[SLP] Mark conditional ops of hor reduction SelectInst as Ignored + propagate IR flags.

[ABataev]

Summary

If the min/max horizontal reduction is found, compare operands of the
SelectInst are considered as instruction that externally use reduction
values. But they must be ignored and their operands must be replaced by
UndefValues.
Also, patch fixes propagation of IR flags. Vectorized compare
instructions did not inherit IR flags from the scalar instructions.

lib/Transforms/Vectorize/SLPVectorizer.cpp

 ///  Or:
 ///    ...
 ///    \  /
 ///     +
 ///     |
 ///   *p =
 ///
 class HorizontalReduction {
-  SmallVector<Value *, 16> ReductionOps;
+  using ReductionOpsType = SmallVector<Value *, 16>;
+  using ReductionOpsListType = SmallVector<ReductionOpsType, 2>;
+  ReductionOpsListType  ReductionOps;
   SmallVector<Value *, 32> ReducedVals;
   // Use map vector to make stable output.
   MapVector<Instruction *, Value *> ExtraArgs;
 
   /// Kind of the reduction data.
   enum ReductionKind {
     RK_None,       /// Not a reduction.
     RK_Arithmetic, /// Binary reduction data.
bool isVectorizable() const {
              ((Kind == RK_Arithmetic &&
                (Opcode == Instruction::Add || Opcode == Instruction::FAdd)) ||
               ((Opcode == Instruction::ICmp || Opcode == Instruction::FCmp) &&
                (Kind == RK_Min || Kind == RK_Max)) ||
               (Opcode == Instruction::ICmp &&
                (Kind == RK_UMin || Kind == RK_UMax)));
     }
 
+    /// Creates reduction operation with the current opcode.
+    Value *createOp(IRBuilder<> &Builder, const Twine &Name) const {
+      assert(isVectorizable() &&
+             "Expected add|fadd or min/max reduction operation.");
+      Value *Cmp;
+      switch (Kind) {
+      case RK_Arithmetic:
+        return Builder.CreateBinOp((Instruction::BinaryOps)Opcode, LHS, RHS,
+                                   Name);
+      case RK_Min:
+        Cmp = Opcode == Instruction::ICmp ? Builder.CreateICmpSLT(LHS, RHS)
+                                          : Builder.CreateFCmpOLT(LHS, RHS);
+        break;
+      case RK_Max:
+        Cmp = Opcode == Instruction::ICmp ? Builder.CreateICmpSGT(LHS, RHS)
+                                          : Builder.CreateFCmpOGT(LHS, RHS);
+        break;
+      case RK_UMin:
+        assert(Opcode == Instruction::ICmp && "Expected integer types.");
+        Cmp = Builder.CreateICmpULT(LHS, RHS);
+        break;
+      case RK_UMax:
+        assert(Opcode == Instruction::ICmp && "Expected integer types.");
+        Cmp = Builder.CreateICmpUGT(LHS, RHS);
+        break;
+      case RK_None:
+        llvm_unreachable("Unknown reduction operation.");
+      }
+      return Builder.CreateSelect(Cmp, LHS, RHS, Name);
+    }
+
   public:
     explicit OperationData() = default;
 
     /// Construction for reduced values. They are identified by opcode only and
     /// don't have associated LHS/RHS values.
     explicit OperationData(Value *V) : Kind(RK_None) {
       if (auto *I = dyn_cast<Instruction>(V))
         Opcode = I->getOpcode();
unsigned getNumberOfOperands() const {
       case RK_UMax:
         return 3;
       case RK_None:
         break;
       }
       llvm_unreachable("Reduction kind is not set");
     }
 
+    /// Checks if the operation has the same parent as \p P.
+    bool hasSameParent(Instruction *I, Value *P, bool IsRedOp) const {
+      assert(Kind != RK_None && !!*this && LHS && RHS &&
+             "Expected reduction operation.");
+      if (!IsRedOp)
+        return I->getParent() == P;
+      switch (Kind) {
+      case RK_Arithmetic:
+        // Arithmetic reduction operation must be used once only.
+        return I->getParent() == P;
+      case RK_Min:
+      case RK_UMin:
+      case RK_Max:
+      case RK_UMax: {
+        // SelectInst must be used twice while the condition op must have single
+        // use only.
+        auto *Cmp = cast<Instruction>(cast<SelectInst>(I)->getCondition());
+        return I->getParent() == P && Cmp && Cmp->getParent() == P;
+      }
+      case RK_None:
+        break;
+      }
+      llvm_unreachable("Reduction kind is not set");
+    }
     /// Expected number of uses for reduction operations/reduced values.
     unsigned getRequiredNumberOfUses() const {
       assert(Kind != RK_None && !!*this && LHS && RHS &&
              "Expected reduction operation.");
       switch (Kind) {
       case RK_Arithmetic:
         return 1;
       case RK_Min:
       case RK_UMin:
       case RK_Max:
       case RK_UMax:
         return 2;
       case RK_None:
         break;
       }
       llvm_unreachable("Reduction kind is not set");
     }
 
+    /// Initializes the list of reduction operations.
+    void initReductionOps(ReductionOpsListType &ReductionOps) {
+      assert(Kind != RK_None && !!*this && LHS && RHS &&
+             "Expected reduction operation.");
+      switch (Kind) {
+      case RK_Arithmetic:
+        ReductionOps.assign(1, ReductionOpsType());
+        break;
+      case RK_Min:
+      case RK_UMin:
+      case RK_Max:
+      case RK_UMax:
+        ReductionOps.assign(2, ReductionOpsType());
+        break;
+      case RK_None:
+        llvm_unreachable("Reduction kind is not set");
+      }
+    }
+    /// Add all reduction operations for the reduction instruction \p I.
+    void addReductionOps(Instruction *I, ReductionOpsListType &ReductionOps) {
+      assert(Kind != RK_None && !!*this && LHS && RHS &&
+             "Expected reduction operation.");
+      switch (Kind) {
+      case RK_Arithmetic:
+        ReductionOps[0].emplace_back(I);
+        break;
+      case RK_Min:
+      case RK_UMin:
+      case RK_Max:
+      case RK_UMax:
+        if (cast<SelectInst>(I)->getCondition()->hasOneUse())
+          ReductionOps[0].emplace_back(cast<SelectInst>(I)->getCondition());
+        ReductionOps[1].emplace_back(I);
+        break;
+      case RK_None:
+        llvm_unreachable("Reduction kind is not set");
+      }
+    }
+
     /// Checks if instruction is associative and can be vectorized.
     bool isAssociative(Instruction *I) const {
       assert(Kind != RK_None && *this && LHS && RHS &&
              "Expected reduction operation.");
       switch (Kind) {
       case RK_Arithmetic:
         return I->isAssociative();
       case RK_Min:
Type *getConditionType() const {
       case RK_UMax:
         return CmpInst::makeCmpResultType(LHS->getType());
       case RK_None:
         break;
       }
       llvm_unreachable("Reduction kind is not set");
     }
 
-    /// Creates reduction operation with the current opcode.
-    Value *createOp(IRBuilder<> &Builder, const Twine &Name = "") const {
+    /// Creates reduction operation with the current opcode with the IR flags
+    /// from \p ReductionOps.
+    Value *createOp(IRBuilder<> &Builder, const Twine &Name,
+                    const ReductionOpsListType &ReductionOps) const {
       assert(isVectorizable() &&
              "Expected add|fadd or min/max reduction operation.");
-      Value *Cmp;
+      auto *Op = createOp(Builder, Name);
       switch (Kind) {
       case RK_Arithmetic:
-        return Builder.CreateBinOp((Instruction::BinaryOps)Opcode, LHS, RHS,
-                                   Name);
+        propagateIRFlags(Op, ReductionOps[0]);
+        return Op;
       case RK_Min:
-        Cmp = Opcode == Instruction::ICmp ? Builder.CreateICmpSLT(LHS, RHS)
-                                          : Builder.CreateFCmpOLT(LHS, RHS);
-        break;
       case RK_Max:
-        Cmp = Opcode == Instruction::ICmp ? Builder.CreateICmpSGT(LHS, RHS)
-                                          : Builder.CreateFCmpOGT(LHS, RHS);
-        break;
       case RK_UMin:
-        assert(Opcode == Instruction::ICmp && "Expected integer types.");
-        Cmp = Builder.CreateICmpULT(LHS, RHS);
-        break;
       case RK_UMax:
-        assert(Opcode == Instruction::ICmp && "Expected integer types.");
-        Cmp = Builder.CreateICmpUGT(LHS, RHS);
-        break;
+        propagateIRFlags(cast<SelectInst>(Op)->getCondition(), ReductionOps[0]);
+        propagateIRFlags(Op, ReductionOps[1]);
+        return Op;
       case RK_None:
-        llvm_unreachable("Unknown reduction operation.");
-      }
-      return Builder.CreateSelect(Cmp, LHS, RHS, Name);
+        break;
+      }
+    /// Creates reduction operation with the current opcode with the IR flags
+    /// from \p I.
+    Value *createOp(IRBuilder<> &Builder, const Twine &Name,
+                    Instruction *I) const {
+      assert(isVectorizable() &&
+             "Expected add|fadd or min/max reduction operation.");
+      auto *Op = createOp(Builder, Name);
+      switch (Kind) {
+      case RK_Arithmetic:
+        propagateIRFlags(Op, I);
+        return Op;
+      case RK_Min:
+      case RK_Max:
+      case RK_UMin:
+      case RK_UMax:
+        propagateIRFlags(cast<SelectInst>(Op)->getCondition(),
+                         cast<SelectInst>(I)->getCondition());
+        propagateIRFlags(Op, I);
+        return Op;
+      case RK_None:
+        break;
       }
+      llvm_unreachable("Unknown reduction operation.");
+    }
+
     TargetTransformInfo::ReductionFlags getFlags() const {
       TargetTransformInfo::ReductionFlags Flags;
       Flags.NoNaN = NoNaN;
       switch (Kind) {
       case RK_Arithmetic:
         break;
       case RK_Min:
         Flags.IsSigned = Opcode == Instruction::ICmp;
bool matchAssociativeReduction(PHINode *Phi, Instruction *B) {
     ReducedValueData.clear();
     ReductionRoot = B;
 
     // Post order traverse the reduction tree starting at B. We only handle true
     // trees containing only binary operators.
     SmallVector<std::pair<Instruction *, unsigned>, 32> Stack;
     Stack.push_back(std::make_pair(B, ReductionData.getFirstOperandIndex()));
     const unsigned NUses = ReductionData.getRequiredNumberOfUses();
+    ReductionData.initReductionOps(ReductionOps);
     while (!Stack.empty()) {
       Instruction *TreeN = Stack.back().first;
       unsigned EdgeToVist = Stack.back().second++;
       OperationData OpData = getOperationData(TreeN);
       bool IsReducedValue = OpData != ReductionData;
 
       // Postorder vist.
       if (IsReducedValue || EdgeToVist == OpData.getNumberOfOperands()) {
while (!Stack.empty()) {
               return false;
             }
             // Yes, TreeN is an extra argument, do not add it to a list of
             // reduction operations.
             // Stack[Stack.size() - 2] always points to the parent operation.
             markExtraArg(Stack[Stack.size() - 2], TreeN);
             ExtraArgs.erase(TreeN);
           } else
-            ReductionOps.push_back(TreeN);
+            ReductionData.addReductionOps(TreeN, ReductionOps);
         }
         // Retract.
         Stack.pop_back();
         continue;
       }
 
       // Visit left or right.
       Value *NextV = TreeN->getOperand(EdgeToVist);
       if (NextV != Phi) {
         auto *I = dyn_cast<Instruction>(NextV);
         OpData = getOperationData(I);
         // Continue analysis if the next operand is a reduction operation or
         // (possibly) a reduced value. If the reduced value opcode is not set,
         // the first met operation != reduction operation is considered as the
         // reduced value class.
         if (I && (!ReducedValueData || OpData == ReducedValueData ||
                   OpData == ReductionData)) {
+          const bool IsReductionOperation = OpData == ReductionData;
           // Only handle trees in the current basic block.
-          if (I->getParent() != B->getParent()) {
+          if (!ReductionData.hasSameParent(I, B->getParent(),
+                                           IsReductionOperation)) {
             // I is an extra argument for TreeN (its parent operation).
             markExtraArg(Stack.back(), I);
             continue;
           }
 
           // Each tree node needs to have minimal number of users except for the
           // ultimate reduction.
           if (!I->hasNUses(NUses) && I != B) {
             // I is an extra argument for TreeN (its parent operation).
             markExtraArg(Stack.back(), I);
             continue;
           }
 
-          if (OpData == ReductionData) {
+          if (IsReductionOperation) {
             // We need to be able to reassociate the reduction operations.
             if (!OpData.isAssociative(I)) {
               // I is an extra argument for TreeN (its parent operation).
               markExtraArg(Stack.back(), I);
               continue;
             }
           } else if (ReducedValueData &&
                      ReducedValueData != OpData) {
bool tryToReduce(BoUpSLP &V, TargetTransformInfo *TTI) {
     Builder.setFastMathFlags(Unsafe);
     unsigned i = 0;
 
     BoUpSLP::ExtraValueToDebugLocsMap ExternallyUsedValues;
     // The same extra argument may be used several time, so log each attempt
     // to use it.
     for (auto &Pair : ExtraArgs)
       ExternallyUsedValues[Pair.second].push_back(Pair.first);
+    SmallVector<Value *, 16> IgnoreList;
+    for (auto &V : ReductionOps)
+      IgnoreList.append(V.begin(), V.end());
     while (i < NumReducedVals - ReduxWidth + 1 && ReduxWidth > 2) {
       auto VL = makeArrayRef(&ReducedVals[i], ReduxWidth);
-      V.buildTree(VL, ExternallyUsedValues, ReductionOps);
+      V.buildTree(VL, ExternallyUsedValues, IgnoreList);
       if (V.shouldReorder()) {
         SmallVector<Value *, 8> Reversed(VL.rbegin(), VL.rend());
-        V.buildTree(Reversed, ExternallyUsedValues, ReductionOps);
+        V.buildTree(Reversed, ExternallyUsedValues, IgnoreList);
       }
       if (V.isTreeTinyAndNotFullyVectorizable())
         break;
 
       V.computeMinimumValueSizes();
 
       // Estimate cost.
       int Cost =
while (i < NumReducedVals - ReduxWidth + 1 && ReduxWidth > 2) {
           << ore::NV("TreeSize", V.getTreeSize()));
 
       // Vectorize a tree.
       DebugLoc Loc = cast<Instruction>(ReducedVals[i])->getDebugLoc();
       Value *VectorizedRoot = V.vectorizeTree(ExternallyUsedValues);
 
       // Emit a reduction.
       Value *ReducedSubTree =
-          emitReduction(VectorizedRoot, Builder, ReduxWidth, ReductionOps, TTI);
+          emitReduction(VectorizedRoot, Builder, ReduxWidth, TTI);
       if (VectorizedTree) {
         Builder.SetCurrentDebugLocation(Loc);
         OperationData VectReductionData(ReductionData.getOpcode(),
                                         VectorizedTree, ReducedSubTree,
                                         ReductionData.getKind());
-        VectorizedTree = VectReductionData.createOp(Builder, "op.rdx");
-        propagateIRFlags(VectorizedTree, ReductionOps);
+        VectorizedTree =
+            VectReductionData.createOp(Builder, "op.rdx", ReductionOps);
       } else
         VectorizedTree = ReducedSubTree;
       i += ReduxWidth;
       ReduxWidth = PowerOf2Floor(NumReducedVals - i);
     }
 
     if (VectorizedTree) {
       // Finish the reduction.
       for (; i < NumReducedVals; ++i) {
         auto *I = cast<Instruction>(ReducedVals[i]);
         Builder.SetCurrentDebugLocation(I->getDebugLoc());
         OperationData VectReductionData(ReductionData.getOpcode(),
                                         VectorizedTree, I,
                                         ReductionData.getKind());
-        VectorizedTree = VectReductionData.createOp(Builder);
-        propagateIRFlags(VectorizedTree, ReductionOps);
+        VectorizedTree = VectReductionData.createOp(Builder, "", ReductionOps);
       }
       for (auto &Pair : ExternallyUsedValues) {
         assert(!Pair.second.empty() &&
                "At least one DebugLoc must be inserted");
         // Add each externally used value to the final reduction.
         for (auto *I : Pair.second) {
           Builder.SetCurrentDebugLocation(I->getDebugLoc());
           OperationData VectReductionData(ReductionData.getOpcode(),
                                           VectorizedTree, Pair.first,
                                           ReductionData.getKind());
-          VectorizedTree = VectReductionData.createOp(Builder, "op.extra");
-          propagateIRFlags(VectorizedTree, I);
+          VectorizedTree = VectReductionData.createOp(Builder, "op.extra", I);
         }
       }
       // Update users.
       ReductionRoot->replaceAllUsesWith(VectorizedTree);
     }
     return VectorizedTree != nullptr;
   }
 
DEBUG(dbgs() << "SLP: Adding cost " << VecReduxCost - ScalarReduxCost
                  << (IsPairwiseReduction ? "pairwise" : "splitting")
                  << " reduction)\n");
 
     return VecReduxCost - ScalarReduxCost;
   }
 
   /// \brief Emit a horizontal reduction of the vectorized value.
   Value *emitReduction(Value *VectorizedValue, IRBuilder<> &Builder,
-                       unsigned ReduxWidth, ArrayRef<Value *> RedOps,
-                       const TargetTransformInfo *TTI) {
+                       unsigned ReduxWidth, const TargetTransformInfo *TTI) {
     assert(VectorizedValue && "Need to have a vectorized tree node");
     assert(isPowerOf2_32(ReduxWidth) &&
            "We only handle power-of-two reductions for now");
 
     if (!IsPairwiseReduction)
       return createSimpleTargetReduction(
           Builder, TTI, ReductionData.getOpcode(), VectorizedValue,
-          ReductionData.getFlags(), RedOps);
+          ReductionData.getFlags(), ReductionOps.back());
 
     Value *TmpVec = VectorizedValue;
     for (unsigned i = ReduxWidth / 2; i != 0; i >>= 1) {
       Value *LeftMask =
           createRdxShuffleMask(ReduxWidth, i, true, true, Builder);
       Value *RightMask =
           createRdxShuffleMask(ReduxWidth, i, true, false, Builder);
 
       Value *LeftShuf = Builder.CreateShuffleVector(
           TmpVec, UndefValue::get(TmpVec->getType()), LeftMask, "rdx.shuf.l");
       Value *RightShuf = Builder.CreateShuffleVector(
           TmpVec, UndefValue::get(TmpVec->getType()), (RightMask),
           "rdx.shuf.r");
       OperationData VectReductionData(ReductionData.getOpcode(), LeftShuf,
                                       RightShuf, ReductionData.getKind());
-      TmpVec = VectReductionData.createOp(Builder, "op.rdx");
-      propagateIRFlags(TmpVec, RedOps);
+      TmpVec = VectReductionData.createOp(Builder, "op.rdx", ReductionOps);
     }
 
     // The result is in the first element of the vector.
     return Builder.CreateExtractElement(TmpVec, Builder.getInt32(0));
   }
 };
 
 } // end anonymous namespace

test/Transforms/SLPVectorizer/X86/horizontal-list.ll

 }
 
 define float @bar() {
 ; CHECK-LABEL: @bar(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x float>, <4 x float>* bitcast ([20 x float]* @arr to <4 x float>*), align 16
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([20 x float]* @arr1 to <4 x float>*), align 16
 ; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast <4 x float> [[TMP1]], [[TMP0]]
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP2]], i32 1
-; CHECK-NEXT:    [[CMP4:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; CHECK-NEXT:    [[CMP4:%.*]] = fcmp fast ogt float undef, undef
 ; CHECK-NEXT:    [[MAX_0_MUL3:%.*]] = select i1 [[CMP4]], float undef, float undef
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
-; CHECK-NEXT:    [[CMP4_1:%.*]] = fcmp fast ogt float [[MAX_0_MUL3]], [[TMP5]]
+; CHECK-NEXT:    [[CMP4_1:%.*]] = fcmp fast ogt float [[MAX_0_MUL3]], undef
 ; CHECK-NEXT:    [[MAX_0_MUL3_1:%.*]] = select i1 [[CMP4_1]], float [[MAX_0_MUL3]], float undef
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP2]], i32 3
-; CHECK-NEXT:    [[CMP4_2:%.*]] = fcmp fast ogt float [[MAX_0_MUL3_1]], [[TMP6]]
+; CHECK-NEXT:    [[CMP4_2:%.*]] = fcmp fast ogt float [[MAX_0_MUL3_1]], undef
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <4 x float> [[TMP2]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x float> [[TMP2]], <4 x float> [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x float> [[RDX_MINMAX_SELECT]], <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <4 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x float> [[RDX_MINMAX_SELECT]], <4 x float> [[RDX_SHUF1]]
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[RDX_MINMAX_SELECT3]], i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[RDX_MINMAX_SELECT3]], i32 0
 ; CHECK-NEXT:    [[MAX_0_MUL3_2:%.*]] = select i1 [[CMP4_2]], float [[MAX_0_MUL3_1]], float undef
-; CHECK-NEXT:    store float [[TMP7]], float* @res, align 4
-; CHECK-NEXT:    ret float [[TMP7]]
+; CHECK-NEXT:    store float [[TMP3]], float* @res, align 4
+; CHECK-NEXT:    ret float [[TMP3]]
 ;
 ; THRESHOLD-LABEL: @bar(
 ; THRESHOLD-NEXT:  entry:
 ; THRESHOLD-NEXT:    [[TMP0:%.*]] = load <4 x float>, <4 x float>* bitcast ([20 x float]* @arr to <4 x float>*), align 16
 ; THRESHOLD-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* bitcast ([20 x float]* @arr1 to <4 x float>*), align 16
 ; THRESHOLD-NEXT:    [[TMP2:%.*]] = fmul fast <4 x float> [[TMP1]], [[TMP0]]
-; THRESHOLD-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
-; THRESHOLD-NEXT:    [[TMP4:%.*]] = extractelement <4 x float> [[TMP2]], i32 1
-; THRESHOLD-NEXT:    [[CMP4:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
+; THRESHOLD-NEXT:    [[CMP4:%.*]] = fcmp fast ogt float undef, undef
 ; THRESHOLD-NEXT:    [[MAX_0_MUL3:%.*]] = select i1 [[CMP4]], float undef, float undef
-; THRESHOLD-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
-; THRESHOLD-NEXT:    [[CMP4_1:%.*]] = fcmp fast ogt float [[MAX_0_MUL3]], [[TMP5]]
+; THRESHOLD-NEXT:    [[CMP4_1:%.*]] = fcmp fast ogt float [[MAX_0_MUL3]], undef
 ; THRESHOLD-NEXT:    [[MAX_0_MUL3_1:%.*]] = select i1 [[CMP4_1]], float [[MAX_0_MUL3]], float undef
-; THRESHOLD-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP2]], i32 3
-; THRESHOLD-NEXT:    [[CMP4_2:%.*]] = fcmp fast ogt float [[MAX_0_MUL3_1]], [[TMP6]]
+; THRESHOLD-NEXT:    [[CMP4_2:%.*]] = fcmp fast ogt float [[MAX_0_MUL3_1]], undef
 ; THRESHOLD-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; THRESHOLD-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <4 x float> [[TMP2]], [[RDX_SHUF]]
 ; THRESHOLD-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x float> [[TMP2]], <4 x float> [[RDX_SHUF]]
 ; THRESHOLD-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x float> [[RDX_MINMAX_SELECT]], <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; THRESHOLD-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <4 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; THRESHOLD-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x float> [[RDX_MINMAX_SELECT]], <4 x float> [[RDX_SHUF1]]
-; THRESHOLD-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[RDX_MINMAX_SELECT3]], i32 0
+; THRESHOLD-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[RDX_MINMAX_SELECT3]], i32 0
 ; THRESHOLD-NEXT:    [[MAX_0_MUL3_2:%.*]] = select i1 [[CMP4_2]], float [[MAX_0_MUL3_1]], float undef
-; THRESHOLD-NEXT:    store float [[TMP7]], float* @res, align 4
-; THRESHOLD-NEXT:    ret float [[TMP7]]
+; THRESHOLD-NEXT:    store float [[TMP3]], float* @res, align 4
+; THRESHOLD-NEXT:    ret float [[TMP3]]
 ;
 entry:
   %0 = load float, float* getelementptr inbounds ([20 x float], [20 x float]* @arr, i64 0, i64 0), align 16
   %1 = load float, float* getelementptr inbounds ([20 x float], [20 x float]* @arr1, i64 0, i64 0), align 16
   %mul = fmul fast float %1, %0
   %2 = load float, float* getelementptr inbounds ([20 x float], [20 x float]* @arr, i64 0, i64 1), align 4
   %3 = load float, float* getelementptr inbounds ([20 x float], [20 x float]* @arr1, i64 0, i64 1), align 4
   %mul3 = fmul fast float %3, %2

test/Transforms/SLPVectorizer/X86/horizontal-minmax.ll

 ; CHECK-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
 ; CHECK-NEXT:    [[TMP21:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
 ; CHECK-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], [[TMP21]]
 ; CHECK-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP20]], i32 [[TMP21]]
 ; CHECK-NEXT:    ret i32 [[TMP23]]
 ;
 ; AVX-LABEL: @maxi8(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
-; AVX-NEXT:    [[TMP3:%.*]] = extractelement <8 x i32> [[TMP2]], i32 0
-; AVX-NEXT:    [[TMP4:%.*]] = extractelement <8 x i32> [[TMP2]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <8 x i32> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <8 x i32> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[TMP2]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[TMP2]], i32 5
-; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX-NEXT:    [[TMP19:%.*]] = extractelement <8 x i32> [[TMP2]], i32 6
-; AVX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; AVX-NEXT:    [[TMP22:%.*]] = extractelement <8 x i32> [[TMP2]], i32 7
-; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; AVX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; AVX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
-; AVX-NEXT:    [[TMP24:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
-; AVX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; AVX-NEXT:    ret i32 [[TMP24]]
+; AVX-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX-NEXT:    [[TMP17:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; AVX-NEXT:    ret i32 [[TMP16]]
 ;
 ; AVX2-LABEL: @maxi8(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
-; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <8 x i32> [[TMP2]], i32 0
-; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <8 x i32> [[TMP2]], i32 1
-; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <8 x i32> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <8 x i32> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[TMP2]], i32 4
-; AVX2-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[TMP2]], i32 5
-; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <8 x i32> [[TMP2]], i32 6
-; AVX2-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; AVX2-NEXT:    [[TMP22:%.*]] = extractelement <8 x i32> [[TMP2]], i32 7
-; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX2-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; AVX2-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; AVX2-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX2-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
-; AVX2-NEXT:    [[TMP24:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
-; AVX2-NEXT:    [[TMP25:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; AVX2-NEXT:    ret i32 [[TMP24]]
+; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX2-NEXT:    [[TMP17:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; AVX2-NEXT:    ret i32 [[TMP16]]
 ;
 ; SKX-LABEL: @maxi8(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <8 x i32>, <8 x i32>* bitcast ([32 x i32]* @arr to <8 x i32>*), align 16
-; SKX-NEXT:    [[TMP3:%.*]] = extractelement <8 x i32> [[TMP2]], i32 0
-; SKX-NEXT:    [[TMP4:%.*]] = extractelement <8 x i32> [[TMP2]], i32 1
-; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <8 x i32> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <8 x i32> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; SKX-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[TMP2]], i32 4
-; SKX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; SKX-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[TMP2]], i32 5
-; SKX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; SKX-NEXT:    [[TMP19:%.*]] = extractelement <8 x i32> [[TMP2]], i32 6
-; SKX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; SKX-NEXT:    [[TMP22:%.*]] = extractelement <8 x i32> [[TMP2]], i32 7
-; SKX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; SKX-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; SKX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; SKX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; SKX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; SKX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; SKX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; SKX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP2]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <8 x i32> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x i32> [[TMP2]], <8 x i32> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x i32> [[RDX_MINMAX_SELECT]], <8 x i32> [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <8 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x i32> [[RDX_MINMAX_SELECT3]], <8 x i32> [[RDX_SHUF4]]
-; SKX-NEXT:    [[TMP24:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
-; SKX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; SKX-NEXT:    ret i32 [[TMP24]]
+; SKX-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[RDX_MINMAX_SELECT6]], i32 0
+; SKX-NEXT:    [[TMP17:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; SKX-NEXT:    ret i32 [[TMP16]]
 ;
   %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
   %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
   %4 = icmp sgt i32 %2, %3
   %5 = select i1 %4, i32 %2, i32 %3
   %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
   %7 = icmp sgt i32 %5, %6
   %8 = select i1 %7, i32 %5, i32 %6
;
   %21 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
   %22 = icmp sgt i32 %20, %21
   %23 = select i1 %22, i32 %20, i32 %21
   ret i32 %23
 }
 
 define i32 @maxi16(i32) {
 ; CHECK-LABEL: @maxi16(
-; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
-; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
-; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
-; CHECK-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], [[TMP6]]
-; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 [[TMP6]]
-; CHECK-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 3), align 4
-; CHECK-NEXT:    [[TMP10:%.*]] = icmp sgt i32 [[TMP8]], [[TMP9]]
-; CHECK-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP8]], i32 [[TMP9]]
-; CHECK-NEXT:    [[TMP12:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
-; CHECK-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP11]], [[TMP12]]
-; CHECK-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP11]], i32 [[TMP12]]
-; CHECK-NEXT:    [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
-; CHECK-NEXT:    [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
-; CHECK-NEXT:    [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
-; CHECK-NEXT:    [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; CHECK-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
-; CHECK-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
-; CHECK-NEXT:    [[TMP21:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
-; CHECK-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP20]], [[TMP21]]
-; CHECK-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP20]], i32 [[TMP21]]
-; CHECK-NEXT:    [[TMP24:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 8), align 16
-; CHECK-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP23]], [[TMP24]]
-; CHECK-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP23]], i32 [[TMP24]]
-; CHECK-NEXT:    [[TMP27:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 9), align 4
-; CHECK-NEXT:    [[TMP28:%.*]] = icmp sgt i32 [[TMP26]], [[TMP27]]
-; CHECK-NEXT:    [[TMP29:%.*]] = select i1 [[TMP28]], i32 [[TMP26]], i32 [[TMP27]]
-; CHECK-NEXT:    [[TMP30:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 10), align 8
-; CHECK-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP29]], [[TMP30]]
-; CHECK-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP29]], i32 [[TMP30]]
-; CHECK-NEXT:    [[TMP33:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 11), align 4
-; CHECK-NEXT:    [[TMP34:%.*]] = icmp sgt i32 [[TMP32]], [[TMP33]]
-; CHECK-NEXT:    [[TMP35:%.*]] = select i1 [[TMP34]], i32 [[TMP32]], i32 [[TMP33]]
-; CHECK-NEXT:    [[TMP36:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 12), align 16
-; CHECK-NEXT:    [[TMP37:%.*]] = icmp sgt i32 [[TMP35]], [[TMP36]]
-; CHECK-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], i32 [[TMP35]], i32 [[TMP36]]
-; CHECK-NEXT:    [[TMP39:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 13), align 4
-; CHECK-NEXT:    [[TMP40:%.*]] = icmp sgt i32 [[TMP38]], [[TMP39]]
-; CHECK-NEXT:    [[TMP41:%.*]] = select i1 [[TMP40]], i32 [[TMP38]], i32 [[TMP39]]
-; CHECK-NEXT:    [[TMP42:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 14), align 8
-; CHECK-NEXT:    [[TMP43:%.*]] = icmp sgt i32 [[TMP41]], [[TMP42]]
-; CHECK-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], i32 [[TMP41]], i32 [[TMP42]]
-; CHECK-NEXT:    [[TMP45:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 15), align 4
-; CHECK-NEXT:    [[TMP46:%.*]] = icmp sgt i32 [[TMP44]], [[TMP45]]
-; CHECK-NEXT:    [[TMP47:%.*]] = select i1 [[TMP46]], i32 [[TMP44]], i32 [[TMP45]]
-; CHECK-NEXT:    ret i32 [[TMP47]]
+; CHECK-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; CHECK-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; CHECK-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; CHECK-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; CHECK-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; CHECK-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; CHECK-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; CHECK-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; CHECK-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; CHECK-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; CHECK-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; CHECK-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; CHECK-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; CHECK-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; CHECK-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; CHECK-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; CHECK-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; CHECK-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; CHECK-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; CHECK-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
+; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
+; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
+; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
+; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
+; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
+; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
+; CHECK-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
+; CHECK-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; CHECK-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; CHECK-NEXT:    ret i32 [[TMP32]]
 ;
 ; AVX-LABEL: @maxi16(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
-; AVX-NEXT:    [[TMP3:%.*]] = extractelement <16 x i32> [[TMP2]], i32 0
-; AVX-NEXT:    [[TMP4:%.*]] = extractelement <16 x i32> [[TMP2]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <16 x i32> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <16 x i32> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <16 x i32> [[TMP2]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <16 x i32> [[TMP2]], i32 5
-; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX-NEXT:    [[TMP19:%.*]] = extractelement <16 x i32> [[TMP2]], i32 6
-; AVX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; AVX-NEXT:    [[TMP22:%.*]] = extractelement <16 x i32> [[TMP2]], i32 7
-; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; AVX-NEXT:    [[TMP25:%.*]] = extractelement <16 x i32> [[TMP2]], i32 8
-; AVX-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; AVX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; AVX-NEXT:    [[TMP28:%.*]] = extractelement <16 x i32> [[TMP2]], i32 9
-; AVX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; AVX-NEXT:    [[TMP31:%.*]] = extractelement <16 x i32> [[TMP2]], i32 10
-; AVX-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; AVX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; AVX-NEXT:    [[TMP34:%.*]] = extractelement <16 x i32> [[TMP2]], i32 11
-; AVX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; AVX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; AVX-NEXT:    [[TMP37:%.*]] = extractelement <16 x i32> [[TMP2]], i32 12
-; AVX-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; AVX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; AVX-NEXT:    [[TMP40:%.*]] = extractelement <16 x i32> [[TMP2]], i32 13
-; AVX-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; AVX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; AVX-NEXT:    [[TMP43:%.*]] = extractelement <16 x i32> [[TMP2]], i32 14
-; AVX-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; AVX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; AVX-NEXT:    [[TMP46:%.*]] = extractelement <16 x i32> [[TMP2]], i32 15
-; AVX-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; AVX-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; AVX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; AVX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; AVX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; AVX-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; AVX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; AVX-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; AVX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; AVX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; AVX-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; AVX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; AVX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; AVX-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
-; AVX-NEXT:    [[TMP48:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
-; AVX-NEXT:    [[TMP49:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; AVX-NEXT:    ret i32 [[TMP48]]
+; AVX-NEXT:    [[TMP32:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; AVX-NEXT:    ret i32 [[TMP32]]
 ;
 ; AVX2-LABEL: @maxi16(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
-; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <16 x i32> [[TMP2]], i32 0
-; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <16 x i32> [[TMP2]], i32 1
-; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <16 x i32> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <16 x i32> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <16 x i32> [[TMP2]], i32 4
-; AVX2-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <16 x i32> [[TMP2]], i32 5
-; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <16 x i32> [[TMP2]], i32 6
-; AVX2-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; AVX2-NEXT:    [[TMP22:%.*]] = extractelement <16 x i32> [[TMP2]], i32 7
-; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; AVX2-NEXT:    [[TMP25:%.*]] = extractelement <16 x i32> [[TMP2]], i32 8
-; AVX2-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; AVX2-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; AVX2-NEXT:    [[TMP28:%.*]] = extractelement <16 x i32> [[TMP2]], i32 9
-; AVX2-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; AVX2-NEXT:    [[TMP31:%.*]] = extractelement <16 x i32> [[TMP2]], i32 10
-; AVX2-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; AVX2-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; AVX2-NEXT:    [[TMP34:%.*]] = extractelement <16 x i32> [[TMP2]], i32 11
-; AVX2-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; AVX2-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; AVX2-NEXT:    [[TMP37:%.*]] = extractelement <16 x i32> [[TMP2]], i32 12
-; AVX2-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; AVX2-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; AVX2-NEXT:    [[TMP40:%.*]] = extractelement <16 x i32> [[TMP2]], i32 13
-; AVX2-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; AVX2-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; AVX2-NEXT:    [[TMP43:%.*]] = extractelement <16 x i32> [[TMP2]], i32 14
-; AVX2-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; AVX2-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; AVX2-NEXT:    [[TMP46:%.*]] = extractelement <16 x i32> [[TMP2]], i32 15
-; AVX2-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; AVX2-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; AVX2-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; AVX2-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX2-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; AVX2-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; AVX2-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; AVX2-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; AVX2-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; AVX2-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; AVX2-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; AVX2-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; AVX2-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; AVX2-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; AVX2-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
-; AVX2-NEXT:    [[TMP48:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
-; AVX2-NEXT:    [[TMP49:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; AVX2-NEXT:    ret i32 [[TMP48]]
+; AVX2-NEXT:    [[TMP32:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX2-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; AVX2-NEXT:    ret i32 [[TMP32]]
 ;
 ; SKX-LABEL: @maxi16(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <16 x i32>, <16 x i32>* bitcast ([32 x i32]* @arr to <16 x i32>*), align 16
-; SKX-NEXT:    [[TMP3:%.*]] = extractelement <16 x i32> [[TMP2]], i32 0
-; SKX-NEXT:    [[TMP4:%.*]] = extractelement <16 x i32> [[TMP2]], i32 1
-; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <16 x i32> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <16 x i32> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; SKX-NEXT:    [[TMP13:%.*]] = extractelement <16 x i32> [[TMP2]], i32 4
-; SKX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; SKX-NEXT:    [[TMP16:%.*]] = extractelement <16 x i32> [[TMP2]], i32 5
-; SKX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; SKX-NEXT:    [[TMP19:%.*]] = extractelement <16 x i32> [[TMP2]], i32 6
-; SKX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; SKX-NEXT:    [[TMP22:%.*]] = extractelement <16 x i32> [[TMP2]], i32 7
-; SKX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; SKX-NEXT:    [[TMP25:%.*]] = extractelement <16 x i32> [[TMP2]], i32 8
-; SKX-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; SKX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; SKX-NEXT:    [[TMP28:%.*]] = extractelement <16 x i32> [[TMP2]], i32 9
-; SKX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; SKX-NEXT:    [[TMP31:%.*]] = extractelement <16 x i32> [[TMP2]], i32 10
-; SKX-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; SKX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; SKX-NEXT:    [[TMP34:%.*]] = extractelement <16 x i32> [[TMP2]], i32 11
-; SKX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; SKX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; SKX-NEXT:    [[TMP37:%.*]] = extractelement <16 x i32> [[TMP2]], i32 12
-; SKX-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; SKX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; SKX-NEXT:    [[TMP40:%.*]] = extractelement <16 x i32> [[TMP2]], i32 13
-; SKX-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; SKX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; SKX-NEXT:    [[TMP43:%.*]] = extractelement <16 x i32> [[TMP2]], i32 14
-; SKX-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; SKX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; SKX-NEXT:    [[TMP46:%.*]] = extractelement <16 x i32> [[TMP2]], i32 15
-; SKX-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
+; SKX-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; SKX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; SKX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; SKX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; SKX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; SKX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; SKX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; SKX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; SKX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; SKX-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; SKX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; SKX-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; SKX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; SKX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; SKX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; SKX-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; SKX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; SKX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; SKX-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x i32> [[TMP2]], <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <16 x i32> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x i32> [[TMP2]], <16 x i32> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x i32> [[RDX_MINMAX_SELECT]], <16 x i32> [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x i32> [[RDX_MINMAX_SELECT3]], <16 x i32> [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <16 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x i32> [[RDX_MINMAX_SELECT6]], <16 x i32> [[RDX_SHUF7]]
-; SKX-NEXT:    [[TMP48:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
-; SKX-NEXT:    [[TMP49:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; SKX-NEXT:    ret i32 [[TMP48]]
+; SKX-NEXT:    [[TMP32:%.*]] = extractelement <16 x i32> [[RDX_MINMAX_SELECT9]], i32 0
+; SKX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; SKX-NEXT:    ret i32 [[TMP32]]
 ;
   %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
   %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
   %4 = icmp sgt i32 %2, %3
   %5 = select i1 %4, i32 %2, i32 %3
   %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
   %7 = icmp sgt i32 %5, %6
   %8 = select i1 %7, i32 %5, i32 %6
;
   %46 = icmp sgt i32 %44, %45
   %47 = select i1 %46, i32 %44, i32 %45
   ret i32 %47
 }
 
 define i32 @maxi32(i32) {
 ; CHECK-LABEL: @maxi32(
 ; CHECK-NEXT:    [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
-; CHECK-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
-; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; CHECK-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
-; CHECK-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; CHECK-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
-; CHECK-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; CHECK-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
-; CHECK-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; CHECK-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
-; CHECK-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; CHECK-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
-; CHECK-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; CHECK-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
-; CHECK-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; CHECK-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
-; CHECK-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; CHECK-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
-; CHECK-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; CHECK-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
-; CHECK-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; CHECK-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
-; CHECK-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; CHECK-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; CHECK-NEXT:    [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
-; CHECK-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; CHECK-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
-; CHECK-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; CHECK-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
-; CHECK-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
-; CHECK-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; CHECK-NEXT:    [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
-; CHECK-NEXT:    [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
-; CHECK-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
-; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
-; CHECK-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
-; CHECK-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
-; CHECK-NEXT:    [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
-; CHECK-NEXT:    [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
-; CHECK-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
-; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
-; CHECK-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
-; CHECK-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
-; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
-; CHECK-NEXT:    [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
-; CHECK-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
-; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
-; CHECK-NEXT:    [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
-; CHECK-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
-; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
-; CHECK-NEXT:    [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
-; CHECK-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
-; CHECK-NEXT:    [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
-; CHECK-NEXT:    [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
-; CHECK-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
-; CHECK-NEXT:    [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
-; CHECK-NEXT:    [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
-; CHECK-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
-; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
-; CHECK-NEXT:    [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
-; CHECK-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
-; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
-; CHECK-NEXT:    [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
-; CHECK-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
-; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
-; CHECK-NEXT:    [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
-; CHECK-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
-; CHECK-NEXT:    [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
-; CHECK-NEXT:    [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
-; CHECK-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
-; CHECK-NEXT:    [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
-; CHECK-NEXT:    [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
-; CHECK-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
-; CHECK-NEXT:    [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
-; CHECK-NEXT:    [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
-; CHECK-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
-; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
-; CHECK-NEXT:    [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; CHECK-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; CHECK-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; CHECK-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; CHECK-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; CHECK-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; CHECK-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; CHECK-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; CHECK-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; CHECK-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; CHECK-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; CHECK-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; CHECK-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; CHECK-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; CHECK-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; CHECK-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; CHECK-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; CHECK-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; CHECK-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; CHECK-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; CHECK-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; CHECK-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; CHECK-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; CHECK-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
+; CHECK-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; CHECK-NEXT:    [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], undef
+; CHECK-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 undef
+; CHECK-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP34]], undef
+; CHECK-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP34]], i32 undef
+; CHECK-NEXT:    [[TMP37:%.*]] = icmp sgt i32 [[TMP36]], undef
+; CHECK-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], i32 [[TMP36]], i32 undef
+; CHECK-NEXT:    [[TMP39:%.*]] = icmp sgt i32 [[TMP38]], undef
+; CHECK-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], i32 [[TMP38]], i32 undef
+; CHECK-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP40]], undef
+; CHECK-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP40]], i32 undef
+; CHECK-NEXT:    [[TMP43:%.*]] = icmp sgt i32 [[TMP42]], undef
+; CHECK-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], i32 [[TMP42]], i32 undef
+; CHECK-NEXT:    [[TMP45:%.*]] = icmp sgt i32 [[TMP44]], undef
+; CHECK-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], i32 [[TMP44]], i32 undef
+; CHECK-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP46]], undef
+; CHECK-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP46]], i32 undef
+; CHECK-NEXT:    [[TMP49:%.*]] = icmp sgt i32 [[TMP48]], undef
+; CHECK-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], i32 [[TMP48]], i32 undef
+; CHECK-NEXT:    [[TMP51:%.*]] = icmp sgt i32 [[TMP50]], undef
+; CHECK-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], i32 [[TMP50]], i32 undef
+; CHECK-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP52]], undef
+; CHECK-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP52]], i32 undef
+; CHECK-NEXT:    [[TMP55:%.*]] = icmp sgt i32 [[TMP54]], undef
+; CHECK-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], i32 [[TMP54]], i32 undef
+; CHECK-NEXT:    [[TMP57:%.*]] = icmp sgt i32 [[TMP56]], undef
+; CHECK-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], i32 [[TMP56]], i32 undef
+; CHECK-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP58]], undef
+; CHECK-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP58]], i32 undef
+; CHECK-NEXT:    [[TMP61:%.*]] = icmp sgt i32 [[TMP60]], undef
+; CHECK-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], i32 [[TMP60]], i32 undef
+; CHECK-NEXT:    [[TMP63:%.*]] = icmp sgt i32 [[TMP62]], undef
 ; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
 ; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
 ; CHECK-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
 ; CHECK-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
 ; CHECK-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; CHECK-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
-; CHECK-NEXT:    [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
-; CHECK-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
-; CHECK-NEXT:    ret i32 [[TMP96]]
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; CHECK-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], i32 [[TMP62]], i32 undef
+; CHECK-NEXT:    ret i32 [[TMP64]]
 ;
 ; AVX-LABEL: @maxi32(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; AVX-NEXT:    [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
-; AVX-NEXT:    [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
-; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX-NEXT:    [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
-; AVX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; AVX-NEXT:    [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
-; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; AVX-NEXT:    [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
-; AVX-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; AVX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; AVX-NEXT:    [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
-; AVX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; AVX-NEXT:    [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
-; AVX-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; AVX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; AVX-NEXT:    [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
-; AVX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; AVX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; AVX-NEXT:    [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
-; AVX-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; AVX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; AVX-NEXT:    [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
-; AVX-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; AVX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; AVX-NEXT:    [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
-; AVX-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; AVX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; AVX-NEXT:    [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
-; AVX-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
-; AVX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; AVX-NEXT:    [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
-; AVX-NEXT:    [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
-; AVX-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
-; AVX-NEXT:    [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
-; AVX-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
-; AVX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
-; AVX-NEXT:    [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
-; AVX-NEXT:    [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
-; AVX-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
-; AVX-NEXT:    [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
-; AVX-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
-; AVX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
-; AVX-NEXT:    [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
-; AVX-NEXT:    [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
-; AVX-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
-; AVX-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
-; AVX-NEXT:    [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
-; AVX-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
-; AVX-NEXT:    [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
-; AVX-NEXT:    [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
-; AVX-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
-; AVX-NEXT:    [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
-; AVX-NEXT:    [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
-; AVX-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
-; AVX-NEXT:    [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
-; AVX-NEXT:    [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
-; AVX-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
-; AVX-NEXT:    [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
-; AVX-NEXT:    [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
-; AVX-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
-; AVX-NEXT:    [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
-; AVX-NEXT:    [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
-; AVX-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
-; AVX-NEXT:    [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
-; AVX-NEXT:    [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
-; AVX-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
-; AVX-NEXT:    [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
-; AVX-NEXT:    [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
-; AVX-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
-; AVX-NEXT:    [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
-; AVX-NEXT:    [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
-; AVX-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
-; AVX-NEXT:    [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
-; AVX-NEXT:    [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
-; AVX-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
-; AVX-NEXT:    [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
-; AVX-NEXT:    [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; AVX-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; AVX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; AVX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; AVX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; AVX-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; AVX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; AVX-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; AVX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; AVX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; AVX-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; AVX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; AVX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; AVX-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
+; AVX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; AVX-NEXT:    [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], undef
+; AVX-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 undef
+; AVX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP34]], undef
+; AVX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP34]], i32 undef
+; AVX-NEXT:    [[TMP37:%.*]] = icmp sgt i32 [[TMP36]], undef
+; AVX-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], i32 [[TMP36]], i32 undef
+; AVX-NEXT:    [[TMP39:%.*]] = icmp sgt i32 [[TMP38]], undef
+; AVX-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], i32 [[TMP38]], i32 undef
+; AVX-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP40]], undef
+; AVX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP40]], i32 undef
+; AVX-NEXT:    [[TMP43:%.*]] = icmp sgt i32 [[TMP42]], undef
+; AVX-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], i32 [[TMP42]], i32 undef
+; AVX-NEXT:    [[TMP45:%.*]] = icmp sgt i32 [[TMP44]], undef
+; AVX-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], i32 [[TMP44]], i32 undef
+; AVX-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP46]], undef
+; AVX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP46]], i32 undef
+; AVX-NEXT:    [[TMP49:%.*]] = icmp sgt i32 [[TMP48]], undef
+; AVX-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], i32 [[TMP48]], i32 undef
+; AVX-NEXT:    [[TMP51:%.*]] = icmp sgt i32 [[TMP50]], undef
+; AVX-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], i32 [[TMP50]], i32 undef
+; AVX-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP52]], undef
+; AVX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP52]], i32 undef
+; AVX-NEXT:    [[TMP55:%.*]] = icmp sgt i32 [[TMP54]], undef
+; AVX-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], i32 [[TMP54]], i32 undef
+; AVX-NEXT:    [[TMP57:%.*]] = icmp sgt i32 [[TMP56]], undef
+; AVX-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], i32 [[TMP56]], i32 undef
+; AVX-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP58]], undef
+; AVX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP58]], i32 undef
+; AVX-NEXT:    [[TMP61:%.*]] = icmp sgt i32 [[TMP60]], undef
+; AVX-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], i32 [[TMP60]], i32 undef
+; AVX-NEXT:    [[TMP63:%.*]] = icmp sgt i32 [[TMP62]], undef
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
 ; AVX-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
-; AVX-NEXT:    [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
-; AVX-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
-; AVX-NEXT:    ret i32 [[TMP96]]
+; AVX-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], i32 [[TMP62]], i32 undef
+; AVX-NEXT:    ret i32 [[TMP64]]
 ;
 ; AVX2-LABEL: @maxi32(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
-; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
-; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
-; AVX2-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
-; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
-; AVX2-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; AVX2-NEXT:    [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
-; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; AVX2-NEXT:    [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
-; AVX2-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; AVX2-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; AVX2-NEXT:    [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
-; AVX2-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; AVX2-NEXT:    [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
-; AVX2-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; AVX2-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; AVX2-NEXT:    [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
-; AVX2-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; AVX2-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; AVX2-NEXT:    [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
-; AVX2-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; AVX2-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; AVX2-NEXT:    [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
-; AVX2-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; AVX2-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; AVX2-NEXT:    [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
-; AVX2-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; AVX2-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; AVX2-NEXT:    [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
-; AVX2-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
-; AVX2-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; AVX2-NEXT:    [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
-; AVX2-NEXT:    [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
-; AVX2-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
-; AVX2-NEXT:    [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
-; AVX2-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
-; AVX2-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
-; AVX2-NEXT:    [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
-; AVX2-NEXT:    [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
-; AVX2-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
-; AVX2-NEXT:    [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
-; AVX2-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
-; AVX2-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
-; AVX2-NEXT:    [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
-; AVX2-NEXT:    [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
-; AVX2-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
-; AVX2-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
-; AVX2-NEXT:    [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
-; AVX2-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
-; AVX2-NEXT:    [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
-; AVX2-NEXT:    [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
-; AVX2-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
-; AVX2-NEXT:    [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
-; AVX2-NEXT:    [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
-; AVX2-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
-; AVX2-NEXT:    [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
-; AVX2-NEXT:    [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
-; AVX2-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
-; AVX2-NEXT:    [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
-; AVX2-NEXT:    [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
-; AVX2-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
-; AVX2-NEXT:    [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
-; AVX2-NEXT:    [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
-; AVX2-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
-; AVX2-NEXT:    [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
-; AVX2-NEXT:    [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
-; AVX2-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
-; AVX2-NEXT:    [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
-; AVX2-NEXT:    [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
-; AVX2-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
-; AVX2-NEXT:    [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
-; AVX2-NEXT:    [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
-; AVX2-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
-; AVX2-NEXT:    [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
-; AVX2-NEXT:    [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
-; AVX2-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
-; AVX2-NEXT:    [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
-; AVX2-NEXT:    [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; AVX2-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; AVX2-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; AVX2-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX2-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; AVX2-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; AVX2-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; AVX2-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; AVX2-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; AVX2-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; AVX2-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; AVX2-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; AVX2-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; AVX2-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; AVX2-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
+; AVX2-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; AVX2-NEXT:    [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], undef
+; AVX2-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 undef
+; AVX2-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP34]], undef
+; AVX2-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP34]], i32 undef
+; AVX2-NEXT:    [[TMP37:%.*]] = icmp sgt i32 [[TMP36]], undef
+; AVX2-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], i32 [[TMP36]], i32 undef
+; AVX2-NEXT:    [[TMP39:%.*]] = icmp sgt i32 [[TMP38]], undef
+; AVX2-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], i32 [[TMP38]], i32 undef
+; AVX2-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP40]], undef
+; AVX2-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP40]], i32 undef
+; AVX2-NEXT:    [[TMP43:%.*]] = icmp sgt i32 [[TMP42]], undef
+; AVX2-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], i32 [[TMP42]], i32 undef
+; AVX2-NEXT:    [[TMP45:%.*]] = icmp sgt i32 [[TMP44]], undef
+; AVX2-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], i32 [[TMP44]], i32 undef
+; AVX2-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP46]], undef
+; AVX2-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP46]], i32 undef
+; AVX2-NEXT:    [[TMP49:%.*]] = icmp sgt i32 [[TMP48]], undef
+; AVX2-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], i32 [[TMP48]], i32 undef
+; AVX2-NEXT:    [[TMP51:%.*]] = icmp sgt i32 [[TMP50]], undef
+; AVX2-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], i32 [[TMP50]], i32 undef
+; AVX2-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP52]], undef
+; AVX2-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP52]], i32 undef
+; AVX2-NEXT:    [[TMP55:%.*]] = icmp sgt i32 [[TMP54]], undef
+; AVX2-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], i32 [[TMP54]], i32 undef
+; AVX2-NEXT:    [[TMP57:%.*]] = icmp sgt i32 [[TMP56]], undef
+; AVX2-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], i32 [[TMP56]], i32 undef
+; AVX2-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP58]], undef
+; AVX2-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP58]], i32 undef
+; AVX2-NEXT:    [[TMP61:%.*]] = icmp sgt i32 [[TMP60]], undef
+; AVX2-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], i32 [[TMP60]], i32 undef
+; AVX2-NEXT:    [[TMP63:%.*]] = icmp sgt i32 [[TMP62]], undef
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
 ; AVX2-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
-; AVX2-NEXT:    [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
-; AVX2-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
-; AVX2-NEXT:    ret i32 [[TMP96]]
+; AVX2-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX2-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], i32 [[TMP62]], i32 undef
+; AVX2-NEXT:    ret i32 [[TMP64]]
 ;
 ; SKX-LABEL: @maxi32(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <32 x i32>, <32 x i32>* bitcast ([32 x i32]* @arr to <32 x i32>*), align 16
-; SKX-NEXT:    [[TMP3:%.*]] = extractelement <32 x i32> [[TMP2]], i32 0
-; SKX-NEXT:    [[TMP4:%.*]] = extractelement <32 x i32> [[TMP2]], i32 1
-; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
-; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <32 x i32> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <32 x i32> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; SKX-NEXT:    [[TMP13:%.*]] = extractelement <32 x i32> [[TMP2]], i32 4
-; SKX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; SKX-NEXT:    [[TMP16:%.*]] = extractelement <32 x i32> [[TMP2]], i32 5
-; SKX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; SKX-NEXT:    [[TMP19:%.*]] = extractelement <32 x i32> [[TMP2]], i32 6
-; SKX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 undef
-; SKX-NEXT:    [[TMP22:%.*]] = extractelement <32 x i32> [[TMP2]], i32 7
-; SKX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
-; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 undef
-; SKX-NEXT:    [[TMP25:%.*]] = extractelement <32 x i32> [[TMP2]], i32 8
-; SKX-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP24]], [[TMP25]]
-; SKX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP24]], i32 undef
-; SKX-NEXT:    [[TMP28:%.*]] = extractelement <32 x i32> [[TMP2]], i32 9
-; SKX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP27]], [[TMP28]]
-; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP27]], i32 undef
-; SKX-NEXT:    [[TMP31:%.*]] = extractelement <32 x i32> [[TMP2]], i32 10
-; SKX-NEXT:    [[TMP32:%.*]] = icmp sgt i32 [[TMP30]], [[TMP31]]
-; SKX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], i32 [[TMP30]], i32 undef
-; SKX-NEXT:    [[TMP34:%.*]] = extractelement <32 x i32> [[TMP2]], i32 11
-; SKX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP33]], [[TMP34]]
-; SKX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP33]], i32 undef
-; SKX-NEXT:    [[TMP37:%.*]] = extractelement <32 x i32> [[TMP2]], i32 12
-; SKX-NEXT:    [[TMP38:%.*]] = icmp sgt i32 [[TMP36]], [[TMP37]]
-; SKX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], i32 [[TMP36]], i32 undef
-; SKX-NEXT:    [[TMP40:%.*]] = extractelement <32 x i32> [[TMP2]], i32 13
-; SKX-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP39]], [[TMP40]]
-; SKX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP39]], i32 undef
-; SKX-NEXT:    [[TMP43:%.*]] = extractelement <32 x i32> [[TMP2]], i32 14
-; SKX-NEXT:    [[TMP44:%.*]] = icmp sgt i32 [[TMP42]], [[TMP43]]
-; SKX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], i32 [[TMP42]], i32 undef
-; SKX-NEXT:    [[TMP46:%.*]] = extractelement <32 x i32> [[TMP2]], i32 15
-; SKX-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP45]], [[TMP46]]
-; SKX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP45]], i32 undef
-; SKX-NEXT:    [[TMP49:%.*]] = extractelement <32 x i32> [[TMP2]], i32 16
-; SKX-NEXT:    [[TMP50:%.*]] = icmp sgt i32 [[TMP48]], [[TMP49]]
-; SKX-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], i32 [[TMP48]], i32 undef
-; SKX-NEXT:    [[TMP52:%.*]] = extractelement <32 x i32> [[TMP2]], i32 17
-; SKX-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP51]], [[TMP52]]
-; SKX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP51]], i32 undef
-; SKX-NEXT:    [[TMP55:%.*]] = extractelement <32 x i32> [[TMP2]], i32 18
-; SKX-NEXT:    [[TMP56:%.*]] = icmp sgt i32 [[TMP54]], [[TMP55]]
-; SKX-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], i32 [[TMP54]], i32 undef
-; SKX-NEXT:    [[TMP58:%.*]] = extractelement <32 x i32> [[TMP2]], i32 19
-; SKX-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP57]], [[TMP58]]
-; SKX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP57]], i32 undef
-; SKX-NEXT:    [[TMP61:%.*]] = extractelement <32 x i32> [[TMP2]], i32 20
-; SKX-NEXT:    [[TMP62:%.*]] = icmp sgt i32 [[TMP60]], [[TMP61]]
-; SKX-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], i32 [[TMP60]], i32 undef
-; SKX-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[TMP2]], i32 21
-; SKX-NEXT:    [[TMP65:%.*]] = icmp sgt i32 [[TMP63]], [[TMP64]]
-; SKX-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], i32 [[TMP63]], i32 undef
-; SKX-NEXT:    [[TMP67:%.*]] = extractelement <32 x i32> [[TMP2]], i32 22
-; SKX-NEXT:    [[TMP68:%.*]] = icmp sgt i32 [[TMP66]], [[TMP67]]
-; SKX-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], i32 [[TMP66]], i32 undef
-; SKX-NEXT:    [[TMP70:%.*]] = extractelement <32 x i32> [[TMP2]], i32 23
-; SKX-NEXT:    [[TMP71:%.*]] = icmp sgt i32 [[TMP69]], [[TMP70]]
-; SKX-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], i32 [[TMP69]], i32 undef
-; SKX-NEXT:    [[TMP73:%.*]] = extractelement <32 x i32> [[TMP2]], i32 24
-; SKX-NEXT:    [[TMP74:%.*]] = icmp sgt i32 [[TMP72]], [[TMP73]]
-; SKX-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], i32 [[TMP72]], i32 undef
-; SKX-NEXT:    [[TMP76:%.*]] = extractelement <32 x i32> [[TMP2]], i32 25
-; SKX-NEXT:    [[TMP77:%.*]] = icmp sgt i32 [[TMP75]], [[TMP76]]
-; SKX-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], i32 [[TMP75]], i32 undef
-; SKX-NEXT:    [[TMP79:%.*]] = extractelement <32 x i32> [[TMP2]], i32 26
-; SKX-NEXT:    [[TMP80:%.*]] = icmp sgt i32 [[TMP78]], [[TMP79]]
-; SKX-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], i32 [[TMP78]], i32 undef
-; SKX-NEXT:    [[TMP82:%.*]] = extractelement <32 x i32> [[TMP2]], i32 27
-; SKX-NEXT:    [[TMP83:%.*]] = icmp sgt i32 [[TMP81]], [[TMP82]]
-; SKX-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], i32 [[TMP81]], i32 undef
-; SKX-NEXT:    [[TMP85:%.*]] = extractelement <32 x i32> [[TMP2]], i32 28
-; SKX-NEXT:    [[TMP86:%.*]] = icmp sgt i32 [[TMP84]], [[TMP85]]
-; SKX-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], i32 [[TMP84]], i32 undef
-; SKX-NEXT:    [[TMP88:%.*]] = extractelement <32 x i32> [[TMP2]], i32 29
-; SKX-NEXT:    [[TMP89:%.*]] = icmp sgt i32 [[TMP87]], [[TMP88]]
-; SKX-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], i32 [[TMP87]], i32 undef
-; SKX-NEXT:    [[TMP91:%.*]] = extractelement <32 x i32> [[TMP2]], i32 30
-; SKX-NEXT:    [[TMP92:%.*]] = icmp sgt i32 [[TMP90]], [[TMP91]]
-; SKX-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], i32 [[TMP90]], i32 undef
-; SKX-NEXT:    [[TMP94:%.*]] = extractelement <32 x i32> [[TMP2]], i32 31
-; SKX-NEXT:    [[TMP95:%.*]] = icmp sgt i32 [[TMP93]], [[TMP94]]
+; SKX-NEXT:    [[TMP3:%.*]] = icmp sgt i32 undef, undef
+; SKX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i32 undef, i32 undef
+; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP4]], undef
+; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 [[TMP4]], i32 undef
+; SKX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; SKX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; SKX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; SKX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; SKX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP14]], undef
+; SKX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP14]], i32 undef
+; SKX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP16]], undef
+; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP16]], i32 undef
+; SKX-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP18]], undef
+; SKX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], i32 [[TMP18]], i32 undef
+; SKX-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; SKX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; SKX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; SKX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; SKX-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; SKX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; SKX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], undef
+; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 undef
+; SKX-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], undef
+; SKX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 undef
+; SKX-NEXT:    [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], undef
+; SKX-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 undef
+; SKX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP34]], undef
+; SKX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], i32 [[TMP34]], i32 undef
+; SKX-NEXT:    [[TMP37:%.*]] = icmp sgt i32 [[TMP36]], undef
+; SKX-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], i32 [[TMP36]], i32 undef
+; SKX-NEXT:    [[TMP39:%.*]] = icmp sgt i32 [[TMP38]], undef
+; SKX-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], i32 [[TMP38]], i32 undef
+; SKX-NEXT:    [[TMP41:%.*]] = icmp sgt i32 [[TMP40]], undef
+; SKX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], i32 [[TMP40]], i32 undef
+; SKX-NEXT:    [[TMP43:%.*]] = icmp sgt i32 [[TMP42]], undef
+; SKX-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], i32 [[TMP42]], i32 undef
+; SKX-NEXT:    [[TMP45:%.*]] = icmp sgt i32 [[TMP44]], undef
+; SKX-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], i32 [[TMP44]], i32 undef
+; SKX-NEXT:    [[TMP47:%.*]] = icmp sgt i32 [[TMP46]], undef
+; SKX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], i32 [[TMP46]], i32 undef
+; SKX-NEXT:    [[TMP49:%.*]] = icmp sgt i32 [[TMP48]], undef
+; SKX-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], i32 [[TMP48]], i32 undef
+; SKX-NEXT:    [[TMP51:%.*]] = icmp sgt i32 [[TMP50]], undef
+; SKX-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], i32 [[TMP50]], i32 undef
+; SKX-NEXT:    [[TMP53:%.*]] = icmp sgt i32 [[TMP52]], undef
+; SKX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], i32 [[TMP52]], i32 undef
+; SKX-NEXT:    [[TMP55:%.*]] = icmp sgt i32 [[TMP54]], undef
+; SKX-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], i32 [[TMP54]], i32 undef
+; SKX-NEXT:    [[TMP57:%.*]] = icmp sgt i32 [[TMP56]], undef
+; SKX-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], i32 [[TMP56]], i32 undef
+; SKX-NEXT:    [[TMP59:%.*]] = icmp sgt i32 [[TMP58]], undef
+; SKX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], i32 [[TMP58]], i32 undef
+; SKX-NEXT:    [[TMP61:%.*]] = icmp sgt i32 [[TMP60]], undef
+; SKX-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], i32 [[TMP60]], i32 undef
+; SKX-NEXT:    [[TMP63:%.*]] = icmp sgt i32 [[TMP62]], undef
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x i32> [[TMP2]], <32 x i32> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <32 x i32> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x i32> [[TMP2]], <32 x i32> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x i32> [[RDX_MINMAX_SELECT]], <32 x i32> [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x i32> [[RDX_MINMAX_SELECT3]], <32 x i32> [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x i32> [[RDX_MINMAX_SELECT6]], <32 x i32> [[RDX_SHUF7]]
 ; SKX-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = icmp sgt <32 x i32> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x i32> [[RDX_MINMAX_SELECT9]], <32 x i32> [[RDX_SHUF10]]
-; SKX-NEXT:    [[TMP96:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
-; SKX-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], i32 [[TMP93]], i32 undef
-; SKX-NEXT:    ret i32 [[TMP96]]
+; SKX-NEXT:    [[TMP64:%.*]] = extractelement <32 x i32> [[RDX_MINMAX_SELECT12]], i32 0
+; SKX-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], i32 [[TMP62]], i32 undef
+; SKX-NEXT:    ret i32 [[TMP64]]
 ;
   %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
   %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
   %4 = icmp sgt i32 %2, %3
   %5 = select i1 %4, i32 %2, i32 %3
   %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
   %7 = icmp sgt i32 %5, %6
   %8 = select i1 %7, i32 %5, i32 %6

 ; CHECK-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP17]], float [[TMP18]]
 ; CHECK-NEXT:    [[TMP21:%.*]] = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 7), align 4
 ; CHECK-NEXT:    [[TMP22:%.*]] = fcmp fast ogt float [[TMP20]], [[TMP21]]
 ; CHECK-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], float [[TMP20]], float [[TMP21]]
 ; CHECK-NEXT:    ret float [[TMP23]]
 ;
 ; AVX-LABEL: @maxf8(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
-; AVX-NEXT:    [[TMP3:%.*]] = extractelement <8 x float> [[TMP2]], i32 0
-; AVX-NEXT:    [[TMP4:%.*]] = extractelement <8 x float> [[TMP2]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <8 x float> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <8 x float> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <8 x float> [[TMP2]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[TMP2]], i32 5
-; AVX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; AVX-NEXT:    [[TMP19:%.*]] = extractelement <8 x float> [[TMP2]], i32 6
-; AVX-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; AVX-NEXT:    [[TMP22:%.*]] = extractelement <8 x float> [[TMP2]], i32 7
-; AVX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; AVX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; AVX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; AVX-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; AVX-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; AVX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; AVX-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; AVX-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
-; AVX-NEXT:    [[TMP24:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
-; AVX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; AVX-NEXT:    ret float [[TMP24]]
+; AVX-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX-NEXT:    [[TMP17:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; AVX-NEXT:    ret float [[TMP16]]
 ;
 ; AVX2-LABEL: @maxf8(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
-; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <8 x float> [[TMP2]], i32 0
-; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <8 x float> [[TMP2]], i32 1
-; AVX2-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <8 x float> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <8 x float> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <8 x float> [[TMP2]], i32 4
-; AVX2-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[TMP2]], i32 5
-; AVX2-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <8 x float> [[TMP2]], i32 6
-; AVX2-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; AVX2-NEXT:    [[TMP22:%.*]] = extractelement <8 x float> [[TMP2]], i32 7
-; AVX2-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; AVX2-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; AVX2-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; AVX2-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; AVX2-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; AVX2-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; AVX2-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; AVX2-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; AVX2-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
-; AVX2-NEXT:    [[TMP24:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
-; AVX2-NEXT:    [[TMP25:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; AVX2-NEXT:    ret float [[TMP24]]
+; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
+; AVX2-NEXT:    [[TMP17:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; AVX2-NEXT:    ret float [[TMP16]]
 ;
 ; SKX-LABEL: @maxf8(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <8 x float>, <8 x float>* bitcast ([32 x float]* @arr1 to <8 x float>*), align 16
-; SKX-NEXT:    [[TMP3:%.*]] = extractelement <8 x float> [[TMP2]], i32 0
-; SKX-NEXT:    [[TMP4:%.*]] = extractelement <8 x float> [[TMP2]], i32 1
-; SKX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <8 x float> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <8 x float> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; SKX-NEXT:    [[TMP13:%.*]] = extractelement <8 x float> [[TMP2]], i32 4
-; SKX-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; SKX-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[TMP2]], i32 5
-; SKX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; SKX-NEXT:    [[TMP19:%.*]] = extractelement <8 x float> [[TMP2]], i32 6
-; SKX-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; SKX-NEXT:    [[TMP22:%.*]] = extractelement <8 x float> [[TMP2]], i32 7
-; SKX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
+; SKX-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; SKX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; SKX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; SKX-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; SKX-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; SKX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; SKX-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; SKX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; SKX-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <8 x float> [[TMP2]], <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <8 x float> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP]], <8 x float> [[TMP2]], <8 x float> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP2]], <8 x float> [[RDX_MINMAX_SELECT]], <8 x float> [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <8 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <8 x i1> [[RDX_MINMAX_CMP5]], <8 x float> [[RDX_MINMAX_SELECT3]], <8 x float> [[RDX_SHUF4]]
-; SKX-NEXT:    [[TMP24:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
-; SKX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; SKX-NEXT:    ret float [[TMP24]]
+; SKX-NEXT:    [[TMP16:%.*]] = extractelement <8 x float> [[RDX_MINMAX_SELECT6]], i32 0
+; SKX-NEXT:    [[TMP17:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; SKX-NEXT:    ret float [[TMP16]]
 ;
   %2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
   %3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
   %4 = fcmp fast ogt float %2, %3
   %5 = select i1 %4, float %2, float %3
   %6 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 2), align 8
   %7 = fcmp fast ogt float %5, %6
   %8 = select i1 %7, float %5, float %6

 ; CHECK-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], float [[TMP41]], float [[TMP42]]
 ; CHECK-NEXT:    [[TMP45:%.*]] = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 15), align 4
 ; CHECK-NEXT:    [[TMP46:%.*]] = fcmp fast ogt float [[TMP44]], [[TMP45]]
 ; CHECK-NEXT:    [[TMP47:%.*]] = select i1 [[TMP46]], float [[TMP44]], float [[TMP45]]
 ; CHECK-NEXT:    ret float [[TMP47]]
 ;
 ; AVX-LABEL: @maxf16(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
-; AVX-NEXT:    [[TMP3:%.*]] = extractelement <16 x float> [[TMP2]], i32 0
-; AVX-NEXT:    [[TMP4:%.*]] = extractelement <16 x float> [[TMP2]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <16 x float> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <16 x float> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <16 x float> [[TMP2]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <16 x float> [[TMP2]], i32 5
-; AVX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; AVX-NEXT:    [[TMP19:%.*]] = extractelement <16 x float> [[TMP2]], i32 6
-; AVX-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; AVX-NEXT:    [[TMP22:%.*]] = extractelement <16 x float> [[TMP2]], i32 7
-; AVX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
-; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; AVX-NEXT:    [[TMP25:%.*]] = extractelement <16 x float> [[TMP2]], i32 8
-; AVX-NEXT:    [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
-; AVX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
-; AVX-NEXT:    [[TMP28:%.*]] = extractelement <16 x float> [[TMP2]], i32 9
-; AVX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
-; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
-; AVX-NEXT:    [[TMP31:%.*]] = extractelement <16 x float> [[TMP2]], i32 10
-; AVX-NEXT:    [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
-; AVX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
-; AVX-NEXT:    [[TMP34:%.*]] = extractelement <16 x float> [[TMP2]], i32 11
-; AVX-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
-; AVX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
-; AVX-NEXT:    [[TMP37:%.*]] = extractelement <16 x float> [[TMP2]], i32 12
-; AVX-NEXT:    [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
-; AVX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
-; AVX-NEXT:    [[TMP40:%.*]] = extractelement <16 x float> [[TMP2]], i32 13
-; AVX-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
-; AVX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
-; AVX-NEXT:    [[TMP43:%.*]] = extractelement <16 x float> [[TMP2]], i32 14
-; AVX-NEXT:    [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
-; AVX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
-; AVX-NEXT:    [[TMP46:%.*]] = extractelement <16 x float> [[TMP2]], i32 15
-; AVX-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; AVX-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; AVX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; AVX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; AVX-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; AVX-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; AVX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; AVX-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; AVX-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
+; AVX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; AVX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP16]], undef
+; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP16]], float undef
+; AVX-NEXT:    [[TMP19:%.*]] = fcmp fast ogt float [[TMP18]], undef
+; AVX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP18]], float undef
+; AVX-NEXT:    [[TMP21:%.*]] = fcmp fast ogt float [[TMP20]], undef
+; AVX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], float [[TMP20]], float undef
+; AVX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP22]], undef
+; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP22]], float undef
+; AVX-NEXT:    [[TMP25:%.*]] = fcmp fast ogt float [[TMP24]], undef
+; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], float [[TMP24]], float undef
+; AVX-NEXT:    [[TMP27:%.*]] = fcmp fast ogt float [[TMP26]], undef
+; AVX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], float [[TMP26]], float undef
+; AVX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP28]], undef
+; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP28]], float undef
+; AVX-NEXT:    [[TMP31:%.*]] = fcmp fast ogt float [[TMP30]], undef
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
-; AVX-NEXT:    [[TMP48:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
-; AVX-NEXT:    [[TMP49:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
-; AVX-NEXT:    ret float [[TMP48]]
+; AVX-NEXT:    [[TMP32:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], float [[TMP30]], float undef
+; AVX-NEXT:    ret float [[TMP32]]
 ;
 ; AVX2-LABEL: @maxf16(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
-; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <16 x float> [[TMP2]], i32 0
-; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <16 x float> [[TMP2]], i32 1
-; AVX2-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <16 x float> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <16 x float> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <16 x float> [[TMP2]], i32 4
-; AVX2-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <16 x float> [[TMP2]], i32 5
-; AVX2-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <16 x float> [[TMP2]], i32 6
-; AVX2-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; AVX2-NEXT:    [[TMP22:%.*]] = extractelement <16 x float> [[TMP2]], i32 7
-; AVX2-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
-; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; AVX2-NEXT:    [[TMP25:%.*]] = extractelement <16 x float> [[TMP2]], i32 8
-; AVX2-NEXT:    [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
-; AVX2-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
-; AVX2-NEXT:    [[TMP28:%.*]] = extractelement <16 x float> [[TMP2]], i32 9
-; AVX2-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
-; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
-; AVX2-NEXT:    [[TMP31:%.*]] = extractelement <16 x float> [[TMP2]], i32 10
-; AVX2-NEXT:    [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
-; AVX2-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
-; AVX2-NEXT:    [[TMP34:%.*]] = extractelement <16 x float> [[TMP2]], i32 11
-; AVX2-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
-; AVX2-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
-; AVX2-NEXT:    [[TMP37:%.*]] = extractelement <16 x float> [[TMP2]], i32 12
-; AVX2-NEXT:    [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
-; AVX2-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
-; AVX2-NEXT:    [[TMP40:%.*]] = extractelement <16 x float> [[TMP2]], i32 13
-; AVX2-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
-; AVX2-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
-; AVX2-NEXT:    [[TMP43:%.*]] = extractelement <16 x float> [[TMP2]], i32 14
-; AVX2-NEXT:    [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
-; AVX2-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
-; AVX2-NEXT:    [[TMP46:%.*]] = extractelement <16 x float> [[TMP2]], i32 15
-; AVX2-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; AVX2-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; AVX2-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; AVX2-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; AVX2-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; AVX2-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; AVX2-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; AVX2-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; AVX2-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
+; AVX2-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; AVX2-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP16]], undef
+; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP16]], float undef
+; AVX2-NEXT:    [[TMP19:%.*]] = fcmp fast ogt float [[TMP18]], undef
+; AVX2-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP18]], float undef
+; AVX2-NEXT:    [[TMP21:%.*]] = fcmp fast ogt float [[TMP20]], undef
+; AVX2-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], float [[TMP20]], float undef
+; AVX2-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP22]], undef
+; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP22]], float undef
+; AVX2-NEXT:    [[TMP25:%.*]] = fcmp fast ogt float [[TMP24]], undef
+; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], float [[TMP24]], float undef
+; AVX2-NEXT:    [[TMP27:%.*]] = fcmp fast ogt float [[TMP26]], undef
+; AVX2-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], float [[TMP26]], float undef
+; AVX2-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP28]], undef
+; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP28]], float undef
+; AVX2-NEXT:    [[TMP31:%.*]] = fcmp fast ogt float [[TMP30]], undef
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
-; AVX2-NEXT:    [[TMP48:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
-; AVX2-NEXT:    [[TMP49:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
-; AVX2-NEXT:    ret float [[TMP48]]
+; AVX2-NEXT:    [[TMP32:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
+; AVX2-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], float [[TMP30]], float undef
+; AVX2-NEXT:    ret float [[TMP32]]
 ;
 ; SKX-LABEL: @maxf16(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <16 x float>, <16 x float>* bitcast ([32 x float]* @arr1 to <16 x float>*), align 16
-; SKX-NEXT:    [[TMP3:%.*]] = extractelement <16 x float> [[TMP2]], i32 0
-; SKX-NEXT:    [[TMP4:%.*]] = extractelement <16 x float> [[TMP2]], i32 1
-; SKX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <16 x float> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <16 x float> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; SKX-NEXT:    [[TMP13:%.*]] = extractelement <16 x float> [[TMP2]], i32 4
-; SKX-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; SKX-NEXT:    [[TMP16:%.*]] = extractelement <16 x float> [[TMP2]], i32 5
-; SKX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; SKX-NEXT:    [[TMP19:%.*]] = extractelement <16 x float> [[TMP2]], i32 6
-; SKX-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; SKX-NEXT:    [[TMP22:%.*]] = extractelement <16 x float> [[TMP2]], i32 7
-; SKX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
-; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; SKX-NEXT:    [[TMP25:%.*]] = extractelement <16 x float> [[TMP2]], i32 8
-; SKX-NEXT:    [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
-; SKX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
-; SKX-NEXT:    [[TMP28:%.*]] = extractelement <16 x float> [[TMP2]], i32 9
-; SKX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
-; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
-; SKX-NEXT:    [[TMP31:%.*]] = extractelement <16 x float> [[TMP2]], i32 10
-; SKX-NEXT:    [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
-; SKX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
-; SKX-NEXT:    [[TMP34:%.*]] = extractelement <16 x float> [[TMP2]], i32 11
-; SKX-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
-; SKX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
-; SKX-NEXT:    [[TMP37:%.*]] = extractelement <16 x float> [[TMP2]], i32 12
-; SKX-NEXT:    [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
-; SKX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
-; SKX-NEXT:    [[TMP40:%.*]] = extractelement <16 x float> [[TMP2]], i32 13
-; SKX-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
-; SKX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
-; SKX-NEXT:    [[TMP43:%.*]] = extractelement <16 x float> [[TMP2]], i32 14
-; SKX-NEXT:    [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
-; SKX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
-; SKX-NEXT:    [[TMP46:%.*]] = extractelement <16 x float> [[TMP2]], i32 15
-; SKX-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
+; SKX-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; SKX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; SKX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; SKX-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; SKX-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; SKX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; SKX-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; SKX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; SKX-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
+; SKX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; SKX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP16]], undef
+; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP16]], float undef
+; SKX-NEXT:    [[TMP19:%.*]] = fcmp fast ogt float [[TMP18]], undef
+; SKX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP18]], float undef
+; SKX-NEXT:    [[TMP21:%.*]] = fcmp fast ogt float [[TMP20]], undef
+; SKX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], float [[TMP20]], float undef
+; SKX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP22]], undef
+; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP22]], float undef
+; SKX-NEXT:    [[TMP25:%.*]] = fcmp fast ogt float [[TMP24]], undef
+; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], float [[TMP24]], float undef
+; SKX-NEXT:    [[TMP27:%.*]] = fcmp fast ogt float [[TMP26]], undef
+; SKX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], float [[TMP26]], float undef
+; SKX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP28]], undef
+; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP28]], float undef
+; SKX-NEXT:    [[TMP31:%.*]] = fcmp fast ogt float [[TMP30]], undef
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <16 x float> [[TMP2]], <16 x float> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <16 x float> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP]], <16 x float> [[TMP2]], <16 x float> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP2]], <16 x float> [[RDX_MINMAX_SELECT]], <16 x float> [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP5]], <16 x float> [[RDX_MINMAX_SELECT3]], <16 x float> [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <16 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <16 x i1> [[RDX_MINMAX_CMP8]], <16 x float> [[RDX_MINMAX_SELECT6]], <16 x float> [[RDX_SHUF7]]
-; SKX-NEXT:    [[TMP48:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
-; SKX-NEXT:    [[TMP49:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
-; SKX-NEXT:    ret float [[TMP48]]
+; SKX-NEXT:    [[TMP32:%.*]] = extractelement <16 x float> [[RDX_MINMAX_SELECT9]], i32 0
+; SKX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP31]], float [[TMP30]], float undef
+; SKX-NEXT:    ret float [[TMP32]]
 ;
   %2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
   %3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
   %4 = fcmp fast ogt float %2, %3
   %5 = select i1 %4, float %2, float %3
   %6 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 2), align 8
   %7 = fcmp fast ogt float %5, %6
   %8 = select i1 %7, float %5, float %6

 ; CHECK-NEXT:    [[TMP92:%.*]] = select i1 [[TMP91]], float [[TMP89]], float [[TMP90]]
 ; CHECK-NEXT:    [[TMP93:%.*]] = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 31), align 4
 ; CHECK-NEXT:    [[TMP94:%.*]] = fcmp fast ogt float [[TMP92]], [[TMP93]]
 ; CHECK-NEXT:    [[TMP95:%.*]] = select i1 [[TMP94]], float [[TMP92]], float [[TMP93]]
 ; CHECK-NEXT:    ret float [[TMP95]]
 ;
 ; AVX-LABEL: @maxf32(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
-; AVX-NEXT:    [[TMP3:%.*]] = extractelement <32 x float> [[TMP2]], i32 0
-; AVX-NEXT:    [[TMP4:%.*]] = extractelement <32 x float> [[TMP2]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <32 x float> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <32 x float> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <32 x float> [[TMP2]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <32 x float> [[TMP2]], i32 5
-; AVX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; AVX-NEXT:    [[TMP19:%.*]] = extractelement <32 x float> [[TMP2]], i32 6
-; AVX-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; AVX-NEXT:    [[TMP22:%.*]] = extractelement <32 x float> [[TMP2]], i32 7
-; AVX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
-; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; AVX-NEXT:    [[TMP25:%.*]] = extractelement <32 x float> [[TMP2]], i32 8
-; AVX-NEXT:    [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
-; AVX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
-; AVX-NEXT:    [[TMP28:%.*]] = extractelement <32 x float> [[TMP2]], i32 9
-; AVX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
-; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
-; AVX-NEXT:    [[TMP31:%.*]] = extractelement <32 x float> [[TMP2]], i32 10
-; AVX-NEXT:    [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
-; AVX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
-; AVX-NEXT:    [[TMP34:%.*]] = extractelement <32 x float> [[TMP2]], i32 11
-; AVX-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
-; AVX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
-; AVX-NEXT:    [[TMP37:%.*]] = extractelement <32 x float> [[TMP2]], i32 12
-; AVX-NEXT:    [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
-; AVX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
-; AVX-NEXT:    [[TMP40:%.*]] = extractelement <32 x float> [[TMP2]], i32 13
-; AVX-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
-; AVX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
-; AVX-NEXT:    [[TMP43:%.*]] = extractelement <32 x float> [[TMP2]], i32 14
-; AVX-NEXT:    [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
-; AVX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
-; AVX-NEXT:    [[TMP46:%.*]] = extractelement <32 x float> [[TMP2]], i32 15
-; AVX-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
-; AVX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
-; AVX-NEXT:    [[TMP49:%.*]] = extractelement <32 x float> [[TMP2]], i32 16
-; AVX-NEXT:    [[TMP50:%.*]] = fcmp fast ogt float [[TMP48]], [[TMP49]]
-; AVX-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], float [[TMP48]], float undef
-; AVX-NEXT:    [[TMP52:%.*]] = extractelement <32 x float> [[TMP2]], i32 17
-; AVX-NEXT:    [[TMP53:%.*]] = fcmp fast ogt float [[TMP51]], [[TMP52]]
-; AVX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP51]], float undef
-; AVX-NEXT:    [[TMP55:%.*]] = extractelement <32 x float> [[TMP2]], i32 18
-; AVX-NEXT:    [[TMP56:%.*]] = fcmp fast ogt float [[TMP54]], [[TMP55]]
-; AVX-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], float [[TMP54]], float undef
-; AVX-NEXT:    [[TMP58:%.*]] = extractelement <32 x float> [[TMP2]], i32 19
-; AVX-NEXT:    [[TMP59:%.*]] = fcmp fast ogt float [[TMP57]], [[TMP58]]
-; AVX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP57]], float undef
-; AVX-NEXT:    [[TMP61:%.*]] = extractelement <32 x float> [[TMP2]], i32 20
-; AVX-NEXT:    [[TMP62:%.*]] = fcmp fast ogt float [[TMP60]], [[TMP61]]
-; AVX-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], float [[TMP60]], float undef
-; AVX-NEXT:    [[TMP64:%.*]] = extractelement <32 x float> [[TMP2]], i32 21
-; AVX-NEXT:    [[TMP65:%.*]] = fcmp fast ogt float [[TMP63]], [[TMP64]]
-; AVX-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], float [[TMP63]], float undef
-; AVX-NEXT:    [[TMP67:%.*]] = extractelement <32 x float> [[TMP2]], i32 22
-; AVX-NEXT:    [[TMP68:%.*]] = fcmp fast ogt float [[TMP66]], [[TMP67]]
-; AVX-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], float [[TMP66]], float undef
-; AVX-NEXT:    [[TMP70:%.*]] = extractelement <32 x float> [[TMP2]], i32 23
-; AVX-NEXT:    [[TMP71:%.*]] = fcmp fast ogt float [[TMP69]], [[TMP70]]
-; AVX-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], float [[TMP69]], float undef
-; AVX-NEXT:    [[TMP73:%.*]] = extractelement <32 x float> [[TMP2]], i32 24
-; AVX-NEXT:    [[TMP74:%.*]] = fcmp fast ogt float [[TMP72]], [[TMP73]]
-; AVX-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], float [[TMP72]], float undef
-; AVX-NEXT:    [[TMP76:%.*]] = extractelement <32 x float> [[TMP2]], i32 25
-; AVX-NEXT:    [[TMP77:%.*]] = fcmp fast ogt float [[TMP75]], [[TMP76]]
-; AVX-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], float [[TMP75]], float undef
-; AVX-NEXT:    [[TMP79:%.*]] = extractelement <32 x float> [[TMP2]], i32 26
-; AVX-NEXT:    [[TMP80:%.*]] = fcmp fast ogt float [[TMP78]], [[TMP79]]
-; AVX-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], float [[TMP78]], float undef
-; AVX-NEXT:    [[TMP82:%.*]] = extractelement <32 x float> [[TMP2]], i32 27
-; AVX-NEXT:    [[TMP83:%.*]] = fcmp fast ogt float [[TMP81]], [[TMP82]]
-; AVX-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], float [[TMP81]], float undef
-; AVX-NEXT:    [[TMP85:%.*]] = extractelement <32 x float> [[TMP2]], i32 28
-; AVX-NEXT:    [[TMP86:%.*]] = fcmp fast ogt float [[TMP84]], [[TMP85]]
-; AVX-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], float [[TMP84]], float undef
-; AVX-NEXT:    [[TMP88:%.*]] = extractelement <32 x float> [[TMP2]], i32 29
-; AVX-NEXT:    [[TMP89:%.*]] = fcmp fast ogt float [[TMP87]], [[TMP88]]
-; AVX-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], float [[TMP87]], float undef
-; AVX-NEXT:    [[TMP91:%.*]] = extractelement <32 x float> [[TMP2]], i32 30
-; AVX-NEXT:    [[TMP92:%.*]] = fcmp fast ogt float [[TMP90]], [[TMP91]]
-; AVX-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], float [[TMP90]], float undef
-; AVX-NEXT:    [[TMP94:%.*]] = extractelement <32 x float> [[TMP2]], i32 31
-; AVX-NEXT:    [[TMP95:%.*]] = fcmp fast ogt float [[TMP93]], [[TMP94]]
+; AVX-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; AVX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; AVX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; AVX-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; AVX-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; AVX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; AVX-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; AVX-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
+; AVX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; AVX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP16]], undef
+; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP16]], float undef
+; AVX-NEXT:    [[TMP19:%.*]] = fcmp fast ogt float [[TMP18]], undef
+; AVX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP18]], float undef
+; AVX-NEXT:    [[TMP21:%.*]] = fcmp fast ogt float [[TMP20]], undef
+; AVX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], float [[TMP20]], float undef
+; AVX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP22]], undef
+; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP22]], float undef
+; AVX-NEXT:    [[TMP25:%.*]] = fcmp fast ogt float [[TMP24]], undef
+; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], float [[TMP24]], float undef
+; AVX-NEXT:    [[TMP27:%.*]] = fcmp fast ogt float [[TMP26]], undef
+; AVX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], float [[TMP26]], float undef
+; AVX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP28]], undef
+; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP28]], float undef
+; AVX-NEXT:    [[TMP31:%.*]] = fcmp fast ogt float [[TMP30]], undef
+; AVX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], float [[TMP30]], float undef
+; AVX-NEXT:    [[TMP33:%.*]] = fcmp fast ogt float [[TMP32]], undef
+; AVX-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], float [[TMP32]], float undef
+; AVX-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP34]], undef
+; AVX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP34]], float undef
+; AVX-NEXT:    [[TMP37:%.*]] = fcmp fast ogt float [[TMP36]], undef
+; AVX-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], float [[TMP36]], float undef
+; AVX-NEXT:    [[TMP39:%.*]] = fcmp fast ogt float [[TMP38]], undef
+; AVX-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], float [[TMP38]], float undef
+; AVX-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP40]], undef
+; AVX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP40]], float undef
+; AVX-NEXT:    [[TMP43:%.*]] = fcmp fast ogt float [[TMP42]], undef
+; AVX-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], float [[TMP42]], float undef
+; AVX-NEXT:    [[TMP45:%.*]] = fcmp fast ogt float [[TMP44]], undef
+; AVX-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], float [[TMP44]], float undef
+; AVX-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP46]], undef
+; AVX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP46]], float undef
+; AVX-NEXT:    [[TMP49:%.*]] = fcmp fast ogt float [[TMP48]], undef
+; AVX-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], float [[TMP48]], float undef
+; AVX-NEXT:    [[TMP51:%.*]] = fcmp fast ogt float [[TMP50]], undef
+; AVX-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], float [[TMP50]], float undef
+; AVX-NEXT:    [[TMP53:%.*]] = fcmp fast ogt float [[TMP52]], undef
+; AVX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP52]], float undef
+; AVX-NEXT:    [[TMP55:%.*]] = fcmp fast ogt float [[TMP54]], undef
+; AVX-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], float [[TMP54]], float undef
+; AVX-NEXT:    [[TMP57:%.*]] = fcmp fast ogt float [[TMP56]], undef
+; AVX-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], float [[TMP56]], float undef
+; AVX-NEXT:    [[TMP59:%.*]] = fcmp fast ogt float [[TMP58]], undef
+; AVX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP58]], float undef
+; AVX-NEXT:    [[TMP61:%.*]] = fcmp fast ogt float [[TMP60]], undef
+; AVX-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], float [[TMP60]], float undef
+; AVX-NEXT:    [[TMP63:%.*]] = fcmp fast ogt float [[TMP62]], undef
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
 ; AVX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
 ; AVX-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
-; AVX-NEXT:    [[TMP96:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
-; AVX-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], float [[TMP93]], float undef
-; AVX-NEXT:    ret float [[TMP96]]
+; AVX-NEXT:    [[TMP64:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], float [[TMP62]], float undef
+; AVX-NEXT:    ret float [[TMP64]]
 ;
 ; AVX2-LABEL: @maxf32(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
-; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <32 x float> [[TMP2]], i32 0
-; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <32 x float> [[TMP2]], i32 1
-; AVX2-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <32 x float> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <32 x float> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <32 x float> [[TMP2]], i32 4
-; AVX2-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <32 x float> [[TMP2]], i32 5
-; AVX2-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <32 x float> [[TMP2]], i32 6
-; AVX2-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; AVX2-NEXT:    [[TMP22:%.*]] = extractelement <32 x float> [[TMP2]], i32 7
-; AVX2-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
-; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; AVX2-NEXT:    [[TMP25:%.*]] = extractelement <32 x float> [[TMP2]], i32 8
-; AVX2-NEXT:    [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
-; AVX2-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
-; AVX2-NEXT:    [[TMP28:%.*]] = extractelement <32 x float> [[TMP2]], i32 9
-; AVX2-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
-; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
-; AVX2-NEXT:    [[TMP31:%.*]] = extractelement <32 x float> [[TMP2]], i32 10
-; AVX2-NEXT:    [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
-; AVX2-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
-; AVX2-NEXT:    [[TMP34:%.*]] = extractelement <32 x float> [[TMP2]], i32 11
-; AVX2-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
-; AVX2-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
-; AVX2-NEXT:    [[TMP37:%.*]] = extractelement <32 x float> [[TMP2]], i32 12
-; AVX2-NEXT:    [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
-; AVX2-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
-; AVX2-NEXT:    [[TMP40:%.*]] = extractelement <32 x float> [[TMP2]], i32 13
-; AVX2-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
-; AVX2-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
-; AVX2-NEXT:    [[TMP43:%.*]] = extractelement <32 x float> [[TMP2]], i32 14
-; AVX2-NEXT:    [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
-; AVX2-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
-; AVX2-NEXT:    [[TMP46:%.*]] = extractelement <32 x float> [[TMP2]], i32 15
-; AVX2-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
-; AVX2-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
-; AVX2-NEXT:    [[TMP49:%.*]] = extractelement <32 x float> [[TMP2]], i32 16
-; AVX2-NEXT:    [[TMP50:%.*]] = fcmp fast ogt float [[TMP48]], [[TMP49]]
-; AVX2-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], float [[TMP48]], float undef
-; AVX2-NEXT:    [[TMP52:%.*]] = extractelement <32 x float> [[TMP2]], i32 17
-; AVX2-NEXT:    [[TMP53:%.*]] = fcmp fast ogt float [[TMP51]], [[TMP52]]
-; AVX2-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP51]], float undef
-; AVX2-NEXT:    [[TMP55:%.*]] = extractelement <32 x float> [[TMP2]], i32 18
-; AVX2-NEXT:    [[TMP56:%.*]] = fcmp fast ogt float [[TMP54]], [[TMP55]]
-; AVX2-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], float [[TMP54]], float undef
-; AVX2-NEXT:    [[TMP58:%.*]] = extractelement <32 x float> [[TMP2]], i32 19
-; AVX2-NEXT:    [[TMP59:%.*]] = fcmp fast ogt float [[TMP57]], [[TMP58]]
-; AVX2-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP57]], float undef
-; AVX2-NEXT:    [[TMP61:%.*]] = extractelement <32 x float> [[TMP2]], i32 20
-; AVX2-NEXT:    [[TMP62:%.*]] = fcmp fast ogt float [[TMP60]], [[TMP61]]
-; AVX2-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], float [[TMP60]], float undef
-; AVX2-NEXT:    [[TMP64:%.*]] = extractelement <32 x float> [[TMP2]], i32 21
-; AVX2-NEXT:    [[TMP65:%.*]] = fcmp fast ogt float [[TMP63]], [[TMP64]]
-; AVX2-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], float [[TMP63]], float undef
-; AVX2-NEXT:    [[TMP67:%.*]] = extractelement <32 x float> [[TMP2]], i32 22
-; AVX2-NEXT:    [[TMP68:%.*]] = fcmp fast ogt float [[TMP66]], [[TMP67]]
-; AVX2-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], float [[TMP66]], float undef
-; AVX2-NEXT:    [[TMP70:%.*]] = extractelement <32 x float> [[TMP2]], i32 23
-; AVX2-NEXT:    [[TMP71:%.*]] = fcmp fast ogt float [[TMP69]], [[TMP70]]
-; AVX2-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], float [[TMP69]], float undef
-; AVX2-NEXT:    [[TMP73:%.*]] = extractelement <32 x float> [[TMP2]], i32 24
-; AVX2-NEXT:    [[TMP74:%.*]] = fcmp fast ogt float [[TMP72]], [[TMP73]]
-; AVX2-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], float [[TMP72]], float undef
-; AVX2-NEXT:    [[TMP76:%.*]] = extractelement <32 x float> [[TMP2]], i32 25
-; AVX2-NEXT:    [[TMP77:%.*]] = fcmp fast ogt float [[TMP75]], [[TMP76]]
-; AVX2-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], float [[TMP75]], float undef
-; AVX2-NEXT:    [[TMP79:%.*]] = extractelement <32 x float> [[TMP2]], i32 26
-; AVX2-NEXT:    [[TMP80:%.*]] = fcmp fast ogt float [[TMP78]], [[TMP79]]
-; AVX2-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], float [[TMP78]], float undef
-; AVX2-NEXT:    [[TMP82:%.*]] = extractelement <32 x float> [[TMP2]], i32 27
-; AVX2-NEXT:    [[TMP83:%.*]] = fcmp fast ogt float [[TMP81]], [[TMP82]]
-; AVX2-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], float [[TMP81]], float undef
-; AVX2-NEXT:    [[TMP85:%.*]] = extractelement <32 x float> [[TMP2]], i32 28
-; AVX2-NEXT:    [[TMP86:%.*]] = fcmp fast ogt float [[TMP84]], [[TMP85]]
-; AVX2-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], float [[TMP84]], float undef
-; AVX2-NEXT:    [[TMP88:%.*]] = extractelement <32 x float> [[TMP2]], i32 29
-; AVX2-NEXT:    [[TMP89:%.*]] = fcmp fast ogt float [[TMP87]], [[TMP88]]
-; AVX2-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], float [[TMP87]], float undef
-; AVX2-NEXT:    [[TMP91:%.*]] = extractelement <32 x float> [[TMP2]], i32 30
-; AVX2-NEXT:    [[TMP92:%.*]] = fcmp fast ogt float [[TMP90]], [[TMP91]]
-; AVX2-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], float [[TMP90]], float undef
-; AVX2-NEXT:    [[TMP94:%.*]] = extractelement <32 x float> [[TMP2]], i32 31
-; AVX2-NEXT:    [[TMP95:%.*]] = fcmp fast ogt float [[TMP93]], [[TMP94]]
+; AVX2-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; AVX2-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; AVX2-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; AVX2-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; AVX2-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; AVX2-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; AVX2-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; AVX2-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
+; AVX2-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; AVX2-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP16]], undef
+; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP16]], float undef
+; AVX2-NEXT:    [[TMP19:%.*]] = fcmp fast ogt float [[TMP18]], undef
+; AVX2-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP18]], float undef
+; AVX2-NEXT:    [[TMP21:%.*]] = fcmp fast ogt float [[TMP20]], undef
+; AVX2-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], float [[TMP20]], float undef
+; AVX2-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP22]], undef
+; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP22]], float undef
+; AVX2-NEXT:    [[TMP25:%.*]] = fcmp fast ogt float [[TMP24]], undef
+; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], float [[TMP24]], float undef
+; AVX2-NEXT:    [[TMP27:%.*]] = fcmp fast ogt float [[TMP26]], undef
+; AVX2-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], float [[TMP26]], float undef
+; AVX2-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP28]], undef
+; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP28]], float undef
+; AVX2-NEXT:    [[TMP31:%.*]] = fcmp fast ogt float [[TMP30]], undef
+; AVX2-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], float [[TMP30]], float undef
+; AVX2-NEXT:    [[TMP33:%.*]] = fcmp fast ogt float [[TMP32]], undef
+; AVX2-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], float [[TMP32]], float undef
+; AVX2-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP34]], undef
+; AVX2-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP34]], float undef
+; AVX2-NEXT:    [[TMP37:%.*]] = fcmp fast ogt float [[TMP36]], undef
+; AVX2-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], float [[TMP36]], float undef
+; AVX2-NEXT:    [[TMP39:%.*]] = fcmp fast ogt float [[TMP38]], undef
+; AVX2-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], float [[TMP38]], float undef
+; AVX2-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP40]], undef
+; AVX2-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP40]], float undef
+; AVX2-NEXT:    [[TMP43:%.*]] = fcmp fast ogt float [[TMP42]], undef
+; AVX2-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], float [[TMP42]], float undef
+; AVX2-NEXT:    [[TMP45:%.*]] = fcmp fast ogt float [[TMP44]], undef
+; AVX2-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], float [[TMP44]], float undef
+; AVX2-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP46]], undef
+; AVX2-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP46]], float undef
+; AVX2-NEXT:    [[TMP49:%.*]] = fcmp fast ogt float [[TMP48]], undef
+; AVX2-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], float [[TMP48]], float undef
+; AVX2-NEXT:    [[TMP51:%.*]] = fcmp fast ogt float [[TMP50]], undef
+; AVX2-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], float [[TMP50]], float undef
+; AVX2-NEXT:    [[TMP53:%.*]] = fcmp fast ogt float [[TMP52]], undef
+; AVX2-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP52]], float undef
+; AVX2-NEXT:    [[TMP55:%.*]] = fcmp fast ogt float [[TMP54]], undef
+; AVX2-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], float [[TMP54]], float undef
+; AVX2-NEXT:    [[TMP57:%.*]] = fcmp fast ogt float [[TMP56]], undef
+; AVX2-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], float [[TMP56]], float undef
+; AVX2-NEXT:    [[TMP59:%.*]] = fcmp fast ogt float [[TMP58]], undef
+; AVX2-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP58]], float undef
+; AVX2-NEXT:    [[TMP61:%.*]] = fcmp fast ogt float [[TMP60]], undef
+; AVX2-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], float [[TMP60]], float undef
+; AVX2-NEXT:    [[TMP63:%.*]] = fcmp fast ogt float [[TMP62]], undef
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
 ; AVX2-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
 ; AVX2-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
-; AVX2-NEXT:    [[TMP96:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
-; AVX2-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], float [[TMP93]], float undef
-; AVX2-NEXT:    ret float [[TMP96]]
+; AVX2-NEXT:    [[TMP64:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
+; AVX2-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], float [[TMP62]], float undef
+; AVX2-NEXT:    ret float [[TMP64]]
 ;
 ; SKX-LABEL: @maxf32(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <32 x float>, <32 x float>* bitcast ([32 x float]* @arr1 to <32 x float>*), align 16
-; SKX-NEXT:    [[TMP3:%.*]] = extractelement <32 x float> [[TMP2]], i32 0
-; SKX-NEXT:    [[TMP4:%.*]] = extractelement <32 x float> [[TMP2]], i32 1
-; SKX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP3]], [[TMP4]]
-; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float undef, float undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <32 x float> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = fcmp fast ogt float [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], float [[TMP6]], float undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <32 x float> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP9]], float undef
-; SKX-NEXT:    [[TMP13:%.*]] = extractelement <32 x float> [[TMP2]], i32 4
-; SKX-NEXT:    [[TMP14:%.*]] = fcmp fast ogt float [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], float [[TMP12]], float undef
-; SKX-NEXT:    [[TMP16:%.*]] = extractelement <32 x float> [[TMP2]], i32 5
-; SKX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP15]], float undef
-; SKX-NEXT:    [[TMP19:%.*]] = extractelement <32 x float> [[TMP2]], i32 6
-; SKX-NEXT:    [[TMP20:%.*]] = fcmp fast ogt float [[TMP18]], [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], float [[TMP18]], float undef
-; SKX-NEXT:    [[TMP22:%.*]] = extractelement <32 x float> [[TMP2]], i32 7
-; SKX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP21]], [[TMP22]]
-; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP21]], float undef
-; SKX-NEXT:    [[TMP25:%.*]] = extractelement <32 x float> [[TMP2]], i32 8
-; SKX-NEXT:    [[TMP26:%.*]] = fcmp fast ogt float [[TMP24]], [[TMP25]]
-; SKX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], float [[TMP24]], float undef
-; SKX-NEXT:    [[TMP28:%.*]] = extractelement <32 x float> [[TMP2]], i32 9
-; SKX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP27]], [[TMP28]]
-; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP27]], float undef
-; SKX-NEXT:    [[TMP31:%.*]] = extractelement <32 x float> [[TMP2]], i32 10
-; SKX-NEXT:    [[TMP32:%.*]] = fcmp fast ogt float [[TMP30]], [[TMP31]]
-; SKX-NEXT:    [[TMP33:%.*]] = select i1 [[TMP32]], float [[TMP30]], float undef
-; SKX-NEXT:    [[TMP34:%.*]] = extractelement <32 x float> [[TMP2]], i32 11
-; SKX-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP33]], [[TMP34]]
-; SKX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP33]], float undef
-; SKX-NEXT:    [[TMP37:%.*]] = extractelement <32 x float> [[TMP2]], i32 12
-; SKX-NEXT:    [[TMP38:%.*]] = fcmp fast ogt float [[TMP36]], [[TMP37]]
-; SKX-NEXT:    [[TMP39:%.*]] = select i1 [[TMP38]], float [[TMP36]], float undef
-; SKX-NEXT:    [[TMP40:%.*]] = extractelement <32 x float> [[TMP2]], i32 13
-; SKX-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP39]], [[TMP40]]
-; SKX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP39]], float undef
-; SKX-NEXT:    [[TMP43:%.*]] = extractelement <32 x float> [[TMP2]], i32 14
-; SKX-NEXT:    [[TMP44:%.*]] = fcmp fast ogt float [[TMP42]], [[TMP43]]
-; SKX-NEXT:    [[TMP45:%.*]] = select i1 [[TMP44]], float [[TMP42]], float undef
-; SKX-NEXT:    [[TMP46:%.*]] = extractelement <32 x float> [[TMP2]], i32 15
-; SKX-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP45]], [[TMP46]]
-; SKX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP45]], float undef
-; SKX-NEXT:    [[TMP49:%.*]] = extractelement <32 x float> [[TMP2]], i32 16
-; SKX-NEXT:    [[TMP50:%.*]] = fcmp fast ogt float [[TMP48]], [[TMP49]]
-; SKX-NEXT:    [[TMP51:%.*]] = select i1 [[TMP50]], float [[TMP48]], float undef
-; SKX-NEXT:    [[TMP52:%.*]] = extractelement <32 x float> [[TMP2]], i32 17
-; SKX-NEXT:    [[TMP53:%.*]] = fcmp fast ogt float [[TMP51]], [[TMP52]]
-; SKX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP51]], float undef
-; SKX-NEXT:    [[TMP55:%.*]] = extractelement <32 x float> [[TMP2]], i32 18
-; SKX-NEXT:    [[TMP56:%.*]] = fcmp fast ogt float [[TMP54]], [[TMP55]]
-; SKX-NEXT:    [[TMP57:%.*]] = select i1 [[TMP56]], float [[TMP54]], float undef
-; SKX-NEXT:    [[TMP58:%.*]] = extractelement <32 x float> [[TMP2]], i32 19
-; SKX-NEXT:    [[TMP59:%.*]] = fcmp fast ogt float [[TMP57]], [[TMP58]]
-; SKX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP57]], float undef
-; SKX-NEXT:    [[TMP61:%.*]] = extractelement <32 x float> [[TMP2]], i32 20
-; SKX-NEXT:    [[TMP62:%.*]] = fcmp fast ogt float [[TMP60]], [[TMP61]]
-; SKX-NEXT:    [[TMP63:%.*]] = select i1 [[TMP62]], float [[TMP60]], float undef
-; SKX-NEXT:    [[TMP64:%.*]] = extractelement <32 x float> [[TMP2]], i32 21
-; SKX-NEXT:    [[TMP65:%.*]] = fcmp fast ogt float [[TMP63]], [[TMP64]]
-; SKX-NEXT:    [[TMP66:%.*]] = select i1 [[TMP65]], float [[TMP63]], float undef
-; SKX-NEXT:    [[TMP67:%.*]] = extractelement <32 x float> [[TMP2]], i32 22
-; SKX-NEXT:    [[TMP68:%.*]] = fcmp fast ogt float [[TMP66]], [[TMP67]]
-; SKX-NEXT:    [[TMP69:%.*]] = select i1 [[TMP68]], float [[TMP66]], float undef
-; SKX-NEXT:    [[TMP70:%.*]] = extractelement <32 x float> [[TMP2]], i32 23
-; SKX-NEXT:    [[TMP71:%.*]] = fcmp fast ogt float [[TMP69]], [[TMP70]]
-; SKX-NEXT:    [[TMP72:%.*]] = select i1 [[TMP71]], float [[TMP69]], float undef
-; SKX-NEXT:    [[TMP73:%.*]] = extractelement <32 x float> [[TMP2]], i32 24
-; SKX-NEXT:    [[TMP74:%.*]] = fcmp fast ogt float [[TMP72]], [[TMP73]]
-; SKX-NEXT:    [[TMP75:%.*]] = select i1 [[TMP74]], float [[TMP72]], float undef
-; SKX-NEXT:    [[TMP76:%.*]] = extractelement <32 x float> [[TMP2]], i32 25
-; SKX-NEXT:    [[TMP77:%.*]] = fcmp fast ogt float [[TMP75]], [[TMP76]]
-; SKX-NEXT:    [[TMP78:%.*]] = select i1 [[TMP77]], float [[TMP75]], float undef
-; SKX-NEXT:    [[TMP79:%.*]] = extractelement <32 x float> [[TMP2]], i32 26
-; SKX-NEXT:    [[TMP80:%.*]] = fcmp fast ogt float [[TMP78]], [[TMP79]]
-; SKX-NEXT:    [[TMP81:%.*]] = select i1 [[TMP80]], float [[TMP78]], float undef
-; SKX-NEXT:    [[TMP82:%.*]] = extractelement <32 x float> [[TMP2]], i32 27
-; SKX-NEXT:    [[TMP83:%.*]] = fcmp fast ogt float [[TMP81]], [[TMP82]]
-; SKX-NEXT:    [[TMP84:%.*]] = select i1 [[TMP83]], float [[TMP81]], float undef
-; SKX-NEXT:    [[TMP85:%.*]] = extractelement <32 x float> [[TMP2]], i32 28
-; SKX-NEXT:    [[TMP86:%.*]] = fcmp fast ogt float [[TMP84]], [[TMP85]]
-; SKX-NEXT:    [[TMP87:%.*]] = select i1 [[TMP86]], float [[TMP84]], float undef
-; SKX-NEXT:    [[TMP88:%.*]] = extractelement <32 x float> [[TMP2]], i32 29
-; SKX-NEXT:    [[TMP89:%.*]] = fcmp fast ogt float [[TMP87]], [[TMP88]]
-; SKX-NEXT:    [[TMP90:%.*]] = select i1 [[TMP89]], float [[TMP87]], float undef
-; SKX-NEXT:    [[TMP91:%.*]] = extractelement <32 x float> [[TMP2]], i32 30
-; SKX-NEXT:    [[TMP92:%.*]] = fcmp fast ogt float [[TMP90]], [[TMP91]]
-; SKX-NEXT:    [[TMP93:%.*]] = select i1 [[TMP92]], float [[TMP90]], float undef
-; SKX-NEXT:    [[TMP94:%.*]] = extractelement <32 x float> [[TMP2]], i32 31
-; SKX-NEXT:    [[TMP95:%.*]] = fcmp fast ogt float [[TMP93]], [[TMP94]]
+; SKX-NEXT:    [[TMP3:%.*]] = fcmp fast ogt float undef, undef
+; SKX-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], float undef, float undef
+; SKX-NEXT:    [[TMP5:%.*]] = fcmp fast ogt float [[TMP4]], undef
+; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float undef
+; SKX-NEXT:    [[TMP7:%.*]] = fcmp fast ogt float [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP6]], float undef
+; SKX-NEXT:    [[TMP9:%.*]] = fcmp fast ogt float [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], float [[TMP8]], float undef
+; SKX-NEXT:    [[TMP11:%.*]] = fcmp fast ogt float [[TMP10]], undef
+; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], float [[TMP10]], float undef
+; SKX-NEXT:    [[TMP13:%.*]] = fcmp fast ogt float [[TMP12]], undef
+; SKX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], float [[TMP12]], float undef
+; SKX-NEXT:    [[TMP15:%.*]] = fcmp fast ogt float [[TMP14]], undef
+; SKX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], float [[TMP14]], float undef
+; SKX-NEXT:    [[TMP17:%.*]] = fcmp fast ogt float [[TMP16]], undef
+; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], float [[TMP16]], float undef
+; SKX-NEXT:    [[TMP19:%.*]] = fcmp fast ogt float [[TMP18]], undef
+; SKX-NEXT:    [[TMP20:%.*]] = select i1 [[TMP19]], float [[TMP18]], float undef
+; SKX-NEXT:    [[TMP21:%.*]] = fcmp fast ogt float [[TMP20]], undef
+; SKX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], float [[TMP20]], float undef
+; SKX-NEXT:    [[TMP23:%.*]] = fcmp fast ogt float [[TMP22]], undef
+; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], float [[TMP22]], float undef
+; SKX-NEXT:    [[TMP25:%.*]] = fcmp fast ogt float [[TMP24]], undef
+; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], float [[TMP24]], float undef
+; SKX-NEXT:    [[TMP27:%.*]] = fcmp fast ogt float [[TMP26]], undef
+; SKX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], float [[TMP26]], float undef
+; SKX-NEXT:    [[TMP29:%.*]] = fcmp fast ogt float [[TMP28]], undef
+; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], float [[TMP28]], float undef
+; SKX-NEXT:    [[TMP31:%.*]] = fcmp fast ogt float [[TMP30]], undef
+; SKX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], float [[TMP30]], float undef
+; SKX-NEXT:    [[TMP33:%.*]] = fcmp fast ogt float [[TMP32]], undef
+; SKX-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], float [[TMP32]], float undef
+; SKX-NEXT:    [[TMP35:%.*]] = fcmp fast ogt float [[TMP34]], undef
+; SKX-NEXT:    [[TMP36:%.*]] = select i1 [[TMP35]], float [[TMP34]], float undef
+; SKX-NEXT:    [[TMP37:%.*]] = fcmp fast ogt float [[TMP36]], undef
+; SKX-NEXT:    [[TMP38:%.*]] = select i1 [[TMP37]], float [[TMP36]], float undef
+; SKX-NEXT:    [[TMP39:%.*]] = fcmp fast ogt float [[TMP38]], undef
+; SKX-NEXT:    [[TMP40:%.*]] = select i1 [[TMP39]], float [[TMP38]], float undef
+; SKX-NEXT:    [[TMP41:%.*]] = fcmp fast ogt float [[TMP40]], undef
+; SKX-NEXT:    [[TMP42:%.*]] = select i1 [[TMP41]], float [[TMP40]], float undef
+; SKX-NEXT:    [[TMP43:%.*]] = fcmp fast ogt float [[TMP42]], undef
+; SKX-NEXT:    [[TMP44:%.*]] = select i1 [[TMP43]], float [[TMP42]], float undef
+; SKX-NEXT:    [[TMP45:%.*]] = fcmp fast ogt float [[TMP44]], undef
+; SKX-NEXT:    [[TMP46:%.*]] = select i1 [[TMP45]], float [[TMP44]], float undef
+; SKX-NEXT:    [[TMP47:%.*]] = fcmp fast ogt float [[TMP46]], undef
+; SKX-NEXT:    [[TMP48:%.*]] = select i1 [[TMP47]], float [[TMP46]], float undef
+; SKX-NEXT:    [[TMP49:%.*]] = fcmp fast ogt float [[TMP48]], undef
+; SKX-NEXT:    [[TMP50:%.*]] = select i1 [[TMP49]], float [[TMP48]], float undef
+; SKX-NEXT:    [[TMP51:%.*]] = fcmp fast ogt float [[TMP50]], undef
+; SKX-NEXT:    [[TMP52:%.*]] = select i1 [[TMP51]], float [[TMP50]], float undef
+; SKX-NEXT:    [[TMP53:%.*]] = fcmp fast ogt float [[TMP52]], undef
+; SKX-NEXT:    [[TMP54:%.*]] = select i1 [[TMP53]], float [[TMP52]], float undef
+; SKX-NEXT:    [[TMP55:%.*]] = fcmp fast ogt float [[TMP54]], undef
+; SKX-NEXT:    [[TMP56:%.*]] = select i1 [[TMP55]], float [[TMP54]], float undef
+; SKX-NEXT:    [[TMP57:%.*]] = fcmp fast ogt float [[TMP56]], undef
+; SKX-NEXT:    [[TMP58:%.*]] = select i1 [[TMP57]], float [[TMP56]], float undef
+; SKX-NEXT:    [[TMP59:%.*]] = fcmp fast ogt float [[TMP58]], undef
+; SKX-NEXT:    [[TMP60:%.*]] = select i1 [[TMP59]], float [[TMP58]], float undef
+; SKX-NEXT:    [[TMP61:%.*]] = fcmp fast ogt float [[TMP60]], undef
+; SKX-NEXT:    [[TMP62:%.*]] = select i1 [[TMP61]], float [[TMP60]], float undef
+; SKX-NEXT:    [[TMP63:%.*]] = fcmp fast ogt float [[TMP62]], undef
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <32 x float> [[TMP2]], <32 x float> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = fcmp fast ogt <32 x float> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP]], <32 x float> [[TMP2]], <32 x float> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP2]], <32 x float> [[RDX_MINMAX_SELECT]], <32 x float> [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_SHUF4:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP5:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT3]], [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT6:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP5]], <32 x float> [[RDX_MINMAX_SELECT3]], <32 x float> [[RDX_SHUF4]]
 ; SKX-NEXT:    [[RDX_SHUF7:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP8:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT6]], [[RDX_SHUF7]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT9:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP8]], <32 x float> [[RDX_MINMAX_SELECT6]], <32 x float> [[RDX_SHUF7]]
 ; SKX-NEXT:    [[RDX_SHUF10:%.*]] = shufflevector <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP11:%.*]] = fcmp fast ogt <32 x float> [[RDX_MINMAX_SELECT9]], [[RDX_SHUF10]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT12:%.*]] = select <32 x i1> [[RDX_MINMAX_CMP11]], <32 x float> [[RDX_MINMAX_SELECT9]], <32 x float> [[RDX_SHUF10]]
-; SKX-NEXT:    [[TMP96:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
-; SKX-NEXT:    [[TMP97:%.*]] = select i1 [[TMP95]], float [[TMP93]], float undef
-; SKX-NEXT:    ret float [[TMP96]]
+; SKX-NEXT:    [[TMP64:%.*]] = extractelement <32 x float> [[RDX_MINMAX_SELECT12]], i32 0
+; SKX-NEXT:    [[TMP65:%.*]] = select i1 [[TMP63]], float [[TMP62]], float undef
+; SKX-NEXT:    ret float [[TMP64]]
 ;
   %2 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 0), align 16
   %3 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 1), align 4
   %4 = fcmp fast ogt float %2, %3
   %5 = select i1 %4, float %2, float %3
   %6 = load float, float* getelementptr inbounds ([32 x float], [32 x float]* @arr1, i64 0, i64 2), align 8
   %7 = fcmp fast ogt float %5, %6
   %8 = select i1 %7, float %5, float %6

 ; CHECK-NEXT:    ret i32 [[TMP23]]
 ;
 ; AVX-LABEL: @maxi8_mutiple_uses(
 ; AVX-NEXT:    [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
 ; AVX-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[TMP2]], i32 0
 ; AVX-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[TMP2]], i32 1
 ; AVX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
 ; AVX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP2]], i32 2
-; AVX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
-; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
-; AVX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 [[TMP13]]
-; AVX-NEXT:    [[TMP16:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
-; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 [[TMP16]]
-; AVX-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; AVX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; AVX-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
+; AVX-NEXT:    [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP10]], i32 [[TMP11]]
+; AVX-NEXT:    [[TMP14:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; AVX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP13]], [[TMP14]]
+; AVX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP13]], i32 [[TMP14]]
+; AVX-NEXT:    [[TMP17:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX-NEXT:    [[TMP18:%.*]] = icmp sgt i32 [[TMP16]], [[TMP17]]
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP2]], <4 x i32> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
-; AVX-NEXT:    [[TMP21:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
-; AVX-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP13]]
-; AVX-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP13]]
-; AVX-NEXT:    [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP16]]
-; AVX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP16]]
-; AVX-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP25]], [[TMP19]]
-; AVX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP25]], i32 [[TMP19]]
-; AVX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
-; AVX-NEXT:    [[TMP29:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
-; AVX-NEXT:    [[TMP30:%.*]] = icmp sgt i32 [[TMP27]], [[TMP29]]
-; AVX-NEXT:    [[TMP31:%.*]] = select i1 [[TMP30]], i32 [[TMP27]], i32 [[TMP29]]
-; AVX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP5]], i32 3, i32 4
-; AVX-NEXT:    store i32 [[TMP32]], i32* @var, align 8
-; AVX-NEXT:    ret i32 [[TMP31]]
+; AVX-NEXT:    [[TMP19:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP19]], [[TMP11]]
+; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP19]], i32 [[TMP11]]
+; AVX-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP14]]
+; AVX-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP14]]
+; AVX-NEXT:    [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP17]]
+; AVX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP17]]
+; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP18]], i32 [[TMP16]], i32 [[TMP17]]
+; AVX-NEXT:    [[TMP27:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX-NEXT:    [[TMP28:%.*]] = icmp sgt i32 [[TMP25]], [[TMP27]]
+; AVX-NEXT:    [[TMP29:%.*]] = select i1 [[TMP28]], i32 [[TMP25]], i32 [[TMP27]]
+; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP5]], i32 3, i32 4
+; AVX-NEXT:    store i32 [[TMP30]], i32* @var, align 8
+; AVX-NEXT:    ret i32 [[TMP29]]
 ;
 ; AVX2-LABEL: @maxi8_mutiple_uses(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
 ; AVX2-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[TMP2]], i32 0
 ; AVX2-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[TMP2]], i32 1
 ; AVX2-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
 ; AVX2-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP2]], i32 2
-; AVX2-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
-; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX2-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
-; AVX2-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 [[TMP13]]
-; AVX2-NEXT:    [[TMP16:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
-; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 [[TMP16]]
-; AVX2-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; AVX2-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; AVX2-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; AVX2-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX2-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; AVX2-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
+; AVX2-NEXT:    [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP10]], i32 [[TMP11]]
+; AVX2-NEXT:    [[TMP14:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; AVX2-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP13]], [[TMP14]]
+; AVX2-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP13]], i32 [[TMP14]]
+; AVX2-NEXT:    [[TMP17:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX2-NEXT:    [[TMP18:%.*]] = icmp sgt i32 [[TMP16]], [[TMP17]]
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP2]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP2]], <4 x i32> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
-; AVX2-NEXT:    [[TMP21:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
-; AVX2-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP13]]
-; AVX2-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP13]]
-; AVX2-NEXT:    [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP16]]
-; AVX2-NEXT:    [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP16]]
-; AVX2-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP25]], [[TMP19]]
-; AVX2-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP25]], i32 [[TMP19]]
-; AVX2-NEXT:    [[TMP28:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
-; AVX2-NEXT:    [[TMP29:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
-; AVX2-NEXT:    [[TMP30:%.*]] = icmp sgt i32 [[TMP27]], [[TMP29]]
-; AVX2-NEXT:    [[TMP31:%.*]] = select i1 [[TMP30]], i32 [[TMP27]], i32 [[TMP29]]
-; AVX2-NEXT:    [[TMP32:%.*]] = select i1 [[TMP5]], i32 3, i32 4
-; AVX2-NEXT:    store i32 [[TMP32]], i32* @var, align 8
-; AVX2-NEXT:    ret i32 [[TMP31]]
+; AVX2-NEXT:    [[TMP19:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX2-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP19]], [[TMP11]]
+; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP19]], i32 [[TMP11]]
+; AVX2-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP14]]
+; AVX2-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP14]]
+; AVX2-NEXT:    [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP17]]
+; AVX2-NEXT:    [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP17]]
+; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP18]], i32 [[TMP16]], i32 [[TMP17]]
+; AVX2-NEXT:    [[TMP27:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX2-NEXT:    [[TMP28:%.*]] = icmp sgt i32 [[TMP25]], [[TMP27]]
+; AVX2-NEXT:    [[TMP29:%.*]] = select i1 [[TMP28]], i32 [[TMP25]], i32 [[TMP27]]
+; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP5]], i32 3, i32 4
+; AVX2-NEXT:    store i32 [[TMP30]], i32* @var, align 8
+; AVX2-NEXT:    ret i32 [[TMP29]]
 ;
 ; SKX-LABEL: @maxi8_mutiple_uses(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* bitcast ([32 x i32]* @arr to <4 x i32>*), align 16
 ; SKX-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[TMP2]], i32 0
 ; SKX-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[TMP2]], i32 1
 ; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
 ; SKX-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], i32 undef, i32 undef
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP2]], i32 2
-; SKX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP6]], [[TMP7]]
-; SKX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP6]], i32 undef
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
-; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; SKX-NEXT:    [[TMP13:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
-; SKX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; SKX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 [[TMP13]]
-; SKX-NEXT:    [[TMP16:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
-; SKX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; SKX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 [[TMP16]]
-; SKX-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; SKX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
+; SKX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP6]], undef
+; SKX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP6]], i32 undef
+; SKX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; SKX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; SKX-NEXT:    [[TMP11:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 4), align 16
+; SKX-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP10]], [[TMP11]]
+; SKX-NEXT:    [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP10]], i32 [[TMP11]]
+; SKX-NEXT:    [[TMP14:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 5), align 4
+; SKX-NEXT:    [[TMP15:%.*]] = icmp sgt i32 [[TMP13]], [[TMP14]]
+; SKX-NEXT:    [[TMP16:%.*]] = select i1 [[TMP15]], i32 [[TMP13]], i32 [[TMP14]]
+; SKX-NEXT:    [[TMP17:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; SKX-NEXT:    [[TMP18:%.*]] = icmp sgt i32 [[TMP16]], [[TMP17]]
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP2]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP2]], <4 x i32> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
-; SKX-NEXT:    [[TMP21:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
-; SKX-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP13]]
-; SKX-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP13]]
-; SKX-NEXT:    [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP16]]
-; SKX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP16]]
-; SKX-NEXT:    [[TMP26:%.*]] = icmp sgt i32 [[TMP25]], [[TMP19]]
-; SKX-NEXT:    [[TMP27:%.*]] = select i1 [[TMP26]], i32 [[TMP25]], i32 [[TMP19]]
-; SKX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
-; SKX-NEXT:    [[TMP29:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
-; SKX-NEXT:    [[TMP30:%.*]] = icmp sgt i32 [[TMP27]], [[TMP29]]
-; SKX-NEXT:    [[TMP31:%.*]] = select i1 [[TMP30]], i32 [[TMP27]], i32 [[TMP29]]
-; SKX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP5]], i32 3, i32 4
-; SKX-NEXT:    store i32 [[TMP32]], i32* @var, align 8
-; SKX-NEXT:    ret i32 [[TMP31]]
+; SKX-NEXT:    [[TMP19:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; SKX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP19]], [[TMP11]]
+; SKX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP19]], i32 [[TMP11]]
+; SKX-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP14]]
+; SKX-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 [[TMP14]]
+; SKX-NEXT:    [[TMP24:%.*]] = icmp sgt i32 [[TMP23]], [[TMP17]]
+; SKX-NEXT:    [[TMP25:%.*]] = select i1 [[TMP24]], i32 [[TMP23]], i32 [[TMP17]]
+; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP18]], i32 [[TMP16]], i32 [[TMP17]]
+; SKX-NEXT:    [[TMP27:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; SKX-NEXT:    [[TMP28:%.*]] = icmp sgt i32 [[TMP25]], [[TMP27]]
+; SKX-NEXT:    [[TMP29:%.*]] = select i1 [[TMP28]], i32 [[TMP25]], i32 [[TMP27]]
+; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP5]], i32 3, i32 4
+; SKX-NEXT:    store i32 [[TMP30]], i32* @var, align 8
+; SKX-NEXT:    ret i32 [[TMP29]]
 ;
   %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
   %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
   %4 = icmp sgt i32 %2, %3
   %5 = select i1 %4, i32 %2, i32 %3
   %6 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2), align 8
   %7 = icmp sgt i32 %5, %6
   %8 = select i1 %7, i32 %5, i32 %6

 ; AVX-LABEL: @maxi8_wrong_parent(
 ; AVX-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
 ; AVX-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
 ; AVX-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
 ; AVX-NEXT:    br label [[PP:%.*]]
 ; AVX:       pp:
 ; AVX-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
 ; AVX-NEXT:    [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
-; AVX-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
-; AVX-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP5]], [[TMP7]]
-; AVX-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP5]], i32 undef
-; AVX-NEXT:    [[TMP10:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
-; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX-NEXT:    [[TMP13:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
-; AVX-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX-NEXT:    [[TMP16:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
-; AVX-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; AVX-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
-; AVX-NEXT:    [[TMP22:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
-; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], undef
+; AVX-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 undef
+; AVX-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX-NEXT:    [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX-NEXT:    [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
+; AVX-NEXT:    [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
+; AVX-NEXT:    [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
 ; AVX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
 ; AVX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; AVX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
-; AVX-NEXT:    [[TMP24:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
-; AVX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], [[TMP19]]
-; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 [[TMP19]]
-; AVX-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], [[TMP22]]
-; AVX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 [[TMP22]]
-; AVX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], [[TMP5]]
-; AVX-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 [[TMP5]]
-; AVX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 [[TMP22]]
+; AVX-NEXT:    [[TMP20:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], [[TMP15]]
+; AVX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 [[TMP15]]
+; AVX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], [[TMP18]]
+; AVX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 [[TMP18]]
+; AVX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], [[TMP5]]
+; AVX-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 [[TMP5]]
+; AVX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
 ; AVX-NEXT:    ret i32 [[OP_EXTRA]]
 ;
 ; AVX2-LABEL: @maxi8_wrong_parent(
 ; AVX2-NEXT:    [[TMP2:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
 ; AVX2-NEXT:    [[TMP3:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
 ; AVX2-NEXT:    [[TMP4:%.*]] = icmp sgt i32 [[TMP2]], [[TMP3]]
 ; AVX2-NEXT:    br label [[PP:%.*]]
 ; AVX2:       pp:
 ; AVX2-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i32 [[TMP2]], i32 [[TMP3]]
 ; AVX2-NEXT:    [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
-; AVX2-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
-; AVX2-NEXT:    [[TMP8:%.*]] = icmp sgt i32 [[TMP5]], [[TMP7]]
-; AVX2-NEXT:    [[TMP9:%.*]] = select i1 [[TMP8]], i32 [[TMP5]], i32 undef
-; AVX2-NEXT:    [[TMP10:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
-; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP9]], [[TMP10]]
-; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP9]], i32 undef
-; AVX2-NEXT:    [[TMP13:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
-; AVX2-NEXT:    [[TMP14:%.*]] = icmp sgt i32 [[TMP12]], [[TMP13]]
-; AVX2-NEXT:    [[TMP15:%.*]] = select i1 [[TMP14]], i32 [[TMP12]], i32 undef
-; AVX2-NEXT:    [[TMP16:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
-; AVX2-NEXT:    [[TMP17:%.*]] = icmp sgt i32 [[TMP15]], [[TMP16]]
-; AVX2-NEXT:    [[TMP18:%.*]] = select i1 [[TMP17]], i32 [[TMP15]], i32 undef
-; AVX2-NEXT:    [[TMP19:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; AVX2-NEXT:    [[TMP20:%.*]] = icmp sgt i32 [[TMP18]], [[TMP19]]
-; AVX2-NEXT:    [[TMP21:%.*]] = select i1 [[TMP20]], i32 [[TMP18]], i32 [[TMP19]]
-; AVX2-NEXT:    [[TMP22:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
-; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP21]], [[TMP22]]
+; AVX2-NEXT:    [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], undef
+; AVX2-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], i32 [[TMP5]], i32 undef
+; AVX2-NEXT:    [[TMP9:%.*]] = icmp sgt i32 [[TMP8]], undef
+; AVX2-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], i32 [[TMP8]], i32 undef
+; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], undef
+; AVX2-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 undef
+; AVX2-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], undef
+; AVX2-NEXT:    [[TMP14:%.*]] = select i1 [[TMP13]], i32 [[TMP12]], i32 undef
+; AVX2-NEXT:    [[TMP15:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; AVX2-NEXT:    [[TMP16:%.*]] = icmp sgt i32 [[TMP14]], [[TMP15]]
+; AVX2-NEXT:    [[TMP17:%.*]] = select i1 [[TMP16]], i32 [[TMP14]], i32 [[TMP15]]
+; AVX2-NEXT:    [[TMP18:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; AVX2-NEXT:    [[TMP19:%.*]] = icmp sgt i32 [[TMP17]], [[TMP18]]
 ; AVX2-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
 ; AVX2-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; AVX2-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; AVX2-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
-; AVX2-NEXT:    [[TMP24:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
-; AVX2-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], [[TMP19]]
-; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 [[TMP19]]
-; AVX2-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], [[TMP22]]
-; AVX2-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 [[TMP22]]
-; AVX2-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], [[TMP5]]
-; AVX2-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 [[TMP5]]
-; AVX2-NEXT:    [[TMP30:%.*]] = select i1 [[TMP23]], i32 [[TMP21]], i32 [[TMP22]]
+; AVX2-NEXT:    [[TMP20:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; AVX2-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], [[TMP15]]
+; AVX2-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 [[TMP15]]
+; AVX2-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], [[TMP18]]
+; AVX2-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 [[TMP18]]
+; AVX2-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], [[TMP5]]
+; AVX2-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 [[TMP5]]
+; AVX2-NEXT:    [[TMP26:%.*]] = select i1 [[TMP19]], i32 [[TMP17]], i32 [[TMP18]]
 ; AVX2-NEXT:    ret i32 [[OP_EXTRA]]
 ;
 ; SKX-LABEL: @maxi8_wrong_parent(
 ; SKX-NEXT:    [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* bitcast ([32 x i32]* @arr to <2 x i32>*), align 16
 ; SKX-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
 ; SKX-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
 ; SKX-NEXT:    [[TMP5:%.*]] = icmp sgt i32 [[TMP3]], [[TMP4]]
 ; SKX-NEXT:    br label [[PP:%.*]]
 ; SKX:       pp:
 ; SKX-NEXT:    [[TMP6:%.*]] = load <4 x i32>, <4 x i32>* bitcast (i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 2) to <4 x i32>*), align 8
-; SKX-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
-; SKX-NEXT:    [[TMP8:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
-; SKX-NEXT:    [[TMP9:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
+; SKX-NEXT:    [[TMP7:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 6), align 8
+; SKX-NEXT:    [[TMP8:%.*]] = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 7), align 4
 ; SKX-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP:%.*]] = icmp sgt <4 x i32> [[TMP6]], [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP]], <4 x i32> [[TMP6]], <4 x i32> [[RDX_SHUF]]
 ; SKX-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
 ; SKX-NEXT:    [[RDX_MINMAX_CMP2:%.*]] = icmp sgt <4 x i32> [[RDX_MINMAX_SELECT]], [[RDX_SHUF1]]
 ; SKX-NEXT:    [[RDX_MINMAX_SELECT3:%.*]] = select <4 x i1> [[RDX_MINMAX_CMP2]], <4 x i32> [[RDX_MINMAX_SELECT]], <4 x i32> [[RDX_SHUF1]]
-; SKX-NEXT:    [[TMP10:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
-; SKX-NEXT:    [[TMP11:%.*]] = icmp sgt i32 [[TMP10]], [[TMP8]]
-; SKX-NEXT:    [[TMP12:%.*]] = select i1 [[TMP11]], i32 [[TMP10]], i32 [[TMP8]]
-; SKX-NEXT:    [[TMP13:%.*]] = icmp sgt i32 [[TMP12]], [[TMP9]]
-; SKX-NEXT:    [[TMP14:%.*]] = insertelement <2 x i1> undef, i1 [[TMP13]], i32 0
-; SKX-NEXT:    [[TMP15:%.*]] = insertelement <2 x i1> [[TMP14]], i1 [[TMP5]], i32 1
-; SKX-NEXT:    [[TMP16:%.*]] = insertelement <2 x i32> undef, i32 [[TMP12]], i32 0
-; SKX-NEXT:    [[TMP17:%.*]] = insertelement <2 x i32> [[TMP16]], i32 [[TMP3]], i32 1
-; SKX-NEXT:    [[TMP18:%.*]] = insertelement <2 x i32> undef, i32 [[TMP9]], i32 0
-; SKX-NEXT:    [[TMP19:%.*]] = insertelement <2 x i32> [[TMP18]], i32 [[TMP4]], i32 1
-; SKX-NEXT:    [[TMP20:%.*]] = select <2 x i1> [[TMP15]], <2 x i32> [[TMP17]], <2 x i32> [[TMP19]]
-; SKX-NEXT:    [[TMP21:%.*]] = extractelement <2 x i32> [[TMP20]], i32 1
-; SKX-NEXT:    [[TMP22:%.*]] = icmp sgt i32 [[TMP21]], [[TMP7]]
-; SKX-NEXT:    [[TMP23:%.*]] = select i1 [[TMP22]], i32 [[TMP21]], i32 undef
-; SKX-NEXT:    [[TMP24:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
-; SKX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP23]], [[TMP24]]
-; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP23]], i32 undef
-; SKX-NEXT:    [[TMP27:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
-; SKX-NEXT:    [[TMP28:%.*]] = icmp sgt i32 [[TMP26]], [[TMP27]]
-; SKX-NEXT:    [[TMP29:%.*]] = select i1 [[TMP28]], i32 [[TMP26]], i32 undef
-; SKX-NEXT:    [[TMP30:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
-; SKX-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP29]], [[TMP30]]
-; SKX-NEXT:    [[TMP32:%.*]] = select i1 [[TMP31]], i32 [[TMP29]], i32 undef
-; SKX-NEXT:    [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], [[TMP8]]
-; SKX-NEXT:    [[TMP34:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 [[TMP8]]
-; SKX-NEXT:    [[TMP35:%.*]] = icmp sgt i32 [[TMP34]], [[TMP9]]
-; SKX-NEXT:    [[TMP36:%.*]] = extractelement <2 x i32> [[TMP20]], i32 0
-; SKX-NEXT:    [[TMP37:%.*]] = icmp sgt i32 [[TMP36]], [[TMP21]]
-; SKX-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP37]], i32 [[TMP36]], i32 [[TMP21]]
-; SKX-NEXT:    [[TMP38:%.*]] = select i1 [[TMP35]], i32 [[TMP34]], i32 [[TMP9]]
+; SKX-NEXT:    [[TMP9:%.*]] = extractelement <4 x i32> [[RDX_MINMAX_SELECT3]], i32 0
+; SKX-NEXT:    [[TMP10:%.*]] = icmp sgt i32 [[TMP9]], [[TMP7]]
+; SKX-NEXT:    [[TMP11:%.*]] = select i1 [[TMP10]], i32 [[TMP9]], i32 [[TMP7]]
+; SKX-NEXT:    [[TMP12:%.*]] = icmp sgt i32 [[TMP11]], [[TMP8]]
+; SKX-NEXT:    [[TMP13:%.*]] = insertelement <2 x i1> undef, i1 [[TMP12]], i32 0
+; SKX-NEXT:    [[TMP14:%.*]] = insertelement <2 x i1> [[TMP13]], i1 [[TMP5]], i32 1
+; SKX-NEXT:    [[TMP15:%.*]] = insertelement <2 x i32> undef, i32 [[TMP11]], i32 0
+; SKX-NEXT:    [[TMP16:%.*]] = insertelement <2 x i32> [[TMP15]], i32 [[TMP3]], i32 1
+; SKX-NEXT:    [[TMP17:%.*]] = insertelement <2 x i32> undef, i32 [[TMP8]], i32 0
+; SKX-NEXT:    [[TMP18:%.*]] = insertelement <2 x i32> [[TMP17]], i32 [[TMP4]], i32 1
+; SKX-NEXT:    [[TMP19:%.*]] = select <2 x i1> [[TMP14]], <2 x i32> [[TMP16]], <2 x i32> [[TMP18]]
+; SKX-NEXT:    [[TMP20:%.*]] = extractelement <2 x i32> [[TMP19]], i32 1
+; SKX-NEXT:    [[TMP21:%.*]] = icmp sgt i32 [[TMP20]], undef
+; SKX-NEXT:    [[TMP22:%.*]] = select i1 [[TMP21]], i32 [[TMP20]], i32 undef
+; SKX-NEXT:    [[TMP23:%.*]] = icmp sgt i32 [[TMP22]], undef
+; SKX-NEXT:    [[TMP24:%.*]] = select i1 [[TMP23]], i32 [[TMP22]], i32 undef
+; SKX-NEXT:    [[TMP25:%.*]] = icmp sgt i32 [[TMP24]], undef
+; SKX-NEXT:    [[TMP26:%.*]] = select i1 [[TMP25]], i32 [[TMP24]], i32 undef
+; SKX-NEXT:    [[TMP27:%.*]] = icmp sgt i32 [[TMP26]], undef
+; SKX-NEXT:    [[TMP28:%.*]] = select i1 [[TMP27]], i32 [[TMP26]], i32 undef
+; SKX-NEXT:    [[TMP29:%.*]] = icmp sgt i32 [[TMP28]], [[TMP7]]
+; SKX-NEXT:    [[TMP30:%.*]] = select i1 [[TMP29]], i32 [[TMP28]], i32 [[TMP7]]
+; SKX-NEXT:    [[TMP31:%.*]] = icmp sgt i32 [[TMP30]], [[TMP8]]
+; SKX-NEXT:    [[TMP32:%.*]] = extractelement <2 x i32> [[TMP19]], i32 0
+; SKX-NEXT:    [[TMP33:%.*]] = icmp sgt i32 [[TMP32]], [[TMP20]]
+; SKX-NEXT:    [[OP_EXTRA:%.*]] = select i1 [[TMP33]], i32 [[TMP32]], i32 [[TMP20]]
+; SKX-NEXT:    [[TMP34:%.*]] = select i1 [[TMP31]], i32 [[TMP30]], i32 [[TMP8]]
 ; SKX-NEXT:    ret i32 [[OP_EXTRA]]
 ;
   %2 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 0), align 16
   %3 = load i32, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @arr, i64 0, i64 1), align 4
   %4 = icmp sgt i32 %2, %3
   br label %pp
 
 pp: