCrossLaneShuffle
+    : SubtargetFeature<"fast-variable-crosslane-shuffle",
+                       "HasFastVariableCrossLaneShuffle",
+                       "true", "Cross-lane shuffles with variable masks are fast">;
+def FeatureFastVariablePerLaneShuffle
+    : SubtargetFeature<"fast-variable-perlane-shuffle",
+                       "HasFastVariablePerLaneShuffle",
+                       "true", "Per-lane shuffles with variable masks are fast">;
+
 // On some X86 processors, a vzeroupper instruction should be inserted after
 // using ymm/zmm registers before executing code that may use SSE instructions.
 def FeatureInsertVZEROUPPER
@@ -639,7 +644,8 @@
                                       FeatureFastScalarFSQRT,
                                       FeatureFastSHLDRotate,
                                       FeatureFast15ByteNOP,
-                                      FeatureFastVariableShuffle,
+                                      FeatureFastVariableCrossLaneShuffle,
+                                      FeatureFastVariablePerLaneShuffle,
                                       FeaturePOPCNTFalseDeps,
                                       FeatureLZCNTFalseDeps,
                                       FeatureInsertVZEROUPPER];
@@ -667,7 +673,8 @@
                                       FeatureFastVectorFSQRT,
                                       FeatureFastSHLDRotate,
                                       FeatureFast15ByteNOP,
-                                      FeatureFastVariableShuffle,
+                                      FeatureFastVariableCrossLaneShuffle,
+                                      FeatureFastVariablePerLaneShuffle,
                                       FeaturePOPCNTFalseDeps,
                                       FeatureInsertVZEROUPPER];
   list<SubtargetFeature> SKLFeatures =
@@ -693,7 +700,8 @@
                                       FeatureFastVectorFSQRT,
                                       FeatureFastSHLDRotate,
                                       FeatureFast15ByteNOP,
-                                      FeatureFastVariableShuffle,
+                                      FeatureFastVariableCrossLaneShuffle,
+                                      FeatureFastVariablePerLaneShuffle,
                                       FeaturePrefer256Bit,
                                       FeaturePOPCNTFalseDeps,
                                       FeatureInsertVZEROUPPER];
@@ -730,7 +738,8 @@
                                       FeatureFastVectorFSQRT,
                                       FeatureFastSHLDRotate,
                                       FeatureFast15ByteNOP,
-                                      FeatureFastVariableShuffle,
+                                      FeatureFastVariableCrossLaneShuffle,
+                                      FeatureFastVariablePerLaneShuffle,
                                       FeaturePrefer256Bit,
                                       FeatureInsertVZEROUPPER];
   list<SubtargetFeature> CNLFeatures =
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -15967,9 +15967,9 @@
   if (SDValue V = getSublanePermute(/*NumSublanes=*/NumLanes * 2))
     return V;
 
-  // If that doesn't work and we have fast variable shuffle,
+  // If that doesn't work and we have fast variable cross-lane shuffle,
   // attempt 32-bit sublanes (vpermd).
-  if (!Subtarget.hasFastVariableShuffle())
+  if (!Subtarget.hasFastVariableCrossLaneShuffle())
     return SDValue();
 
   return getSublanePermute(/*NumSublanes=*/NumLanes * 4);
@@ -16475,7 +16475,7 @@
         if (EltWidth == 32 && NumLowerHalves && HalfVT.is128BitVector() &&
             !is128BitUnpackShuffleMask(HalfMask) &&
             (!isSingleSHUFPSMask(HalfMask) ||
-             Subtarget.hasFastVariableShuffle()))
+             Subtarget.hasFastVariableCrossLaneShuffle()))
           return SDValue();
         // If this is a unary shuffle (assume that the 2nd operand is
         // canonicalized to undef), then we can use vpermpd. Otherwise, we
@@ -35542,7 +35542,8 @@
 
 static SDValue combineX86ShuffleChainWithExtract(
     ArrayRef<SDValue> Inputs, SDValue Root, ArrayRef<int> BaseMask, int Depth,
-    bool HasVariableMask, bool AllowVariableMask, SelectionDAG &DAG,
+    bool HasVariableMask, bool AllowVariableCrossLaneMask,
+    bool AllowVariablePerLaneMask, SelectionDAG &DAG,
     const X86Subtarget &Subtarget);
 
 /// Combine an arbitrary chain of shuffles into a single instruction if
@@ -35557,7 +35558,9 @@
 static SDValue combineX86ShuffleChain(ArrayRef<SDValue> Inputs, SDValue Root,
                                       ArrayRef<int> BaseMask, int Depth,
                                       bool HasVariableMask,
-                                      bool AllowVariableMask, SelectionDAG &DAG,
+                                      bool AllowVariableCrossLaneMask,
+                                      bool AllowVariablePerLaneMask,
+                                      SelectionDAG &DAG,
                                       const X86Subtarget &Subtarget) {
   assert(!BaseMask.empty() && "Cannot combine an empty shuffle mask!");
   assert((Inputs.size() == 1 || Inputs.size() == 2) &&
@@ -36012,8 +36015,14 @@
     return SDValue();
 
   // Depth threshold above which we can efficiently use variable mask shuffles.
-  int VariableShuffleDepth = Subtarget.hasFastVariableShuffle() ? 1 : 2;
-  AllowVariableMask &= (Depth >= VariableShuffleDepth) || HasVariableMask;
+  int VariableCrossLaneShuffleDepth =
+      Subtarget.hasFastVariableCrossLaneShuffle() ? 1 : 2;
+  int VariablePerLaneShuffleDepth =
+      Subtarget.hasFastVariablePerLaneShuffle() ? 1 : 2;
+  AllowVariableCrossLaneMask &=
+      (Depth >= VariableCrossLaneShuffleDepth) || HasVariableMask;
+  AllowVariablePerLaneMask &=
+      (Depth >= VariablePerLaneShuffleDepth) || HasVariableMask;
   // VPERMI2W/VPERMI2B are 3 uops on Skylake and Icelake so we require a
   // higher depth before combining them.
   bool AllowBWIVPERMV3 = (Depth >= 2 || HasVariableMask);
@@ -36022,7 +36031,7 @@
 
   if (is128BitLaneCrossingShuffleMask(MaskVT, Mask)) {
     // If we have a single input lane-crossing shuffle then lower to VPERMV.
-    if (UnaryShuffle && AllowVariableMask && !MaskContainsZeros) {
+    if (UnaryShuffle && AllowVariableCrossLaneMask && !MaskContainsZeros) {
       if (Subtarget.hasAVX2() &&
           (MaskVT == MVT::v8f32 || MaskVT == MVT::v8i32)) {
         SDValue VPermMask = getConstVector(Mask, IntMaskVT, DAG, DL, true);
@@ -36047,7 +36056,7 @@
 
     // Lower a unary+zero lane-crossing shuffle as VPERMV3 with a zero
     // vector as the second source (non-VLX will pad to 512-bit shuffles).
-    if (UnaryShuffle && AllowVariableMask &&
+    if (UnaryShuffle && AllowVariableCrossLaneMask &&
         ((Subtarget.hasAVX512() &&
           (MaskVT == MVT::v8f64 || MaskVT == MVT::v8i64 ||
            MaskVT == MVT::v4f64 || MaskVT == MVT::v4i64 ||
@@ -36070,13 +36079,14 @@
     // If that failed and either input is extracted then try to combine as a
     // shuffle with the larger type.
     if (SDValue WideShuffle = combineX86ShuffleChainWithExtract(
-            Inputs, Root, BaseMask, Depth, HasVariableMask, AllowVariableMask,
-            DAG, Subtarget))
+            Inputs, Root, BaseMask, Depth, HasVariableMask,
+            AllowVariableCrossLaneMask, AllowVariablePerLaneMask, DAG,
+            Subtarget))
       return WideShuffle;
 
     // If we have a dual input lane-crossing shuffle then lower to VPERMV3,
     // (non-VLX will pad to 512-bit shuffles).
-    if (AllowVariableMask && !MaskContainsZeros &&
+    if (AllowVariableCrossLaneMask && !MaskContainsZeros &&
         ((Subtarget.hasAVX512() &&
           (MaskVT == MVT::v8f64 || MaskVT == MVT::v8i64 ||
            MaskVT == MVT::v4f64 || MaskVT == MVT::v4i64 ||
@@ -36096,7 +36106,7 @@
 
   // See if we can combine a single input shuffle with zeros to a bit-mask,
   // which is much simpler than any shuffle.
-  if (UnaryShuffle && MaskContainsZeros && AllowVariableMask &&
+  if (UnaryShuffle && MaskContainsZeros && AllowVariablePerLaneMask &&
       isSequentialOrUndefOrZeroInRange(Mask, 0, NumMaskElts, 0) &&
       DAG.getTargetLoweringInfo().isTypeLegal(MaskVT)) {
     APInt Zero = APInt::getNullValue(MaskEltSizeInBits);
@@ -36124,7 +36134,7 @@
   // If we have a single input shuffle with different shuffle patterns in the
   // the 128-bit lanes use the variable mask to VPERMILPS.
   // TODO Combine other mask types at higher depths.
-  if (UnaryShuffle && AllowVariableMask && !MaskContainsZeros &&
+  if (UnaryShuffle && AllowVariablePerLaneMask && !MaskContainsZeros &&
       ((MaskVT == MVT::v8f32 && Subtarget.hasAVX()) ||
        (MaskVT == MVT::v16f32 && Subtarget.hasAVX512()))) {
     SmallVector<SDValue, 16> VPermIdx;
@@ -36141,7 +36151,7 @@
 
   // With XOP, binary shuffles of 128/256-bit floating point vectors can combine
   // to VPERMIL2PD/VPERMIL2PS.
-  if (AllowVariableMask && Subtarget.hasXOP() &&
+  if (AllowVariablePerLaneMask && Subtarget.hasXOP() &&
       (MaskVT == MVT::v2f64 || MaskVT == MVT::v4f64 || MaskVT == MVT::v4f32 ||
        MaskVT == MVT::v8f32)) {
     // VPERMIL2 Operation.
@@ -36179,7 +36189,7 @@
   // Intel's manuals suggest only using PSHUFB if doing so replacing 5
   // instructions, but in practice PSHUFB tends to be *very* fast so we're
   // more aggressive.
-  if (UnaryShuffle && AllowVariableMask &&
+  if (UnaryShuffle && AllowVariablePerLaneMask &&
       ((RootVT.is128BitVector() && Subtarget.hasSSSE3()) ||
        (RootVT.is256BitVector() && Subtarget.hasAVX2()) ||
        (RootVT.is512BitVector() && Subtarget.hasBWI()))) {
@@ -36210,7 +36220,8 @@
   // With XOP, if we have a 128-bit binary input shuffle we can always combine
   // to VPPERM. We match the depth requirement of PSHUFB - VPPERM is never
   // slower than PSHUFB on targets that support both.
-  if (AllowVariableMask && RootVT.is128BitVector() && Subtarget.hasXOP()) {
+  if (AllowVariablePerLaneMask && RootVT.is128BitVector() &&
+      Subtarget.hasXOP()) {
     // VPPERM Mask Operation
     // Bits[4:0] - Byte Index (0 - 31)
     // Bits[7:5] - Permute Operation (0 - Source byte, 4 - ZERO)
@@ -36241,13 +36252,13 @@
   // If that failed and either input is extracted then try to combine as a
   // shuffle with the larger type.
   if (SDValue WideShuffle = combineX86ShuffleChainWithExtract(
-          Inputs, Root, BaseMask, Depth, HasVariableMask, AllowVariableMask,
-          DAG, Subtarget))
+          Inputs, Root, BaseMask, Depth, HasVariableMask,
+          AllowVariableCrossLaneMask, AllowVariablePerLaneMask, DAG, Subtarget))
     return WideShuffle;
 
   // If we have a dual input shuffle then lower to VPERMV3,
   // (non-VLX will pad to 512-bit shuffles)
-  if (!UnaryShuffle && AllowVariableMask && !MaskContainsZeros &&
+  if (!UnaryShuffle && AllowVariablePerLaneMask && !MaskContainsZeros &&
       ((Subtarget.hasAVX512() &&
         (MaskVT == MVT::v2f64 || MaskVT == MVT::v4f64 || MaskVT == MVT::v8f64 ||
          MaskVT == MVT::v2i64 || MaskVT == MVT::v4i64 || MaskVT == MVT::v8i64 ||
@@ -36255,9 +36266,11 @@
          MaskVT == MVT::v8i32 || MaskVT == MVT::v16f32 ||
          MaskVT == MVT::v16i32)) ||
        (Subtarget.hasBWI() && AllowBWIVPERMV3 &&
-        (MaskVT == MVT::v8i16 || MaskVT == MVT::v16i16 || MaskVT == MVT::v32i16)) ||
+        (MaskVT == MVT::v8i16 || MaskVT == MVT::v16i16 ||
+         MaskVT == MVT::v32i16)) ||
        (Subtarget.hasVBMI() && AllowBWIVPERMV3 &&
-        (MaskVT == MVT::v16i8 || MaskVT == MVT::v32i8 || MaskVT == MVT::v64i8)))) {
+        (MaskVT == MVT::v16i8 || MaskVT == MVT::v32i8 ||
+         MaskVT == MVT::v64i8)))) {
     V1 = CanonicalizeShuffleInput(MaskVT, V1);
     V2 = CanonicalizeShuffleInput(MaskVT, V2);
     Res = lowerShuffleWithPERMV(DL, MaskVT, Mask, V1, V2, Subtarget, DAG);
@@ -36278,7 +36291,8 @@
 // extract_subvector(shuffle(x,y,m2),0)
 static SDValue combineX86ShuffleChainWithExtract(
     ArrayRef<SDValue> Inputs, SDValue Root, ArrayRef<int> BaseMask, int Depth,
-    bool HasVariableMask, bool AllowVariableMask, SelectionDAG &DAG,
+    bool HasVariableMask, bool AllowVariableCrossLaneMask,
+    bool AllowVariablePerLaneMask, SelectionDAG &DAG,
     const X86Subtarget &Subtarget) {
   unsigned NumMaskElts = BaseMask.size();
   unsigned NumInputs = Inputs.size();
@@ -36366,9 +36380,10 @@
   // Attempt to combine wider chain.
   // TODO: Can we use a better Root?
   SDValue WideRoot = WideInputs[0];
-  if (SDValue WideShuffle = combineX86ShuffleChain(
-          WideInputs, WideRoot, WideMask, Depth, HasVariableMask,
-          AllowVariableMask, DAG, Subtarget)) {
+  if (SDValue WideShuffle =
+          combineX86ShuffleChain(WideInputs, WideRoot, WideMask, Depth,
+                                 HasVariableMask, AllowVariableCrossLaneMask,
+                                 AllowVariablePerLaneMask, DAG, Subtarget)) {
     WideShuffle =
         extractSubVector(WideShuffle, 0, DAG, SDLoc(Root), RootSizeInBits);
     return DAG.getBitcast(RootVT, WideShuffle);
@@ -36696,8 +36711,9 @@
 static SDValue combineX86ShufflesRecursively(
     ArrayRef<SDValue> SrcOps, int SrcOpIndex, SDValue Root,
     ArrayRef<int> RootMask, ArrayRef<const SDNode *> SrcNodes, unsigned Depth,
-    unsigned MaxDepth, bool HasVariableMask, bool AllowVariableMask,
-    SelectionDAG &DAG, const X86Subtarget &Subtarget) {
+    unsigned MaxDepth, bool HasVariableMask, bool AllowVariableCrossLaneMask,
+    bool AllowVariablePerLaneMask, SelectionDAG &DAG,
+    const X86Subtarget &Subtarget) {
   assert(RootMask.size() > 0 &&
          (RootMask.size() > 1 || (RootMask[0] == 0 && SrcOpIndex == 0)) &&
          "Illegal shuffle root mask");
@@ -36927,13 +36943,17 @@
       SmallVector<int, 64> ResolvedMask = Mask;
       if (EmptyRoot)
         resolveTargetShuffleFromZeroables(ResolvedMask, OpUndef, OpZero);
-      bool AllowVar = false;
+      bool AllowCrossLaneVar = false;
+      bool AllowPerLaneVar = false;
       if (Ops[i].getNode()->hasOneUse() ||
-          SDNode::areOnlyUsersOf(CombinedNodes, Ops[i].getNode()))
-        AllowVar = AllowVariableMask;
+          SDNode::areOnlyUsersOf(CombinedNodes, Ops[i].getNode())) {
+        AllowCrossLaneVar = AllowVariableCrossLaneMask;
+        AllowPerLaneVar = AllowVariablePerLaneMask;
+      }
       if (SDValue Res = combineX86ShufflesRecursively(
               Ops, i, Root, ResolvedMask, CombinedNodes, Depth + 1, MaxDepth,
-              HasVariableMask, AllowVar, DAG, Subtarget))
+              HasVariableMask, AllowCrossLaneVar, AllowPerLaneVar, DAG,
+              Subtarget))
         return Res;
     }
   }
@@ -36984,23 +37004,25 @@
 
     // Finally, try to combine into a single shuffle instruction.
     return combineX86ShuffleChain(Ops, Root, Mask, Depth, HasVariableMask,
-                                  AllowVariableMask, DAG, Subtarget);
+                                  AllowVariableCrossLaneMask,
+                                  AllowVariablePerLaneMask, DAG, Subtarget);
   }
 
   // If that failed and any input is extracted then try to combine as a
   // shuffle with the larger type.
-  return combineX86ShuffleChainWithExtract(Ops, Root, Mask, Depth,
-                                           HasVariableMask, AllowVariableMask,
-                                           DAG, Subtarget);
+  return combineX86ShuffleChainWithExtract(
+      Ops, Root, Mask, Depth, HasVariableMask, AllowVariableCrossLaneMask,
+      AllowVariablePerLaneMask, DAG, Subtarget);
 }
 
 /// Helper entry wrapper to combineX86ShufflesRecursively.
 static SDValue combineX86ShufflesRecursively(SDValue Op, SelectionDAG &DAG,
                                              const X86Subtarget &Subtarget) {
-  return combineX86ShufflesRecursively({Op}, 0, Op, {0}, {}, /*Depth*/ 0,
-                                       X86::MaxShuffleCombineDepth,
-                                       /*HasVarMask*/ false,
-                                       /*AllowVarMask*/ true, DAG, Subtarget);
+  return combineX86ShufflesRecursively(
+      {Op}, 0, Op, {0}, {}, /*Depth*/ 0, X86::MaxShuffleCombineDepth,
+      /*HasVarMask*/ false,
+      /*AllowCrossLaneVarMask*/ true, /*AllowPerLaneVarMask*/ true, DAG,
+      Subtarget);
 }
 
 /// Get the PSHUF-style mask from PSHUF node.
@@ -37463,7 +37485,8 @@
       if (SDValue Res = combineX86ShufflesRecursively(
               {BC}, 0, BC, DemandedMask, {}, /*Depth*/ 0,
               X86::MaxShuffleCombineDepth,
-              /*HasVarMask*/ false, /*AllowVarMask*/ true, DAG, Subtarget))
+              /*HasVarMask*/ false, /*AllowCrossLaneVarMask*/ true,
+              /*AllowPerLaneVarMask*/ true, DAG, Subtarget))
         return DAG.getNode(X86ISD::VBROADCAST, DL, VT,
                            DAG.getBitcast(SrcVT, Res));
     }
@@ -38958,7 +38981,8 @@
     SDValue NewShuffle = combineX86ShufflesRecursively(
         {Op}, 0, Op, DemandedMask, {}, 0, X86::MaxShuffleCombineDepth - Depth,
         /*HasVarMask*/ false,
-        /*AllowVarMask*/ true, TLO.DAG, Subtarget);
+        /*AllowCrossLaneVarMask*/ true, /*AllowPerLaneVarMask*/ true, TLO.DAG,
+        Subtarget);
     if (NewShuffle)
       return TLO.CombineTo(Op, NewShuffle);
   }
@@ -44630,7 +44654,8 @@
       if (SDValue Shuffle = combineX86ShufflesRecursively(
               {SrcVec}, 0, SrcVec, ShuffleMask, {}, /*Depth*/ 1,
               X86::MaxShuffleCombineDepth,
-              /*HasVarMask*/ false, /*AllowVarMask*/ true, DAG, Subtarget))
+              /*HasVarMask*/ false, /*AllowVarCrossLaneMask*/ true,
+              /*AllowVarPerLaneMask*/ true, DAG, Subtarget))
         return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(N), VT, Shuffle,
                            N->getOperand(0).getOperand(1));
     }
diff --git a/llvm/lib/Target/X86/X86Subtarget.h b/llvm/lib/Target/X86/X86Subtarget.h
--- a/llvm/lib/Target/X86/X86Subtarget.h
+++ b/llvm/lib/Target/X86/X86Subtarget.h
@@ -247,9 +247,13 @@
   /// True if LZCNT/TZCNT instructions have a false dependency on the destination register.
   bool HasLZCNTFalseDeps = false;
 
-  /// True if its preferable to combine to a single shuffle using a variable
-  /// mask over multiple fixed shuffles.
-  bool HasFastVariableShuffle = false;
+  /// True if its preferable to combine to a single cross-lane shuffle
+  /// using a variable mask over multiple fixed shuffles.
+  bool HasFastVariableCrossLaneShuffle = false;
+
+  /// True if its preferable to combine to a single per-lane shuffle
+  /// using a variable mask over multiple fixed shuffles.
+  bool HasFastVariablePerLaneShuffle = false;
 
   /// True if vzeroupper instructions should be inserted after code that uses
   /// ymm or zmm registers.
@@ -704,8 +708,11 @@
   bool useLeaForSP() const { return UseLeaForSP; }
   bool hasPOPCNTFalseDeps() const { return HasPOPCNTFalseDeps; }
   bool hasLZCNTFalseDeps() const { return HasLZCNTFalseDeps; }
-  bool hasFastVariableShuffle() const {
-    return HasFastVariableShuffle;
+  bool hasFastVariableCrossLaneShuffle() const {
+    return HasFastVariableCrossLaneShuffle;
+  }
+  bool hasFastVariablePerLaneShuffle() const {
+    return HasFastVariablePerLaneShuffle;
   }
   bool insertVZEROUPPER() const { return InsertVZEROUPPER; }
   bool hasFastGather() const { return HasFastGather; }
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.h b/llvm/lib/Target/X86/X86TargetTransformInfo.h
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.h
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.h
@@ -55,7 +55,8 @@
       X86::FeatureFastSHLDRotate,
       X86::FeatureFastScalarShiftMasks,
       X86::FeatureFastVectorShiftMasks,
-      X86::FeatureFastVariableShuffle,
+      X86::FeatureFastVariableCrossLaneShuffle,
+      X86::FeatureFastVariablePerLaneShuffle,
       X86::FeatureFastVectorFSQRT,
       X86::FeatureLEAForSP,
       X86::FeatureLEAUsesAG,
diff --git a/llvm/test/CodeGen/X86/avx2-conversions.ll b/llvm/test/CodeGen/X86/avx2-conversions.ll
--- a/llvm/test/CodeGen/X86/avx2-conversions.ll
+++ b/llvm/test/CodeGen/X86/avx2-conversions.ll
@@ -1,8 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X32,X32-SLOW
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=X32,X32-FAST
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X32,X32-FAST-ALL
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X32,X32-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X64,X64-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-PERLANE
 
 define <4 x i32> @trunc4(<4 x i64> %A) nounwind {
 ; X32-SLOW-LABEL: trunc4:
@@ -12,13 +14,20 @@
 ; X32-SLOW-NEXT:    vzeroupper
 ; X32-SLOW-NEXT:    retl
 ;
-; X32-FAST-LABEL: trunc4:
-; X32-FAST:       # %bb.0:
-; X32-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; X32-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; X32-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; X32-FAST-NEXT:    vzeroupper
-; X32-FAST-NEXT:    retl
+; X32-FAST-ALL-LABEL: trunc4:
+; X32-FAST-ALL:       # %bb.0:
+; X32-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; X32-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; X32-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; X32-FAST-ALL-NEXT:    vzeroupper
+; X32-FAST-ALL-NEXT:    retl
+;
+; X32-FAST-PERLANE-LABEL: trunc4:
+; X32-FAST-PERLANE:       # %bb.0:
+; X32-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; X32-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; X32-FAST-PERLANE-NEXT:    vzeroupper
+; X32-FAST-PERLANE-NEXT:    retl
 ;
 ; X64-SLOW-LABEL: trunc4:
 ; X64-SLOW:       # %bb.0:
@@ -27,13 +36,20 @@
 ; X64-SLOW-NEXT:    vzeroupper
 ; X64-SLOW-NEXT:    retq
 ;
-; X64-FAST-LABEL: trunc4:
-; X64-FAST:       # %bb.0:
-; X64-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; X64-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; X64-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; X64-FAST-NEXT:    vzeroupper
-; X64-FAST-NEXT:    retq
+; X64-FAST-ALL-LABEL: trunc4:
+; X64-FAST-ALL:       # %bb.0:
+; X64-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; X64-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; X64-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; X64-FAST-ALL-NEXT:    vzeroupper
+; X64-FAST-ALL-NEXT:    retq
+;
+; X64-FAST-PERLANE-LABEL: trunc4:
+; X64-FAST-PERLANE:       # %bb.0:
+; X64-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; X64-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; X64-FAST-PERLANE-NEXT:    vzeroupper
+; X64-FAST-PERLANE-NEXT:    retq
   %B = trunc <4 x i64> %A to <4 x i32>
   ret <4 x i32>%B
 }
diff --git a/llvm/test/CodeGen/X86/avx2-vector-shifts.ll b/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
--- a/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
+++ b/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
@@ -1,8 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X86,X86-SLOW
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=X86,X86-FAST
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X86,X86-FAST-ALL
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X86,X86-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X64,X64-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-PERLANE
 
 ; AVX2 Logical Shift Left
 
@@ -384,15 +386,25 @@
 ; X86-SLOW-NEXT:    vzeroupper
 ; X86-SLOW-NEXT:    retl
 ;
-; X86-FAST-LABEL: srl_trunc_and_v4i64:
-; X86-FAST:       # %bb.0:
-; X86-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
-; X86-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; X86-FAST-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [8,8,8,8]
-; X86-FAST-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; X86-FAST-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
-; X86-FAST-NEXT:    vzeroupper
-; X86-FAST-NEXT:    retl
+; X86-FAST-ALL-LABEL: srl_trunc_and_v4i64:
+; X86-FAST-ALL:       # %bb.0:
+; X86-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
+; X86-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; X86-FAST-ALL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [8,8,8,8]
+; X86-FAST-ALL-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; X86-FAST-ALL-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; X86-FAST-ALL-NEXT:    vzeroupper
+; X86-FAST-ALL-NEXT:    retl
+;
+; X86-FAST-PERLANE-LABEL: srl_trunc_and_v4i64:
+; X86-FAST-PERLANE:       # %bb.0:
+; X86-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; X86-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; X86-FAST-PERLANE-NEXT:    vbroadcastss {{.*#+}} xmm2 = [8,8,8,8]
+; X86-FAST-PERLANE-NEXT:    vandps %xmm2, %xmm1, %xmm1
+; X86-FAST-PERLANE-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; X86-FAST-PERLANE-NEXT:    vzeroupper
+; X86-FAST-PERLANE-NEXT:    retl
 ;
 ; X64-SLOW-LABEL: srl_trunc_and_v4i64:
 ; X64-SLOW:       # %bb.0:
@@ -404,15 +416,25 @@
 ; X64-SLOW-NEXT:    vzeroupper
 ; X64-SLOW-NEXT:    retq
 ;
-; X64-FAST-LABEL: srl_trunc_and_v4i64:
-; X64-FAST:       # %bb.0:
-; X64-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
-; X64-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; X64-FAST-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [8,8,8,8]
-; X64-FAST-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; X64-FAST-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
-; X64-FAST-NEXT:    vzeroupper
-; X64-FAST-NEXT:    retq
+; X64-FAST-ALL-LABEL: srl_trunc_and_v4i64:
+; X64-FAST-ALL:       # %bb.0:
+; X64-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
+; X64-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; X64-FAST-ALL-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [8,8,8,8]
+; X64-FAST-ALL-NEXT:    vpand %xmm2, %xmm1, %xmm1
+; X64-FAST-ALL-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; X64-FAST-ALL-NEXT:    vzeroupper
+; X64-FAST-ALL-NEXT:    retq
+;
+; X64-FAST-PERLANE-LABEL: srl_trunc_and_v4i64:
+; X64-FAST-PERLANE:       # %bb.0:
+; X64-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; X64-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; X64-FAST-PERLANE-NEXT:    vbroadcastss {{.*#+}} xmm2 = [8,8,8,8]
+; X64-FAST-PERLANE-NEXT:    vandps %xmm2, %xmm1, %xmm1
+; X64-FAST-PERLANE-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; X64-FAST-PERLANE-NEXT:    vzeroupper
+; X64-FAST-PERLANE-NEXT:    retq
   %and = and <4 x i64> %y, <i64 8, i64 8, i64 8, i64 8>
   %trunc = trunc <4 x i64> %and to <4 x i32>
   %sra = lshr <4 x i32> %x, %trunc
diff --git a/llvm/test/CodeGen/X86/avx512-extract-subvector-load-store.ll b/llvm/test/CodeGen/X86/avx512-extract-subvector-load-store.ll
--- a/llvm/test/CodeGen/X86/avx512-extract-subvector-load-store.ll
+++ b/llvm/test/CodeGen/X86/avx512-extract-subvector-load-store.ll
@@ -1,6 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq,+fast-variable-shuffle -O2 | FileCheck %s --check-prefix=AVX512
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+fast-variable-shuffle -O2 | FileCheck %s --check-prefix=AVX512NOTDQ
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle -O2 | FileCheck %s --check-prefixes=AVX512,AVX512-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+avx512dq,+fast-variable-perlane-shuffle -O2 | FileCheck %s --check-prefixes=AVX512,AVX512-FAST-PERLANE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle -O2 | FileCheck %s --check-prefixes=AVX512NOTDQ,AVX512NOTDQ-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl,+fast-variable-perlane-shuffle -O2 | FileCheck %s --check-prefixes=AVX512NOTDQ,AVX512NOTDQ-FAST-PERLANE
 
 define void @load_v8i1_broadcast_4_v2i1(<8 x i1>* %a0,<2 x double> %a1,<2 x double> %a2,<2 x double>* %a3) {
 ; AVX512-LABEL: load_v8i1_broadcast_4_v2i1:
@@ -322,31 +324,57 @@
     ret void
 }
 define void @load_v32i1_broadcast_31_v8i1(<32 x i1>* %a0,<8 x float> %a1,<8 x float> %a2,<8 x float>* %a3) {
-; AVX512-LABEL: load_v32i1_broadcast_31_v8i1:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    kmovb 3(%rdi), %k0
-; AVX512-NEXT:    vpmovm2d %k0, %ymm2
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
-; AVX512-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX512-NEXT:    vpmovd2m %ymm2, %k1
-; AVX512-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
-; AVX512-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512-FAST-LABEL: load_v32i1_broadcast_31_v8i1:
+; AVX512-FAST:       # %bb.0:
+; AVX512-FAST-NEXT:    kmovb 3(%rdi), %k0
+; AVX512-FAST-NEXT:    vpmovm2d %k0, %ymm2
+; AVX512-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
+; AVX512-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX512-FAST-NEXT:    vpmovd2m %ymm2, %k1
+; AVX512-FAST-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512-FAST-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512-FAST-NEXT:    vzeroupper
+; AVX512-FAST-NEXT:    retq
 ;
-; AVX512NOTDQ-LABEL: load_v32i1_broadcast_31_v8i1:
-; AVX512NOTDQ:       # %bb.0:
-; AVX512NOTDQ-NEXT:    movzbl 3(%rdi), %eax
-; AVX512NOTDQ-NEXT:    kmovd %eax, %k1
-; AVX512NOTDQ-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX512NOTDQ-NEXT:    vmovdqa32 %ymm2, %ymm2 {%k1} {z}
-; AVX512NOTDQ-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
-; AVX512NOTDQ-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX512NOTDQ-NEXT:    vptestmd %ymm2, %ymm2, %k1
-; AVX512NOTDQ-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
-; AVX512NOTDQ-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX512NOTDQ-NEXT:    vzeroupper
-; AVX512NOTDQ-NEXT:    retq
+; AVX512-FAST-PERLANE-LABEL: load_v32i1_broadcast_31_v8i1:
+; AVX512-FAST-PERLANE:       # %bb.0:
+; AVX512-FAST-PERLANE-NEXT:    kmovb 3(%rdi), %k0
+; AVX512-FAST-PERLANE-NEXT:    vpmovm2d %k0, %ymm2
+; AVX512-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
+; AVX512-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX512-FAST-PERLANE-NEXT:    vpmovd2m %ymm2, %k1
+; AVX512-FAST-PERLANE-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512-FAST-PERLANE-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512-FAST-PERLANE-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-LABEL: load_v32i1_broadcast_31_v8i1:
+; AVX512NOTDQ-FAST:       # %bb.0:
+; AVX512NOTDQ-FAST-NEXT:    movzbl 3(%rdi), %eax
+; AVX512NOTDQ-FAST-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512NOTDQ-FAST-NEXT:    vmovdqa32 %ymm2, %ymm2 {%k1} {z}
+; AVX512NOTDQ-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
+; AVX512NOTDQ-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX512NOTDQ-FAST-NEXT:    vptestmd %ymm2, %ymm2, %k1
+; AVX512NOTDQ-FAST-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512NOTDQ-FAST-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512NOTDQ-FAST-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-PERLANE-LABEL: load_v32i1_broadcast_31_v8i1:
+; AVX512NOTDQ-FAST-PERLANE:       # %bb.0:
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    movzbl 3(%rdi), %eax
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovdqa32 %ymm2, %ymm2 {%k1} {z}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vptestmd %ymm2, %ymm2, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    retq
     %d0 = load <32 x i1>, <32 x i1>* %a0
     %d1 = shufflevector <32 x i1> %d0,<32 x i1> undef,<8 x i32><i32 31,i32 31,i32 31,i32 31,i32 31,i32 31,i32 31,i32 31>
     %d2 = select <8 x i1> %d1, <8 x float> %a1, <8 x float> %a2
@@ -527,31 +555,57 @@
     ret void
 }
 define void @load_v64i1_broadcast_63_v8i1(<64 x i1>* %a0,<8 x float> %a1,<8 x float> %a2,<8 x float>* %a3) {
-; AVX512-LABEL: load_v64i1_broadcast_63_v8i1:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    kmovb 7(%rdi), %k0
-; AVX512-NEXT:    vpmovm2d %k0, %ymm2
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
-; AVX512-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX512-NEXT:    vpmovd2m %ymm2, %k1
-; AVX512-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
-; AVX512-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512-FAST-LABEL: load_v64i1_broadcast_63_v8i1:
+; AVX512-FAST:       # %bb.0:
+; AVX512-FAST-NEXT:    kmovb 7(%rdi), %k0
+; AVX512-FAST-NEXT:    vpmovm2d %k0, %ymm2
+; AVX512-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
+; AVX512-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX512-FAST-NEXT:    vpmovd2m %ymm2, %k1
+; AVX512-FAST-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512-FAST-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512-FAST-NEXT:    vzeroupper
+; AVX512-FAST-NEXT:    retq
 ;
-; AVX512NOTDQ-LABEL: load_v64i1_broadcast_63_v8i1:
-; AVX512NOTDQ:       # %bb.0:
-; AVX512NOTDQ-NEXT:    movzbl 7(%rdi), %eax
-; AVX512NOTDQ-NEXT:    kmovd %eax, %k1
-; AVX512NOTDQ-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
-; AVX512NOTDQ-NEXT:    vmovdqa32 %ymm2, %ymm2 {%k1} {z}
-; AVX512NOTDQ-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
-; AVX512NOTDQ-NEXT:    vpermd %ymm2, %ymm3, %ymm2
-; AVX512NOTDQ-NEXT:    vptestmd %ymm2, %ymm2, %k1
-; AVX512NOTDQ-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
-; AVX512NOTDQ-NEXT:    vmovaps %ymm1, (%rsi)
-; AVX512NOTDQ-NEXT:    vzeroupper
-; AVX512NOTDQ-NEXT:    retq
+; AVX512-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v8i1:
+; AVX512-FAST-PERLANE:       # %bb.0:
+; AVX512-FAST-PERLANE-NEXT:    kmovb 7(%rdi), %k0
+; AVX512-FAST-PERLANE-NEXT:    vpmovm2d %k0, %ymm2
+; AVX512-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
+; AVX512-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX512-FAST-PERLANE-NEXT:    vpmovd2m %ymm2, %k1
+; AVX512-FAST-PERLANE-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512-FAST-PERLANE-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512-FAST-PERLANE-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-LABEL: load_v64i1_broadcast_63_v8i1:
+; AVX512NOTDQ-FAST:       # %bb.0:
+; AVX512NOTDQ-FAST-NEXT:    movzbl 7(%rdi), %eax
+; AVX512NOTDQ-FAST-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512NOTDQ-FAST-NEXT:    vmovdqa32 %ymm2, %ymm2 {%k1} {z}
+; AVX512NOTDQ-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [7,7,7,7,7,7,7,7]
+; AVX512NOTDQ-FAST-NEXT:    vpermd %ymm2, %ymm3, %ymm2
+; AVX512NOTDQ-FAST-NEXT:    vptestmd %ymm2, %ymm2, %k1
+; AVX512NOTDQ-FAST-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512NOTDQ-FAST-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512NOTDQ-FAST-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v8i1:
+; AVX512NOTDQ-FAST-PERLANE:       # %bb.0:
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    movzbl 7(%rdi), %eax
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpcmpeqd %ymm2, %ymm2, %ymm2
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovdqa32 %ymm2, %ymm2 {%k1} {z}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[3,3,3,3,7,7,7,7]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vptestmd %ymm2, %ymm2, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovaps %ymm0, %ymm1 {%k1}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovaps %ymm1, (%rsi)
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    retq
     %d0 = load <64 x i1>, <64 x i1>* %a0
     %d1 = shufflevector <64 x i1> %d0,<64 x i1> undef,<8 x i32><i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63>
     %d2 = select <8 x i1> %d1, <8 x float> %a1, <8 x float> %a2
@@ -559,29 +613,53 @@
     ret void
 }
 define void @load_v64i1_broadcast_63_v16i1(<64 x i1>* %a0,<16 x float> %a1,<16 x float> %a2,<16 x float>* %a3) {
-; AVX512-LABEL: load_v64i1_broadcast_63_v16i1:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    kmovw 6(%rdi), %k0
-; AVX512-NEXT:    vpmovm2d %k0, %zmm2
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512-NEXT:    vpermd %zmm2, %zmm3, %zmm2
-; AVX512-NEXT:    vpmovd2m %zmm2, %k1
-; AVX512-NEXT:    vmovaps %zmm0, %zmm1 {%k1}
-; AVX512-NEXT:    vmovaps %zmm1, (%rsi)
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512-FAST-LABEL: load_v64i1_broadcast_63_v16i1:
+; AVX512-FAST:       # %bb.0:
+; AVX512-FAST-NEXT:    kmovw 6(%rdi), %k0
+; AVX512-FAST-NEXT:    vpmovm2d %k0, %zmm2
+; AVX512-FAST-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512-FAST-NEXT:    vpermd %zmm2, %zmm3, %zmm2
+; AVX512-FAST-NEXT:    vpmovd2m %zmm2, %k1
+; AVX512-FAST-NEXT:    vmovaps %zmm0, %zmm1 {%k1}
+; AVX512-FAST-NEXT:    vmovaps %zmm1, (%rsi)
+; AVX512-FAST-NEXT:    vzeroupper
+; AVX512-FAST-NEXT:    retq
 ;
-; AVX512NOTDQ-LABEL: load_v64i1_broadcast_63_v16i1:
-; AVX512NOTDQ:       # %bb.0:
-; AVX512NOTDQ-NEXT:    kmovw 6(%rdi), %k1
-; AVX512NOTDQ-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
-; AVX512NOTDQ-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512NOTDQ-NEXT:    vpermd %zmm2, %zmm3, %zmm2
-; AVX512NOTDQ-NEXT:    vptestmd %zmm2, %zmm2, %k1
-; AVX512NOTDQ-NEXT:    vmovaps %zmm0, %zmm1 {%k1}
-; AVX512NOTDQ-NEXT:    vmovaps %zmm1, (%rsi)
-; AVX512NOTDQ-NEXT:    vzeroupper
-; AVX512NOTDQ-NEXT:    retq
+; AVX512-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v16i1:
+; AVX512-FAST-PERLANE:       # %bb.0:
+; AVX512-FAST-PERLANE-NEXT:    kmovw 6(%rdi), %k0
+; AVX512-FAST-PERLANE-NEXT:    vpmovm2d %k0, %zmm2
+; AVX512-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} zmm2 = zmm2[3,3,3,3,7,7,7,7,11,11,11,11,15,15,15,15]
+; AVX512-FAST-PERLANE-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm2[6,7,6,7,6,7,6,7]
+; AVX512-FAST-PERLANE-NEXT:    vpmovd2m %zmm2, %k1
+; AVX512-FAST-PERLANE-NEXT:    vmovaps %zmm0, %zmm1 {%k1}
+; AVX512-FAST-PERLANE-NEXT:    vmovaps %zmm1, (%rsi)
+; AVX512-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512-FAST-PERLANE-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-LABEL: load_v64i1_broadcast_63_v16i1:
+; AVX512NOTDQ-FAST:       # %bb.0:
+; AVX512NOTDQ-FAST-NEXT:    kmovw 6(%rdi), %k1
+; AVX512NOTDQ-FAST-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
+; AVX512NOTDQ-FAST-NEXT:    vpbroadcastd {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512NOTDQ-FAST-NEXT:    vpermd %zmm2, %zmm3, %zmm2
+; AVX512NOTDQ-FAST-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; AVX512NOTDQ-FAST-NEXT:    vmovaps %zmm0, %zmm1 {%k1}
+; AVX512NOTDQ-FAST-NEXT:    vmovaps %zmm1, (%rsi)
+; AVX512NOTDQ-FAST-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v16i1:
+; AVX512NOTDQ-FAST-PERLANE:       # %bb.0:
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovw 6(%rdi), %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpternlogd $255, %zmm2, %zmm2, %zmm2 {%k1} {z}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} zmm2 = zmm2[3,3,3,3,7,7,7,7,11,11,11,11,15,15,15,15]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vshufi64x2 {{.*#+}} zmm2 = zmm2[6,7,6,7,6,7,6,7]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vptestmd %zmm2, %zmm2, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovaps %zmm0, %zmm1 {%k1}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovaps %zmm1, (%rsi)
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    retq
     %d0 = load <64 x i1>, <64 x i1>* %a0
     %d1 = shufflevector <64 x i1> %d0,<64 x i1> undef,<16 x i32><i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63>
     %d2 = select <16 x i1> %d1, <16 x float> %a1, <16 x float> %a2
@@ -1170,30 +1248,55 @@
     ret void
 }
 define void @load_v32i1_broadcast_31_v8i1_store(<32 x i1>* %a0,<8 x i1>* %a1) {
-; AVX512-LABEL: load_v32i1_broadcast_31_v8i1_store:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    kmovb 3(%rdi), %k0
-; AVX512-NEXT:    vpmovm2d %k0, %ymm0
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
-; AVX512-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    vpmovd2m %ymm0, %k0
-; AVX512-NEXT:    kmovb %k0, (%rsi)
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512-FAST-LABEL: load_v32i1_broadcast_31_v8i1_store:
+; AVX512-FAST:       # %bb.0:
+; AVX512-FAST-NEXT:    kmovb 3(%rdi), %k0
+; AVX512-FAST-NEXT:    vpmovm2d %k0, %ymm0
+; AVX512-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
+; AVX512-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512-FAST-NEXT:    vpmovd2m %ymm0, %k0
+; AVX512-FAST-NEXT:    kmovb %k0, (%rsi)
+; AVX512-FAST-NEXT:    vzeroupper
+; AVX512-FAST-NEXT:    retq
 ;
-; AVX512NOTDQ-LABEL: load_v32i1_broadcast_31_v8i1_store:
-; AVX512NOTDQ:       # %bb.0:
-; AVX512NOTDQ-NEXT:    movzbl 3(%rdi), %eax
-; AVX512NOTDQ-NEXT:    kmovd %eax, %k1
-; AVX512NOTDQ-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512NOTDQ-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512NOTDQ-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
-; AVX512NOTDQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512NOTDQ-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512NOTDQ-NEXT:    kmovd %k0, %eax
-; AVX512NOTDQ-NEXT:    movb %al, (%rsi)
-; AVX512NOTDQ-NEXT:    vzeroupper
-; AVX512NOTDQ-NEXT:    retq
+; AVX512-FAST-PERLANE-LABEL: load_v32i1_broadcast_31_v8i1_store:
+; AVX512-FAST-PERLANE:       # %bb.0:
+; AVX512-FAST-PERLANE-NEXT:    kmovb 3(%rdi), %k0
+; AVX512-FAST-PERLANE-NEXT:    vpmovm2d %k0, %ymm0
+; AVX512-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,3,7,7,7,7]
+; AVX512-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,2]
+; AVX512-FAST-PERLANE-NEXT:    vpmovd2m %ymm0, %k0
+; AVX512-FAST-PERLANE-NEXT:    kmovb %k0, (%rsi)
+; AVX512-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512-FAST-PERLANE-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-LABEL: load_v32i1_broadcast_31_v8i1_store:
+; AVX512NOTDQ-FAST:       # %bb.0:
+; AVX512NOTDQ-FAST-NEXT:    movzbl 3(%rdi), %eax
+; AVX512NOTDQ-FAST-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512NOTDQ-FAST-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512NOTDQ-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
+; AVX512NOTDQ-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512NOTDQ-FAST-NEXT:    vptestmd %ymm0, %ymm0, %k0
+; AVX512NOTDQ-FAST-NEXT:    kmovd %k0, %eax
+; AVX512NOTDQ-FAST-NEXT:    movb %al, (%rsi)
+; AVX512NOTDQ-FAST-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-PERLANE-LABEL: load_v32i1_broadcast_31_v8i1_store:
+; AVX512NOTDQ-FAST-PERLANE:       # %bb.0:
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    movzbl 3(%rdi), %eax
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,3,7,7,7,7]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,2]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vptestmd %ymm0, %ymm0, %k0
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovd %k0, %eax
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    movb %al, (%rsi)
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    retq
     %d0 = load <32 x i1>, <32 x i1>* %a0
     %d1 = shufflevector <32 x i1> %d0,<32 x i1> undef,<8 x i32><i32 31,i32 31,i32 31,i32 31,i32 31,i32 31,i32 31,i32 31>
     store <8 x i1> %d1, <8 x i1>* %a1
@@ -1408,57 +1511,104 @@
     ret void
 }
 define void @load_v64i1_broadcast_63_v8i1_store(<64 x i1>* %a0,<8 x i1>* %a1) {
-; AVX512-LABEL: load_v64i1_broadcast_63_v8i1_store:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    kmovb 7(%rdi), %k0
-; AVX512-NEXT:    vpmovm2d %k0, %ymm0
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
-; AVX512-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512-NEXT:    vpmovd2m %ymm0, %k0
-; AVX512-NEXT:    kmovb %k0, (%rsi)
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512-FAST-LABEL: load_v64i1_broadcast_63_v8i1_store:
+; AVX512-FAST:       # %bb.0:
+; AVX512-FAST-NEXT:    kmovb 7(%rdi), %k0
+; AVX512-FAST-NEXT:    vpmovm2d %k0, %ymm0
+; AVX512-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
+; AVX512-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512-FAST-NEXT:    vpmovd2m %ymm0, %k0
+; AVX512-FAST-NEXT:    kmovb %k0, (%rsi)
+; AVX512-FAST-NEXT:    vzeroupper
+; AVX512-FAST-NEXT:    retq
 ;
-; AVX512NOTDQ-LABEL: load_v64i1_broadcast_63_v8i1_store:
-; AVX512NOTDQ:       # %bb.0:
-; AVX512NOTDQ-NEXT:    movzbl 7(%rdi), %eax
-; AVX512NOTDQ-NEXT:    kmovd %eax, %k1
-; AVX512NOTDQ-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512NOTDQ-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512NOTDQ-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
-; AVX512NOTDQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512NOTDQ-NEXT:    vptestmd %ymm0, %ymm0, %k0
-; AVX512NOTDQ-NEXT:    kmovd %k0, %eax
-; AVX512NOTDQ-NEXT:    movb %al, (%rsi)
-; AVX512NOTDQ-NEXT:    vzeroupper
-; AVX512NOTDQ-NEXT:    retq
+; AVX512-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v8i1_store:
+; AVX512-FAST-PERLANE:       # %bb.0:
+; AVX512-FAST-PERLANE-NEXT:    kmovb 7(%rdi), %k0
+; AVX512-FAST-PERLANE-NEXT:    vpmovm2d %k0, %ymm0
+; AVX512-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,3,7,7,7,7]
+; AVX512-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,2]
+; AVX512-FAST-PERLANE-NEXT:    vpmovd2m %ymm0, %k0
+; AVX512-FAST-PERLANE-NEXT:    kmovb %k0, (%rsi)
+; AVX512-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512-FAST-PERLANE-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-LABEL: load_v64i1_broadcast_63_v8i1_store:
+; AVX512NOTDQ-FAST:       # %bb.0:
+; AVX512NOTDQ-FAST-NEXT:    movzbl 7(%rdi), %eax
+; AVX512NOTDQ-FAST-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512NOTDQ-FAST-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512NOTDQ-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [7,7,7,7,7,7,7,7]
+; AVX512NOTDQ-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512NOTDQ-FAST-NEXT:    vptestmd %ymm0, %ymm0, %k0
+; AVX512NOTDQ-FAST-NEXT:    kmovd %k0, %eax
+; AVX512NOTDQ-FAST-NEXT:    movb %al, (%rsi)
+; AVX512NOTDQ-FAST-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v8i1_store:
+; AVX512NOTDQ-FAST-PERLANE:       # %bb.0:
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    movzbl 7(%rdi), %eax
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovd %eax, %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,3,3,3,7,7,7,7]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,2]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vptestmd %ymm0, %ymm0, %k0
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovd %k0, %eax
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    movb %al, (%rsi)
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    retq
     %d0 = load <64 x i1>, <64 x i1>* %a0
     %d1 = shufflevector <64 x i1> %d0,<64 x i1> undef,<8 x i32><i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63>
     store <8 x i1> %d1, <8 x i1>* %a1
     ret void
 }
 define void @load_v64i1_broadcast_63_v16i1_store(<64 x i1>* %a0,<16 x i1>* %a1) {
-; AVX512-LABEL: load_v64i1_broadcast_63_v16i1_store:
-; AVX512:       # %bb.0:
-; AVX512-NEXT:    kmovw 6(%rdi), %k0
-; AVX512-NEXT:    vpmovm2d %k0, %zmm0
-; AVX512-NEXT:    vpbroadcastd {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512-NEXT:    vpermd %zmm0, %zmm1, %zmm0
-; AVX512-NEXT:    vpmovd2m %zmm0, %k0
-; AVX512-NEXT:    kmovw %k0, (%rsi)
-; AVX512-NEXT:    vzeroupper
-; AVX512-NEXT:    retq
+; AVX512-FAST-LABEL: load_v64i1_broadcast_63_v16i1_store:
+; AVX512-FAST:       # %bb.0:
+; AVX512-FAST-NEXT:    kmovw 6(%rdi), %k0
+; AVX512-FAST-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512-FAST-NEXT:    vpbroadcastd {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512-FAST-NEXT:    vpermd %zmm0, %zmm1, %zmm0
+; AVX512-FAST-NEXT:    vpmovd2m %zmm0, %k0
+; AVX512-FAST-NEXT:    kmovw %k0, (%rsi)
+; AVX512-FAST-NEXT:    vzeroupper
+; AVX512-FAST-NEXT:    retq
 ;
-; AVX512NOTDQ-LABEL: load_v64i1_broadcast_63_v16i1_store:
-; AVX512NOTDQ:       # %bb.0:
-; AVX512NOTDQ-NEXT:    kmovw 6(%rdi), %k1
-; AVX512NOTDQ-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512NOTDQ-NEXT:    vpbroadcastd {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; AVX512NOTDQ-NEXT:    vpermd %zmm0, %zmm1, %zmm0
-; AVX512NOTDQ-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512NOTDQ-NEXT:    kmovw %k0, (%rsi)
-; AVX512NOTDQ-NEXT:    vzeroupper
-; AVX512NOTDQ-NEXT:    retq
+; AVX512-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v16i1_store:
+; AVX512-FAST-PERLANE:       # %bb.0:
+; AVX512-FAST-PERLANE-NEXT:    kmovw 6(%rdi), %k0
+; AVX512-FAST-PERLANE-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} zmm0 = zmm0[3,3,3,3,7,7,7,7,11,11,11,11,15,15,15,15]
+; AVX512-FAST-PERLANE-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[6,7,6,7,6,7,6,7]
+; AVX512-FAST-PERLANE-NEXT:    vpmovd2m %zmm0, %k0
+; AVX512-FAST-PERLANE-NEXT:    kmovw %k0, (%rsi)
+; AVX512-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512-FAST-PERLANE-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-LABEL: load_v64i1_broadcast_63_v16i1_store:
+; AVX512NOTDQ-FAST:       # %bb.0:
+; AVX512NOTDQ-FAST-NEXT:    kmovw 6(%rdi), %k1
+; AVX512NOTDQ-FAST-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512NOTDQ-FAST-NEXT:    vpbroadcastd {{.*#+}} zmm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; AVX512NOTDQ-FAST-NEXT:    vpermd %zmm0, %zmm1, %zmm0
+; AVX512NOTDQ-FAST-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512NOTDQ-FAST-NEXT:    kmovw %k0, (%rsi)
+; AVX512NOTDQ-FAST-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-NEXT:    retq
+;
+; AVX512NOTDQ-FAST-PERLANE-LABEL: load_v64i1_broadcast_63_v16i1_store:
+; AVX512NOTDQ-FAST-PERLANE:       # %bb.0:
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovw 6(%rdi), %k1
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} zmm0 = zmm0[3,3,3,3,7,7,7,7,11,11,11,11,15,15,15,15]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[6,7,6,7,6,7,6,7]
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vptestmd %zmm0, %zmm0, %k0
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    kmovw %k0, (%rsi)
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512NOTDQ-FAST-PERLANE-NEXT:    retq
     %d0 = load <64 x i1>, <64 x i1>* %a0
     %d1 = shufflevector <64 x i1> %d0,<64 x i1> undef,<16 x i32><i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63,i32 63>
     store <16 x i1> %d1, <16 x i1>* %a1
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll b/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
--- a/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-vector-int.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512dq,+fast-variable-shuffle %s -o - | FileCheck %s
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle %s -o - | FileCheck %s
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512dq,+fast-variable-perlane-shuffle %s -o - | FileCheck %s
 
 define <4 x i32> @test_2xi32_to_4xi32(<4 x i32> %vec) {
 ; CHECK-LABEL: test_2xi32_to_4xi32:
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll b/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll
--- a/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_mem_shuffle
-; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-shuffle %s -o - | FileCheck %s
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle %s -o - | FileCheck --check-prefixes=CHECK,CHECK-FAST %s
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-perlane-shuffle %s -o - | FileCheck --check-prefixes=CHECK,CHECK-FAST-PERLANE %s
 
 ; FIXME: All cases here should be fixed by PR34380
 
@@ -1758,13 +1759,24 @@
 }
 
 define <4 x i32> @test_16xi32_to_4xi32_perm_mask9(<16 x i32> %vec) {
-; CHECK-LABEL: test_16xi32_to_4xi32_perm_mask9:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} xmm1 = [12,9,4,10]
-; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm0
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_16xi32_to_4xi32_perm_mask9:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} xmm1 = [12,9,4,10]
+; CHECK-FAST-NEXT:    vpermps %zmm0, %zmm1, %zmm0
+; CHECK-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; CHECK-FAST-NEXT:    vzeroupper
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_16xi32_to_4xi32_perm_mask9:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm1 = <4,1,u,2>
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm1, %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,4,3]
+; CHECK-FAST-PERLANE-NEXT:    vpermi2d %xmm2, %xmm1, %xmm0
+; CHECK-FAST-PERLANE-NEXT:    vzeroupper
+; CHECK-FAST-PERLANE-NEXT:    retq
   %res = shufflevector <16 x i32> %vec, <16 x i32> undef, <4 x i32> <i32 12, i32 9, i32 4, i32 10>
   ret <4 x i32> %res
 }
@@ -1937,13 +1949,22 @@
   ret <4 x i64> %res
 }
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mask1(<8 x i64> %vec, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [6,4,6,1]
-; CHECK-NEXT:    vpermq %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; CHECK-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [6,4,6,1]
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3],ymm3[4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 {%k1} = ymm0[2,0,2,1]
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 4, i32 6, i32 1>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %vec2
@@ -1951,26 +1972,43 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mask1(<8 x i64> %vec, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,4,6,1]
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,4,6,1]
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3],ymm2[4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm0[2,0,2,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 4, i32 6, i32 1>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
   ret <4 x i64> %res
 }
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mask2(<8 x i64> %vec, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [6,3,6,3]
-; CHECK-NEXT:    vpermq %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; CHECK-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [6,3,6,3]
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 {%k1} = ymm0[2,3,2,3]
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 3, i32 6, i32 3>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %vec2
@@ -1978,36 +2016,60 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mask2(<8 x i64> %vec, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,3,6,3]
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,3,6,3]
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2,3],ymm2[4,5],ymm0[6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm0[2,3,2,3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 3, i32 6, i32 3>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
   ret <4 x i64> %res
 }
 define <4 x i64> @test_8xi64_to_4xi64_perm_mask3(<8 x i64> %vec) {
-; CHECK-LABEL: test_8xi64_to_4xi64_perm_mask3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} ymm1 = [6,0,0,7]
-; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_8xi64_to_4xi64_perm_mask3:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [6,0,0,7]
+; CHECK-FAST-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_8xi64_to_4xi64_perm_mask3:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,0,0,3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %res = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 0, i32 0, i32 7>
   ret <4 x i64> %res
 }
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mask3(<8 x i64> %vec, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [6,0,0,7]
-; CHECK-NEXT:    vpermq %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; CHECK-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mask3:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [6,0,0,7]
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mask3:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm3[2,3,4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 {%k1} = ymm0[2,0,0,3]
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 0, i32 0, i32 7>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %vec2
@@ -2015,26 +2077,43 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mask3(<8 x i64> %vec, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,0,0,7]
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask3:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [6,0,0,7]
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask3:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm2[2,3,4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm0[2,0,0,3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 6, i32 0, i32 0, i32 7>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
   ret <4 x i64> %res
 }
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mask4(<8 x i64> %vec, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask4:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,7,7,5]
-; CHECK-NEXT:    vpermq %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; CHECK-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mask4:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,7,7,5]
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mask4:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm3[1],ymm0[3],ymm3[3]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 {%k1} = ymm0[2,3,3,1]
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 3, i32 7, i32 7, i32 5>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %vec2
@@ -2042,13 +2121,21 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mask4(<8 x i64> %vec, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask4:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,7,7,5]
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask4:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,7,7,5]
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask4:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm2[1],ymm0[3],ymm2[3]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm0[2,3,3,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 3, i32 7, i32 7, i32 5>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
@@ -2082,23 +2169,39 @@
   ret <4 x i64> %res
 }
 define <4 x i64> @test_8xi64_to_4xi64_perm_mask6(<8 x i64> %vec) {
-; CHECK-LABEL: test_8xi64_to_4xi64_perm_mask6:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,3]
-; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_8xi64_to_4xi64_perm_mask6:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,3]
+; CHECK-FAST-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_8xi64_to_4xi64_perm_mask6:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[3,2,1,3]
+; CHECK-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %res = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 6, i32 5, i32 3>
   ret <4 x i64> %res
 }
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mask6(<8 x i64> %vec, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mask6:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,6,5,3]
-; CHECK-NEXT:    vpermq %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vptestnmq %ymm2, %ymm2, %k1
-; CHECK-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mask6:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,6,5,3]
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mask6:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[3,2,1,3]
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3,4,5],ymm0[6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm2, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpblendmq %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 6, i32 5, i32 3>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> %vec2
@@ -2106,13 +2209,22 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mask6(<8 x i64> %vec, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask6:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,3]
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask6:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,3]
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermq %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mask6:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[3,2,1,3]
+; CHECK-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2,3,4,5],ymm0[6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa64 %ymm0, %ymm0 {%k1} {z}
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 6, i32 5, i32 3>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x i64> %shuf, <4 x i64> zeroinitializer
@@ -2250,14 +2362,22 @@
 }
 
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mem_mask1(<8 x i64>* %vp, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,4]
-; CHECK-NEXT:    vpermi2q (%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [4,3,2,4]
+; CHECK-FAST-NEXT:    vpermi2q (%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpblendd $3, (%rdi), %ymm2, %ymm2 # ymm2 = mem[0,1],ymm2[2,3,4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} = ymm2[0,3,2,0]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 0, i32 7, i32 6, i32 0>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2266,14 +2386,22 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mem_mask1(<8 x i64>* %vp, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,4]
-; CHECK-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; CHECK-NEXT:    vpermi2q (%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovdqa %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,3,2,4]
+; CHECK-FAST-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2q (%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vpblendd $3, (%rdi), %ymm1, %ymm1 # ymm1 = mem[0,1],ymm1[2,3,4,5,6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[0,3,2,0]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 0, i32 7, i32 6, i32 0>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2282,14 +2410,22 @@
 }
 
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mem_mask2(<8 x i64>* %vp, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,5,5,1]
-; CHECK-NEXT:    vpermi2q (%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,5,5,1]
+; CHECK-FAST-NEXT:    vpermi2q (%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpunpckhqdq (%rdi), %ymm2, %ymm2 # ymm2 = ymm2[1],mem[1],ymm2[3],mem[3]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} = ymm2[2,1,1,0]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 1, i32 1, i32 5>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2298,14 +2434,22 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mem_mask2(<8 x i64>* %vp, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm1 = [3,5,5,1]
-; CHECK-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; CHECK-NEXT:    vpermi2q (%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovdqa %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [3,5,5,1]
+; CHECK-FAST-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2q (%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vpunpckhqdq (%rdi), %ymm1, %ymm1 # ymm1 = ymm1[1],mem[1],ymm1[3],mem[3]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[2,1,1,0]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 1, i32 1, i32 5>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2314,25 +2458,40 @@
 }
 
 define <4 x i64> @test_8xi64_to_4xi64_perm_mem_mask3(<8 x i64>* %vp) {
-; CHECK-LABEL: test_8xi64_to_4xi64_perm_mem_mask3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa (%rdi), %ymm1
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm0 = [7,0,0,2]
-; CHECK-NEXT:    vpermi2q 32(%rdi), %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_8xi64_to_4xi64_perm_mem_mask3:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa (%rdi), %ymm1
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [7,0,0,2]
+; CHECK-FAST-NEXT:    vpermi2q 32(%rdi), %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_8xi64_to_4xi64_perm_mem_mask3:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
+; CHECK-FAST-PERLANE-NEXT:    vpalignr $8, 32(%rdi), %ymm0, %ymm0 # ymm0 = mem[8,9,10,11,12,13,14,15],ymm0[0,1,2,3,4,5,6,7],mem[24,25,26,27,28,29,30,31],ymm0[16,17,18,19,20,21,22,23]
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,1,1,3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %res = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 0, i32 0, i32 2>
   ret <4 x i64> %res
 }
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mem_mask3(<8 x i64>* %vp, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa (%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,0,0,2]
-; CHECK-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask3:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [7,0,0,2]
+; CHECK-FAST-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask3:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpalignr $8, 32(%rdi), %ymm2, %ymm2 # ymm2 = mem[8,9,10,11,12,13,14,15],ymm2[0,1,2,3,4,5,6,7],mem[24,25,26,27,28,29,30,31],ymm2[16,17,18,19,20,21,22,23]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} = ymm2[2,1,1,3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 0, i32 0, i32 2>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2341,14 +2500,22 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mem_mask3(<8 x i64>* %vp, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask3:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa (%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm1 = [7,0,0,2]
-; CHECK-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; CHECK-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovdqa %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask3:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [7,0,0,2]
+; CHECK-FAST-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask3:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vpalignr $8, 32(%rdi), %ymm1, %ymm1 # ymm1 = mem[8,9,10,11,12,13,14,15],ymm1[0,1,2,3,4,5,6,7],mem[24,25,26,27,28,29,30,31],ymm1[16,17,18,19,20,21,22,23]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[2,1,1,3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 0, i32 0, i32 2>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2389,14 +2556,22 @@
 }
 
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mem_mask5(<8 x i64>* %vp, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask5:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa (%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,2,7,1]
-; CHECK-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask5:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [0,2,7,1]
+; CHECK-FAST-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask5:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpblendd $192, 32(%rdi), %ymm2, %ymm2 # ymm2 = ymm2[0,1,2,3,4,5],mem[6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} = ymm2[0,2,3,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 0, i32 2, i32 7, i32 1>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2405,14 +2580,22 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mem_mask5(<8 x i64>* %vp, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask5:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa (%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,2,7,1]
-; CHECK-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; CHECK-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovdqa %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask5:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,2,7,1]
+; CHECK-FAST-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2q 32(%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask5:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vpblendd $192, 32(%rdi), %ymm1, %ymm1 # ymm1 = ymm1[0,1,2,3,4,5],mem[6,7]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[0,2,3,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 0, i32 2, i32 7, i32 1>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2464,14 +2647,22 @@
 }
 
 define <4 x i64> @test_masked_8xi64_to_4xi64_perm_mem_mask7(<8 x i64>* %vp, <4 x i64> %vec2, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask7:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,3,1,5]
-; CHECK-NEXT:    vpermi2q (%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vptestnmq %ymm1, %ymm1, %k1
-; CHECK-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask7:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,3,1,5]
+; CHECK-FAST-NEXT:    vpermi2q (%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovdqa64 %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xi64_to_4xi64_perm_mem_mask7:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpunpckhqdq (%rdi), %ymm2, %ymm2 # ymm2 = ymm2[1],mem[1],ymm2[3],mem[3]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm1, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} = ymm2[2,2,0,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 7, i32 5, i32 1>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -2480,14 +2671,22 @@
 }
 
 define <4 x i64> @test_masked_z_8xi64_to_4xi64_perm_mem_mask7(<8 x i64>* %vp, <4 x i64> %mask) {
-; CHECK-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask7:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovdqa 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovdqa {{.*#+}} ymm1 = [3,3,1,5]
-; CHECK-NEXT:    vptestnmq %ymm0, %ymm0, %k1
-; CHECK-NEXT:    vpermi2q (%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovdqa %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask7:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovdqa 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [3,3,1,5]
+; CHECK-FAST-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2q (%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovdqa %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xi64_to_4xi64_perm_mem_mask7:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vpunpckhqdq (%rdi), %ymm1, %ymm1 # ymm1 = ymm1[1],mem[1],ymm1[3],mem[3]
+; CHECK-FAST-PERLANE-NEXT:    vptestnmq %ymm0, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 {%k1} {z} = ymm1[2,2,0,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x i64>, <8 x i64>* %vp
   %shuf = shufflevector <8 x i64> %vec, <8 x i64> undef, <4 x i32> <i32 7, i32 7, i32 5, i32 1>
   %cmp = icmp eq <4 x i64> %mask, zeroinitializer
@@ -3050,15 +3249,27 @@
   ret <4 x float> %res
 }
 define <4 x float> @test_masked_16xfloat_to_4xfloat_perm_mask1(<16 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
-; CHECK-LABEL: test_masked_16xfloat_to_4xfloat_perm_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} xmm3 = [8,6,10,6]
-; CHECK-NEXT:    vpermps %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
-; CHECK-NEXT:    vblendmps %xmm0, %xmm1, %xmm0 {%k1}
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_16xfloat_to_4xfloat_perm_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} xmm3 = [8,6,10,6]
+; CHECK-FAST-NEXT:    vpermps %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
+; CHECK-FAST-NEXT:    vblendmps %xmm0, %xmm1, %xmm0 {%k1}
+; CHECK-FAST-NEXT:    vzeroupper
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_16xfloat_to_4xfloat_perm_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; CHECK-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} xmm4 = [0,6,2,6]
+; CHECK-FAST-PERLANE-NEXT:    vpermi2ps %xmm0, %xmm3, %xmm4
+; CHECK-FAST-PERLANE-NEXT:    vxorps %xmm0, %xmm0, %xmm0
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqps %xmm0, %xmm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vblendmps %xmm4, %xmm1, %xmm0 {%k1}
+; CHECK-FAST-PERLANE-NEXT:    vzeroupper
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <4 x i32> <i32 8, i32 6, i32 10, i32 6>
   %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
@@ -3066,15 +3277,26 @@
 }
 
 define <4 x float> @test_masked_z_16xfloat_to_4xfloat_perm_mask1(<16 x float> %vec, <4 x float> %mask) {
-; CHECK-LABEL: test_masked_z_16xfloat_to_4xfloat_perm_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps {{.*#+}} xmm2 = [8,6,10,6]
-; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqps %xmm3, %xmm1, %k1
-; CHECK-NEXT:    vpermps %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_16xfloat_to_4xfloat_perm_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} xmm2 = [8,6,10,6]
+; CHECK-FAST-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqps %xmm3, %xmm1, %k1
+; CHECK-FAST-NEXT:    vpermps %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
+; CHECK-FAST-NEXT:    vzeroupper
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_16xfloat_to_4xfloat_perm_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} xmm0 = [0,6,2,6]
+; CHECK-FAST-PERLANE-NEXT:    vxorps %xmm4, %xmm4, %xmm4
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqps %xmm4, %xmm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermi2ps %xmm3, %xmm2, %xmm0 {%k1} {z}
+; CHECK-FAST-PERLANE-NEXT:    vzeroupper
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <4 x i32> <i32 8, i32 6, i32 10, i32 6>
   %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
@@ -3234,15 +3456,26 @@
 }
 
 define <8 x float> @test_masked_16xfloat_to_8xfloat_perm_mem_mask2(<16 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
-; CHECK-LABEL: test_masked_16xfloat_to_8xfloat_perm_mem_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovaps {{.*#+}} ymm3 = [9,5,2,3,2,8,8,1]
-; CHECK-NEXT:    vpermi2ps (%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
-; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
-; CHECK-NEXT:    vmovaps %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_16xfloat_to_8xfloat_perm_mem_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} ymm3 = [9,5,2,3,2,8,8,1]
+; CHECK-FAST-NEXT:    vpermi2ps (%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; CHECK-FAST-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovaps %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_16xfloat_to_8xfloat_perm_mem_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovaps (%rdi), %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vmovaps 32(%rdi), %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm4 = [9,5,2,3,2,8,8,1]
+; CHECK-FAST-PERLANE-NEXT:    vpermi2ps %ymm2, %ymm3, %ymm4
+; CHECK-FAST-PERLANE-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vmovaps %ymm4, %ymm0 {%k1}
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <16 x float>, <16 x float>* %vp
   %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <8 x i32> <i32 1, i32 13, i32 10, i32 11, i32 10, i32 0, i32 0, i32 9>
   %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
@@ -3251,15 +3484,26 @@
 }
 
 define <8 x float> @test_masked_z_16xfloat_to_8xfloat_perm_mem_mask2(<16 x float>* %vp, <8 x float> %mask) {
-; CHECK-LABEL: test_masked_z_16xfloat_to_8xfloat_perm_mem_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovaps 32(%rdi), %ymm2
-; CHECK-NEXT:    vmovaps {{.*#+}} ymm1 = [9,5,2,3,2,8,8,1]
-; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqps %ymm3, %ymm0, %k1
-; CHECK-NEXT:    vpermi2ps (%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovaps %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_16xfloat_to_8xfloat_perm_mem_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovaps 32(%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [9,5,2,3,2,8,8,1]
+; CHECK-FAST-NEXT:    vxorps %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqps %ymm3, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2ps (%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovaps %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_16xfloat_to_8xfloat_perm_mem_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovaps (%rdi), %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vmovaps 32(%rdi), %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm1 = [9,5,2,3,2,8,8,1]
+; CHECK-FAST-PERLANE-NEXT:    vxorps %xmm4, %xmm4, %xmm4
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqps %ymm4, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermi2ps %ymm2, %ymm3, %ymm1 {%k1} {z}
+; CHECK-FAST-PERLANE-NEXT:    vmovaps %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <16 x float>, <16 x float>* %vp
   %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <8 x i32> <i32 1, i32 13, i32 10, i32 11, i32 10, i32 0, i32 0, i32 9>
   %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
@@ -3757,15 +4001,25 @@
   ret <4 x double> %res
 }
 define <4 x double> @test_masked_8xdouble_to_4xdouble_perm_mask4(<8 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask4:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm4 = [1,1,5,5]
-; CHECK-NEXT:    vpermi2pd %ymm3, %ymm0, %ymm4
-; CHECK-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vcmpeqpd %ymm0, %ymm2, %k1
-; CHECK-NEXT:    vblendmpd %ymm4, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask4:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm4 = [1,1,5,5]
+; CHECK-FAST-NEXT:    vpermi2pd %ymm3, %ymm0, %ymm4
+; CHECK-FAST-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm0, %ymm2, %k1
+; CHECK-FAST-NEXT:    vblendmpd %ymm4, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask4:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm3[1]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm1 {%k1} = ymm0[0,0,1,1]
+; CHECK-FAST-PERLANE-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 1, i32 1, i32 5, i32 5>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
@@ -3773,15 +4027,24 @@
 }
 
 define <4 x double> @test_masked_z_8xdouble_to_4xdouble_perm_mask4(<8 x double> %vec, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask4:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm2 = [1,1,5,5]
-; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
-; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm1, %k1
-; CHECK-NEXT:    vpermi2pd %ymm3, %ymm0, %ymm2 {%k1} {z}
-; CHECK-NEXT:    vmovapd %ymm2, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask4:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vextractf32x4 $2, %zmm0, %xmm3
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [1,1,5,5]
+; CHECK-FAST-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm4, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermi2pd %ymm3, %ymm0, %ymm2 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovapd %ymm2, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask4:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm2[1]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0,0,1,1]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 1, i32 1, i32 5, i32 5>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
@@ -3817,24 +4080,41 @@
   ret <4 x double> %res
 }
 define <4 x double> @test_8xdouble_to_4xdouble_perm_mask6(<8 x double> %vec) {
-; CHECK-LABEL: test_8xdouble_to_4xdouble_perm_mask6:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm1 = [5,8,7,8]
-; CHECK-NEXT:    vpermt2pd %zmm0, %zmm1, %zmm0
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_8xdouble_to_4xdouble_perm_mask6:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm1 = [5,8,7,8]
+; CHECK-FAST-NEXT:    vpermt2pd %zmm0, %zmm1, %zmm0
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_8xdouble_to_4xdouble_perm_mask6:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vbroadcastsd %xmm0, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm1[1],ymm0[1],ymm1[3],ymm0[3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %res = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 5, i32 0, i32 7, i32 0>
   ret <4 x double> %res
 }
 define <4 x double> @test_masked_8xdouble_to_4xdouble_perm_mask6(<8 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask6:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm3 = [5,8,7,8]
-; CHECK-NEXT:    vpermi2pd %zmm0, %zmm0, %zmm3
-; CHECK-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
-; CHECK-NEXT:    vcmpeqpd %ymm0, %ymm2, %k1
-; CHECK-NEXT:    vblendmpd %ymm3, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask6:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm3 = [5,8,7,8]
+; CHECK-FAST-NEXT:    vpermi2pd %zmm0, %zmm0, %zmm3
+; CHECK-FAST-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm0, %ymm2, %k1
+; CHECK-FAST-NEXT:    vblendmpd %ymm3, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask6:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vbroadcastsd %xmm0, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm4, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vunpckhpd {{.*#+}} ymm1 {%k1} = ymm3[1],ymm0[1],ymm3[3],ymm0[3]
+; CHECK-FAST-PERLANE-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 5, i32 0, i32 7, i32 0>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
@@ -3842,28 +4122,47 @@
 }
 
 define <4 x double> @test_masked_z_8xdouble_to_4xdouble_perm_mask6(<8 x double> %vec, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask6:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm2 = [5,8,7,8]
-; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
-; CHECK-NEXT:    vpermt2pd %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask6:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [5,8,7,8]
+; CHECK-FAST-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermt2pd %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask6:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vbroadcastsd %xmm0, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm2[1],ymm0[1],ymm2[3],ymm0[3]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 5, i32 0, i32 7, i32 0>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
   ret <4 x double> %res
 }
 define <4 x double> @test_masked_8xdouble_to_4xdouble_perm_mask7(<8 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask7:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm3 = [3,5,0,6]
-; CHECK-NEXT:    vpermpd %zmm0, %zmm3, %zmm0
-; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
-; CHECK-NEXT:    vblendmpd %ymm0, %ymm1, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask7:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm3 = [3,5,0,6]
+; CHECK-FAST-NEXT:    vpermpd %zmm0, %zmm3, %zmm0
+; CHECK-FAST-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
+; CHECK-FAST-NEXT:    vblendmpd %ymm0, %ymm1, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xdouble_to_4xdouble_perm_mask7:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm3
+; CHECK-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,0,3]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm4, %ymm2, %k1
+; CHECK-FAST-PERLANE-NEXT:    vshufpd {{.*#+}} ymm1 {%k1} = ymm0[1],ymm3[1],ymm0[2],ymm3[2]
+; CHECK-FAST-PERLANE-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 3, i32 5, i32 0, i32 6>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
@@ -3871,14 +4170,23 @@
 }
 
 define <4 x double> @test_masked_z_8xdouble_to_4xdouble_perm_mask7(<8 x double> %vec, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask7:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm2 = [3,5,0,6]
-; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
-; CHECK-NEXT:    vpermpd %zmm0, %zmm2, %zmm0 {%k1} {z}
-; CHECK-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask7:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [3,5,0,6]
+; CHECK-FAST-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
+; CHECK-FAST-NEXT:    vpermpd %zmm0, %zmm2, %zmm0 {%k1} {z}
+; CHECK-FAST-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mask7:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vextractf64x4 $1, %zmm0, %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,3,0,3]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vshufpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],ymm2[1],ymm0[2],ymm2[2]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 3, i32 5, i32 0, i32 6>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
   %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
@@ -4004,15 +4312,23 @@
 }
 
 define <4 x double> @test_masked_8xdouble_to_4xdouble_perm_mem_mask1(<8 x double>* %vp, <4 x double> %vec2, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_8xdouble_to_4xdouble_perm_mem_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd (%rdi), %ymm2
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm3 = [3,4,2,6]
-; CHECK-NEXT:    vpermi2pd 32(%rdi){1to4}, %ymm2, %ymm3
-; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
-; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
-; CHECK-NEXT:    vmovapd %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xdouble_to_4xdouble_perm_mem_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm3 = [3,4,2,6]
+; CHECK-FAST-NEXT:    vpermi2pd 32(%rdi){1to4}, %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovapd %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xdouble_to_4xdouble_perm_mem_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vpermpd $236, (%rdi), %ymm2 # ymm2 = mem[0,3,2,3]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm3, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vshufpd $1, 32(%rdi){1to4}, %ymm2, %ymm0 {%k1}
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x double>, <8 x double>* %vp
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 3, i32 4, i32 2, i32 4>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
@@ -4021,15 +4337,23 @@
 }
 
 define <4 x double> @test_masked_z_8xdouble_to_4xdouble_perm_mem_mask1(<8 x double>* %vp, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mem_mask1:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd (%rdi), %ymm2
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm1 = [3,4,2,6]
-; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm0, %k1
-; CHECK-NEXT:    vpermi2pd 32(%rdi){1to4}, %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovapd %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mem_mask1:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm1 = [3,4,2,6]
+; CHECK-FAST-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm3, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2pd 32(%rdi){1to4}, %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mem_mask1:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vpermpd $236, (%rdi), %ymm1 # ymm1 = mem[0,3,2,3]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm2, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vshufpd $1, 32(%rdi){1to4}, %ymm1, %ymm0 {%k1} {z}
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x double>, <8 x double>* %vp
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 3, i32 4, i32 2, i32 4>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
@@ -4038,15 +4362,24 @@
 }
 
 define <4 x double> @test_masked_8xdouble_to_4xdouble_perm_mem_mask2(<8 x double>* %vp, <4 x double> %vec2, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_8xdouble_to_4xdouble_perm_mem_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd (%rdi), %ymm2
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm3 = [1,2,3,4]
-; CHECK-NEXT:    vpermi2pd 32(%rdi), %ymm2, %ymm3
-; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
-; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
-; CHECK-NEXT:    vmovapd %ymm3, %ymm0 {%k1}
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_8xdouble_to_4xdouble_perm_mem_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm3 = [1,2,3,4]
+; CHECK-FAST-NEXT:    vpermi2pd 32(%rdi), %ymm2, %ymm3
+; CHECK-FAST-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
+; CHECK-FAST-NEXT:    vmovapd %ymm3, %ymm0 {%k1}
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_8xdouble_to_4xdouble_perm_mem_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovapd (%rdi), %ymm2
+; CHECK-FAST-PERLANE-NEXT:    vperm2f128 $33, 32(%rdi), %ymm2, %ymm3 # ymm3 = ymm2[2,3],mem[0,1]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm4, %ymm1, %k1
+; CHECK-FAST-PERLANE-NEXT:    vshufpd {{.*#+}} ymm0 {%k1} = ymm2[1],ymm3[0],ymm2[3],ymm3[2]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x double>, <8 x double>* %vp
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
@@ -4055,15 +4388,24 @@
 }
 
 define <4 x double> @test_masked_z_8xdouble_to_4xdouble_perm_mem_mask2(<8 x double>* %vp, <4 x double> %mask) {
-; CHECK-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mem_mask2:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vmovapd (%rdi), %ymm2
-; CHECK-NEXT:    vmovapd {{.*#+}} ymm1 = [1,2,3,4]
-; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
-; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm0, %k1
-; CHECK-NEXT:    vpermi2pd 32(%rdi), %ymm2, %ymm1 {%k1} {z}
-; CHECK-NEXT:    vmovapd %ymm1, %ymm0
-; CHECK-NEXT:    retq
+; CHECK-FAST-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mem_mask2:
+; CHECK-FAST:       # %bb.0:
+; CHECK-FAST-NEXT:    vmovapd (%rdi), %ymm2
+; CHECK-FAST-NEXT:    vmovapd {{.*#+}} ymm1 = [1,2,3,4]
+; CHECK-FAST-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-NEXT:    vcmpeqpd %ymm3, %ymm0, %k1
+; CHECK-FAST-NEXT:    vpermi2pd 32(%rdi), %ymm2, %ymm1 {%k1} {z}
+; CHECK-FAST-NEXT:    vmovapd %ymm1, %ymm0
+; CHECK-FAST-NEXT:    retq
+;
+; CHECK-FAST-PERLANE-LABEL: test_masked_z_8xdouble_to_4xdouble_perm_mem_mask2:
+; CHECK-FAST-PERLANE:       # %bb.0:
+; CHECK-FAST-PERLANE-NEXT:    vmovapd (%rdi), %ymm1
+; CHECK-FAST-PERLANE-NEXT:    vperm2f128 $33, 32(%rdi), %ymm1, %ymm2 # ymm2 = ymm1[2,3],mem[0,1]
+; CHECK-FAST-PERLANE-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
+; CHECK-FAST-PERLANE-NEXT:    vcmpeqpd %ymm3, %ymm0, %k1
+; CHECK-FAST-PERLANE-NEXT:    vshufpd {{.*#+}} ymm0 {%k1} {z} = ymm1[1],ymm2[0],ymm1[3],ymm2[2]
+; CHECK-FAST-PERLANE-NEXT:    retq
   %vec = load <8 x double>, <8 x double>* %vp
   %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 4>
   %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
diff --git a/llvm/test/CodeGen/X86/avx512-trunc.ll b/llvm/test/CodeGen/X86/avx512-trunc.ll
--- a/llvm/test/CodeGen/X86/avx512-trunc.ll
+++ b/llvm/test/CodeGen/X86/avx512-trunc.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=KNL
-; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f,+avx512vl,+avx512bw,+avx512dq,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,SKX
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f,+avx512vl,+avx512bw,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,SKX
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx512f,+avx512vl,+avx512bw,+avx512dq,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,SKX
 
  attributes #0 = { nounwind }
 
diff --git a/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll b/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll
--- a/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll
+++ b/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll
@@ -3,8 +3,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512
 
 ;
 ; 128-bit vectors
diff --git a/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-zext.ll b/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-zext.ll
--- a/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-zext.ll
+++ b/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-zext.ll
@@ -3,9 +3,11 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE2-SSSE3,SSSE3
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512VLBW
 
 ;
 ; 128-bit vectors
diff --git a/llvm/test/CodeGen/X86/broadcastm-lowering.ll b/llvm/test/CodeGen/X86/broadcastm-lowering.ll
--- a/llvm/test/CodeGen/X86/broadcastm-lowering.ll
+++ b/llvm/test/CodeGen/X86/broadcastm-lowering.ll
@@ -1,10 +1,16 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CD
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512cd,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CD
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CDBW
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512cd,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CDBW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,avx512cd,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX512VLCDBW
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vl,avx512cd,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX512VLCDBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CD
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CD
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512cd,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CD
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512cd,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CD
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CDBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512cd,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CDBW
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512cd,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CDBW
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512cd,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512CDBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,avx512cd,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512VLCDBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,avx512cd,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512VLCDBW
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vl,avx512cd,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512VLCDBW
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vl,avx512cd,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX512VLCDBW
 
 define <2 x i64> @test_mm_epi64(<8 x i16> %a, <8 x i16> %b) {
 ; AVX512CD-LABEL: test_mm_epi64:
diff --git a/llvm/test/CodeGen/X86/combine-shl.ll b/llvm/test/CodeGen/X86/combine-shl.ll
--- a/llvm/test/CodeGen/X86/combine-shl.ll
+++ b/llvm/test/CodeGen/X86/combine-shl.ll
@@ -2,7 +2,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX-FAST-PERLANE
 
 ; fold (shl 0, x) -> 0
 define <4 x i32> @combine_vec_shl_zero(<4 x i32> %x) {
@@ -145,14 +146,23 @@
 ; AVX-SLOW-NEXT:    vzeroupper
 ; AVX-SLOW-NEXT:    retq
 ;
-; AVX-FAST-LABEL: combine_vec_shl_trunc_and:
-; AVX-FAST:       # %bb.0:
-; AVX-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
-; AVX-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX-FAST-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX-FAST-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
-; AVX-FAST-NEXT:    vzeroupper
-; AVX-FAST-NEXT:    retq
+; AVX-FAST-ALL-LABEL: combine_vec_shl_trunc_and:
+; AVX-FAST-ALL:       # %bb.0:
+; AVX-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
+; AVX-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX-FAST-ALL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX-FAST-ALL-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
+; AVX-FAST-ALL-NEXT:    vzeroupper
+; AVX-FAST-ALL-NEXT:    retq
+;
+; AVX-FAST-PERLANE-LABEL: combine_vec_shl_trunc_and:
+; AVX-FAST-PERLANE:       # %bb.0:
+; AVX-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX-FAST-PERLANE-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX-FAST-PERLANE-NEXT:    vpsllvd %xmm1, %xmm0, %xmm0
+; AVX-FAST-PERLANE-NEXT:    vzeroupper
+; AVX-FAST-PERLANE-NEXT:    retq
   %1 = and <4 x i64> %y, <i64 15, i64 255, i64 4095, i64 65535>
   %2 = trunc <4 x i64> %1 to <4 x i32>
   %3 = shl <4 x i32> %x, %2
diff --git a/llvm/test/CodeGen/X86/combine-sra.ll b/llvm/test/CodeGen/X86/combine-sra.ll
--- a/llvm/test/CodeGen/X86/combine-sra.ll
+++ b/llvm/test/CodeGen/X86/combine-sra.ll
@@ -1,7 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-FAST-PERLANE
 
 ; fold (sra 0, x) -> 0
 define <4 x i32> @combine_vec_ashr_zero(<4 x i32> %x) {
@@ -175,14 +176,23 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: combine_vec_ashr_trunc_and:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-FAST-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: combine_vec_ashr_trunc_and:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-FAST-ALL-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: combine_vec_ashr_trunc_and:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpsravd %xmm1, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %1 = and <4 x i64> %y, <i64 15, i64 255, i64 4095, i64 65535>
   %2 = trunc <4 x i64> %1 to <4 x i32>
   %3 = ashr <4 x i32> %x, %2
@@ -213,13 +223,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: combine_vec_ashr_trunc_lshr:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,3,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpsravd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: combine_vec_ashr_trunc_lshr:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpsravd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: combine_vec_ashr_trunc_lshr:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsravd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %1 = lshr <4 x i64> %x, <i64 32, i64 32, i64 32, i64 32>
   %2 = trunc <4 x i64> %1 to <4 x i32>
   %3 = ashr <4 x i32> %2, <i32 0, i32 1, i32 2, i32 3>
@@ -278,13 +296,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: combine_vec_ashr_trunc_ashr:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,3,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpsravd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: combine_vec_ashr_trunc_ashr:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpsravd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: combine_vec_ashr_trunc_ashr:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsravd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %1 = ashr <4 x i64> %x, <i64 32, i64 32, i64 32, i64 32>
   %2 = trunc <4 x i64> %1 to <4 x i32>
   %3 = ashr <4 x i32> %2, <i32 0, i32 1, i32 2, i32 3>
diff --git a/llvm/test/CodeGen/X86/combine-srl.ll b/llvm/test/CodeGen/X86/combine-srl.ll
--- a/llvm/test/CodeGen/X86/combine-srl.ll
+++ b/llvm/test/CodeGen/X86/combine-srl.ll
@@ -1,7 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2-FAST-PERLANE
 
 ; fold (srl 0, x) -> 0
 define <4 x i32> @combine_vec_lshr_zero(<4 x i32> %x) {
@@ -210,14 +211,23 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: combine_vec_lshr_trunc_lshr1:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: combine_vec_lshr_trunc_lshr1:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: combine_vec_lshr_trunc_lshr1:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %1 = lshr <4 x i64> %x, <i64 32, i64 33, i64 34, i64 35>
   %2 = trunc <4 x i64> %1 to <4 x i32>
   %3 = lshr <4 x i32> %2, <i32 16, i32 17, i32 18, i32 19>
@@ -423,14 +433,23 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: combine_vec_lshr_trunc_and:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
-; AVX2-FAST-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: combine_vec_lshr_trunc_and:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-FAST-ALL-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: combine_vec_lshr_trunc_and:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %1 = and <4 x i64> %y, <i64 15, i64 255, i64 4095, i64 65535>
   %2 = trunc <4 x i64> %1 to <4 x i32>
   %3 = lshr <4 x i32> %x, %2
diff --git a/llvm/test/CodeGen/X86/insertelement-zero.ll b/llvm/test/CodeGen/X86/insertelement-zero.ll
--- a/llvm/test/CodeGen/X86/insertelement-zero.ll
+++ b/llvm/test/CodeGen/X86/insertelement-zero.ll
@@ -5,7 +5,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
 
 define <2 x double> @insert_v2f64_z1(<2 x double> %a) {
 ; SSE2-LABEL: insert_v2f64_z1:
diff --git a/llvm/test/CodeGen/X86/oddshuffles.ll b/llvm/test/CodeGen/X86/oddshuffles.ll
--- a/llvm/test/CodeGen/X86/oddshuffles.ll
+++ b/llvm/test/CodeGen/X86/oddshuffles.ll
@@ -3,7 +3,8 @@
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+sse4.2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE42
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+xop | FileCheck %s --check-prefix=XOP
 
 define void @v3i64(<2 x i64> %a, <2 x i64> %b, <3 x i64>* %p) nounwind {
@@ -582,20 +583,35 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: v12i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <0,4,u,1,5,u,2,6>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm2
-; AVX2-FAST-NEXT:    vbroadcastsd %xmm1, %ymm3
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm3 = <u,3,7,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm3, %ymm0
-; AVX2-FAST-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
-; AVX2-FAST-NEXT:    vmovaps %xmm0, 32(%rdi)
-; AVX2-FAST-NEXT:    vmovaps %ymm2, (%rdi)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: v12i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <0,4,u,1,5,u,2,6>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm2
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd %xmm1, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm3 = <u,3,7,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm3, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
+; AVX2-FAST-ALL-NEXT:    vmovaps %xmm0, 32(%rdi)
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, (%rdi)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: v12i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm2 = <0,4,u,1,5,u,2,6>
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm0, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd %xmm1, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3],xmm3[3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %xmm0, 32(%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, (%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; XOP-LABEL: v12i32:
 ; XOP:       # %bb.0:
@@ -1155,27 +1171,50 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: interleave_24i16_in:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqu (%rsi), %xmm0
-; AVX2-FAST-NEXT:    vmovdqu (%rdx), %xmm1
-; AVX2-FAST-NEXT:    vmovdqu (%rcx), %xmm2
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,0,0,u,1,1,u,2>
-; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,4,1,5,1,5,2,6]
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm5, %ymm3
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[0,1,4,5,u,u,2,3,6,7,u,u,8,9,12,13,u,u,18,19,22,23,u,u,24,25,28,29,u,u,26,27]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,u,u,10,11,8,9,u,u,14,15,12,13,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
-; AVX2-FAST-NEXT:    vmovdqu %xmm0, 32(%rdi)
-; AVX2-FAST-NEXT:    vmovdqu %ymm3, (%rdi)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: interleave_24i16_in:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqu (%rsi), %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqu (%rdx), %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqu (%rcx), %xmm2
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,0,0,u,1,1,u,2>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm2, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,4,1,5,1,5,2,6]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm3, %ymm5, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[0,1,4,5,u,u,2,3,6,7,u,u,8,9,12,13,u,u,18,19,22,23,u,u,24,25,28,29,u,u,26,27]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,u,u,10,11,8,9,u,u,14,15,12,13,u,u]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %xmm0, 32(%rdi)
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm3, (%rdi)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: interleave_24i16_in:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu (%rsi), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu (%rdx), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu (%rcx), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm3[0,1,u,u,6,7,2,3,u,u,8,9,4,5,u,u,16,17,u,u,22,23,18,19,u,u,24,25,20,21,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,0,1,u,u,u,u,2,3,u,u,u,u,4,5,u,u,22,23,u,u,u,u,24,25,u,u,u,u,26,27]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7],ymm4[8],ymm3[9],ymm4[10,11],ymm3[12],ymm4[13,14],ymm3[15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,0,0,u,1,1,u,2>
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm4, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,u,u,10,11,8,9,u,u,14,15,12,13,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %xmm0, 32(%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm3, (%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; XOP-LABEL: interleave_24i16_in:
 ; XOP:       # %bb.0:
@@ -1349,35 +1388,65 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: interleave_24i32_out:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovups (%rdi), %ymm0
-; AVX2-FAST-NEXT:    vmovups 32(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vmovups 64(%rdi), %ymm2
-; AVX2-FAST-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [21474836482,21474836482,21474836482,21474836482]
-; AVX2-FAST-NEXT:    vpermps %ymm2, %ymm3, %ymm3
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm5 = <0,3,6,1,4,7,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
-; AVX2-FAST-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm2, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm6 = <1,4,7,2,5,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm5, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm5 = <u,u,u,u,u,1,4,7>
-; AVX2-FAST-NEXT:    vpermps %ymm2, %ymm5, %ymm2
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <2,5,0,3,6,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
-; AVX2-FAST-NEXT:    vmovups %ymm3, (%rsi)
-; AVX2-FAST-NEXT:    vmovups %ymm4, (%rdx)
-; AVX2-FAST-NEXT:    vmovups %ymm0, (%rcx)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: interleave_24i32_out:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovups (%rdi), %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovups 32(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovups 64(%rdi), %ymm2
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [21474836482,21474836482,21474836482,21474836482]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm3, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm5 = <0,3,6,1,4,7,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm4, %ymm5, %ymm4
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
+; AVX2-FAST-ALL-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm6 = <1,4,7,2,5,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm5, %ymm6, %ymm5
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm5 = <u,u,u,u,u,1,4,7>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm5, %ymm2
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <2,5,0,3,6,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm2[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovups %ymm3, (%rsi)
+; AVX2-FAST-ALL-NEXT:    vmovups %ymm4, (%rdx)
+; AVX2-FAST-ALL-NEXT:    vmovups %ymm0, (%rcx)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: interleave_24i32_out:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovups (%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovups 32(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovups 64(%rdi), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd {{.*#+}} ymm3 = [21474836482,21474836482,21474836482,21474836482]
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm2, %ymm3, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm5 = <0,3,6,1,4,7,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm4, %ymm5, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastf128 {{.*#+}} ymm4 = [0,0,3,6,0,0,3,6]
+; AVX2-FAST-PERLANE-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm2, %ymm4, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm5 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm6 = <1,4,7,2,5,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm5, %ymm6, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0,1,2,3,4],ymm4[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm1 = <2,5,0,3,6,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm1 = ymm2[0,1,0,3,4,5,4,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,1,0,3]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovups %ymm3, (%rsi)
+; AVX2-FAST-PERLANE-NEXT:    vmovups %ymm4, (%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vmovups %ymm0, (%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; XOP-LABEL: interleave_24i32_out:
 ; XOP:       # %bb.0:
@@ -1572,34 +1641,62 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: interleave_24i32_in:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovups (%rsi), %ymm0
-; AVX2-FAST-NEXT:    vmovups (%rdx), %ymm1
-; AVX2-FAST-NEXT:    vmovups (%rcx), %ymm2
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [1,0,2,2,1,0,2,2]
-; AVX2-FAST-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm3, %ymm3
-; AVX2-FAST-NEXT:    vpermpd {{.*#+}} ymm4 = ymm0[0,0,2,1]
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
-; AVX2-FAST-NEXT:    vbroadcastsd (%rcx), %ymm4
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm4 = <5,u,u,6,u,u,7,u>
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vpermpd {{.*#+}} ymm5 = ymm2[2,1,3,3]
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3],ymm5[4],ymm4[5,6],ymm5[7]
-; AVX2-FAST-NEXT:    vbroadcastsd 24(%rsi), %ymm5
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
-; AVX2-FAST-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[1,1,2,2]
-; AVX2-FAST-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[1,1,2,2]
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2,3],ymm0[4],ymm2[5,6],ymm0[7]
-; AVX2-FAST-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[0,0,3,3,4,4,7,7]
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
-; AVX2-FAST-NEXT:    vmovups %ymm0, 32(%rdi)
-; AVX2-FAST-NEXT:    vmovups %ymm4, 64(%rdi)
-; AVX2-FAST-NEXT:    vmovups %ymm3, (%rdi)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: interleave_24i32_in:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovups (%rsi), %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovups (%rdx), %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovups (%rcx), %ymm2
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm3 = [1,0,2,2,1,0,2,2]
+; AVX2-FAST-ALL-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm3, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpermpd {{.*#+}} ymm4 = ymm0[0,0,2,1]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd (%rcx), %ymm4
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm4 = <5,u,u,6,u,u,7,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpermpd {{.*#+}} ymm5 = ymm2[2,1,3,3]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3],ymm5[4],ymm4[5,6],ymm5[7]
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd 24(%rsi), %ymm5
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
+; AVX2-FAST-ALL-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[1,1,2,2]
+; AVX2-FAST-ALL-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[1,1,2,2]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2,3],ymm0[4],ymm2[5,6],ymm0[7]
+; AVX2-FAST-ALL-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[0,0,3,3,4,4,7,7]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovups %ymm0, 32(%rdi)
+; AVX2-FAST-ALL-NEXT:    vmovups %ymm4, 64(%rdi)
+; AVX2-FAST-ALL-NEXT:    vmovups %ymm3, (%rdi)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: interleave_24i32_in:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovups (%rsi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovups (%rdx), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovups (%rcx), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm3 = mem[1,0,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm3 = ymm3[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm4 = ymm0[0,0,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd (%rcx), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm4 = ymm2[2,1,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm5 = ymm1[1,2,3,3,5,6,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm5 = ymm5[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6],ymm4[7]
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd 24(%rsi), %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[1,1,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm2 = ymm2[1,1,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2,3],ymm0[4],ymm2[5,6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[0,0,3,3,4,4,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovups %ymm0, 32(%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vmovups %ymm4, 64(%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vmovups %ymm3, (%rdi)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; XOP-LABEL: interleave_24i32_in:
 ; XOP:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/oddsubvector.ll b/llvm/test/CodeGen/X86/oddsubvector.ll
--- a/llvm/test/CodeGen/X86/oddsubvector.ll
+++ b/llvm/test/CodeGen/X86/oddsubvector.ll
@@ -3,9 +3,11 @@
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f | FileCheck %s --check-prefix=AVX512
-; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+xop | FileCheck %s --check-prefixes=AVX,XOP
 
 define void @insert_v7i8_v2i16_2(<7 x i8> *%a0, <2 x i16> *%a1) nounwind {
diff --git a/llvm/test/CodeGen/X86/paddus.ll b/llvm/test/CodeGen/X86/paddus.ll
--- a/llvm/test/CodeGen/X86/paddus.ll
+++ b/llvm/test/CodeGen/X86/paddus.ll
@@ -4,7 +4,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512
 
 ; FIXME: should be paddusb
 define <16 x i8> @test1(<16 x i8> %x) {
diff --git a/llvm/test/CodeGen/X86/phaddsub.ll b/llvm/test/CodeGen/X86/phaddsub.ll
--- a/llvm/test/CodeGen/X86/phaddsub.ll
+++ b/llvm/test/CodeGen/X86/phaddsub.ll
@@ -5,7 +5,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx,fast-hops   | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX1-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2            | FileCheck %s --check-prefixes=AVX,AVX-SLOW,AVX2-SLOW
 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,fast-hops  | FileCheck %s --check-prefixes=AVX,AVX-FAST,AVX2-FAST
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-SHUF
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-SHUF
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-SHUF
 
 define <8 x i16> @phaddw1(<8 x i16> %x, <8 x i16> %y) {
 ; SSSE3-LABEL: phaddw1:
diff --git a/llvm/test/CodeGen/X86/psubus.ll b/llvm/test/CodeGen/X86/psubus.ll
--- a/llvm/test/CodeGen/X86/psubus.ll
+++ b/llvm/test/CodeGen/X86/psubus.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-PERLANE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512
 
 define <8 x i16> @test1(<8 x i16> %x) nounwind {
 ; SSE-LABEL: test1:
@@ -2591,24 +2593,42 @@
 ; AVX2-SLOW-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: test33:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
-; AVX2-FAST-NEXT:    vpxor %ymm3, %ymm1, %ymm4
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [4294967295,4294967295,4294967295,4294967295]
-; AVX2-FAST-NEXT:    vblendvpd %ymm4, %ymm1, %ymm6, %ymm1
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm4 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vpxor %ymm3, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm2, %ymm6, %ymm2
-; AVX2-FAST-NEXT:    vpermps %ymm2, %ymm4, %ymm2
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: test33:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm3, %ymm1, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm4, %ymm1, %ymm6, %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm4 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm4, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm3, %ymm2, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm2, %ymm6, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm4, %ymm2
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: test33:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm3, %ymm2, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm4, %ymm2, %ymm6, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm3, %ymm1, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm1, %ymm6, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm1[2,3],ymm2[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2],ymm3[0,2],ymm1[4,6],ymm3[4,6]
+; AVX2-FAST-PERLANE-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: test33:
 ; AVX512:       # %bb.0:
@@ -2811,26 +2831,46 @@
 ; AVX2-SLOW-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: test34:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1]
-; AVX2-FAST-NEXT:    vpand %ymm3, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
-; AVX2-FAST-NEXT:    vpxor %ymm3, %ymm1, %ymm4
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [4294967295,4294967295,4294967295,4294967295]
-; AVX2-FAST-NEXT:    vblendvpd %ymm4, %ymm1, %ymm6, %ymm1
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm4 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm4, %ymm1
-; AVX2-FAST-NEXT:    vpxor %ymm3, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm2, %ymm6, %ymm2
-; AVX2-FAST-NEXT:    vpermps %ymm2, %ymm4, %ymm2
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: test34:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1]
+; AVX2-FAST-ALL-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm3, %ymm1, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm4, %ymm1, %ymm6, %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm4 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm4, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm3, %ymm2, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm2, %ymm6, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm2, %ymm4, %ymm2
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: test34:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastd {{.*#+}} ymm3 = [1,1,1,1,1,1,1,1]
+; AVX2-FAST-PERLANE-NEXT:    vpand %ymm3, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm3, %ymm2, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd {{.*#+}} ymm6 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm4, %ymm2, %ymm6, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm3, %ymm1, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm1, %ymm6, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm1[2,3],ymm2[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm1 = ymm1[0,2],ymm3[0,2],ymm1[4,6],ymm3[4,6]
+; AVX2-FAST-PERLANE-NEXT:    vpmaxud %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpsubd %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: test34:
 ; AVX512:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/sadd_sat_vec.ll b/llvm/test/CodeGen/X86/sadd_sat_vec.ll
--- a/llvm/test/CodeGen/X86/sadd_sat_vec.ll
+++ b/llvm/test/CodeGen/X86/sadd_sat_vec.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
 
 declare <1 x i8> @llvm.sadd.sat.v1i8(<1 x i8>, <1 x i8>)
 declare <2 x i8> @llvm.sadd.sat.v2i8(<2 x i8>, <2 x i8>)
diff --git a/llvm/test/CodeGen/X86/shuffle-of-splat-multiuses.ll b/llvm/test/CodeGen/X86/shuffle-of-splat-multiuses.ll
--- a/llvm/test/CodeGen/X86/shuffle-of-splat-multiuses.ll
+++ b/llvm/test/CodeGen/X86/shuffle-of-splat-multiuses.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX2,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX2,AVX2-FAST-PERLANE
 ; PR32449
 
 define <2 x double> @foo2(<2 x double> %v, <2 x double> *%p) nounwind {
@@ -35,12 +36,19 @@
 ; AVX2-SLOW-NEXT:    vmovaps %ymm0, (%rdi)
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: foo8:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vbroadcastss {{.*#+}} ymm1 = [5,5,5,5,5,5,5,5]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovaps %ymm0, (%rdi)
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: foo8:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vbroadcastss {{.*#+}} ymm1 = [5,5,5,5,5,5,5,5]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm0, (%rdi)
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: foo8:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, (%rdi)
+; AVX2-FAST-PERLANE-NEXT:    retq
   %res = shufflevector <8 x float> %v, <8 x float> undef, <8 x i32> <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
   %res1 = shufflevector<8 x float> %res, <8 x float> undef, <8 x i32> <i32 2, i32 0, i32 undef, i32 undef, i32 5, i32 1, i32 3, i32 7>
   store <8 x float> %res, <8 x float>* %p
diff --git a/llvm/test/CodeGen/X86/shuffle-strided-with-offset-128.ll b/llvm/test/CodeGen/X86/shuffle-strided-with-offset-128.ll
--- a/llvm/test/CodeGen/X86/shuffle-strided-with-offset-128.ll
+++ b/llvm/test/CodeGen/X86/shuffle-strided-with-offset-128.ll
@@ -3,11 +3,15 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 
 define void @shuffle_v16i8_to_v8i8_1(<16 x i8>* %L, <8 x i8>* %S) nounwind {
 ; SSE2-LABEL: shuffle_v16i8_to_v8i8_1:
diff --git a/llvm/test/CodeGen/X86/shuffle-strided-with-offset-256.ll b/llvm/test/CodeGen/X86/shuffle-strided-with-offset-256.ll
--- a/llvm/test/CodeGen/X86/shuffle-strided-with-offset-256.ll
+++ b/llvm/test/CodeGen/X86/shuffle-strided-with-offset-256.ll
@@ -1,11 +1,15 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 
 define void @shuffle_v32i8_to_v16i8_1(<32 x i8>* %L, <16 x i8>* %S) nounwind {
 ; AVX-LABEL: shuffle_v32i8_to_v16i8_1:
diff --git a/llvm/test/CodeGen/X86/shuffle-strided-with-offset-512.ll b/llvm/test/CodeGen/X86/shuffle-strided-with-offset-512.ll
--- a/llvm/test/CodeGen/X86/shuffle-strided-with-offset-512.ll
+++ b/llvm/test/CodeGen/X86/shuffle-strided-with-offset-512.ll
@@ -1,8 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL,AVX512BWVL-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL,AVX512BWVL-FAST-PERLANE
 
 define void @shuffle_v64i8_to_v32i8_1(<64 x i8>* %L, <32 x i8>* %S) nounwind {
 ; AVX512F-LABEL: shuffle_v64i8_to_v32i8_1:
@@ -17,41 +20,29 @@
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuffle_v64i8_to_v32i8_1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX512VL-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]
-; AVX512VL-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]
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,5,7]
-; AVX512VL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa %ymm2, (%rsi)
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512BWVL-FAST-ALL-LABEL: shuffle_v64i8_to_v32i8_1:
+; AVX512BWVL-FAST-ALL:       # %bb.0:
+; AVX512BWVL-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512BWVL-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX512BWVL-FAST-ALL-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-FAST-ALL-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-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,5,7]
+; AVX512BWVL-FAST-ALL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
+; AVX512BWVL-FAST-ALL-NEXT:    vmovdqa %ymm2, (%rsi)
+; AVX512BWVL-FAST-ALL-NEXT:    vzeroupper
+; AVX512BWVL-FAST-ALL-NEXT:    retq
 ;
-; AVX512BW-LABEL: shuffle_v64i8_to_v32i8_1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm0 = [0,2,9,11]
-; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm1
-; AVX512BW-NEXT:    vmovdqa 32(%rdi), %ymm2
-; AVX512BW-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[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 {{.*#+}} ymm1 = ymm1[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:    vpermt2q %zmm2, %zmm0, %zmm1
-; AVX512BW-NEXT:    vmovdqa %ymm1, (%rsi)
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512BWVL-LABEL: shuffle_v64i8_to_v32i8_1:
-; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vmovdqa 32(%rdi), %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:    vmovdqa {{.*#+}} ymm2 = [0,2,5,7]
-; AVX512BWVL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
-; AVX512BWVL-NEXT:    vmovdqa %ymm2, (%rsi)
-; AVX512BWVL-NEXT:    vzeroupper
-; AVX512BWVL-NEXT:    retq
+; AVX512BWVL-FAST-PERLANE-LABEL: shuffle_v64i8_to_v32i8_1:
+; AVX512BWVL-FAST-PERLANE:       # %bb.0:
+; AVX512BWVL-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512BWVL-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX512BWVL-FAST-PERLANE-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-FAST-PERLANE-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-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
+; AVX512BWVL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512BWVL-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%rsi)
+; AVX512BWVL-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512BWVL-FAST-PERLANE-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
@@ -71,18 +62,6 @@
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuffle_v32i16_to_v16i16_1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX512VL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,2,3,6,7,10,11,14,15,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31]
-; AVX512VL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,6,7,10,11,14,15,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u]
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,5,7]
-; AVX512VL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa %ymm2, (%rsi)
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
-;
 ; AVX512BW-LABEL: shuffle_v32i16_to_v16i16_1:
 ; AVX512BW:       # %bb.0:
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm0 = [1,3,5,7,33,35,37,39,9,11,13,15,41,43,45,47]
@@ -118,33 +97,23 @@
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuffle_v16i32_to_v8i32_1:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,5,7,9,11,13,15]
-; AVX512VL-NEXT:    vpermi2d 32(%rdi), %ymm0, %ymm1
-; AVX512VL-NEXT:    vmovdqa %ymm1, (%rsi)
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512BWVL-FAST-ALL-LABEL: shuffle_v16i32_to_v8i32_1:
+; AVX512BWVL-FAST-ALL:       # %bb.0:
+; AVX512BWVL-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512BWVL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,5,7,9,11,13,15]
+; AVX512BWVL-FAST-ALL-NEXT:    vpermi2d 32(%rdi), %ymm0, %ymm1
+; AVX512BWVL-FAST-ALL-NEXT:    vmovdqa %ymm1, (%rsi)
+; AVX512BWVL-FAST-ALL-NEXT:    vzeroupper
+; AVX512BWVL-FAST-ALL-NEXT:    retq
 ;
-; AVX512BW-LABEL: shuffle_v16i32_to_v8i32_1:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa {{.*#+}} ymm0 = [1,3,5,7,17,19,21,23]
-; AVX512BW-NEXT:    vmovdqa (%rdi), %ymm1
-; AVX512BW-NEXT:    vmovdqa 32(%rdi), %ymm2
-; AVX512BW-NEXT:    vpermt2d %zmm2, %zmm0, %zmm1
-; AVX512BW-NEXT:    vmovdqa %ymm1, (%rsi)
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512BWVL-LABEL: shuffle_v16i32_to_v8i32_1:
-; AVX512BWVL:       # %bb.0:
-; AVX512BWVL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512BWVL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,5,7,9,11,13,15]
-; AVX512BWVL-NEXT:    vpermi2d 32(%rdi), %ymm0, %ymm1
-; AVX512BWVL-NEXT:    vmovdqa %ymm1, (%rsi)
-; AVX512BWVL-NEXT:    vzeroupper
-; AVX512BWVL-NEXT:    retq
+; AVX512BWVL-FAST-PERLANE-LABEL: shuffle_v16i32_to_v8i32_1:
+; AVX512BWVL-FAST-PERLANE:       # %bb.0:
+; AVX512BWVL-FAST-PERLANE-NEXT:    vmovaps (%rdi), %ymm0
+; AVX512BWVL-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],mem[1,3],ymm0[5,7],mem[5,7]
+; AVX512BWVL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512BWVL-FAST-PERLANE-NEXT:    vmovaps %ymm0, (%rsi)
+; AVX512BWVL-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512BWVL-FAST-PERLANE-NEXT:    retq
   %vec = load <16 x i32>, <16 x i32>* %L
   %strided.vec = shufflevector <16 x i32> %vec, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   store <8 x i32> %strided.vec, <8 x i32>* %S
diff --git a/llvm/test/CodeGen/X86/shuffle-vs-trunc-128.ll b/llvm/test/CodeGen/X86/shuffle-vs-trunc-128.ll
--- a/llvm/test/CodeGen/X86/shuffle-vs-trunc-128.ll
+++ b/llvm/test/CodeGen/X86/shuffle-vs-trunc-128.ll
@@ -3,12 +3,17 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 
 ; PR31551
 ; Pairs of shufflevector:trunc functions with functional equivalence.
diff --git a/llvm/test/CodeGen/X86/shuffle-vs-trunc-256.ll b/llvm/test/CodeGen/X86/shuffle-vs-trunc-256.ll
--- a/llvm/test/CodeGen/X86/shuffle-vs-trunc-256.ll
+++ b/llvm/test/CodeGen/X86/shuffle-vs-trunc-256.ll
@@ -1,12 +1,17 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512BWVL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512VBMIVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512VBMIVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512VBMIVL
 
 ; PR31551
 ; Pairs of shufflevector:trunc functions with functional equivalence.
@@ -309,13 +314,20 @@
 ; AVX2-SLOW-NEXT:    vmovaps %xmm0, (%rsi)
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_v4i64_to_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm0 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps (%rdi), %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovaps %xmm0, (%rsi)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_v4i64_to_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm0 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps (%rdi), %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps %xmm0, (%rsi)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_v4i64_to_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovaps (%rdi), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],mem[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %xmm0, (%rsi)
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v4i64_to_v4i32:
 ; AVX512F:       # %bb.0:
@@ -790,13 +802,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v4i64_to_v4i16_with_zext_return_v8i16:
 ; AVX512F:       # %bb.0:
@@ -853,13 +873,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_v4i64_to_v4i16_via_v4i32_return_v8i16:
 ; AVX512F:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll b/llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll
--- a/llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll
+++ b/llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll
@@ -1,10 +1,15 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMI
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMIVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL,AVX512VL-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL,AVX512VL-FAST-PERLANE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMIVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VBMIVL
 
 ; PR31551
 ; Pairs of shufflevector:trunc functions with functional equivalence.
@@ -23,17 +28,29 @@
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuffle_v64i8_to_v32i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX512VL-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX512VL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u,16,18,20,22,24,26,28,30]
-; AVX512VL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u,16,18,20,22,24,26,28,30,u,u,u,u,u,u,u,u]
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,5,7]
-; AVX512VL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
-; AVX512VL-NEXT:    vmovdqa %ymm2, (%rsi)
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v64i8_to_v32i8:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX512VL-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u,16,18,20,22,24,26,28,30]
+; AVX512VL-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u,16,18,20,22,24,26,28,30,u,u,u,u,u,u,u,u]
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,5,7]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa %ymm2, (%rsi)
+; AVX512VL-FAST-ALL-NEXT:    vzeroupper
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v64i8_to_v32i8:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u,16,18,20,22,24,26,28,30]
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u,16,18,20,22,24,26,28,30,u,u,u,u,u,u,u,u]
+; AVX512VL-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%rsi)
+; AVX512VL-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512VL-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_to_v32i8:
 ; AVX512BW:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/ssub_sat_vec.ll b/llvm/test/CodeGen/X86/ssub_sat_vec.ll
--- a/llvm/test/CodeGen/X86/ssub_sat_vec.ll
+++ b/llvm/test/CodeGen/X86/ssub_sat_vec.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
 
 declare <1 x i8> @llvm.ssub.sat.v1i8(<1 x i8>, <1 x i8>)
 declare <2 x i8> @llvm.ssub.sat.v2i8(<2 x i8>, <2 x i8>)
diff --git a/llvm/test/CodeGen/X86/uadd_sat_vec.ll b/llvm/test/CodeGen/X86/uadd_sat_vec.ll
--- a/llvm/test/CodeGen/X86/uadd_sat_vec.ll
+++ b/llvm/test/CodeGen/X86/uadd_sat_vec.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
 
 declare <1 x i8> @llvm.uadd.sat.v1i8(<1 x i8>, <1 x i8>)
 declare <2 x i8> @llvm.uadd.sat.v2i8(<2 x i8>, <2 x i8>)
diff --git a/llvm/test/CodeGen/X86/usub_sat_vec.ll b/llvm/test/CodeGen/X86/usub_sat_vec.ll
--- a/llvm/test/CodeGen/X86/usub_sat_vec.ll
+++ b/llvm/test/CodeGen/X86/usub_sat_vec.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
 
 declare <1 x i8> @llvm.usub.sat.v1i8(<1 x i8>, <1 x i8>)
 declare <2 x i8> @llvm.usub.sat.v2i8(<2 x i8>, <2 x i8>)
diff --git a/llvm/test/CodeGen/X86/vec_saddo.ll b/llvm/test/CodeGen/X86/vec_saddo.ll
--- a/llvm/test/CodeGen/X86/vec_saddo.ll
+++ b/llvm/test/CodeGen/X86/vec_saddo.ll
@@ -4,7 +4,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
 
 declare {<1 x i32>, <1 x i1>} @llvm.sadd.with.overflow.v1i32(<1 x i32>, <1 x i32>)
 declare {<2 x i32>, <2 x i1>} @llvm.sadd.with.overflow.v2i32(<2 x i32>, <2 x i32>)
diff --git a/llvm/test/CodeGen/X86/vec_smulo.ll b/llvm/test/CodeGen/X86/vec_smulo.ll
--- a/llvm/test/CodeGen/X86/vec_smulo.ll
+++ b/llvm/test/CodeGen/X86/vec_smulo.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
 
 declare {<1 x i32>, <1 x i1>} @llvm.smul.with.overflow.v1i32(<1 x i32>, <1 x i32>)
 declare {<2 x i32>, <2 x i1>} @llvm.smul.with.overflow.v2i32(<2 x i32>, <2 x i32>)
diff --git a/llvm/test/CodeGen/X86/vec_ssubo.ll b/llvm/test/CodeGen/X86/vec_ssubo.ll
--- a/llvm/test/CodeGen/X86/vec_ssubo.ll
+++ b/llvm/test/CodeGen/X86/vec_ssubo.ll
@@ -4,7 +4,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
 
 declare {<1 x i32>, <1 x i1>} @llvm.ssub.with.overflow.v1i32(<1 x i32>, <1 x i32>)
 declare {<2 x i32>, <2 x i1>} @llvm.ssub.with.overflow.v2i32(<2 x i32>, <2 x i32>)
diff --git a/llvm/test/CodeGen/X86/vec_uaddo.ll b/llvm/test/CodeGen/X86/vec_uaddo.ll
--- a/llvm/test/CodeGen/X86/vec_uaddo.ll
+++ b/llvm/test/CodeGen/X86/vec_uaddo.ll
@@ -4,7 +4,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
 
 declare {<1 x i32>, <1 x i1>} @llvm.uadd.with.overflow.v1i32(<1 x i32>, <1 x i32>)
 declare {<2 x i32>, <2 x i1>} @llvm.uadd.with.overflow.v2i32(<2 x i32>, <2 x i32>)
diff --git a/llvm/test/CodeGen/X86/vec_umulo.ll b/llvm/test/CodeGen/X86/vec_umulo.ll
--- a/llvm/test/CodeGen/X86/vec_umulo.ll
+++ b/llvm/test/CodeGen/X86/vec_umulo.ll
@@ -4,8 +4,10 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
 
 declare {<1 x i32>, <1 x i1>} @llvm.umul.with.overflow.v1i32(<1 x i32>, <1 x i32>)
 declare {<2 x i32>, <2 x i1>} @llvm.umul.with.overflow.v2i32(<2 x i32>, <2 x i32>)
diff --git a/llvm/test/CodeGen/X86/vec_usubo.ll b/llvm/test/CodeGen/X86/vec_usubo.ll
--- a/llvm/test/CodeGen/X86/vec_usubo.ll
+++ b/llvm/test/CodeGen/X86/vec_usubo.ll
@@ -4,7 +4,8 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX512
 
 declare {<1 x i32>, <1 x i1>} @llvm.usub.with.overflow.v1i32(<1 x i32>, <1 x i32>)
 declare {<2 x i32>, <2 x i1>} @llvm.usub.with.overflow.v2i32(<2 x i32>, <2 x i32>)
diff --git a/llvm/test/CodeGen/X86/vector-half-conversions.ll b/llvm/test/CodeGen/X86/vector-half-conversions.ll
--- a/llvm/test/CodeGen/X86/vector-half-conversions.ll
+++ b/llvm/test/CodeGen/X86/vector-half-conversions.ll
@@ -1,9 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX1
 ; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
-; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c,+fast-variable-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
 ; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512
-; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512
 
 ;
 ; Half to Float
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-2.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-2.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-2.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-2.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2,AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
 
 ; These patterns are produced by LoopVectorizer for interleaved loads.
 
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-3.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-3.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-3.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-3.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2,AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
 
 ; These patterns are produced by LoopVectorizer for interleaved loads.
 
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-4.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-4.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-4.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-4.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2,AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST,AVX2-FAST-ALL %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST,AVX2-FAST-PERLANE %s
 
 ; These patterns are produced by LoopVectorizer for interleaved loads.
 
@@ -322,80 +323,180 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf16:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa 64(%rdi), %ymm8
-; AVX2-FAST-NEXT:    vmovdqa 96(%rdi), %ymm13
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm14
-; AVX2-FAST-NEXT:    vmovdqa 16(%rdi), %xmm1
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %xmm3
-; AVX2-FAST-NEXT:    vmovdqa 48(%rdi), %xmm5
-; AVX2-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm7 = xmm5[0],xmm2[1,2,3],xmm5[4],xmm2[5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm4 = xmm3[0],xmm2[1,2,3],xmm3[4],xmm2[5,6,7]
-; AVX2-FAST-NEXT:    vpackusdw %xmm7, %xmm4, %xmm4
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm7 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm2 = xmm14[0],xmm2[1,2,3],xmm14[4],xmm2[5,6,7]
-; AVX2-FAST-NEXT:    vpackusdw %xmm7, %xmm2, %xmm2
-; AVX2-FAST-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,2,2,3,0,2,4,6]
-; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm4, %ymm7
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm10 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
-; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm4, %ymm12
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm10 = ymm12[0,1,2,3,4,5],ymm10[6,7]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm10[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,2,3,10,11,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm5, %xmm6
-; AVX2-FAST-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm6[0],xmm2[1],xmm6[1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm6 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT:    vpshufb %xmm6, %xmm1, %xmm0
-; AVX2-FAST-NEXT:    vpshufb %xmm6, %xmm14, %xmm6
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31>
-; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm7, %ymm6
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm12 = <2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm9 = ymm0[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [1,3,2,3,1,3,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm13, %ymm0, %ymm4
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
-; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm0, %ymm8
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm6[6,7]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm5[0,1,2,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm3[0,1,2,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm7[0],xmm2[0],xmm7[1],xmm2[1]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm1[2,0,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm6 = xmm14[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm6[2,0,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm7[0],xmm2[1],xmm7[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm11[2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vpshufb %ymm12, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm5[0,1,3,1,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[0,1,3,1,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm6[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa %ymm10, (%rsi)
-; AVX2-FAST-NEXT:    vmovdqa %ymm9, (%rdx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm2, (%rcx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%r8)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf16:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa 64(%rdi), %ymm8
+; AVX2-FAST-ALL-NEXT:    vmovdqa 96(%rdi), %ymm13
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm14
+; AVX2-FAST-ALL-NEXT:    vmovdqa 16(%rdi), %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %xmm3
+; AVX2-FAST-ALL-NEXT:    vmovdqa 48(%rdi), %xmm5
+; AVX2-FAST-ALL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm7 = xmm5[0],xmm2[1,2,3],xmm5[4],xmm2[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm4 = xmm3[0],xmm2[1,2,3],xmm3[4],xmm2[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm7, %xmm4, %xmm4
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm7 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm2 = xmm14[0],xmm2[1,2,3],xmm14[4],xmm2[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm7, %xmm2, %xmm2
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm4, %xmm2, %xmm2
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm4 = [0,2,2,3,0,2,4,6]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm13, %ymm4, %ymm7
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm10 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm8, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm11 = <0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm4, %ymm12
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm10 = ymm12[0,1,2,3,4,5],ymm10[6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm10 = ymm2[0,1,2,3],ymm10[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,2,3,2,3,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm2, %xmm5, %xmm6
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm2, %xmm3, %xmm2
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm6[0],xmm2[1],xmm6[1]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm6 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm6, %xmm1, %xmm0
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm6, %xmm14, %xmm6
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm15, %ymm7, %ymm6
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm12 = <2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm12, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm9 = ymm0[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm0 = [1,3,2,3,1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm13, %ymm0, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm8, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm0, %ymm8
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm6[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm5[0,1,2,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm3[0,1,2,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm7[0],xmm2[0],xmm7[1],xmm2[1]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm1[2,0,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm6 = xmm14[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm6[2,0,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm2[0],xmm7[0],xmm2[1],xmm7[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm2 = xmm2[0,1],xmm11[2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm15, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm12, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm5[0,1,3,1,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[0,1,3,1,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm6[3,1,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm10, (%rsi)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm9, (%rdx)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm2, (%rcx)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf16:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm10
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 16(%rdi), %xmm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %xmm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 48(%rdi), %xmm13
+; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm13[0],xmm0[1,2,3],xmm13[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm5 = xmm12[0],xmm0[1,2,3],xmm12[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm5, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm5 = xmm11[0],xmm0[1,2,3],xmm11[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm6 = xmm10[0],xmm0[1,2,3],xmm10[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm5, %xmm6, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm5, %xmm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 112(%rdi), %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm6[0],xmm0[1,2,3],xmm6[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rdi), %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm5 = xmm7[0],xmm0[1,2,3],xmm7[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm5, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 80(%rdi), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0],xmm0[1,2,3],xmm2[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rdi), %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0],xmm0[1,2,3],xmm5[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm0, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm8[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,2,3,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm6, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm7, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm3 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm2, %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm5, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm9[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm13, %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm12, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm4[0],xmm0[1],xmm4[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm11, %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm3, %xmm10, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm3[0],xmm4[0],xmm3[1],xmm4[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm0 = xmm3[0,1],xmm0[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm9 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm15 = xmm6[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm15[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm8 = xmm7[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm8[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm2[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm5[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm6[0],xmm4[0],xmm6[1],xmm4[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm14 = ymm4[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm13 = xmm13[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm13[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm7 = xmm12[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm7[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm6 = xmm11[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm6[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm4 = xmm10[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm4[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm14[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm15[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm8[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm5[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm13[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm7[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm6[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, (%rsi)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm9, (%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, (%r8)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %wide.vec = load <64 x i16>, <64 x i16>* %in.vec, align 32
 
   %strided.vec0 = shufflevector <64 x i16> %wide.vec, <64 x i16> poison, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60>
@@ -656,180 +757,415 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    subq $216, %rsp
-; AVX2-FAST-NEXT:    vmovdqa 64(%rdi), %ymm5
-; AVX2-FAST-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 96(%rdi), %ymm6
-; AVX2-FAST-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm2
-; AVX2-FAST-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 16(%rdi), %xmm11
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %xmm12
-; AVX2-FAST-NEXT:    vmovdqa 48(%rdi), %xmm9
-; AVX2-FAST-NEXT:    vpxor %xmm8, %xmm8, %xmm8
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm9[0],xmm8[1,2,3],xmm9[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm12[0],xmm8[1,2,3],xmm12[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vpackusdw %xmm0, %xmm1, %xmm0
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm11[0],xmm8[1,2,3],xmm11[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm4 = xmm2[0],xmm8[1,2,3],xmm2[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vpackusdw %xmm1, %xmm4, %xmm1
-; AVX2-FAST-NEXT:    vpackusdw %xmm0, %xmm1, %xmm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = [0,2,2,3,0,2,4,6]
-; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm15, %ymm2
-; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29>
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, %ymm3
-; AVX2-FAST-NEXT:    vpermd %ymm5, %ymm15, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = <0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %ymm0, %ymm2, %ymm5
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, %ymm14
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6,7]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm4[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 176(%rdi), %xmm13
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm5 = xmm13[0],xmm8[1,2,3],xmm13[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm13, (%rsp) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 160(%rdi), %xmm0
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm7 = xmm0[0],xmm8[1,2,3],xmm0[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vpackusdw %xmm5, %xmm7, %xmm10
-; AVX2-FAST-NEXT:    vmovdqa 144(%rdi), %xmm5
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0],xmm8[1,2,3],xmm5[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 128(%rdi), %xmm4
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm6 = xmm4[0],xmm8[1,2,3],xmm4[4],xmm8[5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
-; AVX2-FAST-NEXT:    vpackusdw %xmm0, %xmm6, %xmm0
-; AVX2-FAST-NEXT:    vpackusdw %xmm10, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa 192(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa 224(%rdi), %ymm10
-; AVX2-FAST-NEXT:    vpermd %ymm10, %ymm15, %ymm6
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm15, %ymm1
-; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm6, %ymm15
-; AVX2-FAST-NEXT:    vpshufb %ymm14, %ymm1, %ymm14
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm14[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm14 = [0,1,2,3,2,3,10,11,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT:    vpshufb %xmm14, %xmm9, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa %xmm12, %xmm7
-; AVX2-FAST-NEXT:    vpshufb %xmm14, %xmm12, %xmm3
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm15 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm3 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
-; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm11, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm8, %xmm12
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm12[0],xmm0[0],xmm12[1],xmm0[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm15[2,3]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31>
-; AVX2-FAST-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm12, %ymm12
-; AVX2-FAST-NEXT:    vmovdqa %ymm11, %ymm15
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm2, %ymm2
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm12[6,7]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpshufb %xmm14, %xmm13, %xmm0
-; AVX2-FAST-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
-; AVX2-FAST-NEXT:    vpshufb %xmm14, %xmm13, %xmm2
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
-; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm5, %xmm2
-; AVX2-FAST-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3]
-; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm6, %ymm2
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,2,3,1,3,5,7]
-; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm14 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %ymm2, %ymm14, %ymm2
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm11 = xmm9[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm9 = xmm7[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm11[0,1,2,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm9[0,1,2,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
-; AVX2-FAST-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm12 = mem[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm8[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm12[2,0,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm4[2,0,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm5[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpermd %ymm10, %ymm1, %ymm2
-; AVX2-FAST-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm10 # 32-byte Folded Reload
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm10[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm5[6,7]
-; AVX2-FAST-NEXT:    vpshufd $231, (%rsp), %xmm3 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm3 = mem[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm13 = xmm13[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm3[0,1,2,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm13[0,1,2,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm7[0],xmm1[0],xmm7[1],xmm1[1]
-; AVX2-FAST-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm7 = mem[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
-; AVX2-FAST-NEXT:    # xmm5 = mem[3,1,2,3]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm7[2,0,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm5[2,0,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm8[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31>
-; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm6, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm14[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm11[0,1,3,1,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm9[0,1,3,1,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm12[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-NEXT:    vpshufb %ymm15, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm10[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm3[0,1,3,1,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm13[0,1,3,1,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm7[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm5[3,1,2,3,4,5,6,7]
-; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, 32(%rsi)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, (%rsi)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, 32(%rdx)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, (%rdx)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, 32(%rcx)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm2, (%rcx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, 32(%r8)
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%r8)
-; AVX2-FAST-NEXT:    addq $216, %rsp
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    subq $216, %rsp
+; AVX2-FAST-ALL-NEXT:    vmovdqa 64(%rdi), %ymm5
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm5, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa 96(%rdi), %ymm6
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm6, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm2
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa 16(%rdi), %xmm11
+; AVX2-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %xmm12
+; AVX2-FAST-ALL-NEXT:    vmovdqa 48(%rdi), %xmm9
+; AVX2-FAST-ALL-NEXT:    vpxor %xmm8, %xmm8, %xmm8
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm9[0],xmm8[1,2,3],xmm9[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm12[0],xmm8[1,2,3],xmm12[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm0, %xmm1, %xmm0
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm11[0],xmm8[1,2,3],xmm11[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm4 = xmm2[0],xmm8[1,2,3],xmm2[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm1, %xmm4, %xmm1
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm0, %xmm1, %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm15 = [0,2,2,3,0,2,4,6]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm6, %ymm15, %ymm2
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm0 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm0, %ymm2, %ymm4
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm5, %ymm15, %ymm2
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm0 = <0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm0, %ymm2, %ymm5
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, %ymm14
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0,1,2,3,4,5],ymm4[6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm4[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa 176(%rdi), %xmm13
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm5 = xmm13[0],xmm8[1,2,3],xmm13[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm13, (%rsp) # 16-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa 160(%rdi), %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm7 = xmm0[0],xmm8[1,2,3],xmm0[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm5, %xmm7, %xmm10
+; AVX2-FAST-ALL-NEXT:    vmovdqa 144(%rdi), %xmm5
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0],xmm8[1,2,3],xmm5[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa 128(%rdi), %xmm4
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm6 = xmm4[0],xmm8[1,2,3],xmm4[4],xmm8[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm0, %xmm6, %xmm0
+; AVX2-FAST-ALL-NEXT:    vpackusdw %xmm10, %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa 192(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa 224(%rdi), %ymm10
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm10, %ymm15, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm15, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm3, %ymm6, %ymm15
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm14, %ymm1, %ymm14
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm14 = ymm14[0,1,2,3,4,5],ymm15[6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm14[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm14 = [0,1,2,3,2,3,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm14, %xmm9, %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm12, %xmm7
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm14, %xmm12, %xmm3
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm15 = xmm3[0],xmm0[0],xmm3[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm3 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm3, %xmm11, %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm3, %xmm8, %xmm12
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm12[0],xmm0[0],xmm12[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm15[2,3]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31>
+; AVX2-FAST-ALL-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm12 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm12, %ymm12
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm11, %ymm15
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm11 = <2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm2, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm12[6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm14, %xmm13, %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm14, %xmm13, %xmm2
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm3, %xmm5, %xmm2
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm3, %xmm4, %xmm3
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm15, %ymm6, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm1, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,2,3,1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm6 # 32-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm14 # 32-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm2, %ymm14, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm5 = ymm2[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm11 = xmm9[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm9 = xmm7[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm11[0,1,2,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm9[0,1,2,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX2-FAST-ALL-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    # xmm12 = mem[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm4 = xmm8[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm12[2,0,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm4[2,0,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm5[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm10, %ymm1, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpermd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm10 # 32-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm5 = ymm2[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm8 = ymm10[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3,4,5],ymm5[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd $231, (%rsp), %xmm3 # 16-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    # xmm3 = mem[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm13 = xmm13[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm3[0,1,2,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm13[0,1,2,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm7[0],xmm1[0],xmm7[1],xmm1[1]
+; AVX2-FAST-ALL-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    # xmm7 = mem[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Folded Reload
+; AVX2-FAST-ALL-NEXT:    # xmm5 = mem[3,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm7[2,0,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm5[2,0,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm8[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm15 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,22,23,26,27,30,31>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm15, %ymm6, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm8 = ymm14[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm8[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm11[0,1,3,1,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm9[0,1,3,1,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm12[3,1,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm6[0],xmm4[1],xmm6[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm15, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm2 = ymm10[2,3,6,7,10,11,14,15,14,15,10,11,12,13,14,15,18,19,22,23,26,27,30,31,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm3[0,1,3,1,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm13[0,1,3,1,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm7[3,1,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm5[3,1,2,3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, 32(%rsi)
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, (%rsi)
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, 32(%rdx)
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, (%rdx)
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, 32(%rcx)
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm2, (%rcx)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm1, 32(%r8)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-FAST-ALL-NEXT:    addq $216, %rsp
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    subq $280, %rsp # imm = 0x118
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 16(%rdi), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 48(%rdi), %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm0, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0],xmm0[1,2,3],xmm3[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm3, %xmm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm3, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm3 = xmm5[0],xmm0[1,2,3],xmm5[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm5, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm3, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm3 = xmm1[0],xmm0[1,2,3],xmm1[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, %xmm9
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm4 = xmm6[0],xmm0[1,2,3],xmm6[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm3, %xmm4, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm3, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 112(%rdi), %xmm15
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm3 = xmm15[0],xmm0[1,2,3],xmm15[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rdi), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm4 = xmm1[0],xmm0[1,2,3],xmm1[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm3, %xmm4, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm3, %xmm3, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm3, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 80(%rdi), %xmm12
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm7 = xmm12[0],xmm0[1,2,3],xmm12[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rdi), %xmm14
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm14[0],xmm0[1,2,3],xmm14[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm7, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm1, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 176(%rdi), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0],xmm0[1,2,3],xmm1[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 160(%rdi), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0],xmm0[1,2,3],xmm2[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm1, %xmm2, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 144(%rdi), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0],xmm0[1,2,3],xmm2[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 128(%rdi), %xmm10
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm3 = xmm10[0],xmm0[1,2,3],xmm10[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm3, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm1, %xmm2, %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 240(%rdi), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm1[0],xmm0[1,2,3],xmm1[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 224(%rdi), %xmm13
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm3 = xmm13[0],xmm0[1,2,3],xmm13[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm13, (%rsp) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm2, %xmm3, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 208(%rdi), %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm3[0],xmm0[1,2,3],xmm3[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 192(%rdi), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1,2,3],xmm2[4],xmm0[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm7, %xmm7, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpackusdw %xmm0, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,2,3,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm15, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm6, %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm4 = [2,3,10,11,8,9,10,11,8,9,10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm12, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm14, %xmm11
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm7 = xmm11[0],xmm7[0],xmm11[1],xmm7[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm7[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm8, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm5, %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm9, %xmm8
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm9, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm9, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm5[0],xmm7[0],xmm5[1],xmm7[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm5 = xmm5[0,1],xmm6[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm7, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm13, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm5[0],xmm1[0],xmm5[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm3, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm2, %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm5 = xmm6[0],xmm5[0],xmm6[1],xmm5[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm5[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm6, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm13, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm5 # 16-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm5, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm4, %xmm10, %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm0 = xmm4[0,1],xmm0[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm0 = xmm15[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm2 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm2[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm1 = xmm12[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm4 = xmm14[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm4, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm4[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm10 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm14 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm10[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm14[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm12 = xmm4[0],xmm1[0],xmm4[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm5 = xmm8[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm4 = xmm9[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm5[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm15 = xmm4[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm15[0],xmm0[0],xmm15[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm12[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm11[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm8 = xmm7[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, (%rsp), %xmm9 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm9 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm8[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm9[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm11 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm2 = xmm3[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm3 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm3 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm2[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm3[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm7[0],xmm0[0],xmm7[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm11, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm7 = ymm0[0,1,2,3,4,5],ymm7[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm11 = xmm6[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm15 = xmm13[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm11[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm15[0,1,2,0,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm12 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm1 = mem[3,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm12[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm13 = xmm1[2,0,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm13[0],xmm0[0],xmm13[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm6[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm13 = ymm0[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw $116, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm0 = mem[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw $116, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm6 = mem[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm6[0],xmm0[0],xmm6[1],xmm0[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm6 = mem[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw $231, {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Folded Reload
+; AVX2-FAST-PERLANE-NEXT:    # xmm7 = mem[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm6 = xmm10[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm7 = xmm14[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm6 = xmm7[0],xmm6[0],xmm7[1],xmm6[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm5[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm4[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm4 = xmm4[0,1],xmm6[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm4[0,1,2,3],ymm0[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm8[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm5 = xmm9[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm3[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm4, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3,4,5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm3 = xmm11[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm15[0,1,3,1,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm4 = xmm12[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[3,1,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm1 = xmm1[0,1],xmm3[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3],ymm2[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, 32(%rsi)
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, (%rsi)
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, 32(%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, (%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm13, 32(%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm2, (%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 32(%r8)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-FAST-PERLANE-NEXT:    addq $280, %rsp # imm = 0x118
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %wide.vec = load <128 x i16>, <128 x i16>* %in.vec, align 32
 
   %strided.vec0 = shufflevector <128 x i16> %wide.vec, <128 x i16> poison, <32 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 32, i32 36, i32 40, i32 44, i32 48, i32 52, i32 56, i32 60, i32 64, i32 68, i32 72, i32 76, i32 80, i32 84, i32 88, i32 92, i32 96, i32 100, i32 104, i32 108, i32 112, i32 116, i32 120, i32 124>
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-5.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-ALL %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-PERLANE %s
 
 ; These patterns are produced by LoopVectorizer for interleaved loads.
 
@@ -329,94 +330,181 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf16:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa 128(%rdi), %ymm0
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm3
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm4
-; AVX2-FAST-NEXT:    vmovdqa 64(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vmovdqa 96(%rdi), %ymm2
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm5 = ymm3[0],ymm4[1],ymm3[2,3],ymm4[4],ymm3[5],ymm4[6],ymm3[7,8],ymm4[9],ymm3[10,11],ymm4[12],ymm3[13],ymm4[14],ymm3[15]
-; AVX2-FAST-NEXT:    vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0],xmm6[1,2,3],xmm5[4,5],xmm6[6,7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[0,1,10,11,4,5,14,15,8,9,2,3,12,13,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm1[0],ymm2[1,2],ymm1[3],ymm2[4],ymm1[5],ymm2[6,7],ymm1[8],ymm2[9,10],ymm1[11],ymm2[12],ymm1[13],ymm2[14,15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [1,3,0,2,4,6,1,3]
-; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm7, %ymm6
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[0,1,6,7,8,9,14,15,4,5,14,15,4,5,2,3,16,17,22,23,24,25,30,31,20,21,30,31,20,21,18,19]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm8 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
-; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm5, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,0,3,5,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm6, %ymm6
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,26,27>
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm6, %ymm6
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm9 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm4[0],ymm3[1],ymm4[2],ymm3[3],ymm4[4,5],ymm3[6],ymm4[7,8],ymm3[9],ymm4[10],ymm3[11],ymm4[12,13],ymm3[14],ymm4[15]
-; AVX2-FAST-NEXT:    vextracti128 $1, %ymm6, %xmm5
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3],xmm6[4,5,6],xmm5[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,3,12,13,6,7,0,1,10,11,4,5,14,15,10,11]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5],ymm1[6],ymm2[7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13],ymm1[14],ymm2[15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <2,u,u,u,4,7,1,6>
-; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm10, %ymm6
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[2,3,4,5,10,11,0,1,14,15,2,3,12,13,0,1,18,19,20,21,26,27,16,17,30,31,18,19,28,29,16,17]
-; AVX2-FAST-NEXT:    vpblendvb %ymm8, %ymm5, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,1,3,6,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm6, %ymm6
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25>
-; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm5 = ymm4[0,1],ymm3[2],ymm4[3],ymm3[4],ymm4[5,6],ymm3[7],ymm4[8,9],ymm3[10],ymm4[11],ymm3[12],ymm4[13,14],ymm3[15]
-; AVX2-FAST-NEXT:    vextracti128 $1, %ymm5, %xmm7
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2],xmm7[3,4],xmm5[5,6,7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[4,5,14,15,8,9,2,3,12,13,6,7,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm7 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5],ymm2[6],ymm1[7,8],ymm2[9],ymm1[10,11],ymm2[12],ymm1[13],ymm2[14],ymm1[15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,2,u,u,5,7,2,4>
-; AVX2-FAST-NEXT:    vpermd %ymm7, %ymm10, %ymm7
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,0,1,6,7,16,17,22,23,24,25,30,31,20,21,22,23,16,17,22,23]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm7[3,4,5,6,7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [1,4,6,0,1,4,6,0]
-; AVX2-FAST-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm7, %ymm7
-; AVX2-FAST-NEXT:    vpshufb %ymm11, %ymm7, %ymm7
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm7 = ymm5[0,1,2,3,4],ymm7[5,6,7],ymm5[8,9,10,11,12],ymm7[13,14,15]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm7 = ymm5[0,1,2,3],ymm7[4,5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm5 = ymm3[0],ymm4[1,2],ymm3[3],ymm4[4],ymm3[5],ymm4[6,7],ymm3[8],ymm4[9,10],ymm3[11],ymm4[12],ymm3[13],ymm4[14,15]
-; AVX2-FAST-NEXT:    vextracti128 $1, %ymm5, %xmm6
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0],xmm5[1],xmm6[2],xmm5[3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[6,7,0,1,10,11,4,5,14,15,8,9,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7,8],ymm1[9],ymm2[10],ymm1[11],ymm2[12,13],ymm1[14],ymm2[15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,3,u,u,5,0,2,7>
-; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm10, %ymm6
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,2,3,4,5,18,19,20,21,26,27,16,17,30,31,30,31,18,19,20,21]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3,4,5,6,7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [2,4,7,0,2,4,7,0]
-; AVX2-FAST-NEXT:    # ymm6 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm6, %ymm6
-; AVX2-FAST-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm6[4,5,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5],ymm3[6],ymm4[7,8],ymm3[9],ymm4[10,11],ymm3[12],ymm4[13],ymm3[14],ymm4[15]
-; AVX2-FAST-NEXT:    vextracti128 $1, %ymm3, %xmm4
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2],xmm3[3,4],xmm4[5,6,7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[8,9,2,3,12,13,6,7,0,1,10,11,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3],ymm1[4],ymm2[5,6],ymm1[7],ymm2[8,9],ymm1[10],ymm2[11],ymm1[12],ymm2[13,14],ymm1[15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <1,3,u,u,6,0,3,5>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,0,1,6,7,16,17,22,23,24,25,30,31,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5],ymm3[6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,0,2,5,7>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
-; AVX2-FAST-NEXT:    vmovdqa %ymm9, (%rsi)
-; AVX2-FAST-NEXT:    vmovdqa %ymm12, (%rdx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm7, (%rcx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm5, (%r8)
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%r9)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf16:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa 128(%rdi), %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm3
+; AVX2-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %ymm4
+; AVX2-FAST-ALL-NEXT:    vmovdqa 64(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa 96(%rdi), %ymm2
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm5 = ymm3[0],ymm4[1],ymm3[2,3],ymm4[4],ymm3[5],ymm4[6],ymm3[7,8],ymm4[9],ymm3[10,11],ymm4[12],ymm3[13],ymm4[14],ymm3[15]
+; AVX2-FAST-ALL-NEXT:    vextracti128 $1, %ymm5, %xmm6
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0],xmm6[1,2,3],xmm5[4,5],xmm6[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[0,1,10,11,4,5,14,15,8,9,2,3,12,13,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm1[0],ymm2[1,2],ymm1[3],ymm2[4],ymm1[5],ymm2[6,7],ymm1[8],ymm2[9,10],ymm1[11],ymm2[12],ymm1[13],ymm2[14,15]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm7 = [1,3,0,2,4,6,1,3]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm6, %ymm7, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[0,1,6,7,8,9,14,15,4,5,14,15,4,5,2,3,16,17,22,23,24,25,30,31,20,21,30,31,20,21,18,19]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm8 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm6, %ymm5
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,0,3,5,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm6, %ymm6
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm11 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,20,21,26,27>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm6, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm9 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm4[0],ymm3[1],ymm4[2],ymm3[3],ymm4[4,5],ymm3[6],ymm4[7,8],ymm3[9],ymm4[10],ymm3[11],ymm4[12,13],ymm3[14],ymm4[15]
+; AVX2-FAST-ALL-NEXT:    vextracti128 $1, %ymm6, %xmm5
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0,1],xmm5[2,3],xmm6[4,5,6],xmm5[7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[2,3,12,13,6,7,0,1,10,11,4,5,14,15,10,11]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5],ymm1[6],ymm2[7,8],ymm1[9],ymm2[10,11],ymm1[12],ymm2[13],ymm1[14],ymm2[15]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm10 = <2,u,u,u,4,7,1,6>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm6, %ymm10, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[2,3,4,5,10,11,0,1,14,15,2,3,12,13,0,1,18,19,20,21,26,27,16,17,30,31,18,19,28,29,16,17]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm8, %ymm5, %ymm6, %ymm5
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,u,u,1,3,6,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm6, %ymm6
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm8 = <u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm12 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm5 = ymm4[0,1],ymm3[2],ymm4[3],ymm3[4],ymm4[5,6],ymm3[7],ymm4[8,9],ymm3[10],ymm4[11],ymm3[12],ymm4[13,14],ymm3[15]
+; AVX2-FAST-ALL-NEXT:    vextracti128 $1, %ymm5, %xmm7
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm5 = xmm5[0,1,2],xmm7[3,4],xmm5[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[4,5,14,15,8,9,2,3,12,13,6,7,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm7 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5],ymm2[6],ymm1[7,8],ymm2[9],ymm1[10,11],ymm2[12],ymm1[13],ymm2[14],ymm1[15]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,2,u,u,5,7,2,4>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm7, %ymm10, %ymm7
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,0,1,6,7,16,17,22,23,24,25,30,31,20,21,22,23,16,17,22,23]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [1,4,6,0,1,4,6,0]
+; AVX2-FAST-ALL-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm7, %ymm7
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm11, %ymm7, %ymm7
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm7 = ymm5[0,1,2,3,4],ymm7[5,6,7],ymm5[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm7 = ymm5[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm5 = ymm3[0],ymm4[1,2],ymm3[3],ymm4[4],ymm3[5],ymm4[6,7],ymm3[8],ymm4[9,10],ymm3[11],ymm4[12],ymm3[13],ymm4[14,15]
+; AVX2-FAST-ALL-NEXT:    vextracti128 $1, %ymm5, %xmm6
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm5 = xmm6[0],xmm5[1],xmm6[2],xmm5[3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[6,7,0,1,10,11,4,5,14,15,8,9,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4,5],ymm1[6],ymm2[7,8],ymm1[9],ymm2[10],ymm1[11],ymm2[12,13],ymm1[14],ymm2[15]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm10 = <0,3,u,u,5,0,2,7>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm6, %ymm10, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[u,u,u,u,u,u,u,u,u,u,u,u,2,3,4,5,18,19,20,21,26,27,16,17,30,31,30,31,18,19,20,21]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2],ymm6[3,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm6 = [2,4,7,0,2,4,7,0]
+; AVX2-FAST-ALL-NEXT:    # ymm6 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm6, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm8, %ymm6, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2,3,4],ymm6[5,6,7],ymm5[8,9,10,11,12],ymm6[13,14,15]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm6[4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5],ymm3[6],ymm4[7,8],ymm3[9],ymm4[10,11],ymm3[12],ymm4[13],ymm3[14],ymm4[15]
+; AVX2-FAST-ALL-NEXT:    vextracti128 $1, %ymm3, %xmm4
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm3 = xmm4[0,1,2],xmm3[3,4],xmm4[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[8,9,2,3,12,13,6,7,0,1,10,11,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3],ymm1[4],ymm2[5,6],ymm1[7],ymm2[8,9],ymm1[10],ymm2[11],ymm1[12],ymm2[13,14],ymm1[15]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <1,3,u,u,6,0,3,5>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,u,u,u,u,0,1,6,7,16,17,22,23,24,25,30,31,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm3[0,1,2],ymm1[3,4,5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,0,2,5,7>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,16,17,22,23,24,25,30,31]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5],ymm0[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm9, (%rsi)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm12, (%rdx)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm7, (%rcx)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm5, (%r8)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, (%r9)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf16:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 64(%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 96(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm4 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5],ymm3[6],ymm2[7,8],ymm3[9],ymm2[10,11],ymm3[12],ymm2[13],ymm3[14],ymm2[15]
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm4, %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0],xmm5[1,2,3],xmm4[4,5],xmm5[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[0,1,10,11,4,5,14,15,8,9,2,3,12,13,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm5 = ymm0[0],ymm1[1,2],ymm0[3],ymm1[4],ymm0[5],ymm1[6,7],ymm0[8],ymm1[9,10],ymm0[11],ymm1[12],ymm0[13],ymm1[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3,4],ymm6[5],ymm5[6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[0,1,10,11,4,5,14,15,8,9,10,11,4,5,6,7,16,17,26,27,20,21,30,31,24,25,26,27,20,21,22,23]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm8 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,0]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm4, %ymm5, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 144(%rdi), %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 128(%rdi), %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm7 = xmm5[0],xmm4[1],xmm5[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[u,u,u,u,u,u,u,u,u,u,2,3,12,13,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm6[0,1,2,3,4],ymm7[5,6,7],ymm6[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm9 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm3[0],ymm2[1],ymm3[2],ymm2[3],ymm3[4,5],ymm2[6],ymm3[7,8],ymm2[9],ymm3[10],ymm2[11],ymm3[12,13],ymm2[14],ymm3[15]
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm7, %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm6 = xmm7[0,1],xmm6[2,3],xmm7[4,5,6],xmm6[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[2,3,12,13,6,7,0,1,10,11,4,5,14,15,10,11]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm1[0],ymm0[1],ymm1[2,3],ymm0[4],ymm1[5],ymm0[6],ymm1[7,8],ymm0[9],ymm1[10,11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm10 = ymm7[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4],ymm10[5],ymm7[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[2,3,12,13,6,7,0,1,10,11,6,7,8,9,8,9,18,19,28,29,22,23,16,17,26,27,22,23,24,25,24,25]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm8, %ymm6, %ymm7, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm7 = xmm5[0,1],xmm4[2],xmm5[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[u,u,u,u,u,u,u,u,u,u,4,5,14,15,8,9]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm6[0,1,2,3,4],ymm7[5,6,7],ymm6[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm10 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm3[0,1],ymm2[2],ymm3[3],ymm2[4],ymm3[5,6],ymm2[7],ymm3[8,9],ymm2[10],ymm3[11],ymm2[12],ymm3[13,14],ymm2[15]
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm6, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm6 = xmm6[0,1,2],xmm7[3,4],xmm6[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[4,5,14,15,8,9,2,3,12,13,6,7,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5],ymm1[6],ymm0[7,8],ymm1[9],ymm0[10,11],ymm1[12],ymm0[13],ymm1[14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm7[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3,4,5],ymm8[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,0,1,10,11,20,21,30,31,24,25,18,19,28,29,26,27,16,17,26,27]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm7 = xmm4[0],xmm5[1],xmm4[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[u,u,u,u,u,u,u,u,u,u,6,7,0,1,10,11]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm6[0,1,2,3,4],ymm7[5,6,7],ymm6[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm8 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm2[0],ymm3[1,2],ymm2[3],ymm3[4],ymm2[5],ymm3[6,7],ymm2[8],ymm3[9,10],ymm2[11],ymm3[12],ymm2[13],ymm3[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm6, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm6 = xmm7[0],xmm6[1],xmm7[2],xmm6[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[6,7,0,1,10,11,4,5,14,15,8,9,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8],ymm0[9],ymm1[10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm11 = ymm7[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm7 = ymm7[0,1,2,3],ymm11[4],ymm7[5],ymm11[6],ymm7[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm7[u,u,u,u,u,u,u,u,u,u,u,u,2,3,12,13,22,23,16,17,26,27,20,21,30,31,30,31,18,19,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2],ymm7[3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm7 = xmm4[0,1],xmm5[2],xmm4[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm7 = xmm7[u,u,u,u,u,u,u,u,u,u,8,9,2,3,12,13]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm7, %ymm0, %ymm7
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm7 = ymm6[0,1,2,3,4],ymm7[5,6,7],ymm6[8,9,10,11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm6 = ymm6[0,1,2,3],ymm7[4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5],ymm2[6],ymm3[7,8],ymm2[9],ymm3[10,11],ymm2[12],ymm3[13],ymm2[14],ymm3[15]
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1,2],xmm2[3,4],xmm3[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[8,9,2,3,12,13,6,7,0,1,10,11,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3],ymm0[4],ymm1[5,6],ymm0[7],ymm1[8,9],ymm0[10],ymm1[11],ymm0[12],ymm1[13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,4,5,14,15,24,25,18,19,28,29,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1,2],ymm0[3,4,5],ymm2[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm4[12,13,14,15,4,5,14,15,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm2 = xmm5[0,1,2,3,0,1,10,11,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm9, (%rsi)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, (%rdx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm8, (%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm6, (%r8)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%r9)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %wide.vec = load <80 x i16>, <80 x i16>* %in.vec, align 32
 
   %strided.vec0 = shufflevector <80 x i16> %wide.vec, <80 x i16> poison, <16 x i32> <i32 0, i32 5, i32 10, i32 15, i32 20, i32 25, i32 30, i32 35, i32 40, i32 45, i32 50, i32 55, i32 60, i32 65, i32 70, i32 75>
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST %s
 
 ; These patterns are produced by LoopVectorizer for interleaved loads.
 
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-2.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-2.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-2.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-2.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2 %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2 %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2 %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2 %s
 
 ; These patterns are produced by LoopVectorizer for interleaved stores.
 
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2,AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST,AVX2-FAST-ALL %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST,AVX2-FAST-PERLANE %s
 
 ; These patterns are produced by LoopVectorizer for interleaved stores.
 
@@ -83,27 +84,50 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf8:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm1
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm2
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,0,0,u,1,1,u,2>
-; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,4,1,5,1,5,2,6]
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm5, %ymm3
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[0,1,4,5,u,u,2,3,6,7,u,u,8,9,12,13,u,u,18,19,22,23,u,u,24,25,28,29,u,u,26,27]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,u,u,10,11,8,9,u,u,14,15,12,13,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, 32(%rcx)
-; AVX2-FAST-NEXT:    vmovdqa %ymm3, (%rcx)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf8:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rsi), %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %xmm2
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,0,0,u,1,1,u,2>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm2, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm5 = [0,4,1,5,1,5,2,6]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm3, %ymm5, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[0,1,4,5,u,u,2,3,6,7,u,u,8,9,12,13,u,u,18,19,22,23,u,u,24,25,28,29,u,u,26,27]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,u,u,10,11,8,9,u,u,14,15,12,13,u,u]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm0, 32(%rcx)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm3, (%rcx)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf8:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm3[0,1,u,u,6,7,2,3,u,u,8,9,4,5,u,u,16,17,u,u,22,23,18,19,u,u,24,25,20,21,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,0,1,u,u,u,u,2,3,u,u,u,u,4,5,u,u,22,23,u,u,u,u,24,25,u,u,u,u,26,27]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7],ymm4[8],ymm3[9],ymm4[10,11],ymm3[12],ymm4[13,14],ymm3[15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm4 = <u,0,0,u,1,1,u,2>
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm2, %ymm4, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm5 = [255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255,255,255,0,0,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm5, %ymm3, %ymm4, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,u,u,10,11,8,9,u,u,14,15,12,13,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm1 = xmm2[2,2,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm1[1],xmm0[2,3],xmm1[4],xmm0[5,6],xmm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, 32(%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, (%rcx)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <8 x i16>, <8 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <8 x i16>, <8 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <8 x i16>, <8 x i16>* %in.vecptr2, align 32
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-4.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-4.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-4.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-4.ll
@@ -1,20 +1,32 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2,AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2-SLOW %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-ALL %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-PERLANE %s
 
 ; These patterns are produced by LoopVectorizer for interleaved stores.
 
 define void @vf2(<2 x i16>* %in.vecptr0, <2 x i16>* %in.vecptr1, <2 x i16>* %in.vecptr2, <2 x i16>* %in.vecptr3, <8 x i16>* %out.vec) nounwind {
-; AVX2-LABEL: vf2:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa (%rdx), %xmm1
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; AVX2-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,2,3,10,11,4,5,12,13,6,7,14,15]
-; AVX2-NEXT:    vmovdqa %xmm0, (%r8)
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: vf2:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,2,3,10,11,4,5,12,13,6,7,14,15]
+; AVX2-SLOW-NEXT:    vmovdqa %xmm0, (%r8)
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: vf2:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,8,9,2,3,10,11,4,5,12,13,6,7,14,15]
+; AVX2-FAST-NEXT:    vmovdqa %xmm0, (%r8)
+; AVX2-FAST-NEXT:    retq
   %in.vec0 = load <2 x i16>, <2 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <2 x i16>, <2 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <2 x i16>, <2 x i16>* %in.vecptr2, align 32
@@ -48,21 +60,38 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf4:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,2,4,6,1,3,5,7]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,2,3,6,7,10,11,14,15,16,17,20,21,24,25,28,29,18,19,22,23,26,27,30,31]
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%r8)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf4:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,2,4,6,1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,2,3,6,7,10,11,14,15,16,17,20,21,24,25,28,29,18,19,22,23,26,27,30,31]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf4:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u,22,23,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,0,1,8,9,u,u,u,u,2,3,10,11,20,21,28,29,u,u,u,u,22,23,30,31,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <4 x i16>, <4 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <4 x i16>, <4 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <4 x i16>, <4 x i16>* %in.vecptr2, align 32
@@ -79,28 +108,51 @@
 }
 
 define void @vf8(<8 x i16>* %in.vecptr0, <8 x i16>* %in.vecptr1, <8 x i16>* %in.vecptr2, <8 x i16>* %in.vecptr3, <32 x i16>* %out.vec) nounwind {
-; AVX2-LABEL: vf8:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa (%rdx), %xmm1
-; AVX2-NEXT:    vinserti128 $1, (%rsi), %ymm0, %ymm0
-; AVX2-NEXT:    vinserti128 $1, (%rcx), %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[0,2,0,2]
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,4,5,12,13,u,u,u,u,6,7,14,15>
-; AVX2-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
-; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,0,2]
-; AVX2-NEXT:    vmovdqa {{.*#+}} ymm5 = <0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,4,5,12,13,u,u,u,u,6,7,14,15,u,u,u,u>
-; AVX2-NEXT:    vpshufb %ymm5, %ymm4, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0],ymm2[1],ymm4[2],ymm2[3],ymm4[4],ymm2[5],ymm4[6],ymm2[7]
-; AVX2-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[1,3,1,3]
-; AVX2-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,3,1,3]
-; AVX2-NEXT:    vpshufb %ymm5, %ymm0, %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
-; AVX2-NEXT:    vmovdqa %ymm0, 32(%r8)
-; AVX2-NEXT:    vmovdqa %ymm2, (%r8)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: vf8:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-SLOW-NEXT:    vinserti128 $1, (%rsi), %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vinserti128 $1, (%rcx), %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[0,2,0,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,4,5,12,13,u,u,u,u,6,7,14,15>
+; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,0,2]
+; AVX2-SLOW-NEXT:    vmovdqa {{.*#+}} ymm5 = <0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,4,5,12,13,u,u,u,u,6,7,14,15,u,u,u,u>
+; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm4, %ymm4
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0],ymm2[1],ymm4[2],ymm2[3],ymm4[4],ymm2[5],ymm4[6],ymm2[7]
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[1,3,1,3]
+; AVX2-SLOW-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,3,1,3]
+; AVX2-SLOW-NEXT:    vpshufb %ymm5, %ymm0, %ymm0
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, (%r8)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: vf8:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-FAST-NEXT:    vinserti128 $1, (%rsi), %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vinserti128 $1, (%rcx), %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[0,2,0,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,4,5,12,13,u,u,u,u,6,7,14,15>
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm2, %ymm2
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,0,2]
+; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm5 = <0,1,8,9,u,u,u,u,2,3,10,11,u,u,u,u,4,5,12,13,u,u,u,u,6,7,14,15,u,u,u,u>
+; AVX2-FAST-NEXT:    vpshufb %ymm5, %ymm4, %ymm4
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0],ymm2[1],ymm4[2],ymm2[3],ymm4[4],ymm2[5],ymm4[6],ymm2[7]
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[1,3,1,3]
+; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,3,1,3]
+; AVX2-FAST-NEXT:    vpshufb %ymm5, %ymm0, %ymm0
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
+; AVX2-FAST-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm2, (%r8)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
   %in.vec0 = load <8 x i16>, <8 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <8 x i16>, <8 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <8 x i16>, <8 x i16>* %in.vecptr2, align 32
@@ -117,58 +169,111 @@
 }
 
 define void @vf16(<16 x i16>* %in.vecptr0, <16 x i16>* %in.vecptr1, <16 x i16>* %in.vecptr2, <16 x i16>* %in.vecptr3, <64 x i16>* %out.vec) nounwind {
-; AVX2-LABEL: vf16:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rcx), %xmm5
-; AVX2-NEXT:    vmovdqa 16(%rcx), %xmm8
-; AVX2-NEXT:    vmovdqa (%rdx), %xmm6
-; AVX2-NEXT:    vmovdqa 16(%rdx), %xmm9
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
-; AVX2-NEXT:    vmovdqa (%rsi), %xmm7
-; AVX2-NEXT:    vmovdqa 16(%rsi), %xmm3
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-NEXT:    vmovdqa 16(%rdi), %xmm4
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm1[0],zero,xmm1[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm10, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0],ymm2[1],ymm1[2],ymm2[3],ymm1[4],ymm2[5],ymm1[6],ymm2[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm10 = xmm1[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm10, %ymm1
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm2[0],zero,xmm2[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm10, %ymm2
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4],ymm1[5],ymm2[6],ymm1[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm5, %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3],ymm3[4],ymm2[5],ymm3[6],ymm2[7]
-; AVX2-NEXT:    vmovdqa %ymm2, 96(%r8)
-; AVX2-NEXT:    vmovdqa %ymm0, (%r8)
-; AVX2-NEXT:    vmovdqa %ymm1, 32(%r8)
-; AVX2-NEXT:    vmovdqa %ymm11, 64(%r8)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: vf16:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %xmm5
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rcx), %xmm8
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm6
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rdx), %xmm9
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm7
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rsi), %xmm3
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rdi), %xmm4
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm1[0],zero,xmm1[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm10, %ymm1
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0],ymm2[1],ymm1[2],ymm2[3],ymm1[4],ymm2[5],ymm1[6],ymm2[7]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm10 = xmm1[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm10, %ymm1
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm2[0],zero,xmm2[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm2, %ymm10, %ymm2
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4],ymm1[5],ymm2[6],ymm1[7]
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm0[0],zero,xmm0[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm0, %ymm5, %ymm0
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3],ymm3[4],ymm2[5],ymm3[6],ymm2[7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, 96(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm1, 32(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm11, 64(%r8)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: vf16:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm5
+; AVX2-FAST-NEXT:    vmovdqa 16(%rcx), %xmm8
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm6
+; AVX2-FAST-NEXT:    vmovdqa 16(%rdx), %xmm9
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm9[0],xmm8[0],xmm9[1],xmm8[1],xmm9[2],xmm8[2],xmm9[3],xmm8[3]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
+; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm7
+; AVX2-FAST-NEXT:    vmovdqa 16(%rsi), %xmm3
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-NEXT:    vmovdqa 16(%rdi), %xmm4
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm1[0],zero,xmm1[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm10, %ymm1
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm11 = ymm1[0],ymm2[1],ymm1[2],ymm2[3],ymm1[4],ymm2[5],ymm1[6],ymm2[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm6[4],xmm5[4],xmm6[5],xmm5[5],xmm6[6],xmm5[6],xmm6[7],xmm5[7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm10 = xmm1[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm10, %ymm1
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm2[0],zero,xmm2[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm2, %ymm10, %ymm2
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2],ymm1[3],ymm2[4],ymm1[5],ymm2[6],ymm1[7]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm0[0],zero,xmm0[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm5, %ymm0
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm9[4],xmm8[4],xmm9[5],xmm8[5],xmm9[6],xmm8[6],xmm9[7],xmm8[7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm2[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm2, %ymm5, %ymm2
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3],ymm3[4],ymm2[5],ymm3[6],ymm2[7]
+; AVX2-FAST-NEXT:    vmovdqa %ymm2, 96(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, 32(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm11, 64(%r8)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
   %in.vec0 = load <16 x i16>, <16 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <16 x i16>, <16 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <16 x i16>, <16 x i16>* %in.vecptr2, align 32
@@ -185,108 +290,211 @@
 }
 
 define void @vf32(<32 x i16>* %in.vecptr0, <32 x i16>* %in.vecptr1, <32 x i16>* %in.vecptr2, <32 x i16>* %in.vecptr3, <128 x i16>* %out.vec) nounwind {
-; AVX2-LABEL: vf32:
-; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rcx), %xmm15
-; AVX2-NEXT:    vmovdqa 16(%rcx), %xmm12
-; AVX2-NEXT:    vmovdqa 32(%rcx), %xmm11
-; AVX2-NEXT:    vmovdqa 48(%rcx), %xmm2
-; AVX2-NEXT:    vmovdqa (%rdx), %xmm6
-; AVX2-NEXT:    vmovdqa 16(%rdx), %xmm13
-; AVX2-NEXT:    vmovdqa 32(%rdx), %xmm1
-; AVX2-NEXT:    vmovdqa 48(%rdx), %xmm7
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm5, %ymm8
-; AVX2-NEXT:    vmovdqa 16(%rsi), %xmm14
-; AVX2-NEXT:    vmovdqa 32(%rsi), %xmm3
-; AVX2-NEXT:    vmovdqa 16(%rdi), %xmm5
-; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm4
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm14[0],xmm5[1],xmm14[1],xmm5[2],xmm14[2],xmm5[3],xmm14[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm9, %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2],ymm8[3],ymm0[4],ymm8[5],ymm0[6],ymm8[7]
-; AVX2-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm0[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm8, %ymm8
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm10, %ymm0
-; AVX2-NEXT:    vmovdqa 48(%rsi), %xmm10
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm0[0],ymm8[1],ymm0[2],ymm8[3],ymm0[4],ymm8[5],ymm0[6],ymm8[7]
-; AVX2-NEXT:    vmovdqa 48(%rdi), %xmm0
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm8 = xmm1[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm8, %ymm1
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm3[0],ymm1[1],ymm3[2],ymm1[3],ymm3[4],ymm1[5],ymm3[6],ymm1[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm3, %ymm1
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm0[4],xmm10[4],xmm0[5],xmm10[5],xmm0[6],xmm10[6],xmm0[7],xmm10[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; AVX2-NEXT:    vmovdqa (%rsi), %xmm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm11 = ymm3[0],ymm1[1],ymm3[2],ymm1[3],ymm3[4],ymm1[5],ymm3[6],ymm1[7]
-; AVX2-NEXT:    vmovdqa (%rdi), %xmm1
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm10[0],xmm0[1],xmm10[1],xmm0[2],xmm10[2],xmm0[3],xmm10[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm15[4],xmm6[5],xmm15[5],xmm6[6],xmm15[6],xmm6[7],xmm15[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm3[0],zero,xmm3[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm7, %ymm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0],ymm0[1],ymm3[2],ymm0[3],ymm3[4],ymm0[5],ymm3[6],ymm0[7]
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm15[0],xmm6[1],xmm15[1],xmm6[2],xmm15[2],xmm6[3],xmm15[3]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm6 = xmm3[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm6, %ymm3
-; AVX2-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm4, %ymm1
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2],ymm3[3],ymm1[4],ymm3[5],ymm1[6],ymm3[7]
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[0,0,1,1]
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
-; AVX2-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm14[4],xmm5[5],xmm14[5],xmm5[6],xmm14[6],xmm5[7],xmm14[7]
-; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm4[0],zero,xmm4[1],zero
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,2,3,3]
-; AVX2-NEXT:    vinserti128 $1, %xmm4, %ymm5, %ymm4
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2],ymm3[3],ymm4[4],ymm3[5],ymm4[6],ymm3[7]
-; AVX2-NEXT:    vmovdqa %ymm3, 96(%r8)
-; AVX2-NEXT:    vmovdqa %ymm1, (%r8)
-; AVX2-NEXT:    vmovdqa %ymm0, 32(%r8)
-; AVX2-NEXT:    vmovdqa %ymm2, 192(%r8)
-; AVX2-NEXT:    vmovdqa %ymm11, 224(%r8)
-; AVX2-NEXT:    vmovdqa %ymm8, 128(%r8)
-; AVX2-NEXT:    vmovdqa %ymm9, 160(%r8)
-; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-NEXT:    vmovaps %ymm0, 64(%r8)
-; AVX2-NEXT:    vzeroupper
-; AVX2-NEXT:    retq
+; AVX2-SLOW-LABEL: vf32:
+; AVX2-SLOW:       # %bb.0:
+; AVX2-SLOW-NEXT:    vmovdqa (%rcx), %xmm15
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rcx), %xmm12
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rcx), %xmm11
+; AVX2-SLOW-NEXT:    vmovdqa 48(%rcx), %xmm2
+; AVX2-SLOW-NEXT:    vmovdqa (%rdx), %xmm6
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rdx), %xmm13
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rdx), %xmm1
+; AVX2-SLOW-NEXT:    vmovdqa 48(%rdx), %xmm7
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm5, %ymm8
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rsi), %xmm14
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rsi), %xmm3
+; AVX2-SLOW-NEXT:    vmovdqa 16(%rdi), %xmm5
+; AVX2-SLOW-NEXT:    vmovdqa 32(%rdi), %xmm4
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm14[0],xmm5[1],xmm14[1],xmm5[2],xmm14[2],xmm5[3],xmm14[3]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm0[0],zero,xmm0[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm0, %ymm9, %ymm0
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2],ymm8[3],ymm0[4],ymm8[5],ymm0[6],ymm8[7]
+; AVX2-SLOW-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm8 = xmm0[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm0, %ymm8, %ymm8
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm0[0],zero,xmm0[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm0, %ymm10, %ymm0
+; AVX2-SLOW-NEXT:    vmovdqa 48(%rsi), %xmm10
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm9 = ymm0[0],ymm8[1],ymm0[2],ymm8[3],ymm0[4],ymm8[5],ymm0[6],ymm8[7]
+; AVX2-SLOW-NEXT:    vmovdqa 48(%rdi), %xmm0
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm8 = xmm1[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm8, %ymm1
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm8 = ymm3[0],ymm1[1],ymm3[2],ymm1[3],ymm3[4],ymm1[5],ymm3[6],ymm1[7]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm3, %ymm1
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm0[4],xmm10[4],xmm0[5],xmm10[5],xmm0[6],xmm10[6],xmm0[7],xmm10[7]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT:    vmovdqa (%rsi), %xmm4
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm11 = ymm3[0],ymm1[1],ymm3[2],ymm1[3],ymm3[4],ymm1[5],ymm3[6],ymm1[7]
+; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm1
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm10[0],xmm0[1],xmm10[1],xmm0[2],xmm10[2],xmm0[3],xmm10[3]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm2 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm15[4],xmm6[5],xmm15[5],xmm6[6],xmm15[6],xmm6[7],xmm15[7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm3[0],zero,xmm3[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm7, %ymm3
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0],ymm0[1],ymm3[2],ymm0[3],ymm3[4],ymm0[5],ymm3[6],ymm0[7]
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm15[0],xmm6[1],xmm15[1],xmm6[2],xmm15[2],xmm6[3],xmm15[3]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm6 = xmm3[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm6, %ymm3
+; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm1, %ymm4, %ymm1
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2],ymm3[3],ymm1[4],ymm3[5],ymm1[6],ymm3[7]
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[0,0,1,1]
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-SLOW-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm14[4],xmm5[5],xmm14[5],xmm5[6],xmm14[6],xmm5[7],xmm14[7]
+; AVX2-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm4[0],zero,xmm4[1],zero
+; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,2,3,3]
+; AVX2-SLOW-NEXT:    vinserti128 $1, %xmm4, %ymm5, %ymm4
+; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2],ymm3[3],ymm4[4],ymm3[5],ymm4[6],ymm3[7]
+; AVX2-SLOW-NEXT:    vmovdqa %ymm3, 96(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm1, (%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm2, 192(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm11, 224(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm8, 128(%r8)
+; AVX2-SLOW-NEXT:    vmovdqa %ymm9, 160(%r8)
+; AVX2-SLOW-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-SLOW-NEXT:    vmovaps %ymm0, 64(%r8)
+; AVX2-SLOW-NEXT:    vzeroupper
+; AVX2-SLOW-NEXT:    retq
+;
+; AVX2-FAST-LABEL: vf32:
+; AVX2-FAST:       # %bb.0:
+; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm15
+; AVX2-FAST-NEXT:    vmovdqa 16(%rcx), %xmm12
+; AVX2-FAST-NEXT:    vmovdqa 32(%rcx), %xmm11
+; AVX2-FAST-NEXT:    vmovdqa 48(%rcx), %xmm2
+; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm6
+; AVX2-FAST-NEXT:    vmovdqa 16(%rdx), %xmm13
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdx), %xmm1
+; AVX2-FAST-NEXT:    vmovdqa 48(%rdx), %xmm7
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm13[0],xmm12[0],xmm13[1],xmm12[1],xmm13[2],xmm12[2],xmm13[3],xmm12[3]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = xmm3[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm5, %ymm8
+; AVX2-FAST-NEXT:    vmovdqa 16(%rsi), %xmm14
+; AVX2-FAST-NEXT:    vmovdqa 32(%rsi), %xmm3
+; AVX2-FAST-NEXT:    vmovdqa 16(%rdi), %xmm5
+; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %xmm4
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm5[0],xmm14[0],xmm5[1],xmm14[1],xmm5[2],xmm14[2],xmm5[3],xmm14[3]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm9 = xmm0[0],zero,xmm0[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm9, %ymm0
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm8[1],ymm0[2],ymm8[3],ymm0[4],ymm8[5],ymm0[6],ymm8[7]
+; AVX2-FAST-NEXT:    vmovdqu %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm1[4],xmm11[4],xmm1[5],xmm11[5],xmm1[6],xmm11[6],xmm1[7],xmm11[7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm8 = xmm0[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm8, %ymm8
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm4[4],xmm3[4],xmm4[5],xmm3[5],xmm4[6],xmm3[6],xmm4[7],xmm3[7]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm10 = xmm0[0],zero,xmm0[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm10, %ymm0
+; AVX2-FAST-NEXT:    vmovdqa 48(%rsi), %xmm10
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm9 = ymm0[0],ymm8[1],ymm0[2],ymm8[3],ymm0[4],ymm8[5],ymm0[6],ymm8[7]
+; AVX2-FAST-NEXT:    vmovdqa 48(%rdi), %xmm0
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm11[0],xmm1[1],xmm11[1],xmm1[2],xmm11[2],xmm1[3],xmm11[3]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm8 = xmm1[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm8, %ymm1
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm4[0],xmm3[0],xmm4[1],xmm3[1],xmm4[2],xmm3[2],xmm4[3],xmm3[3]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm8 = ymm3[0],ymm1[1],ymm3[2],ymm1[3],ymm3[4],ymm1[5],ymm3[6],ymm1[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm7[4],xmm2[4],xmm7[5],xmm2[5],xmm7[6],xmm2[6],xmm7[7],xmm2[7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm1[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm3, %ymm1
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm0[4],xmm10[4],xmm0[5],xmm10[5],xmm0[6],xmm10[6],xmm0[7],xmm10[7]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm3[0],zero,xmm3[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm4
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm11 = ymm3[0],ymm1[1],ymm3[2],ymm1[3],ymm3[4],ymm1[5],ymm3[6],ymm1[7]
+; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm1
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm2[0],xmm7[1],xmm2[1],xmm7[2],xmm2[2],xmm7[3],xmm2[3]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm2[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm10[0],xmm0[1],xmm10[1],xmm0[2],xmm10[2],xmm0[3],xmm10[3]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm3 = xmm0[0],zero,xmm0[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm0[0],ymm2[1],ymm0[2],ymm2[3],ymm0[4],ymm2[5],ymm0[6],ymm2[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm6[4],xmm15[4],xmm6[5],xmm15[5],xmm6[6],xmm15[6],xmm6[7],xmm15[7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm3, %ymm0
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm4[4],xmm1[5],xmm4[5],xmm1[6],xmm4[6],xmm1[7],xmm4[7]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm7 = xmm3[0],zero,xmm3[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm7, %ymm3
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0],ymm0[1],ymm3[2],ymm0[3],ymm3[4],ymm0[5],ymm3[6],ymm0[7]
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm15[0],xmm6[1],xmm15[1],xmm6[2],xmm15[2],xmm6[3],xmm15[3]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm6 = xmm3[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm6, %ymm3
+; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm4[0],xmm1[1],xmm4[1],xmm1[2],xmm4[2],xmm1[3],xmm4[3]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm4 = xmm1[0],zero,xmm1[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm4, %ymm1
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm3[1],ymm1[2],ymm3[3],ymm1[4],ymm3[5],ymm1[6],ymm3[7]
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm13[4],xmm12[4],xmm13[5],xmm12[5],xmm13[6],xmm12[6],xmm13[7],xmm12[7]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm3[0,0,1,1]
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm4, %ymm3
+; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm5[4],xmm14[4],xmm5[5],xmm14[5],xmm5[6],xmm14[6],xmm5[7],xmm14[7]
+; AVX2-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm5 = xmm4[0],zero,xmm4[1],zero
+; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[2,2,3,3]
+; AVX2-FAST-NEXT:    vinserti128 $1, %xmm4, %ymm5, %ymm4
+; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2],ymm3[3],ymm4[4],ymm3[5],ymm4[6],ymm3[7]
+; AVX2-FAST-NEXT:    vmovdqa %ymm3, 96(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm1, (%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm0, 32(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm2, 192(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm11, 224(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm8, 128(%r8)
+; AVX2-FAST-NEXT:    vmovdqa %ymm9, 160(%r8)
+; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-NEXT:    vmovaps %ymm0, 64(%r8)
+; AVX2-FAST-NEXT:    vzeroupper
+; AVX2-FAST-NEXT:    retq
   %in.vec0 = load <32 x i16>, <32 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <32 x i16>, <32 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <32 x i16>, <32 x i16>* %in.vecptr2, align 32
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-5.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-ALL %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-PERLANE %s
 
 ; These patterns are produced by LoopVectorizer for interleaved stores.
 
@@ -178,46 +179,90 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf8:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm1
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm0
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm2
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm3
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm4
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm5
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,0]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,ymm5[0,1,8,9,12,13],zero,zero,zero,zero,ymm5[2,3,18,19,18,19],zero,zero,zero,zero,ymm5[28,29,20,21,28,29],zero,zero
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm6
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,0,2]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[0,1,8,9],zero,zero,zero,zero,zero,zero,ymm6[2,3,10,11],zero,zero,zero,zero,zero,zero,ymm6[20,21,28,29],zero,zero,zero,zero,zero,zero,ymm6[22,23]
-; AVX2-FAST-NEXT:    vpor %ymm5, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vpbroadcastq (%r8), %ymm6
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm4, %ymm3, %ymm6
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm8 = <1,5,2,u,6,2,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,ymm6[2,3,6,7,6,7],zero,zero,zero,zero,ymm6[8,9,16,17,18,19],zero,zero,zero,zero,ymm6[22,23,18,19,18,19],zero,zero
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm9 = <1,5,2,6,2,6,3,u>
-; AVX2-FAST-NEXT:    vpermd %ymm8, %ymm9, %ymm8
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm8 = ymm8[2,3,6,7],zero,zero,zero,zero,zero,zero,ymm8[8,9,12,13],zero,zero,zero,zero,zero,zero,ymm8[18,19,22,23],zero,zero,zero,zero,zero,zero,ymm8[24,25]
-; AVX2-FAST-NEXT:    vpor %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT:    vpbroadcastq 8(%rdi), %ymm8
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm6, %ymm8, %ymm6
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm1[4],xmm4[5],xmm1[5],xmm4[6],xmm1[6],xmm4[7],xmm1[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,8,9,14,15,u,u,u,u,u,u,12,13]
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[8,9,10,11,u,u,u,u,u,u,12,13,14,15,u,u]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5,6],xmm1[7]
-; AVX2-FAST-NEXT:    vpsrlq $48, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4],xmm1[5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, 64(%r9)
-; AVX2-FAST-NEXT:    vmovdqa %ymm6, 32(%r9)
-; AVX2-FAST-NEXT:    vmovdqa %ymm5, (%r9)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf8:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rsi), %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %xmm2
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rcx), %xmm3
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r8), %xmm4
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm3, %ymm2, %ymm5
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,2,2,0]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm5 = zero,zero,zero,zero,ymm5[0,1,8,9,12,13],zero,zero,zero,zero,ymm5[2,3,18,19,18,19],zero,zero,zero,zero,ymm5[28,29,20,21,28,29],zero,zero
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,0,2]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = ymm6[0,1,8,9],zero,zero,zero,zero,zero,zero,ymm6[2,3,10,11],zero,zero,zero,zero,zero,zero,ymm6[20,21,28,29],zero,zero,zero,zero,zero,zero,ymm6[22,23]
+; AVX2-FAST-ALL-NEXT:    vpor %ymm5, %ymm6, %ymm5
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq (%r8), %ymm6
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm4, %ymm3, %ymm6
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm8 = <1,5,2,u,6,2,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm6, %ymm8, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,ymm6[2,3,6,7,6,7],zero,zero,zero,zero,ymm6[8,9,16,17,18,19],zero,zero,zero,zero,ymm6[22,23,18,19,18,19],zero,zero
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm8
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm9 = <1,5,2,6,2,6,3,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm8, %ymm9, %ymm8
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm8 = ymm8[2,3,6,7],zero,zero,zero,zero,zero,zero,ymm8[8,9,12,13],zero,zero,zero,zero,zero,zero,ymm8[18,19,22,23],zero,zero,zero,zero,zero,zero,ymm8[24,25]
+; AVX2-FAST-ALL-NEXT:    vpor %ymm6, %ymm8, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq 8(%rdi), %ymm8
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm6, %ymm8, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm4[4],xmm1[4],xmm4[5],xmm1[5],xmm4[6],xmm1[6],xmm4[7],xmm1[7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,8,9,14,15,u,u,u,u,u,u,12,13]
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[8,9,10,11,u,u,u,u,u,u,12,13,14,15,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5,6],xmm1[7]
+; AVX2-FAST-ALL-NEXT:    vpsrlq $48, %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm0, 64(%r9)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm6, 32(%r9)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm5, (%r9)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf8:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm4
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm6 = ymm5[6,7,u,u,u,u,10,11,u,u,8,9,u,u,u,u,22,23,u,u,u,u,26,27,u,u,24,25,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,6,7,10,11,u,u,8,9,u,u,8,9,12,13,u,u,22,23,26,27,u,u,24,25,u,u,24,25,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm5 = ymm6[0],ymm5[1,2],ymm6[3],ymm5[4],ymm6[5],ymm5[6,7],ymm6[8],ymm5[9,10],ymm6[11],ymm5[12],ymm6[13],ymm5[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm4, %ymm3, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm7 = ymm6[8,9,10,11,6,7,u,u,u,u,10,11,12,13,8,9,24,25,26,27,22,23,u,u,u,u,26,27,28,29,24,25]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufhw {{.*#+}} ymm6 = ymm6[0,1,2,3,5,5,5,5,8,9,10,11,13,13,13,13]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm6 = ymm7[0,1,2],ymm6[3,4],ymm7[5,6,7,8,9,10],ymm6[11,12],ymm7[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = <255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq 8(%rdi), %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm5, %ymm6, %ymm5
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm3, %ymm1, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,2,2,0]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm6 = zero,zero,zero,zero,ymm6[0,1,8,9,12,13],zero,zero,zero,zero,ymm6[2,3,18,19,18,19],zero,zero,zero,zero,ymm6[28,29,20,21,28,29],zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm0, %ymm2, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm8[0,2,0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm8 = ymm8[0,1,8,9],zero,zero,zero,zero,zero,zero,ymm8[2,3,10,11],zero,zero,zero,zero,zero,zero,ymm8[20,21,28,29],zero,zero,zero,zero,zero,zero,ymm8[22,23]
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm6, %ymm8, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq (%r8), %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm6, %ymm8, %ymm6
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm2 = xmm4[4],xmm2[4],xmm4[5],xmm2[5],xmm4[6],xmm2[6],xmm4[7],xmm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[u,u,u,u,8,9,14,15,u,u,u,u,u,u,12,13]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm1 = xmm1[4],xmm3[4],xmm1[5],xmm3[5],xmm1[6],xmm3[6],xmm1[7],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[8,9,10,11,u,u,u,u,u,u,12,13,14,15,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5,6],xmm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vpsrlq $48, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4],xmm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, 64(%r9)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm6, (%r9)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm5, 32(%r9)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <8 x i16>, <8 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <8 x i16>, <8 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <8 x i16>, <8 x i16>* %in.vecptr2, align 32
@@ -315,79 +360,153 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf16:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm12
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm11
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm3
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm4
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm2
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm5
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm6
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[0,1,2,3,8,9,10,11,4,5,4,5,6,7,12,13]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm7 = ymm6[0,1,0,1]
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm6
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm0
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,2,3,2,3,0,1,10,11,8,9,4,5,6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm10 = <255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255>
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm7, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpbroadcastq (%r8), %ymm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm1, %ymm8, %ymm9
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[1,2,2,2]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[6,7,u,u,10,11,u,u,8,9,8,9,u,u,12,13]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0],xmm1[1],xmm5[2],xmm1[3],xmm5[4,5],xmm1[6],xmm5[7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm5 = mem[2,1,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,u,u,6,7,u,u,10,11,u,u,u,u,8,9]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1],xmm0[2],xmm5[3],xmm0[4],xmm5[5,6],xmm0[7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpbroadcastq 8(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm8
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm2[0,1,0,1,4,5,4,5]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,u,u,22,23,u,u,24,25,20,21,u,u,24,25]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8],ymm0[9],ymm1[10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[0,1,2,1,4,5,6,5]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,18,19,u,u,20,21,u,u,24,25,24,25,u,u]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0,1],ymm1[2],ymm5[3],ymm1[4],ymm5[5,6],ymm1[7],ymm5[8,9],ymm1[10],ymm5[11],ymm1[12],ymm5[13,14],ymm1[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [25769803781,25769803781,25769803781,25769803781]
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[2,3,2,3,6,7,6,7]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[2,3,2,3,6,7,6,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0],ymm1[1],ymm5[2],ymm1[3],ymm5[4,5],ymm1[6],ymm5[7,8],ymm1[9],ymm5[10],ymm1[11],ymm5[12,13],ymm1[14],ymm5[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,u,u,30,31,u,u,u,u,28,29]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm3[3,2,3,3,7,6,7,7]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm5 = ymm6[0,1],ymm5[2],ymm6[3],ymm5[4],ymm6[5,6],ymm5[7],ymm6[8,9],ymm5[10],ymm6[11],ymm5[12],ymm6[13,14],ymm5[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm1, %ymm5, %ymm1
-; AVX2-FAST-NEXT:    vpsrldq {{.*#+}} ymm5 = ymm4[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm4[26,27,28,29,30,31],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,2]
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm1, %ymm5, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,12,13,u,u,0,1,u,u,u,u,14,15,u,u,u,u,28,29,u,u,16,17,u,u,u,u,30,31,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,0,3,0,7,4,7,4]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3],ymm3[4,5],ymm4[6],ymm3[7,8],ymm4[9],ymm3[10],ymm4[11],ymm3[12,13],ymm4[14],ymm3[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm12[1,1,2,2]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,1,2,2]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2],ymm4[3],ymm2[4],ymm4[5],ymm2[6,7],ymm4[8],ymm2[9,10],ymm4[11],ymm2[12],ymm4[13],ymm2[14,15]
-; AVX2-FAST-NEXT:    vpblendvb %ymm10, %ymm3, %ymm2, %ymm2
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm11[0,1,0,1,0,1,0,1,14,15,2,3,2,3,14,15,16,17,16,17,16,17,16,17,30,31,18,19,18,19,30,31]
-; AVX2-FAST-NEXT:    vpblendvb %ymm7, %ymm2, %ymm3, %ymm2
-; AVX2-FAST-NEXT:    vmovdqa %ymm2, 64(%r9)
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, 128(%r9)
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, 96(%r9)
-; AVX2-FAST-NEXT:    vmovdqa %ymm8, 32(%r9)
-; AVX2-FAST-NEXT:    vmovdqa %ymm9, (%r9)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf16:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm12
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rsi), %ymm11
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %ymm3
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rcx), %ymm4
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r8), %ymm2
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rsi), %xmm5
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm6
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[0,1,2,3,8,9,10,11,4,5,4,5,6,7,12,13]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm7 = ymm6[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %xmm6
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rcx), %xmm0
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,2,3,2,3,0,1,10,11,8,9,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm10 = <255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255>
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm10, %ymm7, %ymm1, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq (%r8), %ymm8
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm1, %ymm8, %ymm9
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[1,2,2,2]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[6,7,u,u,10,11,u,u,8,9,8,9,u,u,12,13]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0],xmm1[1],xmm5[2],xmm1[3],xmm5[4,5],xmm1[6],xmm5[7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm5 = mem[2,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,u,u,6,7,u,u,10,11,u,u,u,u,8,9]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1],xmm0[2],xmm5[3],xmm0[4],xmm5[5,6],xmm0[7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm10, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq 8(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm8
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm0 = ymm2[0,1,0,1,4,5,4,5]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,u,u,22,23,u,u,24,25,20,21,u,u,24,25]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8],ymm0[9],ymm1[10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[0,1,2,1,4,5,6,5]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,18,19,u,u,20,21,u,u,24,25,24,25,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0,1],ymm1[2],ymm5[3],ymm1[4],ymm5[5,6],ymm1[7],ymm5[8,9],ymm1[10],ymm5[11],ymm1[12],ymm5[13,14],ymm1[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [25769803781,25769803781,25769803781,25769803781]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm3, %ymm1, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[2,3,2,3,6,7,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[2,3,2,3,6,7,6,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0],ymm1[1],ymm5[2],ymm1[3],ymm5[4,5],ymm1[6],ymm5[7,8],ymm1[9],ymm5[10],ymm1[11],ymm5[12,13],ymm1[14],ymm5[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,u,u,30,31,u,u,u,u,28,29]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm6 = ymm3[3,2,3,3,7,6,7,7]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm5 = ymm6[0,1],ymm5[2],ymm6[3],ymm5[4],ymm6[5,6],ymm5[7],ymm6[8,9],ymm5[10],ymm6[11],ymm5[12],ymm6[13,14],ymm5[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm10, %ymm1, %ymm5, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpsrldq {{.*#+}} ymm5 = ymm4[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm4[26,27,28,29,30,31],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,2]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm1, %ymm5, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,12,13,u,u,0,1,u,u,u,u,14,15,u,u,u,u,28,29,u,u,16,17,u,u,u,u,30,31,u,u]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,0,3,0,7,4,7,4]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3],ymm3[4,5],ymm4[6],ymm3[7,8],ymm4[9],ymm3[10],ymm4[11],ymm3[12,13],ymm4[14],ymm3[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm4 = ymm12[1,1,2,2]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,1,2,2]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2],ymm4[3],ymm2[4],ymm4[5],ymm2[6,7],ymm4[8],ymm2[9,10],ymm4[11],ymm2[12],ymm4[13],ymm2[14,15]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm10, %ymm3, %ymm2, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm3 = ymm11[0,1,0,1,0,1,0,1,14,15,2,3,2,3,14,15,16,17,16,17,16,17,16,17,30,31,18,19,18,19,30,31]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm7, %ymm2, %ymm3, %ymm2
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm2, 64(%r9)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm1, 128(%r9)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, 96(%r9)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm8, 32(%r9)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm9, (%r9)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf16:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm5
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm5[0],xmm6[1],xmm5[1],xmm6[2],xmm5[2],xmm6[3],xmm5[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[0,1,2,3,8,9,10,11,4,5,4,5,6,7,12,13]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm7 = ymm6[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm0[0],xmm6[0],xmm0[1],xmm6[1],xmm0[2],xmm6[2],xmm0[3],xmm6[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[4,5,2,3,2,3,0,1,10,11,8,9,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm10 = <255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255,255,255,0,0,0,0,u,u,255,255>
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm7, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq (%r8), %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm7 = [255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255,255,255,255,255,255,255,0,0,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm1, %ymm8, %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm1 = xmm6[1,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[6,7,u,u,10,11,u,u,8,9,8,9,u,u,12,13]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm1 = xmm5[0],xmm1[1],xmm5[2],xmm1[3],xmm5[4,5],xmm1[6],xmm5[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm5 = mem[2,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[u,u,u,u,6,7,u,u,10,11,u,u,u,u,8,9]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1],xmm0[2],xmm5[3],xmm0[4],xmm5[5,6],xmm0[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq 8(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm8
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm2[0,1,0,1,4,5,4,5]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm4[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,18,19,u,u,22,23,u,u,24,25,20,21,u,u,24,25]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4,5],ymm0[6],ymm1[7,8],ymm0[9],ymm1[10],ymm0[11],ymm1[12,13],ymm0[14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[0,1,2,1,4,5,6,5]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,22,23,18,19,u,u,20,21,u,u,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0,1],ymm1[2],ymm5[3],ymm1[4],ymm5[5,6],ymm1[7],ymm5[8,9],ymm1[10],ymm5[11],ymm1[12],ymm5[13,14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm3[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm12[2,3,2,3,6,7,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm5 = ymm2[2,3,2,3,6,7,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0],ymm1[1],ymm5[2],ymm1[3],ymm5[4,5],ymm1[6],ymm5[7,8],ymm1[9],ymm5[10],ymm1[11],ymm5[12,13],ymm1[14],ymm5[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm11[u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,26,27,u,u,30,31,u,u,u,u,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm6 = ymm3[3,2,3,3,7,6,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm5 = ymm6[0,1],ymm5[2],ymm6[3],ymm5[4],ymm6[5,6],ymm5[7],ymm6[8,9],ymm5[10],ymm6[11],ymm5[12],ymm6[13,14],ymm5[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm1, %ymm5, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpsrldq {{.*#+}} ymm5 = ymm4[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm4[26,27,28,29,30,31],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm1, %ymm5, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,12,13,u,u,0,1,u,u,u,u,14,15,u,u,u,u,28,29,u,u,16,17,u,u,u,u,30,31,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm3[3,0,3,0,7,4,7,4]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3],ymm3[4,5],ymm4[6],ymm3[7,8],ymm4[9],ymm3[10],ymm4[11],ymm3[12,13],ymm4[14],ymm3[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm12[1,1,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,1,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm2 = ymm4[0],ymm2[1,2],ymm4[3],ymm2[4],ymm4[5],ymm2[6,7],ymm4[8],ymm2[9,10],ymm4[11],ymm2[12],ymm4[13],ymm2[14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm10, %ymm3, %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm11[0,1,0,1,0,1,0,1,14,15,2,3,2,3,14,15,16,17,16,17,16,17,16,17,30,31,18,19,18,19,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm7, %ymm2, %ymm3, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, 64(%r9)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 128(%r9)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 96(%r9)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm8, 32(%r9)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm9, (%r9)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <16 x i16>, <16 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <16 x i16>, <16 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <16 x i16>, <16 x i16>* %in.vecptr2, align 32
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-6.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-6.ll
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-6.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-6.ll
@@ -1,6 +1,7 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2-SLOW %s
-; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-shuffle | FileCheck --check-prefixes=AVX2-FAST %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-ALL %s
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2-FAST,AVX2-FAST-PERLANE %s
 
 ; These patterns are produced by LoopVectorizer for interleaved stores.
 
@@ -100,37 +101,67 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf4:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm3 = mem[0],zero
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm4
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm2[0],xmm3[0]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm6 = ymm4[0,1,8,9,u,u,u,u,u,u,u,u,2,3,10,11,18,19,26,27,u,u,u,u,u,u,u,u,20,21,28,29]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [4,6,1,3,4,6,1,3]
-; AVX2-FAST-NEXT:    # ymm7 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm4, %ymm7, %ymm4
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,0,1,4,5,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,28,29,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3,4,5],ymm4[6],ymm6[7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
-; AVX2-FAST-NEXT:    vpbroadcastq %xmm2, %ymm2
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2],ymm4[3,4],ymm2[5],ymm4[6,7]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[6,7,14,15,4,5,6,7,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[12,13,14,15,6,7,14,15,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm1 = xmm5[4,5,12,13,u,u,u,u,u,u,u,u,6,7,14,15]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
-; AVX2-FAST-NEXT:    vmovdqa %xmm0, 32(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm2, (%rax)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf4:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vmovq {{.*#+}} xmm3 = mem[0],zero
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm2[0],xmm3[0]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm6 = ymm4[0,1,8,9,u,u,u,u,u,u,u,u,2,3,10,11,18,19,26,27,u,u,u,u,u,u,u,u,20,21,28,29]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm7 = [4,6,1,3,4,6,1,3]
+; AVX2-FAST-ALL-NEXT:    # ymm7 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm4, %ymm7, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,0,1,4,5,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,24,25,28,29,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3,4,5],ymm4[6],ymm6[7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq %xmm2, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2],ymm4[3,4],ymm2[5],ymm4[6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[6,7,14,15,4,5,6,7,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[12,13,14,15,6,7,14,15,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} xmm1 = xmm5[4,5,12,13,u,u,u,u,u,u,u,u,6,7,14,15]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %xmm0, 32(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm2, (%rax)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf4:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm2 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vmovq {{.*#+}} xmm3 = mem[0],zero
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} xmm5 = xmm2[0],xmm3[0]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm6 = ymm4[0,1,8,9,u,u,u,u,u,u,u,u,2,3,10,11,18,19,26,27,u,u,u,u,u,u,u,u,20,21,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,0,1,8,9,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm4 = ymm6[0],ymm4[1],ymm6[2,3,4,5],ymm4[6],ymm6[7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq %xmm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm2 = ymm4[0,1],ymm2[2],ymm4[3,4],ymm2[5],ymm4[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[6,7,14,15,4,5,6,7,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[12,13,14,15,6,7,14,15,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm1 = xmm5[4,5,12,13,u,u,u,u,u,u,u,u,6,7,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %xmm0, 32(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, (%rax)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <4 x i16>, <4 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <4 x i16>, <4 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <4 x i16>, <4 x i16>* %in.vecptr2, align 32
@@ -191,54 +222,93 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf8:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm1
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm2
-; AVX2-FAST-NEXT:    vinserti128 $1, (%rcx), %ymm1, %ymm1
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,4,1,5,0,4,1,5]
-; AVX2-FAST-NEXT:    # ymm3 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm3, %ymm3
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,0,1,4,5,u,u,u,u,u,u,u,u,18,19,22,23,u,u,u,u,u,u,u,u,24,25,28,29]
-; AVX2-FAST-NEXT:    vinserti128 $1, (%rsi), %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,0,2]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[0,1,8,9,u,u,u,u,u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
-; AVX2-FAST-NEXT:    vinserti128 $1, (%r9), %ymm2, %ymm2
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [0,4,4,0,0,4,4,0]
-; AVX2-FAST-NEXT:    # ymm4 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm4, %ymm4
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,12,13,8,9,u,u,u,u,u,u,u,u,18,19,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,1,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u,u,u,u,u,18,19,26,27]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm5 = ymm2[0,2,1,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[4,5,12,13,u,u,u,u,u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6],ymm4[7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [2,6,1,5,2,6,1,5]
-; AVX2-FAST-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm5, %ymm5
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,10,11,14,15,u,u,u,u,u,u,u,u,16,17,20,21,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,3,1,3]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u,u,u,u,u,22,23,30,31]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [2,6,3,7,2,6,3,7]
-; AVX2-FAST-NEXT:    # ymm5 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm5, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[2,3,6,7,u,u,u,u,u,u,u,u,8,9,12,13,u,u,u,u,u,u,u,u,26,27,30,31,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [7,3,3,7,7,3,3,7]
-; AVX2-FAST-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,8,9,12,13,u,u,u,u,u,u,u,u,22,23,18,19,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, 64(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm4, 32(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm3, (%rax)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf8:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r8), %xmm2
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, (%rcx), %ymm1, %ymm1
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,4,1,5,0,4,1,5]
+; AVX2-FAST-ALL-NEXT:    # ymm3 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm3, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,0,1,4,5,u,u,u,u,u,u,u,u,18,19,22,23,u,u,u,u,u,u,u,u,24,25,28,29]
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, (%rsi), %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,0,2]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[0,1,8,9,u,u,u,u,u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, (%r9), %ymm2, %ymm2
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm4 = [0,4,4,0,0,4,4,0]
+; AVX2-FAST-ALL-NEXT:    # ymm4 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm2, %ymm4, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,12,13,8,9,u,u,u,u,u,u,u,u,18,19,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,1,3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u,u,u,u,u,18,19,26,27]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm5 = ymm2[0,2,1,3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[4,5,12,13,u,u,u,u,u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6],ymm4[7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [2,6,1,5,2,6,1,5]
+; AVX2-FAST-ALL-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm5, %ymm5
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,10,11,14,15,u,u,u,u,u,u,u,u,16,17,20,21,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,3,1,3]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u,u,u,u,u,22,23,30,31]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm5 = [2,6,3,7,2,6,3,7]
+; AVX2-FAST-ALL-NEXT:    # ymm5 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm5, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[2,3,6,7,u,u,u,u,u,u,u,u,8,9,12,13,u,u,u,u,u,u,u,u,26,27,30,31,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [7,3,3,7,7,3,3,7]
+; AVX2-FAST-ALL-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,8,9,12,13,u,u,u,u,u,u,u,u,22,23,18,19,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, 64(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm4, 32(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm3, (%rax)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf8:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, (%rcx), %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm1[0,2,0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm3 = ymm3[u,u,u,u,0,1,8,9,u,u,u,u,u,u,u,u,18,19,26,27,u,u,u,u,u,u,u,u,20,21,28,29]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, (%rsi), %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[0,1,8,9,u,u,u,u,u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm3 = ymm4[0],ymm3[1],ymm4[2,3],ymm3[4],ymm4[5,6],ymm3[7]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, (%r9), %ymm2, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm2[0,2,0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,u,u,u,u,0,1,8,9,u,u,u,u,u,u,u,u,18,19,26,27,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2],ymm3[3,4],ymm4[5],ymm3[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm0[0,2,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm4 = ymm4[u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u,u,u,u,u,18,19,26,27]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm2[0,2,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[4,5,12,13,u,u,u,u,u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm4 = ymm5[0],ymm4[1],ymm5[2,3],ymm4[4],ymm5[5,6],ymm4[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm1[0,2,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[u,u,u,u,u,u,u,u,6,7,14,15,u,u,u,u,u,u,u,u,16,17,24,25,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[1,3,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[u,u,u,u,2,3,10,11,u,u,u,u,u,u,u,u,20,21,28,29,u,u,u,u,u,u,u,u,22,23,30,31]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[1,3,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[2,3,10,11,u,u,u,u,u,u,u,u,4,5,12,13,u,u,u,u,u,u,u,u,22,23,30,31,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[1,3,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,4,5,12,13,u,u,u,u,u,u,u,u,22,23,30,31,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 64(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm4, 32(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm3, (%rax)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <8 x i16>, <8 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <8 x i16>, <8 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <8 x i16>, <8 x i16>* %in.vecptr2, align 32
@@ -356,98 +426,192 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: vf16:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm10
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %ymm11
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %ymm14
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %ymm13
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %ymm15
-; AVX2-FAST-NEXT:    vmovdqa (%rsi), %xmm8
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,6,7,8,9,6,7,4,5,10,11]
-; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm8, %xmm1
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm7
-; AVX2-FAST-NEXT:    vpshufb %xmm0, %xmm7, %xmm0
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[0,1,0,1]
-; AVX2-FAST-NEXT:    vmovdqa (%r9), %xmm5
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[1,2,2,3]
-; AVX2-FAST-NEXT:    vmovdqa (%r8), %xmm0
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,2,2,3]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %xmm1
-; AVX2-FAST-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vmovdqa (%rdx), %xmm4
-; AVX2-FAST-NEXT:    vpsrldq {{.*#+}} xmm6 = xmm4[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3]
-; AVX2-FAST-NEXT:    vpbroadcastq %xmm3, %ymm3
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
-; AVX2-FAST-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm4[4],xmm1[4],xmm4[5],xmm1[5],xmm4[6],xmm1[6],xmm4[7],xmm1[7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm6 = <1,u,u,2,u,u,3,u>
-; AVX2-FAST-NEXT:    vpermd %ymm3, %ymm6, %ymm3
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0],ymm6[1],ymm3[2,3],ymm6[4],ymm3[5,6],ymm6[7]
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[1,1,1,1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm9 = ymm3[0,1],ymm6[2],ymm3[3,4],ymm6[5],ymm3[6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,4,5,10,11,u,u,u,u,u,u,u,u,24,25,22,23,20,21,26,27,u,u,u,u,u,u,u,u>
-; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm11, %ymm6
-; AVX2-FAST-NEXT:    vpshufb %ymm3, %ymm10, %ymm3
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm3[0],ymm6[0],ymm3[1],ymm6[1],ymm3[2],ymm6[2],ymm3[3],ymm6[3],ymm3[8],ymm6[8],ymm3[9],ymm6[9],ymm3[10],ymm6[10],ymm3[11],ymm6[11]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm6 = ymm15[1,2,2,3,5,6,6,7]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm12 = ymm13[1,2,2,3,5,6,6,7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm12[0],ymm6[0],ymm12[1],ymm6[1],ymm12[2],ymm6[2],ymm12[3],ymm6[3],ymm12[8],ymm6[8],ymm12[9],ymm6[9],ymm12[10],ymm6[10],ymm12[11],ymm6[11]
-; AVX2-FAST-NEXT:    vmovdqa (%rcx), %ymm12
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,2,3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,2,2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2,3],ymm3[4],ymm6[5,6],ymm3[7]
-; AVX2-FAST-NEXT:    vpsrldq {{.*#+}} ymm6 = ymm12[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,ymm12[22,23,24,25,26,27,28,29,30,31],zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpsrldq {{.*#+}} ymm2 = ymm14[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,ymm14[22,23,24,25,26,27,28,29,30,31],zero,zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm6[0],ymm2[1],ymm6[1],ymm2[2],ymm6[2],ymm2[3],ymm6[3],ymm2[8],ymm6[8],ymm2[9],ymm6[9],ymm2[10],ymm6[10],ymm2[11],ymm6[11]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [1,0,2,2,1,0,2,2]
-; AVX2-FAST-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3]
-; AVX2-FAST-NEXT:    vpbroadcastq %xmm0, %ymm0
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm14[0],ymm12[0],ymm14[1],ymm12[1],ymm14[2],ymm12[2],ymm14[3],ymm12[3],ymm14[8],ymm12[8],ymm14[9],ymm12[9],ymm14[10],ymm12[10],ymm14[11],ymm12[11]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,4,u,u,5,u,u,6>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm10[0],ymm11[0],ymm10[1],ymm11[1],ymm10[2],ymm11[2],ymm10[3],ymm11[3],ymm10[8],ymm11[8],ymm10[9],ymm11[9],ymm10[10],ymm11[10],ymm10[11],ymm11[11]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0],ymm15[0],ymm13[1],ymm15[1],ymm13[2],ymm15[2],ymm13[3],ymm15[3],ymm13[8],ymm15[8],ymm13[9],ymm15[9],ymm13[10],ymm15[10],ymm13[11],ymm15[11]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm14[4],ymm12[4],ymm14[5],ymm12[5],ymm14[6],ymm12[6],ymm14[7],ymm12[7],ymm14[12],ymm12[12],ymm14[13],ymm12[13],ymm14[14],ymm12[14],ymm14[15],ymm12[15]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm4 = <5,u,u,6,u,u,7,u>
-; AVX2-FAST-NEXT:    vpermd %ymm2, %ymm4, %ymm2
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm13[4],ymm15[4],ymm13[5],ymm15[5],ymm13[6],ymm15[6],ymm13[7],ymm15[7],ymm13[12],ymm15[12],ymm13[13],ymm15[13],ymm13[14],ymm15[14],ymm13[15],ymm15[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3],ymm4[4],ymm2[5,6],ymm4[7]
-; AVX2-FAST-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm10[4],ymm11[4],ymm10[5],ymm11[5],ymm10[6],ymm11[6],ymm10[7],ymm11[7],ymm10[12],ymm11[12],ymm10[13],ymm11[13],ymm10[14],ymm11[14],ymm10[15],ymm11[15]
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[3,3,3,3]
-; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm4[2],ymm2[3,4],ymm4[5],ymm2[6,7]
-; AVX2-FAST-NEXT:    movq {{[0-9]+}}(%rsp), %rax
-; AVX2-FAST-NEXT:    vmovdqa %ymm2, 160(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm1, 96(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm0, (%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm3, 128(%rax)
-; AVX2-FAST-NEXT:    vmovdqa %ymm9, 64(%rax)
-; AVX2-FAST-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
-; AVX2-FAST-NEXT:    vmovaps %ymm0, 32(%rax)
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: vf16:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm10
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rsi), %ymm11
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %ymm14
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r8), %ymm13
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r9), %ymm15
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rsi), %xmm8
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,6,7,8,9,6,7,4,5,10,11]
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm0, %xmm8, %xmm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %xmm7
+; AVX2-FAST-ALL-NEXT:    vpshufb %xmm0, %xmm7, %xmm0
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r9), %xmm5
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm2 = xmm5[1,2,2,3]
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%r8), %xmm0
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,2,2,3]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rcx), %xmm1
+; AVX2-FAST-ALL-NEXT:    vpsrldq {{.*#+}} xmm3 = xmm1[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdx), %xmm4
+; AVX2-FAST-ALL-NEXT:    vpsrldq {{.*#+}} xmm6 = xmm4[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm3 = xmm6[0],xmm3[0],xmm6[1],xmm3[1],xmm6[2],xmm3[2],xmm6[3],xmm3[3]
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq %xmm3, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqu %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm3 = xmm4[4],xmm1[4],xmm4[5],xmm1[5],xmm4[6],xmm1[6],xmm4[7],xmm1[7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm6 = <1,u,u,2,u,u,3,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm3, %ymm6, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm0[4],xmm5[4],xmm0[5],xmm5[5],xmm0[6],xmm5[6],xmm0[7],xmm5[7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[0,1,1,1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0],ymm6[1],ymm3[2,3],ymm6[4],ymm3[5,6],ymm6[7]
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} xmm6 = xmm7[4],xmm8[4],xmm7[5],xmm8[5],xmm7[6],xmm8[6],xmm7[7],xmm8[7]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[1,1,1,1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm9 = ymm3[0,1],ymm6[2],ymm3[3,4],ymm6[5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm3 = <u,u,u,u,4,5,10,11,u,u,u,u,u,u,u,u,24,25,22,23,20,21,26,27,u,u,u,u,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm3, %ymm11, %ymm6
+; AVX2-FAST-ALL-NEXT:    vpshufb %ymm3, %ymm10, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm3[0],ymm6[0],ymm3[1],ymm6[1],ymm3[2],ymm6[2],ymm3[3],ymm6[3],ymm3[8],ymm6[8],ymm3[9],ymm6[9],ymm3[10],ymm6[10],ymm3[11],ymm6[11]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm6 = ymm15[1,2,2,3,5,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm12 = ymm13[1,2,2,3,5,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm6 = ymm12[0],ymm6[0],ymm12[1],ymm6[1],ymm12[2],ymm6[2],ymm12[3],ymm6[3],ymm12[8],ymm6[8],ymm12[9],ymm6[9],ymm12[10],ymm6[10],ymm12[11],ymm6[11]
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rcx), %ymm12
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,1,2,3]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm6 = ymm6[2,2,2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm3 = ymm6[0],ymm3[1],ymm6[2,3],ymm3[4],ymm6[5,6],ymm3[7]
+; AVX2-FAST-ALL-NEXT:    vpsrldq {{.*#+}} ymm6 = ymm12[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,ymm12[22,23,24,25,26,27,28,29,30,31],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vpsrldq {{.*#+}} ymm2 = ymm14[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,ymm14[22,23,24,25,26,27,28,29,30,31],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm6[0],ymm2[1],ymm6[1],ymm2[2],ymm6[2],ymm2[3],ymm6[3],ymm2[8],ymm6[8],ymm2[9],ymm6[9],ymm2[10],ymm6[10],ymm2[11],ymm6[11]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm2[2],ymm3[3,4],ymm2[5],ymm3[6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm4[0],xmm1[0],xmm4[1],xmm1[1],xmm4[2],xmm1[2],xmm4[3],xmm1[3]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm2 = [1,0,2,2,1,0,2,2]
+; AVX2-FAST-ALL-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm7[0],xmm8[0],xmm7[1],xmm8[1],xmm7[2],xmm8[2],xmm7[3],xmm8[3]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm5[0],xmm0[1],xmm5[1],xmm0[2],xmm5[2],xmm0[3],xmm5[3]
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq %xmm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm14[0],ymm12[0],ymm14[1],ymm12[1],ymm14[2],ymm12[2],ymm14[3],ymm12[3],ymm14[8],ymm12[8],ymm14[9],ymm12[9],ymm14[10],ymm12[10],ymm14[11],ymm12[11]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,4,u,u,5,u,u,6>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm10[0],ymm11[0],ymm10[1],ymm11[1],ymm10[2],ymm11[2],ymm10[3],ymm11[3],ymm10[8],ymm11[8],ymm10[9],ymm11[9],ymm10[10],ymm11[10],ymm10[11],ymm11[11]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm13[0],ymm15[0],ymm13[1],ymm15[1],ymm13[2],ymm15[2],ymm13[3],ymm15[3],ymm13[8],ymm15[8],ymm13[9],ymm15[9],ymm13[10],ymm15[10],ymm13[11],ymm15[11]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,2]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm14[4],ymm12[4],ymm14[5],ymm12[5],ymm14[6],ymm12[6],ymm14[7],ymm12[7],ymm14[12],ymm12[12],ymm14[13],ymm12[13],ymm14[14],ymm12[14],ymm14[15],ymm12[15]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm4 = <5,u,u,6,u,u,7,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm2, %ymm4, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm13[4],ymm15[4],ymm13[5],ymm15[5],ymm13[6],ymm15[6],ymm13[7],ymm15[7],ymm13[12],ymm15[12],ymm13[13],ymm15[13],ymm13[14],ymm15[14],ymm13[15],ymm15[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[2,1,3,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0],ymm4[1],ymm2[2,3],ymm4[4],ymm2[5,6],ymm4[7]
+; AVX2-FAST-ALL-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm10[4],ymm11[4],ymm10[5],ymm11[5],ymm10[6],ymm11[6],ymm10[7],ymm11[7],ymm10[12],ymm11[12],ymm10[13],ymm11[13],ymm10[14],ymm11[14],ymm10[15],ymm11[15]
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[3,3,3,3]
+; AVX2-FAST-ALL-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm4[2],ymm2[3,4],ymm4[5],ymm2[6,7]
+; AVX2-FAST-ALL-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm2, 160(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm1, 96(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm0, (%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm3, 128(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovdqa %ymm9, 64(%rax)
+; AVX2-FAST-ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-ALL-NEXT:    vmovaps %ymm0, 32(%rax)
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: vf16:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm13
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %ymm15
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %ymm11
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %ymm12
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %ymm14
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %ymm9
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rsi), %xmm6
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} xmm0 = [0,1,2,3,4,5,6,7,8,9,6,7,4,5,10,11]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm6, %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %xmm0, %xmm1, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm0 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm8 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r9), %xmm7
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm2 = xmm7[1,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%r8), %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm3 = xmm0[1,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm3[0],xmm2[0],xmm3[1],xmm2[1],xmm3[2],xmm2[2],xmm3[3],xmm2[3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm8 = ymm2[0],ymm8[1],ymm2[2,3],ymm8[4],ymm2[5,6],ymm8[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rcx), %xmm3
+; AVX2-FAST-PERLANE-NEXT:    vpsrldq {{.*#+}} xmm4 = xmm3[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdx), %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpsrldq {{.*#+}} xmm5 = xmm2[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm4 = xmm5[0],xmm4[0],xmm5[1],xmm4[1],xmm5[2],xmm4[2],xmm5[3],xmm4[3]
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq %xmm4, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm4 = ymm8[0,1],ymm4[2],ymm8[3,4],ymm4[5],ymm8[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqu %ymm4, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm0[4],xmm7[4],xmm0[5],xmm7[5],xmm0[6],xmm7[6],xmm0[7],xmm7[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm5 = xmm5[2,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm4 = xmm2[4],xmm3[4],xmm2[5],xmm3[5],xmm2[6],xmm3[6],xmm2[7],xmm3[7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm4 = xmm4[1,2,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm4 = ymm4[0,0,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm4 = ymm4[0],ymm5[1],ymm4[2,3],ymm5[4],ymm4[5,6],ymm5[7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} xmm5 = xmm1[4],xmm6[4],xmm1[5],xmm6[5],xmm1[6],xmm6[6],xmm1[7],xmm6[7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm5 = ymm5[1,1,1,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm10 = ymm4[0,1],ymm5[2],ymm4[3,4],ymm5[5],ymm4[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm6[0],xmm1[1],xmm6[1],xmm1[2],xmm6[2],xmm1[3],xmm6[3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1],xmm2[2],xmm3[2],xmm2[3],xmm3[3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm2 = xmm2[1,0,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3],ymm2[4],ymm1[5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm7[0],xmm0[1],xmm7[1],xmm0[2],xmm7[2],xmm0[3],xmm7[3]
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq %xmm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm8 = ymm1[0,1],ymm0[2],ymm1[3,4],ymm0[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm13[0],ymm15[0],ymm13[1],ymm15[1],ymm13[2],ymm15[2],ymm13[3],ymm15[3],ymm13[8],ymm15[8],ymm13[9],ymm15[9],ymm13[10],ymm15[10],ymm13[11],ymm15[11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm11[0],ymm12[0],ymm11[1],ymm12[1],ymm11[2],ymm12[2],ymm11[3],ymm12[3],ymm11[8],ymm12[8],ymm11[9],ymm12[9],ymm11[10],ymm12[10],ymm11[11],ymm12[11]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[1,0,2,2,5,4,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3],ymm1[4],ymm0[5,6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm14[0],ymm9[0],ymm14[1],ymm9[1],ymm14[2],ymm9[2],ymm14[3],ymm9[3],ymm14[8],ymm9[8],ymm14[9],ymm9[9],ymm14[10],ymm9[10],ymm14[11],ymm9[11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2],ymm0[3,4],ymm1[5],ymm0[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm1 = ymm14[4],ymm9[4],ymm14[5],ymm9[5],ymm14[6],ymm9[6],ymm14[7],ymm9[7],ymm14[12],ymm9[12],ymm14[13],ymm9[13],ymm14[14],ymm9[14],ymm14[15],ymm9[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,1,2,3,6,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm11[4],ymm12[4],ymm11[5],ymm12[5],ymm11[6],ymm12[6],ymm11[7],ymm12[7],ymm11[12],ymm12[12],ymm11[13],ymm12[13],ymm11[14],ymm12[14],ymm11[15],ymm12[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm2 = ymm2[1,2,3,3,5,6,7,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0],ymm1[1],ymm2[2,3],ymm1[4],ymm2[5,6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpckhwd {{.*#+}} ymm2 = ymm13[4],ymm15[4],ymm13[5],ymm15[5],ymm13[6],ymm15[6],ymm13[7],ymm15[7],ymm13[12],ymm15[12],ymm13[13],ymm15[13],ymm13[14],ymm15[14],ymm13[15],ymm15[15]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[3,3,3,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm2[2],ymm1[3,4],ymm2[5],ymm1[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = <u,u,u,u,4,5,10,11,u,u,u,u,u,u,u,u,24,25,22,23,20,21,26,27,u,u,u,u,u,u,u,u>
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm15, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm13, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm3[0],ymm2[1],ymm3[1],ymm2[2],ymm3[2],ymm2[3],ymm3[3],ymm2[8],ymm3[8],ymm2[9],ymm3[9],ymm2[10],ymm3[10],ymm2[11],ymm3[11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[2,1,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm3 = ymm9[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm4 = ymm14[1,2,2,3,5,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2,3],ymm2[4],ymm3[5,6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vpsrldq {{.*#+}} ymm3 = ymm12[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,ymm12[22,23,24,25,26,27,28,29,30,31],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpsrldq {{.*#+}} ymm4 = ymm11[6,7,8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,ymm11[22,23,24,25,26,27,28,29,30,31],zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm3 = ymm3[2,2,2,2]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2],ymm2[3,4],ymm3[5],ymm2[6,7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm2, 128(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm1, 160(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm0, 96(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm8, (%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa %ymm10, 64(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
+; AVX2-FAST-PERLANE-NEXT:    vmovaps %ymm0, 32(%rax)
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
   %in.vec0 = load <16 x i16>, <16 x i16>* %in.vecptr0, align 32
   %in.vec1 = load <16 x i16>, <16 x i16>* %in.vecptr1, align 32
   %in.vec2 = load <16 x i16>, <16 x i16>* %in.vecptr2, align 32
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll
@@ -1,9 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=ALL,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL
 
 define <2 x i64> @unpckh_unary_extracted_v4i64(<4 x i64> %x) {
 ; AVX1-LABEL: unpckh_unary_extracted_v4i64:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-v16.ll
@@ -4,9 +4,12 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=ALL,SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX2OR512VL,AVX512VL,AVX512VLBW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+avx512vbmi,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX2OR512VL,AVX512VL,AVX512VLVBMI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX2OR512VL,AVX512VL,AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX2OR512VL,AVX512VL,AVX512VLBW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+avx512vbmi,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX2OR512VL,AVX512VL,AVX512VLVBMI
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+avx512vbmi,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX,AVX2OR512VL,AVX512VL,AVX512VLVBMI
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,XOP,XOPAVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=ALL,AVX,AVX1OR2,XOP,XOPAVX2
 
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-v4.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-v4.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-v4.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-v4.ll
@@ -5,8 +5,10 @@
 ; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX1
 ; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-SLOW
-; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
-; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL
+; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL
 
 define <4 x i32> @shuffle_v4i32_0001(<4 x i32> %a, <4 x i32> %b) {
 ; SSE-LABEL: shuffle_v4i32_0001:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-v8.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-v8.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-128-v8.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-128-v8.ll
@@ -4,9 +4,11 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX2,AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL,AVX512VL-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL,AVX512VL-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL,AVX512VL-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2OR512VL,AVX512VL,AVX512VL-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=AVX,XOP,XOPAVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=AVX,XOP,XOPAVX2
 
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v16.ll
@@ -1,9 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=ALL,AVX1OR2,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,AVX1OR2,AVX2OR512VL,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX1OR2,AVX2OR512VL,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VL-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VL-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VL-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VL-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=ALL,XOP,XOPAVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=ALL,XOP,XOPAVX2
 
@@ -52,12 +54,18 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,1,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,1,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
 ; AVX512VL:       # %bb.0:
@@ -101,12 +109,18 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,2,4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,2,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
 ; AVX512VL:       # %bb.0:
@@ -150,12 +164,18 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,3,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,3,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
 ; AVX512VL:       # %bb.0:
@@ -987,13 +1007,19 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_00_00_01_00_00_00_00_00_00_00_01_00:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,1,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,1,0,0,0,1]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_00_00_01_00_00_00_00_00_00_00_01_00:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,1,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,1,0,0,0,1]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_00_00_01_00_00_00_00_00_00_00_01_00:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,0,1,2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_00_00_01_00_00_00_00_00_00_00_01_00:
 ; AVX512VL:       # %bb.0:
@@ -1034,13 +1060,19 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_00_02_00_00_00_00_00_00_00_02_00_00:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,2,4,5,6,7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,1,0,0,0,1,0]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_00_02_00_00_00_00_00_00_00_02_00_00:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,2,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,1,0,0,0,1,0]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_00_02_00_00_00_00_00_00_00_02_00_00:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,0,1,4,5,0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_00_02_00_00_00_00_00_00_00_02_00_00:
 ; AVX512VL:       # %bb.0:
@@ -1081,13 +1113,19 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_03_00_00_00_00_00_00_00_03_00_00_00:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,3,0,4,5,6,7]
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,1,0,0,0,1,0]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_03_00_00_00_00_00_00_00_03_00_00_00:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,3,0,4,5,6,7]
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,1,0,0,0,1,0]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_03_00_00_00_00_00_00_00_03_00_00_00:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,0,1,0,1,0,1,6,7,0,1,0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_03_00_00_00_00_00_00_00_03_00_00_00:
 ; AVX512VL:       # %bb.0:
@@ -3482,12 +3520,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_00_00_04_04_04_12_08_08_08_08_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,4,5,4,5,4,5,8,9,16,17,16,17,16,17,16,17,20,21,20,21,20,21,20,21]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_00_00_04_04_04_12_08_08_08_08_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,4,5,4,5,4,5,8,9,16,17,16,17,16,17,16,17,20,21,20,21,20,21,20,21]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_00_00_04_04_04_12_08_08_08_08_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,0,1,0,1,0,1,8,9,8,9,8,9,u,u,16,17,16,17,16,17,16,17,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_00_00_04_04_04_12_08_08_08_08_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -3667,12 +3713,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_04_04_04_04_00_00_00_08_12_12_12_12_08_08_08_08:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,4,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,4,5,4,5,4,5,8,9,16,17,16,17,16,17,16,17,20,21,20,21,20,21,20,21]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_04_04_04_04_00_00_00_08_12_12_12_12_08_08_08_08:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,4,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,0,1,0,1,4,5,4,5,4,5,8,9,16,17,16,17,16,17,16,17,20,21,20,21,20,21,20,21]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_04_04_04_04_00_00_00_08_12_12_12_12_08_08_08_08:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[8,9,8,9,8,9,8,9,0,1,0,1,0,1,u,u,24,25,24,25,24,25,24,25,16,17,16,17,16,17,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpslldq {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0,1],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_04_04_04_04_00_00_00_08_12_12_12_12_08_08_08_08:
 ; AVX512VL:       # %bb.0:
@@ -3896,12 +3950,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_01_00_05_04_05_04_01_08_09_08_13_12_13_12_09_08:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,6,7,4,5,2,3,8,9,18,19,16,17,22,23,20,21,22,23,20,21,18,19,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_01_00_05_04_05_04_01_08_09_08_13_12_13_12_09_08:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,6,7,4,5,2,3,8,9,18,19,16,17,22,23,20,21,22,23,20,21,18,19,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_01_00_05_04_05_04_01_08_09_08_13_12_13_12_09_08:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[2,3,0,1,10,11,8,9,10,11,8,9,2,3,u,u,18,19,16,17,26,27,24,25,26,27,24,25,18,19,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpslldq {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0,1],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_01_00_05_04_05_04_01_08_09_08_13_12_13_12_09_08:
 ; AVX512VL:       # %bb.0:
@@ -3947,12 +4009,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_05_04_01_00_05_04_01_08_13_12_09_08_13_12_09_08:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,4,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,2,3,0,1,6,7,8,9,18,19,16,17,22,23,20,21,18,19,16,17,22,23,20,21]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_05_04_01_00_05_04_01_08_13_12_09_08_13_12_09_08:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,4,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,2,3,0,1,6,7,8,9,18,19,16,17,22,23,20,21,18,19,16,17,22,23,20,21]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_05_04_01_00_05_04_01_08_13_12_09_08_13_12_09_08:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[10,11,8,9,2,3,0,1,10,11,8,9,2,3,u,u,26,27,24,25,18,19,16,17,26,27,24,25,18,19,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpslldq {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0,1],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_05_04_01_00_05_04_01_08_13_12_09_08_13_12_09_08:
 ; AVX512VL:       # %bb.0:
@@ -3998,12 +4068,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_05_04_01_00_01_00_05_12_13_12_09_08_09_08_13_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,6,7,4,5,2,3,8,9,18,19,16,17,22,23,20,21,22,23,20,21,18,19,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_05_04_01_00_01_00_05_12_13_12_09_08_09_08_13_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5,6,7,4,5,2,3,8,9,18,19,16,17,22,23,20,21,22,23,20,21,18,19,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_05_04_01_00_01_00_05_12_13_12_09_08_09_08_13_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[10,11,8,9,2,3,0,1,2,3,0,1,10,11,u,u,26,27,24,25,18,19,16,17,18,19,16,17,26,27,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_05_04_01_00_01_00_05_12_13_12_09_08_09_08_13_12:
 ; AVX512VL:       # %bb.0:
@@ -4049,12 +4127,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_04_04_00_00_04_04_08_08_12_12_08_08_12_12_08:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,0,1,4,5,4,5,8,9,16,17,20,21,20,21,16,17,16,17,20,21,20,21,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_04_04_00_00_04_04_08_08_12_12_08_08_12_12_08:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,0,1,4,5,4,5,8,9,16,17,20,21,20,21,16,17,16,17,20,21,20,21,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_04_04_00_00_04_04_08_08_12_12_08_08_12_12_08:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,8,9,8,9,0,1,0,1,8,9,8,9,u,u,16,17,24,25,24,25,16,17,16,17,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpslldq {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0,1],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_04_04_00_00_04_04_08_08_12_12_08_08_12_12_08:
 ; AVX512VL:       # %bb.0:
@@ -4100,12 +4186,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_04_00_00_04_04_00_00_12_12_08_08_12_12_08_08_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,0,1,4,5,4,5,8,9,16,17,20,21,20,21,16,17,16,17,20,21,20,21,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_04_00_00_04_04_00_00_12_12_08_08_12_12_08_08_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,0,1,4,5,4,5,8,9,16,17,20,21,20,21,16,17,16,17,20,21,20,21,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_04_00_00_04_04_00_00_12_12_08_08_12_12_08_08_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[8,9,0,1,0,1,8,9,8,9,0,1,0,1,u,u,24,25,16,17,16,17,24,25,24,25,16,17,16,17,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_04_00_00_04_04_00_00_12_12_08_08_12_12_08_08_12:
 ; AVX512VL:       # %bb.0:
@@ -4276,12 +4370,19 @@
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[12,13,12,13,14,15,10,11,2,3,12,13,8,9,6,7,28,29,28,29,30,31,26,27,18,19,28,29,24,25,22,23]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_06_06_07_05_01_06_04_11_14_14_15_13_09_14_12_11:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [3,2,0,5,7,6,4,5]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,2,3,6,7,10,11,0,1,4,5,14,15,16,17,16,17,18,19,22,23,26,27,16,17,20,21,30,31]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_06_06_07_05_01_06_04_11_14_14_15_13_09_14_12_11:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [3,2,0,5,7,6,4,5]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,2,3,6,7,10,11,0,1,4,5,14,15,16,17,16,17,18,19,22,23,26,27,16,17,20,21,30,31]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_06_06_07_05_01_06_04_11_14_14_15_13_09_14_12_11:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,5,2,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[12,13,12,13,14,15,10,11,2,3,12,13,8,9,6,7,28,29,28,29,30,31,26,27,18,19,28,29,24,25,22,23]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_06_06_07_05_01_06_04_11_14_14_15_13_09_14_12_11:
 ; AVX512VL:       # %bb.0:
@@ -4326,12 +4427,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_00_04_04_04_04_04_12_08_08_12_12_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5,4,5,4,5,4,5,8,9,16,17,16,17,20,21,20,21,20,21,20,21,20,21,20,21]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_00_04_04_04_04_04_12_08_08_12_12_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5,4,5,4,5,4,5,8,9,16,17,16,17,20,21,20,21,20,21,20,21,20,21,20,21]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_00_04_04_04_04_04_12_08_08_12_12_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,0,1,8,9,8,9,8,9,8,9,8,9,u,u,16,17,16,17,24,25,24,25,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_00_04_04_04_04_04_12_08_08_12_12_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -4377,12 +4486,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_04_04_00_00_04_04_04_12_12_12_08_08_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5,0,1,0,1,0,1,8,9,16,17,16,17,20,21,20,21,16,17,16,17,16,17,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_04_04_00_00_04_04_04_12_12_12_08_08_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5,0,1,0,1,0,1,8,9,16,17,16,17,20,21,20,21,16,17,16,17,16,17,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_04_04_00_00_04_04_04_12_12_12_08_08_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[8,9,8,9,0,1,0,1,8,9,8,9,8,9,u,u,24,25,24,25,16,17,16,17,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_04_04_00_00_04_04_04_12_12_12_08_08_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -4428,12 +4545,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_04_04_00_04_04_04_12_08_12_12_08_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,4,5,4,5,4,5,8,9,16,17,20,21,20,21,16,17,20,21,20,21,20,21,20,21]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_04_04_00_04_04_04_12_08_12_12_08_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,4,5,4,5,4,5,8,9,16,17,20,21,20,21,16,17,20,21,20,21,20,21,20,21]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_04_04_00_04_04_04_12_08_12_12_08_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,8,9,8,9,0,1,8,9,8,9,8,9,u,u,16,17,24,25,24,25,16,17,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_04_04_00_04_04_04_12_08_12_12_08_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -4479,12 +4604,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_04_04_00_00_00_00_08_08_12_12_08_08_08_08_08:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,0,1,0,1,0,1,8,9,16,17,20,21,20,21,16,17,16,17,16,17,16,17,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_04_04_00_00_00_00_08_08_12_12_08_08_08_08_08:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,0,1,0,1,0,1,8,9,16,17,20,21,20,21,16,17,16,17,16,17,16,17,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_04_04_00_00_00_00_08_08_12_12_08_08_08_08_08:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,8,9,8,9,0,1,0,1,0,1,0,1,u,u,16,17,24,25,24,25,16,17,16,17,16,17,16,17,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpslldq {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0,1],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_04_04_00_00_00_00_08_08_12_12_08_08_08_08_08:
 ; AVX512VL:       # %bb.0:
@@ -4531,12 +4664,19 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_04_04_00_04_05_06_15_08_12_12_08_12_13_14_15:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,3,7,4,6,7,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,4,5,6,7,8,9,14,15,16,17,20,21,20,21,16,17,20,21,22,23,24,25,26,27]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_04_04_00_04_05_06_15_08_12_12_08_12_13_14_15:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,3,7,4,6,7,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,4,5,0,1,4,5,6,7,8,9,14,15,16,17,20,21,20,21,16,17,20,21,22,23,24,25,26,27]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_04_04_00_04_05_06_15_08_12_12_08_12_13_14_15:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,8,9,8,9,0,1,8,9,10,11,12,13,u,u,16,17,24,25,24,25,16,17,24,25,26,27,28,29,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_04_04_00_04_05_06_15_08_12_12_08_12_13_14_15:
 ; AVX512VL:       # %bb.0:
@@ -4583,12 +4723,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_uu_04_04_04_04_04_12_08_uu_12_12_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,u,u,4,5,4,5,4,5,4,5,4,5,8,9,16,17,u,u,20,21,20,21,20,21,20,21,20,21,20,21]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_uu_04_04_04_04_04_12_08_uu_12_12_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,6,u,4,6,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,u,u,4,5,4,5,4,5,4,5,4,5,8,9,16,17,u,u,20,21,20,21,20,21,20,21,20,21,20,21]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_uu_04_04_04_04_04_12_08_uu_12_12_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,2,3,8,9,8,9,8,9,8,9,8,9,u,u,16,17,18,19,24,25,24,25,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_uu_04_04_04_04_04_12_08_uu_12_12_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -4634,12 +4782,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_04_04_uu_00_04_04_04_12_12_12_uu_08_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,u,u,4,5,0,1,0,1,0,1,8,9,16,17,16,17,u,u,20,21,16,17,16,17,16,17,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_04_04_uu_00_04_04_04_12_12_12_uu_08_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,0,1,u,u,4,5,0,1,0,1,0,1,8,9,16,17,16,17,u,u,20,21,16,17,16,17,16,17,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_04_04_uu_00_04_04_04_12_12_12_uu_08_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[8,9,8,9,8,9,0,1,8,9,8,9,8,9,u,u,24,25,24,25,24,25,16,17,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_04_04_uu_00_04_04_04_12_12_12_uu_08_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -4685,12 +4841,20 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_uu_04_04_00_04_04_04_12_uu_12_12_08_12_12_12_12:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,0,1,0,1,4,5,0,1,0,1,0,1,8,9,u,u,16,17,16,17,20,21,16,17,16,17,16,17,16,17]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_uu_04_04_00_04_04_04_12_uu_12_12_08_12_12_12_12:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <2,0,6,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,0,1,0,1,4,5,0,1,0,1,0,1,8,9,u,u,16,17,16,17,20,21,16,17,16,17,16,17,16,17]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_uu_04_04_00_04_04_04_12_uu_12_12_08_12_12_12_12:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm0[0,1,8,9,8,9,0,1,8,9,8,9,8,9,u,u,16,17,24,25,24,25,16,17,24,25,24,25,24,25,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpsllq $48, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0,1,2,3,4,5,6],ymm0[7],ymm1[8,9,10,11,12,13,14],ymm0[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_uu_04_04_00_04_04_04_12_uu_12_12_08_12_12_12_12:
 ; AVX512VL:       # %bb.0:
@@ -4952,12 +5116,19 @@
 ; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[6,7,14,15,2,3,0,1,4,5,14,15,6,7,10,11,22,23,30,31,18,19,16,17,20,21,30,31,22,23,26,27]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_03_07_01_00_02_07_03_13_11_15_09_08_10_15_11_13:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,0,6,5,7,4,6]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,6,7,10,11,8,9,0,1,6,7,2,3,14,15,18,19,22,23,26,27,24,25,16,17,22,23,18,19,30,31]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_03_07_01_00_02_07_03_13_11_15_09_08_10_15_11_13:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,0,6,5,7,4,6]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,6,7,10,11,8,9,0,1,6,7,2,3,14,15,18,19,22,23,26,27,24,25,16,17,22,23,18,19,30,31]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_03_07_01_00_02_07_03_13_11_15_09_08_10_15_11_13:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,1,6,3,4,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[6,7,14,15,2,3,0,1,4,5,14,15,6,7,10,11,22,23,30,31,18,19,16,17,20,21,30,31,22,23,26,27]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_03_07_01_00_02_07_03_13_11_15_09_08_10_15_11_13:
 ; AVX512VL:       # %bb.0:
@@ -5197,14 +5368,23 @@
 ; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_16_01_17_06_22_07_31_08_24_09_25_14_30_15_31:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,3,7,u,4,7,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,10,11,u,u,u,u,u,u,u,u,16,17,18,19,20,21,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,3,2,3,4,7,6,7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_16_01_17_06_22_07_31_08_24_09_25_14_30_15_31:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,3,7,u,4,7,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,10,11,u,u,u,u,u,u,u,u,16,17,18,19,20,21,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,3,2,3,4,7,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_16_01_17_06_22_07_31_08_24_09_25_14_30_15_31:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,6],ymm2[7],ymm1[8,9,10,11,12,13,14],ymm2[15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,3,2,3,4,7,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,3,2,3,4,7,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_16_01_17_06_22_07_31_08_24_09_25_14_30_15_31:
 ; AVX512VL:       # %bb.0:
@@ -5259,14 +5439,23 @@
 ; AVX2-SLOW-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_00_20_01_21_06_16_07_25_08_28_09_29_14_24_15_25:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <2,0,4,u,6,4,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,10,11,u,u,u,u,u,u,u,u,16,17,18,19,20,21,22,23,u,u,u,u,u,u,u,u]
-; AVX2-FAST-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,3,2,3,4,7,6,7]
-; AVX2-FAST-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_00_20_01_21_06_16_07_25_08_28_09_29_14_24_15_25:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <2,0,4,u,6,4,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5,10,11,u,u,u,u,u,u,u,u,16,17,18,19,20,21,22,23,u,u,u,u,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,3,2,3,4,7,6,7]
+; AVX2-FAST-ALL-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_00_20_01_21_06_16_07_25_08_28_09_29_14_24_15_25:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm1 = ymm1[0],ymm2[1],ymm1[2,3,4,5,6,7,8],ymm2[9],ymm1[10,11,12,13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[2,0,2,3,6,4,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[0,3,2,3,4,7,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_00_20_01_21_06_16_07_25_08_28_09_29_14_24_15_25:
 ; AVX512VL:       # %bb.0:
@@ -5864,13 +6053,21 @@
 ; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v16i16_uu_uu_uu_01_uu_05_07_25_uu_uu_uu_09_uu_13_15_25:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [4,4,4,4,4,4,4,4]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,2,3,2,3,2,3,8,9,10,11,14,15,u,u,18,19,18,19,18,19,18,19,24,25,26,27,30,31,u,u]
-; AVX2-FAST-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v16i16_uu_uu_uu_01_uu_05_07_25_uu_uu_uu_09_uu_13_15_25:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [4,4,4,4,4,4,4,4]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,2,3,2,3,2,3,8,9,10,11,14,15,u,u,18,19,18,19,18,19,18,19,24,25,26,27,30,31,u,u]
+; AVX2-FAST-ALL-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v16i16_uu_uu_uu_01_uu_05_07_25_uu_uu_uu_09_uu_13_15_25:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[2,3,2,3,2,3,2,3,8,9,10,11,14,15,u,u,18,19,18,19,18,19,18,19,24,25,26,27,30,31,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,0,0,4,4,4,4]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5,6],ymm1[7],ymm0[8,9,10,11,12,13,14],ymm1[15]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v16i16_uu_uu_uu_01_uu_05_07_25_uu_uu_uu_09_uu_13_15_25:
 ; AVX512VL:       # %bb.0:
@@ -7248,16 +7445,30 @@
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: PR24935:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,5,u,u,0,4,6,2>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,ymm0[2,3],zero,zero,zero,zero,zero,zero,ymm0[6,7],zero,zero,ymm0[18,19,22,23],zero,zero,zero,zero,ymm0[26,27,28,29,16,17],zero,zero
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <5,6,3,0,0,6,4,u>
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[2,3,0,1],zero,zero,ymm1[6,7,0,1,10,11],zero,zero,ymm1[12,13],zero,zero,zero,zero,ymm1[16,17,20,21],zero,zero,zero,zero,zero,zero,ymm1[24,25]
-; AVX2-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: PR24935:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <0,5,u,u,0,4,6,2>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,ymm0[2,3],zero,zero,zero,zero,zero,zero,ymm0[6,7],zero,zero,ymm0[18,19,22,23],zero,zero,zero,zero,ymm0[26,27,28,29,16,17],zero,zero
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <5,6,3,0,0,6,4,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[2,3,0,1],zero,zero,ymm1[6,7,0,1,10,11],zero,zero,ymm1[12,13],zero,zero,zero,zero,ymm1[16,17,20,21],zero,zero,zero,zero,zero,zero,ymm1[24,25]
+; AVX2-FAST-ALL-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: PR24935:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,ymm1[8,9],zero,zero,zero,zero,ymm1[14,15,12,13,0,1,24,25,24,25],zero,zero,ymm1[24,25,16,17,30,31,28,29,16,17]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = ymm1[6,7,4,5],zero,zero,ymm1[10,11,4,5],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16,17],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm2, %ymm1, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,2,3,2,3,u,u,10,11,u,u,u,u,u,u,u,u,18,19,18,19,u,u,26,27,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,u,u,u,u,6,7,u,u,18,19,u,u,u,u,u,u,u,u,24,25,16,17,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0],ymm2[1,2],ymm0[3],ymm2[4],ymm0[5,6,7,8],ymm2[9,10],ymm0[11],ymm2[12],ymm0[13,14,15]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,255,255,0,0,255,255,255,255,255,255,0,0,255,255,0,0,0,0,255,255,255,255,0,0,0,0,0,0,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: PR24935:
 ; AVX512VL:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
@@ -1,11 +1,14 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=ALL,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLBW,AVX512VLBW-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLBW,AVX512VLBW-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLBW,AVX512VLBW-FAST,AVX512VLBW-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLBW,AVX512VLBW-FAST,AVX512VLBW-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512vbmi | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLVBMI,AVX512VLVBMI-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512vbmi,+fast-variable-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLVBMI,AVX512VLVBMI-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512vbmi,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLVBMI,AVX512VLVBMI-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512vbmi,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=ALL,AVX2OR512VL,AVX512VL,AVX512VLVBMI,AVX512VLVBMI-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx | FileCheck %s --check-prefixes=ALL,XOP,XOPAVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=ALL,XOP,XOPAVX2
 
@@ -64,11 +67,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
 ; XOPAVX1:       # %bb.0:
@@ -114,11 +123,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -164,11 +179,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,3,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,3,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,3,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_03_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -214,11 +235,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_04_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,4,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_04_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,4,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_04_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,4,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_04_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -264,11 +291,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_05_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,5,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_05_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,5,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_05_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,5,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_05_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -314,11 +347,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_06_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,6,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_06_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,6,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_06_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,6,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_06_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -364,11 +403,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -414,11 +459,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,8]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -464,11 +515,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_09_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_09_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_09_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,9,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_09_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -514,11 +571,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_10_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_10_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_10_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,10,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_10_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -564,11 +627,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_11_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_11_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_11_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,11,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_11_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -614,11 +683,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_12_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_12_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_12_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,12,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_12_00_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -664,11 +739,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_13_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_13_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_13_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,13,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_13_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -714,11 +795,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,14,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -764,11 +851,17 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -1822,12 +1915,18 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0]
-; AVX512VLVBMI-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_01_00:
 ; XOPAVX1:       # %bb.0:
@@ -1869,12 +1968,18 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0]
-; AVX512VLVBMI-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_02_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -1916,12 +2021,18 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,0,7,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_07_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -1963,12 +2074,18 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_08_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -2010,12 +2127,18 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_14_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -2057,12 +2180,18 @@
 ; AVX512VLVBMI-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VLVBMI-SLOW-NEXT:    retq
 ;
-; AVX512VLVBMI-FAST-LABEL: shuffle_v32i8_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLVBMI-FAST:       # %bb.0:
-; AVX512VLVBMI-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
-; AVX512VLVBMI-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VLVBMI-FAST-NEXT:    vpermb %ymm0, %ymm1, %ymm0
-; AVX512VLVBMI-FAST-NEXT:    retq
+; AVX512VLVBMI-FAST-ALL-LABEL: shuffle_v32i8_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-ALL:       # %bb.0:
+; AVX512VLVBMI-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VLVBMI-FAST-ALL-NEXT:    vpermb %ymm0, %ymm1, %ymm0
+; AVX512VLVBMI-FAST-ALL-NEXT:    retq
+;
+; AVX512VLVBMI-FAST-PERLANE-LABEL: shuffle_v32i8_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLVBMI-FAST-PERLANE:       # %bb.0:
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VLVBMI-FAST-PERLANE-NEXT:    retq
 ;
 ; XOPAVX1-LABEL: shuffle_v32i8_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_15_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; XOPAVX1:       # %bb.0:
@@ -3276,18 +3405,32 @@
 ; AVX2-SLOW-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm2 = ymm1[10,13],zero,zero,ymm1[3,3],zero,ymm1[8],zero,zero,zero,ymm1[12,1],zero,zero,zero,zero,zero,ymm1[20],zero,ymm1[17,22],zero,zero,ymm1[16],zero,ymm1[27],zero,zero,zero,zero,zero
-; AVX2-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,ymm1[u,u],zero,zero,ymm1[12],zero,ymm1[u,u,u],zero,zero,ymm1[u,0,3,u,u],zero,ymm1[u],zero,zero,ymm1[21,16],zero,ymm1[26],zero,ymm1[u,20,18,20,23]
-; AVX2-FAST-NEXT:    vpor %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <3,4,5,7,5,4,1,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,0,1,u,u,u,u,5,10,13,u,u,0,u,u,16,23,u,23,u,u,u,u,u,u,u,27,u,u,u,u]
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,255,255,255,255,0,0,0,255,255,0,255,255,0,0,255,0,255,255,255,255,255,255,255,0,255,255,255,255]
-; AVX2-FAST-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm2 = ymm1[10,13],zero,zero,ymm1[3,3],zero,ymm1[8],zero,zero,zero,ymm1[12,1],zero,zero,zero,zero,zero,ymm1[20],zero,ymm1[17,22],zero,zero,ymm1[16],zero,ymm1[27],zero,zero,zero,zero,zero
+; AVX2-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,ymm1[u,u],zero,zero,ymm1[12],zero,ymm1[u,u,u],zero,zero,ymm1[u,0,3,u,u],zero,ymm1[u],zero,zero,ymm1[21,16],zero,ymm1[26],zero,ymm1[u,20,18,20,23]
+; AVX2-FAST-ALL-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <3,4,5,7,5,4,1,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,0,1,u,u,u,u,5,10,13,u,u,0,u,u,16,23,u,23,u,u,u,u,u,u,u,27,u,u,u,u]
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,255,255,255,255,0,0,0,255,255,0,255,255,0,0,255,0,255,255,255,255,255,255,255,0,255,255,255,255]
+; AVX2-FAST-ALL-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm1[10,13],zero,zero,ymm1[3,3],zero,ymm1[8],zero,zero,zero,ymm1[12,1],zero,zero,zero,zero,zero,ymm1[20],zero,ymm1[17,22],zero,zero,ymm1[16],zero,ymm1[27],zero,zero,zero,zero,zero
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,ymm1[u,u],zero,zero,ymm1[12],zero,ymm1[u,u,u],zero,zero,ymm1[u,0,3,u,u],zero,ymm1[u],zero,zero,ymm1[21,16],zero,ymm1[26],zero,ymm1[u,20,18,20,23]
+; AVX2-FAST-PERLANE-NEXT:    vpor %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,12,13,u,u,u,u,u,u,u,u,u,12,u,u,20,19,u,19,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,1,6,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,23,u,u,u,u]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0,1],ymm0[2],ymm2[3,4,5],ymm0[6],ymm2[7]
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa {{.*#+}} ymm2 = [255,255,0,0,255,255,255,255,0,0,0,255,255,0,255,255,0,0,255,0,255,255,255,255,255,255,255,0,255,255,255,255]
+; AVX2-FAST-PERLANE-NEXT:    vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
 ; AVX512VLBW-SLOW:       # %bb.0:
@@ -3304,18 +3447,33 @@
 ; AVX512VLBW-SLOW-NEXT:    vmovdqu8 %ymm2, %ymm0 {%k1}
 ; AVX512VLBW-SLOW-NEXT:    retq
 ;
-; AVX512VLBW-FAST-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
-; AVX512VLBW-FAST:       # %bb.0:
-; AVX512VLBW-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = <3,4,5,7,5,4,1,u>
-; AVX512VLBW-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm2
-; AVX512VLBW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[10,13,u,u,3,3],zero,ymm1[8,u,u,u,12,1,u],zero,zero,ymm1[u,u,20,u,17,22],zero,zero,ymm1[16],zero,ymm1[27,u],zero,zero,zero,zero
-; AVX512VLBW-FAST-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
-; AVX512VLBW-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,ymm1[12],zero,zero,zero,zero,zero,zero,zero,ymm1[0,3],zero,zero,zero,zero,zero,zero,ymm1[21,16],zero,ymm1[26],zero,zero,ymm1[20,18,20,23]
-; AVX512VLBW-FAST-NEXT:    vpor %ymm0, %ymm1, %ymm0
-; AVX512VLBW-FAST-NEXT:    movl $134948620, %eax # imm = 0x80B270C
-; AVX512VLBW-FAST-NEXT:    kmovd %eax, %k1
-; AVX512VLBW-FAST-NEXT:    vpshufb {{.*#+}} ymm0 {%k1} = ymm2[u,u,0,1,u,u,u,u,5,10,13,u,u,0,u,u,16,23,u,23,u,u,u,u,u,u,u,27,u,u,u,u]
-; AVX512VLBW-FAST-NEXT:    retq
+; AVX512VLBW-FAST-ALL-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
+; AVX512VLBW-FAST-ALL:       # %bb.0:
+; AVX512VLBW-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = <3,4,5,7,5,4,1,u>
+; AVX512VLBW-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm2
+; AVX512VLBW-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[10,13,u,u,3,3],zero,ymm1[8,u,u,u,12,1,u],zero,zero,ymm1[u,u,20,u,17,22],zero,zero,ymm1[16],zero,ymm1[27,u],zero,zero,zero,zero
+; AVX512VLBW-FAST-ALL-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX512VLBW-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,ymm1[12],zero,zero,zero,zero,zero,zero,zero,ymm1[0,3],zero,zero,zero,zero,zero,zero,ymm1[21,16],zero,ymm1[26],zero,zero,ymm1[20,18,20,23]
+; AVX512VLBW-FAST-ALL-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-FAST-ALL-NEXT:    movl $134948620, %eax # imm = 0x80B270C
+; AVX512VLBW-FAST-ALL-NEXT:    kmovd %eax, %k1
+; AVX512VLBW-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm0 {%k1} = ymm2[u,u,0,1,u,u,u,u,5,10,13,u,u,0,u,u,16,23,u,23,u,u,u,u,u,u,u,27,u,u,u,u]
+; AVX512VLBW-FAST-ALL-NEXT:    retq
+;
+; AVX512VLBW-FAST-PERLANE-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
+; AVX512VLBW-FAST-PERLANE:       # %bb.0:
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm2 = ymm0[u,u,12,13,u,u,u,u,u,u,u,u,u,12,u,u,20,19,u,19,u,u,u,u,u,u,u,u,u,u,u,u]
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[u,u,u,u,u,u,u,u,1,6,13,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,u,23,u,u,u,u]
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm0[2],ymm2[3,4,5],ymm0[6],ymm2[7]
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm0 = ymm1[10,13,u,u,3,3],zero,ymm1[8,u,u,u,12,1,u],zero,zero,ymm1[u,u,20,u,17,22],zero,zero,ymm1[16],zero,ymm1[27,u],zero,zero,zero,zero
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,ymm1[12],zero,zero,zero,zero,zero,zero,zero,ymm1[0,3],zero,zero,zero,zero,zero,zero,ymm1[21,16],zero,ymm1[26],zero,zero,ymm1[20,18,20,23]
+; AVX512VLBW-FAST-PERLANE-NEXT:    vpor %ymm0, %ymm1, %ymm0
+; AVX512VLBW-FAST-PERLANE-NEXT:    movl $134948620, %eax # imm = 0x80B270C
+; AVX512VLBW-FAST-PERLANE-NEXT:    kmovd %eax, %k1
+; AVX512VLBW-FAST-PERLANE-NEXT:    vmovdqu8 %ymm2, %ymm0 {%k1}
+; AVX512VLBW-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VLVBMI-LABEL: shuffle_v32i8_42_45_12_13_35_35_60_40_17_22_29_44_33_12_48_51_20_19_52_19_49_54_37_32_48_42_59_07_36_34_36_39:
 ; AVX512VLVBMI:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v4.ll
@@ -1,9 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2 --check-prefix=AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2 --check-prefix=AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2 --check-prefix=AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512VL-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST --check-prefix=AVX512VL-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST --check-prefix=AVX512VL-FAST-PERLANE
 
 define <4 x double> @shuffle_v4f64_0000(<4 x double> %a, <4 x double> %b) {
 ; AVX1-LABEL: shuffle_v4f64_0000:
@@ -457,11 +459,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4f64_1054:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [1,0,5,4]
-; AVX512VL-FAST-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4f64_1054:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [1,0,5,4]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4f64_1054:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 1, i32 0, i32 5, i32 4>
   ret <4 x double> %shuffle
 }
@@ -479,11 +487,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4f64_3254:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [3,2,5,4]
-; AVX512VL-FAST-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4f64_3254:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [3,2,5,4]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4f64_3254:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 3, i32 2, i32 5, i32 4>
   ret <4 x double> %shuffle
 }
@@ -501,11 +515,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4f64_3276:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [3,2,7,6]
-; AVX512VL-FAST-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4f64_3276:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [3,2,7,6]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4f64_3276:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 3, i32 2, i32 7, i32 6>
   ret <4 x double> %shuffle
 }
@@ -636,12 +656,18 @@
 ; AVX512VL-SLOW-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[3],ymm1[2]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4f64_0456:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [4,0,1,2]
-; AVX512VL-FAST-NEXT:    vpermi2pd %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovapd %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4f64_0456:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [4,0,1,2]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2pd %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovapd %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4f64_0456:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[3],ymm1[2]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 0, i32 4, i32 5, i32 6>
   ret <4 x double> %shuffle
 }
@@ -685,11 +711,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4f64_0044:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [0,0,4,4]
-; AVX512VL-FAST-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4f64_0044:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [0,0,4,4]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2pd %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4f64_0044:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %1 = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 0, i32 0, i32 4, i32 4>
   ret <4 x double> %1
 }
@@ -954,11 +986,17 @@
 ; AVX512VL-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4i64_0124:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,4]
-; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4i64_0124:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,1,2,4]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4i64_0124:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vbroadcastsd %xmm1, %ymm1
+; AVX512VL-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 4>
   ret <4 x i64> %shuffle
 }
@@ -1029,11 +1067,17 @@
 ; AVX512VL-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4i64_4012:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,0,1,2]
-; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4i64_4012:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,0,1,2]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4i64_4012:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,2]
+; AVX512VL-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5,6,7]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 4, i32 0, i32 1, i32 2>
   ret <4 x i64> %shuffle
 }
@@ -1124,11 +1168,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,1,3,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4i64_2u35:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,5,3,5]
-; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4i64_2u35:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,5,3,5]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4i64_2u35:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,1,3,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 2, i32 undef, i32 3, i32 5>
   ret <4 x i64> %shuffle
 }
@@ -1176,11 +1226,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4i64_1054:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [1,0,5,4]
-; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4i64_1054:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [1,0,5,4]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4i64_1054:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 1, i32 0, i32 5, i32 4>
   ret <4 x i64> %shuffle
 }
@@ -1204,11 +1260,17 @@
 ; AVX512VL-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4i64_3254:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,2,5,4]
-; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4i64_3254:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,2,5,4]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4i64_3254:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 3, i32 2, i32 5, i32 4>
   ret <4 x i64> %shuffle
 }
@@ -1232,11 +1294,17 @@
 ; AVX512VL-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v4i64_3276:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,2,7,6]
-; AVX512VL-FAST-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v4i64_3276:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,2,7,6]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2q %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v4i64_3276:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 3, i32 2, i32 7, i32 6>
   ret <4 x i64> %shuffle
 }
@@ -1876,14 +1944,23 @@
 ; AVX512VL-SLOW-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: add_v4i64_0246_1357:
-; AVX512VL-FAST:       # %bb.0: # %entry
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6]
-; AVX512VL-FAST-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,3,5,7]
-; AVX512VL-FAST-NEXT:    vpermi2q %ymm1, %ymm0, %ymm3
-; AVX512VL-FAST-NEXT:    vpaddq %ymm3, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: add_v4i64_0246_1357:
+; AVX512VL-FAST-ALL:       # %bb.0: # %entry
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,3,5,7]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2q %ymm1, %ymm0, %ymm3
+; AVX512VL-FAST-ALL-NEXT:    vpaddq %ymm3, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: add_v4i64_0246_1357:
+; AVX512VL-FAST-PERLANE:       # %bb.0: # %entry
+; AVX512VL-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    retq
 entry:
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
   %shuffle1 = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
@@ -1923,14 +2000,23 @@
 ; AVX512VL-SLOW-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: add_v4i64_4602_5713:
-; AVX512VL-FAST:       # %bb.0: # %entry
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6]
-; AVX512VL-FAST-NEXT:    vpermi2q %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,3,5,7]
-; AVX512VL-FAST-NEXT:    vpermi2q %ymm0, %ymm1, %ymm3
-; AVX512VL-FAST-NEXT:    vpaddq %ymm3, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: add_v4i64_4602_5713:
+; AVX512VL-FAST-ALL:       # %bb.0: # %entry
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2q %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm3 = [1,3,5,7]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2q %ymm0, %ymm1, %ymm3
+; AVX512VL-FAST-ALL-NEXT:    vpaddq %ymm3, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: add_v4i64_4602_5713:
+; AVX512VL-FAST-PERLANE:       # %bb.0: # %entry
+; AVX512VL-FAST-PERLANE-NEXT:    vpunpcklqdq {{.*#+}} ymm2 = ymm1[0],ymm0[0],ymm1[2],ymm0[2]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm2 = ymm2[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpunpckhqdq {{.*#+}} ymm0 = ymm1[1],ymm0[1],ymm1[3],ymm0[3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpaddq %ymm0, %ymm2, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    retq
 entry:
   %shuffle = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 4, i32 6, i32 0, i32 2>
   %shuffle1 = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 5, i32 7, i32 1, i32 3>
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
@@ -1,9 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2OR512VL --check-prefix=AVX2 --check-prefix=AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2OR512VL --check-prefix=AVX2 --check-prefix=AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2OR512VL --check-prefix=AVX2 --check-prefix=AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX1OR2 --check-prefix=AVX2OR512VL --check-prefix=AVX2 --check-prefix=AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefix=ALL --check-prefix=AVX2OR512VL --check-prefix=AVX512VL --check-prefix=AVX512VL-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX2OR512VL --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX2OR512VL --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST --check-prefix=AVX512VL-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+avx512dq,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=ALL --check-prefix=AVX2OR512VL --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST --check-prefix=AVX512VL-FAST-PERLANE
 
 define <8 x float> @shuffle_v8f32_00000000(<8 x float> %a, <8 x float> %b) {
 ; AVX1-LABEL: shuffle_v8f32_00000000:
@@ -34,11 +36,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_00000010:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_00000010:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_00000010:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_00000010:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -46,11 +54,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_00000010:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_00000010:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_00000010:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0>
   ret <8 x float> %shuffle
 }
@@ -69,11 +83,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_00000200:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_00000200:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_00000200:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_00000200:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -81,11 +101,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_00000200:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_00000200:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_00000200:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,2]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0>
   ret <8 x float> %shuffle
 }
@@ -104,11 +130,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_00003000:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_00003000:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_00003000:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,3,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_00003000:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -116,11 +148,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_00003000:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_00003000:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_00003000:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,3,0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0>
   ret <8 x float> %shuffle
 }
@@ -234,11 +272,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_00001111:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_00001111:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_00001111:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_00001111:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -246,11 +290,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_00001111:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_00001111:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_00001111:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,1]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
   ret <8 x float> %shuffle
 }
@@ -284,11 +334,17 @@
 ; AVX512VL-SLOW-NEXT:    vbroadcastsd %xmm0, %ymm0
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_08080808:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [34359738368,34359738368,34359738368,34359738368]
-; AVX512VL-FAST-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_08080808:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [34359738368,34359738368,34359738368,34359738368]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_08080808:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512VL-FAST-PERLANE-NEXT:    vbroadcastsd %xmm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8>
   ret <8 x float> %shuffle
 }
@@ -396,14 +452,23 @@
 ; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_08991abb:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_08991abb:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_08991abb:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovaps {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8f32_08991abb:
 ; AVX512VL:       # %bb.0:
@@ -455,12 +520,19 @@
 ; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_09ab1def:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_09ab1def:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_09ab1def:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8f32_09ab1def:
 ; AVX512VL:       # %bb.0:
@@ -781,15 +853,25 @@
 ; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3,4,5,6],ymm0[7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_c348cda0:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <u,3,4,u,u,u,u,0>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [4,5,2,0,4,5,2,0]
-; AVX2-FAST-NEXT:    # ymm2 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3,4,5,6],ymm0[7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_c348cda0:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <u,3,4,u,u,u,u,0>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [4,5,2,0,4,5,2,0]
+; AVX2-FAST-ALL-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3,4,5,6],ymm0[7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_c348cda0:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastf128 {{.*#+}} ymm2 = [4,5,2,0,4,5,2,0]
+; AVX2-FAST-PERLANE-NEXT:    # ymm2 = mem[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,3,2,0,4,7,6,4]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,2,1]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3,4,5,6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8f32_c348cda0:
 ; AVX512VL:       # %bb.0:
@@ -821,14 +903,23 @@
 ; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_f511235a:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <7,u,u,u,u,u,u,2>
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <u,5,1,1,2,3,5,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_f511235a:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <7,u,u,u,u,u,u,2>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <u,5,1,1,2,3,5,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_f511235a:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[3,2,2,3,7,6,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[2,1,2,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[1,1,2,3,5,5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,0,1,2]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6],ymm1[7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8f32_f511235a:
 ; AVX512VL:       # %bb.0:
@@ -852,12 +943,18 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_32103210:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_32103210:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_32103210:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_32103210:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -865,12 +962,18 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_32103210:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
-; AVX512VL-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VL-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_32103210:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
+; AVX512VL-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VL-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_32103210:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>
   ret <8 x float> %shuffle
 }
@@ -888,12 +991,18 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_76547654:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_76547654:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_76547654:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_76547654:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -901,12 +1010,18 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_76547654:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
-; AVX512VL-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VL-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_76547654:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
+; AVX512VL-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VL-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_76547654:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 7, i32 6, i32 5, i32 4>
   ret <8 x float> %shuffle
 }
@@ -924,11 +1039,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8f32_76543210:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8f32_76543210:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8f32_76543210:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8f32_76543210:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -936,11 +1057,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_76543210:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_76543210:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_76543210:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
   ret <8 x float> %shuffle
 }
@@ -958,11 +1085,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_3210ba98:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [3,2,1,0,11,10,9,8]
-; AVX512VL-FAST-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_3210ba98:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [3,2,1,0,11,10,9,8]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_3210ba98:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 11, i32 10, i32 9, i32 8>
   ret <8 x float> %shuffle
 }
@@ -1002,11 +1135,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_7654fedc:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
-; AVX512VL-FAST-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_7654fedc:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_7654fedc:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 15, i32 14, i32 13, i32 12>
   ret <8 x float> %shuffle
 }
@@ -1024,12 +1163,18 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_fedc7654:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
-; AVX512VL-FAST-NEXT:    vpermi2ps %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovaps %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_fedc7654:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2ps %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovaps %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_fedc7654:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 15, i32 14, i32 13, i32 12, i32 7, i32 6, i32 5, i32 4>
   ret <8 x float> %shuffle
 }
@@ -1054,11 +1199,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: PR21138:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [1,3,5,7,9,11,13,15]
-; AVX512VL-FAST-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: PR21138:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [1,3,5,7,9,11,13,15]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: PR21138:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm1[1,3],ymm0[5,7],ymm1[5,7]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %truc, <8 x float> %tchose, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   ret <8 x float> %shuffle
 }
@@ -1099,12 +1250,18 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_ba983210:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [3,2,1,0,11,10,9,8]
-; AVX512VL-FAST-NEXT:    vpermi2ps %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovaps %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_ba983210:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [3,2,1,0,11,10,9,8]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2ps %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovaps %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_ba983210:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 11, i32 10, i32 9, i32 8, i32 3, i32 2, i32 1, i32 0>
   ret <8 x float> %shuffle
 }
@@ -1153,11 +1310,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_084c195d:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [0,8,4,12,1,9,5,13]
-; AVX512VL-FAST-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_084c195d:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [0,8,4,12,1,9,5,13]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2ps %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_084c195d:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 8, i32 4, i32 12, i32 1, i32 9, i32 5, i32 13>
   ret <8 x float> %shuffle
 }
@@ -1219,12 +1382,18 @@
 ; AVX512VL-SLOW-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[3],ymm1[2]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8f32_0189abcd:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [4,0,1,2]
-; AVX512VL-FAST-NEXT:    vpermi2pd %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovapd %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8f32_0189abcd:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [4,0,1,2]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2pd %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovapd %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8f32_0189abcd:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vshufpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[3],ymm1[2]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13>
   ret <8 x float> %shuffle
 }
@@ -1368,11 +1537,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_00000010:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_00000010:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_00000010:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_00000010:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -1380,11 +1555,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_00000010:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_00000010:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,0,1,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_00000010:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0>
   ret <8 x i32> %shuffle
 }
@@ -1403,11 +1584,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_00000200:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_00000200:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_00000200:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_00000200:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -1415,11 +1602,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_00000200:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_00000200:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,0,2,0,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_00000200:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,2]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0>
   ret <8 x i32> %shuffle
 }
@@ -1438,11 +1631,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_00003000:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_00003000:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_00003000:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,3,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_00003000:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -1450,11 +1649,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_00003000:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_00003000:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,3,0,0,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_00003000:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,3,0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0>
   ret <8 x i32> %shuffle
 }
@@ -1556,11 +1761,17 @@
 ; AVX2-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,1,1,4,4,5,5]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_00112233:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,1,1,2,2,3,3]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_00112233:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,1,1,2,2,3,3]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_00112233:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,1,1,4,4,5,5]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_00112233:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -1568,11 +1779,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,1,1,4,4,5,5]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_00112233:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,1,1,2,2,3,3]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_00112233:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,1,1,2,2,3,3]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_00112233:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,0,1,1,4,4,5,5]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3>
   ret <8 x i32> %shuffle
 }
@@ -1591,11 +1808,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_00001111:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_00001111:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_00001111:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,1]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_00001111:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -1603,11 +1826,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_00001111:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_00001111:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [0,0,0,0,1,1,1,1]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_00001111:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[0,0,1,1]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1>
   ret <8 x i32> %shuffle
 }
@@ -1641,11 +1870,17 @@
 ; AVX512VL-SLOW-NEXT:    vbroadcastsd %xmm0, %ymm0
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_08080808:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [34359738368,34359738368,34359738368,34359738368]
-; AVX512VL-FAST-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_08080808:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [34359738368,34359738368,34359738368,34359738368]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_08080808:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; AVX512VL-FAST-PERLANE-NEXT:    vbroadcastsd %xmm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8>
   ret <8 x i32> %shuffle
 }
@@ -1759,14 +1994,23 @@
 ; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_08991abb:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_08991abb:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <u,0,1,1,u,2,3,3>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_08991abb:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm1 = ymm1[0,2,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm1 = ymm1[0,0,1,1,4,4,5,5]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8i32_08991abb:
 ; AVX512VL:       # %bb.0:
@@ -1799,11 +2043,17 @@
 ; AVX512VL-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_091b2d3f:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,9,1,11,2,13,3,15]
-; AVX512VL-FAST-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_091b2d3f:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,9,1,11,2,13,3,15]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_091b2d3f:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
+; AVX512VL-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15>
   ret <8 x i32> %shuffle
 }
@@ -1823,12 +2073,19 @@
 ; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_09ab1def:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_09ab1def:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <0,u,u,u,1,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_09ab1def:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3]
+; AVX2-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8i32_09ab1def:
 ; AVX512VL:       # %bb.0:
@@ -2270,13 +2527,21 @@
 ; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6],ymm0[7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_6caa87e5:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = <u,4,2,2,0,u,6,u>
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[3,1,3,2]
-; AVX2-FAST-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6],ymm0[7]
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_6caa87e5:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = <u,4,2,2,0,u,6,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[3,1,3,2]
+; AVX2-FAST-ALL-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6],ymm0[7]
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_6caa87e5:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[3,1,3,2]
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm1 = ymm1[0,0,2,2,4,4,6,6]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[2,1,0,3]
+; AVX2-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4],ymm0[5],ymm1[6],ymm0[7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-LABEL: shuffle_v8i32_6caa87e5:
 ; AVX512VL:       # %bb.0:
@@ -2301,12 +2566,18 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_32103210:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_32103210:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_32103210:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_32103210:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -2314,12 +2585,18 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_32103210:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
-; AVX512VL-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VL-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_32103210:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [3,2,1,0,3,2,1,0]
+; AVX512VL-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VL-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_32103210:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,2,1,0]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 3, i32 2, i32 1, i32 0>
   ret <8 x i32> %shuffle
 }
@@ -2337,12 +2614,18 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_76547654:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_76547654:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vbroadcastf128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_76547654:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_76547654:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -2350,12 +2633,18 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_76547654:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
-; AVX512VL-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512VL-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_76547654:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
+; AVX512VL-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX512VL-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_76547654:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 7, i32 6, i32 5, i32 4>
   ret <8 x i32> %shuffle
 }
@@ -2373,11 +2662,17 @@
 ; AVX2-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: shuffle_v8i32_76543210:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: shuffle_v8i32_76543210:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: shuffle_v8i32_76543210:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX2-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512VL-SLOW-LABEL: shuffle_v8i32_76543210:
 ; AVX512VL-SLOW:       # %bb.0:
@@ -2385,11 +2680,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_76543210:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
-; AVX512VL-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_76543210:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = [7,6,5,4,3,2,1,0]
+; AVX512VL-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_76543210:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
   ret <8 x i32> %shuffle
 }
@@ -2407,11 +2708,17 @@
 ; AVX512VL-SLOW-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_3210ba98:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,2,1,0,11,10,9,8]
-; AVX512VL-FAST-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_3210ba98:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,2,1,0,11,10,9,8]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_3210ba98:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 11, i32 10, i32 9, i32 8>
   ret <8 x i32> %shuffle
 }
@@ -2451,11 +2758,17 @@
 ; AVX512VL-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_7654fedc:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
-; AVX512VL-FAST-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_7654fedc:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
+; AVX512VL-FAST-ALL-NEXT:    vpermt2d %ymm1, %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_7654fedc:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 15, i32 14, i32 13, i32 12>
   ret <8 x i32> %shuffle
 }
@@ -2473,12 +2786,18 @@
 ; AVX512VL-SLOW-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_fedc7654:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
-; AVX512VL-FAST-NEXT:    vpermi2d %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovdqa %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_fedc7654:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [7,6,5,4,15,14,13,12]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2d %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_fedc7654:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vperm2i128 {{.*#+}} ymm0 = ymm1[2,3],ymm0[2,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 15, i32 14, i32 13, i32 12, i32 7, i32 6, i32 5, i32 4>
   ret <8 x i32> %shuffle
 }
@@ -2552,12 +2871,18 @@
 ; AVX512VL-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_089abcde:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,0,1,2,3,4,5,6]
-; AVX512VL-FAST-NEXT:    vpermi2d %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovdqa %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_089abcde:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,0,1,2,3,4,5,6]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2d %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_089abcde:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    valignd {{.*#+}} ymm1 = ymm1[7,0,1,2,3,4,5,6]
+; AVX512VL-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14>
   ret <8 x i32> %shuffle
 }
@@ -2581,12 +2906,18 @@
 ; AVX512VL-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
 ; AVX512VL-SLOW-NEXT:    retq
 ;
-; AVX512VL-FAST-LABEL: shuffle_v8i32_0189abcd:
-; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,0,1,2]
-; AVX512VL-FAST-NEXT:    vpermi2q %ymm0, %ymm1, %ymm2
-; AVX512VL-FAST-NEXT:    vmovdqa %ymm2, %ymm0
-; AVX512VL-FAST-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuffle_v8i32_0189abcd:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [4,0,1,2]
+; AVX512VL-FAST-ALL-NEXT:    vpermi2q %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa %ymm2, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    retq
+;
+; AVX512VL-FAST-PERLANE-LABEL: shuffle_v8i32_0189abcd:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[0,0,1,2]
+; AVX512VL-FAST-PERLANE-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VL-FAST-PERLANE-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13>
   ret <8 x i32> %shuffle
 }
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-combining.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-combining.ll
@@ -4,7 +4,8 @@
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+sse4.1 | FileCheck %s --check-prefixes=CHECK,SSE,SSE41
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mcpu=x86-64 -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mcpu=x86-64 -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mcpu=x86-64 -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ;
 ; Verify that the DAG combiner correctly folds bitwise operations across
 ; shuffles, nested shuffles with undef, pairs of nested shuffles, and other
@@ -2401,13 +2402,20 @@
 ; AVX2-SLOW-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: combine_unneeded_subvector1:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
-; AVX2-FAST-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: combine_unneeded_subvector1:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [7,6,5,4,7,6,5,4]
+; AVX2-FAST-ALL-NEXT:    # ymm1 = mem[0,1,0,1]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: combine_unneeded_subvector1:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]
+; AVX2-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[2,3,2,3]
+; AVX2-FAST-PERLANE-NEXT:    retq
   %b = add <8 x i32> %a, <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
   %c = shufflevector <8 x i32> %b, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 7, i32 6, i32 5, i32 4>
   ret <8 x i32> %c
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-v1.ll b/llvm/test/CodeGen/X86/vector-shuffle-v1.ll
--- a/llvm/test/CodeGen/X86/vector-shuffle-v1.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-v1.ll
@@ -1,7 +1,9 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+avx512dq,+fast-variable-shuffle | FileCheck %s --check-prefix=VL_BW_DQ
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX512VL --check-prefix=AVX512VL-FAST-PERLANE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=VL_BW_DQ --check-prefix=VL_BW_DQ-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+avx512dq,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=VL_BW_DQ --check-prefix=VL_BW_DQ-FAST-PERLANE
 
 define <2 x i1> @shuf2i1_1_0(<2 x i1> %a) {
 ; AVX512F-LABEL: shuf2i1_1_0:
@@ -495,30 +497,55 @@
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    kmovw %edi, %k1
-; AVX512VL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [2,2,2,2,2,2,2,2]
-; AVX512VL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX512VL-NEXT:    vpslld $31, %ymm1, %ymm1
-; AVX512VL-NEXT:    vptestmd %ymm1, %ymm1, %k1
-; AVX512VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
-; AVX512VL-NEXT:    vpmovdw %ymm0, %xmm0
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
+; AVX512VL-FAST-ALL-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
+; AVX512VL-FAST-ALL:       # %bb.0:
+; AVX512VL-FAST-ALL-NEXT:    kmovw %edi, %k1
+; AVX512VL-FAST-ALL-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512VL-FAST-ALL-NEXT:    vpbroadcastd {{.*#+}} ymm2 = [2,2,2,2,2,2,2,2]
+; AVX512VL-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX512VL-FAST-ALL-NEXT:    vpslld $31, %ymm1, %ymm1
+; AVX512VL-FAST-ALL-NEXT:    vptestmd %ymm1, %ymm1, %k1
+; AVX512VL-FAST-ALL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-FAST-ALL-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-FAST-ALL-NEXT:    vzeroupper
+; AVX512VL-FAST-ALL-NEXT:    retq
 ;
-; VL_BW_DQ-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
-; VL_BW_DQ:       # %bb.0:
-; VL_BW_DQ-NEXT:    kmovd %edi, %k0
-; VL_BW_DQ-NEXT:    vpmovm2d %k0, %ymm0
-; VL_BW_DQ-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [2,2,2,2,2,2,2,2]
-; VL_BW_DQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; VL_BW_DQ-NEXT:    vpmovd2m %ymm0, %k0
-; VL_BW_DQ-NEXT:    vpmovm2w %k0, %xmm0
-; VL_BW_DQ-NEXT:    vzeroupper
-; VL_BW_DQ-NEXT:    retq
+; AVX512VL-FAST-PERLANE-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
+; AVX512VL-FAST-PERLANE:       # %bb.0:
+; AVX512VL-FAST-PERLANE-NEXT:    kmovw %edi, %k1
+; AVX512VL-FAST-PERLANE-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-FAST-PERLANE-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512VL-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX512VL-FAST-PERLANE-NEXT:    vpbroadcastq %xmm1, %ymm1
+; AVX512VL-FAST-PERLANE-NEXT:    vpslld $31, %ymm1, %ymm1
+; AVX512VL-FAST-PERLANE-NEXT:    vptestmd %ymm1, %ymm1, %k1
+; AVX512VL-FAST-PERLANE-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-FAST-PERLANE-NEXT:    vpmovdw %ymm0, %xmm0
+; AVX512VL-FAST-PERLANE-NEXT:    vzeroupper
+; AVX512VL-FAST-PERLANE-NEXT:    retq
+;
+; VL_BW_DQ-FAST-ALL-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
+; VL_BW_DQ-FAST-ALL:       # %bb.0:
+; VL_BW_DQ-FAST-ALL-NEXT:    kmovd %edi, %k0
+; VL_BW_DQ-FAST-ALL-NEXT:    vpmovm2d %k0, %ymm0
+; VL_BW_DQ-FAST-ALL-NEXT:    vpbroadcastd {{.*#+}} ymm1 = [2,2,2,2,2,2,2,2]
+; VL_BW_DQ-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; VL_BW_DQ-FAST-ALL-NEXT:    vpmovd2m %ymm0, %k0
+; VL_BW_DQ-FAST-ALL-NEXT:    vpmovm2w %k0, %xmm0
+; VL_BW_DQ-FAST-ALL-NEXT:    vzeroupper
+; VL_BW_DQ-FAST-ALL-NEXT:    retq
+;
+; VL_BW_DQ-FAST-PERLANE-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
+; VL_BW_DQ-FAST-PERLANE:       # %bb.0:
+; VL_BW_DQ-FAST-PERLANE-NEXT:    kmovd %edi, %k0
+; VL_BW_DQ-FAST-PERLANE-NEXT:    vpmovm2d %k0, %ymm0
+; VL_BW_DQ-FAST-PERLANE-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; VL_BW_DQ-FAST-PERLANE-NEXT:    vpbroadcastq %xmm0, %ymm0
+; VL_BW_DQ-FAST-PERLANE-NEXT:    vpmovd2m %ymm0, %k0
+; VL_BW_DQ-FAST-PERLANE-NEXT:    vpmovm2w %k0, %xmm0
+; VL_BW_DQ-FAST-PERLANE-NEXT:    vzeroupper
+; VL_BW_DQ-FAST-PERLANE-NEXT:    retq
   %b = bitcast i8 %a to <8 x i1>
   %c = shufflevector < 8 x i1> %b, <8 x i1>undef, <8 x i32> <i32 undef, i32 2, i32 undef, i32 undef, i32 2, i32 undef, i32 2, i32 undef>
   ret <8 x i1> %c
diff --git a/llvm/test/CodeGen/X86/vector-trunc-math.ll b/llvm/test/CodeGen/X86/vector-trunc-math.ll
--- a/llvm/test/CodeGen/X86/vector-trunc-math.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-math.ll
@@ -2,10 +2,13 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512DQ
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512DQ
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512DQ
 
 ;
 ; add
@@ -37,14 +40,22 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_add_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_add_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_add_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpaddq %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_add_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -472,13 +483,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_add_const_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_add_const_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_add_const_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_add_const_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -803,14 +822,22 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_sub_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpsubq %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_sub_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpsubq %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_sub_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpsubq %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_sub_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -1206,13 +1233,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_sub_const_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpsubd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_sub_const_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpsubd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_sub_const_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpsubd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_sub_const_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -1574,14 +1609,24 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_mul_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_mul_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_mul_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm1, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm1 = xmm1[0,2],xmm2[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm2[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpmulld %xmm1, %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_mul_v4i64_v4i32:
 ; AVX512F:       # %bb.0:
@@ -2118,13 +2163,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_mul_const_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpmulld {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_mul_const_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpmulld {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_mul_const_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vpmulld {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_mul_const_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -2529,14 +2582,22 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_and_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vandps %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_and_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vandps %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_and_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vandps %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_and_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -2885,13 +2946,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_and_const_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_and_const_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_and_const_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_and_const_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -3214,14 +3283,22 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_xor_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vxorps %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_xor_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_xor_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vxorps %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_xor_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -3570,13 +3647,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_xor_const_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_xor_const_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_xor_const_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_xor_const_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -3899,14 +3984,22 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_or_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vorps %ymm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_or_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vorps %ymm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_or_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vorps %ymm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_or_v4i64_v4i32:
 ; AVX512:       # %bb.0:
@@ -4255,13 +4348,21 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_or_const_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_or_const_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_or_const_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_or_const_v4i64_v4i32:
 ; AVX512:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-trunc-packus.ll b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
--- a/llvm/test/CodeGen/X86/vector-trunc-packus.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-packus.ll
@@ -4,11 +4,15 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skx | FileCheck %s --check-prefixes=SKX
 
 ;
@@ -502,19 +506,32 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_packus_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm2
-; AVX2-FAST-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm1
-; AVX2-FAST-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_packus_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm2
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_packus_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_packus_v4i64_v4i32:
 ; AVX512F:       # %bb.0:
@@ -923,25 +940,44 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_packus_v8i64_v8i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
-; AVX2-FAST-NEXT:    vpand %ymm1, %ymm3, %ymm1
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm2
-; AVX2-FAST-NEXT:    vpand %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermd %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermd %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_packus_v8i64_v8i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpand %ymm1, %ymm3, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm2
+; AVX2-FAST-ALL-NEXT:    vpand %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermd %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_packus_v8i64_v8i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpand %ymm1, %ymm3, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vpand %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vperm2i128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_packus_v8i64_v8i32:
 ; AVX512:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
--- a/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-ssat.ll
@@ -4,11 +4,15 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skx | FileCheck %s --check-prefixes=SKX
 
 ;
@@ -508,19 +512,32 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_ssat_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm2
-; AVX2-FAST-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm2
-; AVX2-FAST-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_ssat_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm2
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm2
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_ssat_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [2147483647,2147483647,2147483647,2147483647]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm0, %ymm1, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm2
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm2, %ymm0, %ymm1, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_ssat_v4i64_v4i32:
 ; AVX512F:       # %bb.0:
@@ -945,25 +962,44 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_ssat_v8i64_v8i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2147483647,2147483647,2147483647,2147483647]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_ssat_v8i64_v8i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2147483647,2147483647,2147483647,2147483647]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_ssat_v8i64_v8i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2147483647,2147483647,2147483647,2147483647]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm0, %ymm2, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [18446744071562067968,18446744071562067968,18446744071562067968,18446744071562067968]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_ssat_v8i64_v8i32:
 ; AVX512:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-trunc-usat.ll b/llvm/test/CodeGen/X86/vector-trunc-usat.ll
--- a/llvm/test/CodeGen/X86/vector-trunc-usat.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc-usat.ll
@@ -1,14 +1,18 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE41
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+ssse3 | FileCheck %s --check-prefix=SSSE3
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE41
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skx | FileCheck %s --check-prefixes=SKX
 
 ;
@@ -349,19 +353,32 @@
 ; AVX2-SLOW-NEXT:    vzeroupper
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_usat_v4i64_v4i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
-; AVX2-FAST-NEXT:    vpxor %ymm1, %ymm0, %ymm1
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,429496729]
-; AVX2-FAST-NEXT:    vblendvpd %ymm1, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm1, %ymm0
-; AVX2-FAST-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; AVX2-FAST-NEXT:    vzeroupper
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_usat_v4i64_v4i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm1, %ymm0, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,429496729]
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm1, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm1 = <0,2,4,6,u,u,u,u>
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-FAST-ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-FAST-ALL-NEXT:    vzeroupper
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_usat_v4i64_v4i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm1 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm1, %ymm0, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vmovapd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,429496729]
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm1, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
+; AVX2-FAST-PERLANE-NEXT:    vzeroupper
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc_usat_v4i64_v4i32:
 ; AVX512F:       # %bb.0:
@@ -645,24 +662,42 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc_usat_v8i64_v8i32:
-; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovdqa (%rdi), %ymm0
-; AVX2-FAST-NEXT:    vmovdqa 32(%rdi), %ymm1
-; AVX2-FAST-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
-; AVX2-FAST-NEXT:    vpxor %ymm3, %ymm1, %ymm4
-; AVX2-FAST-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
-; AVX2-FAST-NEXT:    vblendvpd %ymm4, %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vpxor %ymm3, %ymm0, %ymm3
-; AVX2-FAST-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
-; AVX2-FAST-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vmovapd {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc_usat_v8i64_v8i32:
+; AVX2-FAST-ALL:       # %bb.0:
+; AVX2-FAST-ALL-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-ALL-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm3, %ymm1, %ymm4
+; AVX2-FAST-ALL-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm4, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vpxor %ymm3, %ymm0, %ymm3
+; AVX2-FAST-ALL-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
+; AVX2-FAST-ALL-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vmovapd {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc_usat_v8i64_v8i32:
+; AVX2-FAST-PERLANE:       # %bb.0:
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa (%rdi), %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vmovdqa 32(%rdi), %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4294967295,4294967295,4294967295,4294967295]
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [9223372036854775808,9223372036854775808,9223372036854775808,9223372036854775808]
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm3, %ymm1, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vpbroadcastq {{.*#+}} ymm5 = [9223372041149743103,9223372041149743103,9223372041149743103,9223372041149743103]
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm4, %ymm5, %ymm4
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm4, %ymm1, %ymm2, %ymm1
+; AVX2-FAST-PERLANE-NEXT:    vpxor %ymm3, %ymm0, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vpcmpgtq %ymm3, %ymm5, %ymm3
+; AVX2-FAST-PERLANE-NEXT:    vblendvpd %ymm3, %ymm0, %ymm2, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc_usat_v8i64_v8i32:
 ; AVX512:       # %bb.0:
diff --git a/llvm/test/CodeGen/X86/vector-trunc.ll b/llvm/test/CodeGen/X86/vector-trunc.ll
--- a/llvm/test/CodeGen/X86/vector-trunc.ll
+++ b/llvm/test/CodeGen/X86/vector-trunc.ll
@@ -4,11 +4,15 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX2,AVX2-FAST,AVX2-FAST-PERLANE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL
 
 define <8 x i32> @trunc8i64_8i32(<8 x i64> %a) {
 ; SSE-LABEL: trunc8i64_8i32:
@@ -32,13 +36,20 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc8i64_8i32:
-; AVX2-FAST:       # %bb.0: # %entry
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc8i64_8i32:
+; AVX2-FAST-ALL:       # %bb.0: # %entry
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc8i64_8i32:
+; AVX2-FAST-PERLANE:       # %bb.0: # %entry
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc8i64_8i32:
 ; AVX512:       # %bb.0: # %entry
@@ -71,13 +82,20 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc8i64_8i32_ashr:
-; AVX2-FAST:       # %bb.0: # %entry
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} xmm2 = [1,3,5,7]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc8i64_8i32_ashr:
+; AVX2-FAST-ALL:       # %bb.0: # %entry
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} xmm2 = [1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc8i64_8i32_ashr:
+; AVX2-FAST-PERLANE:       # %bb.0: # %entry
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc8i64_8i32_ashr:
 ; AVX512:       # %bb.0: # %entry
@@ -112,13 +130,20 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc8i64_8i32_lshr:
-; AVX2-FAST:       # %bb.0: # %entry
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} xmm2 = [1,3,5,7]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc8i64_8i32_lshr:
+; AVX2-FAST-ALL:       # %bb.0: # %entry
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} xmm2 = [1,3,5,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc8i64_8i32_lshr:
+; AVX2-FAST-PERLANE:       # %bb.0: # %entry
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[1,3],ymm2[1,3],ymm0[5,7],ymm2[5,7]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512-LABEL: trunc8i64_8i32_lshr:
 ; AVX512:       # %bb.0: # %entry
@@ -1354,13 +1379,20 @@
 ; AVX2-SLOW-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
 ; AVX2-SLOW-NEXT:    retq
 ;
-; AVX2-FAST-LABEL: trunc2x4i64_8i32:
-; AVX2-FAST:       # %bb.0: # %entry
-; AVX2-FAST-NEXT:    vmovaps {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
-; AVX2-FAST-NEXT:    vpermps %ymm0, %ymm2, %ymm0
-; AVX2-FAST-NEXT:    vpermps %ymm1, %ymm2, %ymm1
-; AVX2-FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; AVX2-FAST-NEXT:    retq
+; AVX2-FAST-ALL-LABEL: trunc2x4i64_8i32:
+; AVX2-FAST-ALL:       # %bb.0: # %entry
+; AVX2-FAST-ALL-NEXT:    vmovaps {{.*#+}} ymm2 = [0,2,4,6,4,6,6,7]
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm0, %ymm2, %ymm0
+; AVX2-FAST-ALL-NEXT:    vpermps %ymm1, %ymm2, %ymm1
+; AVX2-FAST-ALL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-ALL-NEXT:    retq
+;
+; AVX2-FAST-PERLANE-LABEL: trunc2x4i64_8i32:
+; AVX2-FAST-PERLANE:       # %bb.0: # %entry
+; AVX2-FAST-PERLANE-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm1[2,3]
+; AVX2-FAST-PERLANE-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX2-FAST-PERLANE-NEXT:    vshufps {{.*#+}} ymm0 = ymm0[0,2],ymm2[0,2],ymm0[4,6],ymm2[4,6]
+; AVX2-FAST-PERLANE-NEXT:    retq
 ;
 ; AVX512F-LABEL: trunc2x4i64_8i32:
 ; AVX512F:       # %bb.0: # %entry
diff --git a/llvm/test/CodeGen/X86/vector-zext.ll b/llvm/test/CodeGen/X86/vector-zext.ll
--- a/llvm/test/CodeGen/X86/vector-zext.ll
+++ b/llvm/test/CodeGen/X86/vector-zext.ll
@@ -4,9 +4,11 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 --check-prefix=AVX2-FAST
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512F
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+fast-variable-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW
 
 define <8 x i16> @zext_16i8_to_8i16(<16 x i8> %A) nounwind uwtable readnone ssp {
 ; SSE2-LABEL: zext_16i8_to_8i16: