[VectorCombine] Improve shuffle select shuffle-of-shuffles

[dmgreen]

Summary

This in an extension to the code added in D123911 which added vector combine folding of shuffle-select patterns, attempting to reduce the total amount of shuffling required in patterns like:

%x = shuffle %i1, %i2
%y = shuffle %i1, %i2
%a = binop %x, %y
%b = binop %x, %y
shuffle %a, %b, selectmask

This patch extends the handing of shuffles that are dependent on one another, which can arise from the SLP vectorizer, as-in:

%x = shuffle %i1, %i2
%y = shuffle %x

The input shuffles can also be emitted, in which case they are treated like identity shuffles. This patch also attempts to calculate a better ordering of input shuffles, which can help getting lower cost input shuffles, pushing complex shuffles further down the tree.

[samtebbs]

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

1421–1429

Do you think it would be worth making an anonymous function for this similar behaviour?

1501–1523

This has the same structure as GetBaseMaskValue, so I wonder if GetMaskValue can use this. If it makes it messier then having both is OK.

llvm/test/Transforms/VectorCombine/AArch64/select-shuffle.ll

578

Looks like more instructions are added here. Is that fine because it gets lowered to something better or of lower cost?

[dmgreen]

Updates for review comments.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

1501–1523

Hmm. Good point, but I'm not sure how easy it is for them to share logic, unfortunately. They are each returning quite different values, and the parameters are different between the two methods.

llvm/test/Transforms/VectorCombine/AArch64/select-shuffle.ll

578

Yeah - More instructions is better in this case, as the shuffles are each simpler. It can be hard to see where this is an improvement and where it isn't from the tests. In general it can be a little difficult to be very precise, but most of the changes are improvements (and this one especially is much better as the shuffles are a lot easier to materialize).

[samtebbs]

LGTM now, cheers.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

1501–1523

Yeah looks like it would be messier to do that. Looks good to me as it is then.

[ebevhan]

Hi!

We get a memory corruption issue in our downstream testing with this patch. I've attached a reproducer, which can be run with opt -passes=vector-combine reduced.ll:

reduced.ll29 KBDownload

The issue occurs in X86TTIImpl::getShuffleCost:

unsigned E = *NumOfDests.getValue();
unsigned NormalizedVF =
    LegalVT.getVectorNumElements() * std::max(NumOfSrcs, E);
unsigned NumOfSrcRegs = NormalizedVF / LegalVT.getVectorNumElements();
unsigned NumOfDestRegs = NormalizedVF / LegalVT.getVectorNumElements();
SmallVector<int> NormalizedMask(NormalizedVF, UndefMaskElem);
copy(Mask, NormalizedMask.begin());

Mask has a size of 64, but NormalizedVF is only 16, so the copy tramples over the heap.

[dmgreen]

Thanks for the reproducer - I'll take a look now.

[dmgreen]

There is hopefully a fix in rG4b7913c35733. Let me know if anything else shows up.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

bool VectorCombine::foldSelectShuffle(Instruction &I, bool FromReduction) {
   auto *VT = dyn_cast<FixedVectorType>(I.getType());
   if (!SVI || !VT)
     return false;
   auto *Op0 = dyn_cast<Instruction>(SVI->getOperand(0));
   auto *Op1 = dyn_cast<Instruction>(SVI->getOperand(1));
   if (!Op0 || !Op1 || Op0 == Op1 || !Op0->isBinaryOp() || !Op1->isBinaryOp() ||
       VT != Op0->getType())
     return false;
-  auto *SVI0A = dyn_cast<ShuffleVectorInst>(Op0->getOperand(0));
-  auto *SVI0B = dyn_cast<ShuffleVectorInst>(Op0->getOperand(1));
-  auto *SVI1A = dyn_cast<ShuffleVectorInst>(Op1->getOperand(0));
-  auto *SVI1B = dyn_cast<ShuffleVectorInst>(Op1->getOperand(1));
+  auto *SVI0A = dyn_cast<Instruction>(Op0->getOperand(0));
+  auto *SVI0B = dyn_cast<Instruction>(Op0->getOperand(1));
+  auto *SVI1A = dyn_cast<Instruction>(Op1->getOperand(0));
+  auto *SVI1B = dyn_cast<Instruction>(Op1->getOperand(1));
+  SmallPtrSet<Instruction *, 4> InputShuffles({SVI0A, SVI0B, SVI1A, SVI1B});
   auto checkSVNonOpUses = [&](Instruction *I) {
     if (!I || I->getOperand(0)->getType() != VT)
       return true;
-    return any_of(I->users(), [&](User *U) { return U != Op0 && U != Op1; });
+    return any_of(I->users(), [&](User *U) {
+      return U != Op0 && U != Op1 &&
+             !(isa<ShuffleVectorInst>(U) &&
+               (InputShuffles.contains(cast<Instruction>(U)) ||
+                isInstructionTriviallyDead(cast<Instruction>(U))));
+    });
   };
   if (checkSVNonOpUses(SVI0A) || checkSVNonOpUses(SVI0B) ||
       checkSVNonOpUses(SVI1A) || checkSVNonOpUses(SVI1B))
     return false;
 
   // Collect all the uses that are shuffles that we can transform together. We
   // may not have a single shuffle, but a group that can all be transformed
   // together profitably.
bool VectorCombine::foldSelectShuffle(Instruction &I, bool FromReduction) {
   };
   if (!collectShuffles(Op0) || !collectShuffles(Op1))
     return false;
   // From a reduction, we need to be processing a single shuffle, otherwise the
   // other uses will not be lane-invariant.
   if (FromReduction && Shuffles.size() > 1)
     return false;
 
+  // Add any shuffle uses for the shuffles we have found, to include them in our
+  // cost calculations.
+  if (!FromReduction) {
+    for (ShuffleVectorInst *SV : Shuffles) {
+      for (auto U : SV->users()) {
+        ShuffleVectorInst *SSV = dyn_cast<ShuffleVectorInst>(U);
+        if (SSV && isa<UndefValue>(SSV->getOperand(1)))
+          Shuffles.push_back(SSV);
+      }
+    }
+  }
+
   // For each of the output shuffles, we try to sort all the first vector
   // elements to the beginning, followed by the second array elements at the
   // end. If the binops are legalized to smaller vectors, this may reduce total
   // number of binops. We compute the ReconstructMask mask needed to convert
   // back to the original lane order.
-  SmallVector<int> V1, V2;
-  SmallVector<SmallVector<int>> ReconstructMasks;
+  SmallVector<std::pair<int, int>> V1, V2;
+  SmallVector<SmallVector<int>> OrigReconstructMasks;
   int MaxV1Elt = 0, MaxV2Elt = 0;
   unsigned NumElts = VT->getNumElements();
   for (ShuffleVectorInst *SVN : Shuffles) {
     SmallVector<int> Mask;
     SVN->getShuffleMask(Mask);
 
     // Check the operands are the same as the original, or reversed (in which
     // case we need to commute the mask).
     Value *SVOp0 = SVN->getOperand(0);
     Value *SVOp1 = SVN->getOperand(1);
+    if (isa<UndefValue>(SVOp1)) {
+      auto *SSV = cast<ShuffleVectorInst>(SVOp0);
+      SVOp0 = SSV->getOperand(0);
+      SVOp1 = SSV->getOperand(1);
+      for (unsigned I = 0, E = Mask.size(); I != E; I++)
+        Mask[I] = Mask[I] < 0 ? Mask[I] : SSV->getMaskValue(Mask[I]);
+    }
     if (SVOp0 == Op1 && SVOp1 == Op0) {
       std::swap(SVOp0, SVOp1);
       ShuffleVectorInst::commuteShuffleMask(Mask, NumElts);
     }
     if (SVOp0 != Op0 || SVOp1 != Op1)
       return false;
 
     // Calculate the reconstruction mask for this shuffle, as the mask needed to
     // take the packed values from Op0/Op1 and reconstructing to the original
     // order.
     SmallVector<int> ReconstructMask;
     for (unsigned I = 0; I < Mask.size(); I++) {
       if (Mask[I] < 0) {
         ReconstructMask.push_back(-1);
       } else if (Mask[I] < static_cast<int>(NumElts)) {
         MaxV1Elt = std::max(MaxV1Elt, Mask[I]);
-        auto It = find(V1, Mask[I]);
+        auto It = find_if(V1, [&](const std::pair<int, int> &A) {
+          return Mask[I] == A.first;
+        });
         if (It != V1.end())
           ReconstructMask.push_back(It - V1.begin());
         else {
           ReconstructMask.push_back(V1.size());
-          V1.push_back(Mask[I]);
+          V1.emplace_back(Mask[I], V1.size());
         }
       } else {
         MaxV2Elt = std::max<int>(MaxV2Elt, Mask[I] - NumElts);
-        auto It = find(V2, Mask[I] - NumElts);
+        auto It = find_if(V2, [&](const std::pair<int, int> &A) {
+          return Mask[I] - static_cast<int>(NumElts) == A.first;
+        });
         if (It != V2.end())
           ReconstructMask.push_back(NumElts + It - V2.begin());
         else {
           ReconstructMask.push_back(NumElts + V2.size());
-          V2.push_back(Mask[I] - NumElts);
+          V2.emplace_back(Mask[I] - NumElts, NumElts + V2.size());
         }
       }
     }
 
     // For reductions, we know that the lane ordering out doesn't alter the
     // result. In-order can help simplify the shuffle away.
     if (FromReduction)
       sort(ReconstructMask);
-    ReconstructMasks.push_back(ReconstructMask);
+    OrigReconstructMasks.push_back(std::move(ReconstructMask));
   }
 
   // If the Maximum element used from V1 and V2 are not larger than the new
   // vectors, the vectors are already packes and performing the optimization
   // again will likely not help any further. This also prevents us from getting
   // stuck in a cycle in case the costs do not also rule it out.
   if (V1.empty() || V2.empty() ||
       (MaxV1Elt == static_cast<int>(V1.size()) - 1 &&
        MaxV2Elt == static_cast<int>(V2.size()) - 1))
     return false;
 
+  // GetBaseMaskValue takes one of the inputs, which may either be a shuffle, a
+  // shuffle of another shuffle, or not a shuffle (that is treated like a
+  // identity shuffle).
+  auto GetBaseMaskValue = [&](Instruction *I, int M) {
+    auto *SV = dyn_cast<ShuffleVectorInst>(I);
+    if (!SV)
+      return M;
+    if (isa<UndefValue>(SV->getOperand(1)))
+      if (auto *SSV = dyn_cast<ShuffleVectorInst>(SV->getOperand(0)))
+        if (InputShuffles.contains(SSV))
+          return SSV->getMaskValue(SV->getMaskValue(M));
+    return SV->getMaskValue(M);
+  };
+
+  // Attempt to sort the inputs my ascending mask values to make simpler input
+  // shuffles and push complex shuffles down to the uses. We sort on the first
+  // of the two input shuffle orders, to try and get at least one input into a
+  // nice order.
+  auto SortBase = [&](Instruction *A, std::pair<int, int> X,
+                      std::pair<int, int> Y) {
+    int MXA = GetBaseMaskValue(A, X.first);
+    int MYA = GetBaseMaskValue(A, Y.first);
+    return MXA < MYA;
+  };
+  stable_sort(V1, [&](std::pair<int, int> A, std::pair<int, int> B) {
+    return SortBase(SVI0A, A, B);
+  });
+  stable_sort(V2, [&](std::pair<int, int> A, std::pair<int, int> B) {
+    return SortBase(SVI1A, A, B);
+  });
+  // Calculate our ReconstructMasks from the OrigReconstructMasks and the
+  // modified order of the input shuffles.
+  SmallVector<SmallVector<int>> ReconstructMasks;
+  for (auto Mask : OrigReconstructMasks) {
+    SmallVector<int> ReconstructMask;
+    for (int M : Mask) {
+      auto FindIndex = [](const SmallVector<std::pair<int, int>> &V, int M) {
+        auto It = find_if(V, [M](auto A) { return A.second == M; });
+        assert(It != V.end() && "Expected all entries in Mask");
+        return std::distance(V.begin(), It);
+      };
+      if (M < 0)
+        ReconstructMask.push_back(-1);
+      else if (M < static_cast<int>(NumElts)) {
+        ReconstructMask.push_back(FindIndex(V1, M));
+      } else {
+        ReconstructMask.push_back(NumElts + FindIndex(V2, M));
+      }
+    }
+    ReconstructMasks.push_back(std::move(ReconstructMask));
+  }
+
   // Calculate the masks needed for the new input shuffles, which get padded
   // with undef
   SmallVector<int> V1A, V1B, V2A, V2B;
   for (unsigned I = 0; I < V1.size(); I++) {
-    V1A.push_back(SVI0A->getMaskValue(V1[I]));
-    V1B.push_back(SVI0B->getMaskValue(V1[I]));
+    V1A.push_back(GetBaseMaskValue(SVI0A, V1[I].first));
+    V1B.push_back(GetBaseMaskValue(SVI0B, V1[I].first));
   }
   for (unsigned I = 0; I < V2.size(); I++) {
-    V2A.push_back(SVI1A->getMaskValue(V2[I]));
-    V2B.push_back(SVI1B->getMaskValue(V2[I]));
+    V2A.push_back(GetBaseMaskValue(SVI1A, V2[I].first));
+    V2B.push_back(GetBaseMaskValue(SVI1B, V2[I].first));
   }
   while (V1A.size() < NumElts) {
     V1A.push_back(UndefMaskElem);
     V1B.push_back(UndefMaskElem);
   }
   while (V2A.size() < NumElts) {
     V2A.push_back(UndefMaskElem);
     V2B.push_back(UndefMaskElem);
   }
 
-  auto AddShuffleCost = [&](InstructionCost C, ShuffleVectorInst *SV) {
-    return C +
-           TTI.getShuffleCost(TTI::SK_PermuteTwoSrc, VT, SV->getShuffleMask());
+  auto AddShuffleCost = [&](InstructionCost C, Instruction *I) {
+    auto *SV = dyn_cast<ShuffleVectorInst>(I);
+    if (!SV)
+      return C;
+    return C + TTI.getShuffleCost(isa<UndefValue>(SV->getOperand(1))
+                                      ? TTI::SK_PermuteSingleSrc
+                                      : TTI::SK_PermuteTwoSrc,
+                                  VT, SV->getShuffleMask());
   };
   auto AddShuffleMaskCost = [&](InstructionCost C, ArrayRef<int> Mask) {
     return C + TTI.getShuffleCost(TTI::SK_PermuteTwoSrc, VT, Mask);
   };
 
   // Get the costs of the shuffles + binops before and after with the new
   // shuffle masks.
   InstructionCost CostBefore =
       TTI.getArithmeticInstrCost(Op0->getOpcode(), VT) +
       TTI.getArithmeticInstrCost(Op1->getOpcode(), VT);
   CostBefore += std::accumulate(Shuffles.begin(), Shuffles.end(),
                                 InstructionCost(0), AddShuffleCost);
-  // This set helps us only cost each unique shuffle once.
-  SmallPtrSet<ShuffleVectorInst *, 4> InputShuffles(
-      {SVI0A, SVI0B, SVI1A, SVI1B});
   CostBefore += std::accumulate(InputShuffles.begin(), InputShuffles.end(),
                                 InstructionCost(0), AddShuffleCost);
 
   // The new binops will be unused for lanes past the used shuffle lengths.
   // These types attempt to get the correct cost for that from the target.
   FixedVectorType *Op0SmallVT =
       FixedVectorType::get(VT->getScalarType(), V1.size());
   FixedVectorType *Op1SmallVT =
       FixedVectorType::get(VT->getScalarType(), V2.size());
   InstructionCost CostAfter =
       TTI.getArithmeticInstrCost(Op0->getOpcode(), Op0SmallVT) +
       TTI.getArithmeticInstrCost(Op1->getOpcode(), Op1SmallVT);
   CostAfter += std::accumulate(ReconstructMasks.begin(), ReconstructMasks.end(),
                                InstructionCost(0), AddShuffleMaskCost);
   std::set<SmallVector<int>> OutputShuffleMasks({V1A, V1B, V2A, V2B});
   CostAfter +=
       std::accumulate(OutputShuffleMasks.begin(), OutputShuffleMasks.end(),
                       InstructionCost(0), AddShuffleMaskCost);
 
+  LLVM_DEBUG(dbgs() << "Found a binop select shuffle pattern: " << I << "\n");
+  LLVM_DEBUG(dbgs() << "  CostBefore: " << CostBefore
+                    << " vs CostAfter: " << CostAfter << "\n");
   if (CostBefore <= CostAfter)
     return false;
 
   // The cost model has passed, create the new instructions.
-  Builder.SetInsertPoint(SVI0A);
-  Value *NSV0A = Builder.CreateShuffleVector(SVI0A->getOperand(0),
-                                             SVI0A->getOperand(1), V1A);
-  Builder.SetInsertPoint(SVI0B);
-  Value *NSV0B = Builder.CreateShuffleVector(SVI0B->getOperand(0),
-                                             SVI0B->getOperand(1), V1B);
-  Builder.SetInsertPoint(SVI1A);
-  Value *NSV1A = Builder.CreateShuffleVector(SVI1A->getOperand(0),
-                                             SVI1A->getOperand(1), V2A);
-  Builder.SetInsertPoint(SVI1B);
-  Value *NSV1B = Builder.CreateShuffleVector(SVI1B->getOperand(0),
-                                             SVI1B->getOperand(1), V2B);
+  auto GetShuffleOperand = [&](Instruction *I, unsigned Op) -> Value * {
+    auto *SV = dyn_cast<ShuffleVectorInst>(I);
+    if (!SV)
+      return I;
+    if (isa<UndefValue>(SV->getOperand(1)))
+      if (auto *SSV = dyn_cast<ShuffleVectorInst>(SV->getOperand(0)))
+        if (InputShuffles.contains(SSV))
+          return SSV->getOperand(Op);
+    return SV->getOperand(Op);
+  };
+  Builder.SetInsertPoint(SVI0A->getNextNode());
+  Value *NSV0A = Builder.CreateShuffleVector(GetShuffleOperand(SVI0A, 0),
+                                             GetShuffleOperand(SVI0A, 1), V1A);
+  Builder.SetInsertPoint(SVI0B->getNextNode());
+  Value *NSV0B = Builder.CreateShuffleVector(GetShuffleOperand(SVI0B, 0),
+                                             GetShuffleOperand(SVI0B, 1), V1B);
+  Builder.SetInsertPoint(SVI1A->getNextNode());
+  Value *NSV1A = Builder.CreateShuffleVector(GetShuffleOperand(SVI1A, 0),
+                                             GetShuffleOperand(SVI1A, 1), V2A);
+  Builder.SetInsertPoint(SVI1B->getNextNode());
+  Value *NSV1B = Builder.CreateShuffleVector(GetShuffleOperand(SVI1B, 0),
+                                             GetShuffleOperand(SVI1B, 1), V2B);
   Builder.SetInsertPoint(Op0);
   Value *NOp0 = Builder.CreateBinOp((Instruction::BinaryOps)Op0->getOpcode(),
                                     NSV0A, NSV0B);
   if (auto *I = dyn_cast<Instruction>(NOp0))
     I->copyIRFlags(Op0, true);
   Builder.SetInsertPoint(Op1);
   Value *NOp1 = Builder.CreateBinOp((Instruction::BinaryOps)Op1->getOpcode(),
                                     NSV1A, NSV1B);
   if (auto *I = dyn_cast<Instruction>(NOp1))

llvm/test/Transforms/VectorCombine/AArch64/select-shuffle.ll

;
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 21, i32 22, i32 7, i32 24, i32 25, i32 10, i32 27, i32 28, i32 13, i32 30, i32 31>
   ret <16 x i32> %s3
 }
 
 define i32 @test1_reduce(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @test1_reduce(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 23, i32 26, i32 29, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 21, i32 6, i32 24, i32 9, i32 27, i32 12, i32 30, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 8, i32 11, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 4, i32 19, i32 7, i32 18, i32 10, i32 17, i32 13, i32 16, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP1]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 6, i32 9, i32 12, i32 15, i32 21, i32 24, i32 27, i32 30, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 23, i32 26, i32 29, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 19, i32 18, i32 17, i32 16, i32 4, i32 7, i32 10, i32 13, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 8, i32 11, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
 ; CHECK-NEXT:    [[R:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[S3]])
 ; CHECK-NEXT:    ret i32 [[R]]
 ;
   %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
   %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
;
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
   ret <16 x i32> %s3
 }
 
 define <16 x i32> @test2_2(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @test2_2(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 21, i32 20, i32 24, i32 23, i32 11, i32 10, i32 14, i32 13, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 15, i32 12, i32 25, i32 22, i32 19, i32 18, i32 1, i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 6, i32 3, i32 9, i32 2, i32 28, i32 17, i32 31, i32 16, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 30, i32 27, i32 8, i32 5, i32 4, i32 7, i32 26, i32 29, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP1]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP2]], [[TMP4]]
-; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 0, i32 1, i32 20, i32 2, i32 3, i32 21, i32 4, i32 5, i32 22, i32 6, i32 7, i32 23>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 12, i32 15, i32 18, i32 19, i32 22, i32 25, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 10, i32 11, i32 13, i32 14, i32 20, i32 21, i32 23, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 27, i32 30, i32 7, i32 4, i32 5, i32 8, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 17, i32 28, i32 16, i32 31, i32 3, i32 6, i32 2, i32 9, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 19, i32 18, i32 23, i32 22, i32 5, i32 4, i32 21, i32 7, i32 6, i32 20, i32 1, i32 0, i32 17, i32 3, i32 2, i32 16>
 ; CHECK-NEXT:    ret <16 x i32> [[S3]]
 ;
   %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
   ret <16 x i32> %s3
;
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 6, i32 23, i32 24, i32 9, i32 26, i32 27, i32 12, i32 29, i32 30, i32 15>
   %s4 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 31, i32 28, i32 25, i32 22, i32 5, i32 4, i32 19, i32 8, i32 7, i32 18, i32 11, i32 10, i32 17, i32 14, i32 13, i32 16>
   %c = mul <16 x i32> %s3, %s4
   ret <16 x i32> %c
 }
 
 define <16 x i32> @test3_1(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @test3_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 18, i32 3, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP1]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP2]], [[TMP4]]
-; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
+; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
+; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
+; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
+; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:    ret <16 x i32> [[S3]]
 ;
   %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
   %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   ret <16 x i32> %s3
 }
 
 define <16 x i32> @test3_2(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @test3_2(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 6, i32 23, i32 19, i32 31, i32 27, i32 2, i32 14, i32 10, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 4, i32 21, i32 17, i32 29, i32 25, i32 0, i32 12, i32 8, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 20, i32 5, i32 1, i32 13, i32 9, i32 16, i32 28, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 22, i32 7, i32 3, i32 15, i32 11, i32 18, i32 30, i32 26, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP1]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP2]], [[TMP4]]
-; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 16, i32 17, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 17, i32 21, i32 25, i32 29, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 19, i32 23, i32 27, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 18, i32 22, i32 26, i32 30, i32 3, i32 7, i32 11, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 16, i32 20, i32 24, i32 28, i32 1, i32 5, i32 9, i32 13, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 17, i32 21, i32 1, i32 5, i32 4, i32 7, i32 6, i32 0, i32 3, i32 2, i32 20, i32 23, i32 22, i32 16, i32 19, i32 18>
 ; CHECK-NEXT:    ret <16 x i32> [[S3]]
 ;
   %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
   %s2 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
   ret <16 x i32> %s3
;
   %s4 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
   %c = add <16 x i32> %s3, %s4
   ret <16 x i32> %c
 }
 
 
 define <16 x i32> @test23(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @test23(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 18, i32 0, i32 20, i32 24, i32 12, i32 22, i32 10, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 19, i32 1, i32 21, i32 25, i32 13, i32 23, i32 11, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 7, i32 29, i32 3, i32 9, i32 27, i32 5, i32 17, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 4, i32 26, i32 6, i32 8, i32 16, i32 2, i32 28, i32 30, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP1]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP2]], [[TMP4]]
-; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 1, i32 17, i32 0, i32 16, i32 21, i32 22, i32 23, i32 5, i32 6, i32 7, i32 18, i32 19, i32 20, i32 2, i32 3, i32 4>
-; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 18, i32 19, i32 20, i32 2, i32 3, i32 4, i32 21, i32 22, i32 23, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[S10:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[S20:%.*]] = shufflevector <16 x i32> [[Y]], <16 x i32> [[X]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
+; CHECK-NEXT:    [[A0:%.*]] = add nsw <16 x i32> [[S10]], [[S20]]
+; CHECK-NEXT:    [[B0:%.*]] = sub nsw <16 x i32> [[S10]], [[S20]]
+; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[A0]], <16 x i32> [[B0]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
+; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[B0]], <16 x i32> [[A0]], <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
 ; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
 ; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
 ; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
 ; CHECK-NEXT:    [[C:%.*]] = add <16 x i32> [[S3]], [[S4]]
 ; CHECK-NEXT:    ret <16 x i32> [[C]]
 ;
   %s10 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %s20 = shufflevector <16 x i32> %y, <16 x i32> %x, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
   %a0 = add nsw <16 x i32> %s10, %s20
   %b0 = sub nsw <16 x i32> %s10, %s20
   %s1 = shufflevector <16 x i32> %a0, <16 x i32> %b0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 23, i32 27, i32 31, i32 6, i32 10, i32 14, i32 21, i32 25, i32 29, i32 4, i32 8, i32 12>
   %s2 = shufflevector <16 x i32> %b0, <16 x i32> %a0, <16 x i32> <i32 18, i32 3, i32 16, i32 1, i32 5, i32 9, i32 13, i32 20, i32 24, i32 28, i32 7, i32 11, i32 15, i32 22, i32 26, i32 30>
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 1, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   %s4 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 29, i32 26, i32 7, i32 4, i32 3, i32 6, i32 5, i32 2, i32 9, i32 8, i32 17, i32 28, i32 27, i32 16, i32 31, i32 30>
   %c = add <16 x i32> %s3, %s4
   ret <16 x i32> %c
 }
 
 define <16 x i32> @testgood(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @testgood(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP1]], [[TMP3]]
-; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
 ; CHECK-NEXT:    ret <16 x i32> [[S3]]
 ;
   %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %a = add nsw <16 x i32> %s1, %s2
   %b = sub nsw <16 x i32> %s1, %s2
   %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   ret <16 x i32> %s3
 }
 
+define <16 x i32> @test_shufshufin(<16 x i32> %x, <16 x i32> %y) {
+; CHECK-LABEL: @test_shufshufin(
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
+; CHECK-NEXT:    ret <16 x i32> [[S3]]
+;
+  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %s2 = shufflevector <16 x i32> %s1, <16 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+  %a = add nsw <16 x i32> %s1, %s2
+  %b = sub nsw <16 x i32> %s1, %s2
+  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x i32> %s3
+}
+
+define <16 x i32> @testshufshufout(<16 x i32> %x, <16 x i32> %y) {
+; CHECK-LABEL: @testshufshufout(
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <16 x i32> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <16 x i32> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
+; CHECK-NEXT:    [[S4:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 16, i32 0, i32 17, i32 1, i32 18, i32 2, i32 19, i32 3, i32 20, i32 4, i32 21, i32 5, i32 22, i32 6, i32 23, i32 7>
+; CHECK-NEXT:    [[R:%.*]] = add nsw <16 x i32> [[S3]], [[S4]]
+; CHECK-NEXT:    ret <16 x i32> [[R]]
+;
+  %s1 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %s2 = shufflevector <16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %a = add nsw <16 x i32> %s1, %s2
+  %b = sub nsw <16 x i32> %s1, %s2
+  %s3 = shufflevector <16 x i32> %a, <16 x i32> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %s4 = shufflevector <16 x i32> %s3, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
+  %r = add nsw <16 x i32> %s3, %s4
+  ret <16 x i32> %r
+}
+
 declare void @use(<16 x i32>)
 define <16 x i32> @test_extrashuffleuse(<16 x i32> %x, <16 x i32> %y) {
 ; CHECK-LABEL: @test_extrashuffleuse(
 ; CHECK-NEXT:    [[S1:%.*]] = shufflevector <16 x i32> [[X:%.*]], <16 x i32> [[Y:%.*]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
 ; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[X]], <16 x i32> [[Y]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
 ; CHECK-NEXT:    [[A:%.*]] = add nsw <16 x i32> [[S1]], [[S2]]
 ; CHECK-NEXT:    [[B:%.*]] = sub nsw <16 x i32> [[S1]], [[S2]]
 ; CHECK-NEXT:    [[S3:%.*]] = shufflevector <16 x i32> [[A]], <16 x i32> [[B]], <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
;
   %t1 = shufflevector <4 x i32> %add, <4 x i32> %add, <4 x i32> <i32 6, i32 0, i32 1, i32 7>
   %h1 = getelementptr <4 x i32>, ptr %dst, i32 1
   store <4 x i32> %t1, ptr %h1, align 16
   ret void
 }
 
 define <16 x i32> @test_1651256324(<16 x i32> %l0, <16 x i32> %l1, <16 x i32> %l6, <16 x i32> %l7) {
 ; CHECK-LABEL: @test_1651256324(
-; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[L0:%.*]], <16 x i32> [[L6:%.*]], <16 x i32> <i32 30, i32 10, i32 1, i32 20, i32 10, i32 0, i32 20, i32 3, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[L1:%.*]], <16 x i32> [[L1]], <16 x i32> <i32 24, i32 1, i32 10, i32 0, i32 5, i32 7, i32 11, i32 11, i32 4, i32 12, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[L1]], <16 x i32> [[L1]], <16 x i32> <i32 24, i32 1, i32 10, i32 0, i32 5, i32 7, i32 11, i32 11, i32 4, i32 12, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i32> [[L0:%.*]], <16 x i32> [[L6:%.*]], <16 x i32> <i32 0, i32 1, i32 3, i32 10, i32 10, i32 15, i32 20, i32 20, i32 30, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[L1:%.*]], <16 x i32> [[L1]], <16 x i32> <i32 0, i32 1, i32 4, i32 5, i32 7, i32 10, i32 11, i32 11, i32 12, i32 22, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i32> [[L1]], <16 x i32> [[L1]], <16 x i32> <i32 0, i32 1, i32 4, i32 5, i32 7, i32 10, i32 11, i32 11, i32 12, i32 22, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[S2:%.*]] = shufflevector <16 x i32> [[L7:%.*]], <16 x i32> [[L7]], <16 x i32> <i32 23, i32 20, i32 29, i32 25, i32 14, i32 21, i32 11, i32 9, i32 2, i32 7, i32 5, i32 15, i32 24, i32 30, i32 26, i32 5>
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[L6]], <16 x i32> [[L7]], <16 x i32> <i32 4, i32 24, i32 13, i32 15, i32 0, i32 29, i32 8, i32 30, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP5:%.*]] = add <16 x i32> [[TMP2]], [[TMP3]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i32> [[L6]], <16 x i32> [[L7]], <16 x i32> <i32 29, i32 13, i32 30, i32 24, i32 0, i32 15, i32 15, i32 8, i32 4, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP5:%.*]] = add <16 x i32> [[TMP3]], [[TMP2]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = sub <16 x i32> [[TMP1]], [[TMP4]]
-; CHECK-NEXT:    [[T0:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 18, i32 7, i32 22, i32 1, i32 8, i32 1, i32 9, i32 0, i32 6, i32 6, i32 23, i32 16, i32 18, i32 10, i32 24, i32 21>
-; CHECK-NEXT:    [[T1:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 0, i32 1, i32 2, i32 16, i32 17, i32 3, i32 4, i32 3, i32 18, i32 19, i32 3, i32 5, i32 1, i32 20, i32 21, i32 6>
+; CHECK-NEXT:    [[T0:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 17, i32 7, i32 23, i32 1, i32 2, i32 1, i32 8, i32 10, i32 6, i32 6, i32 18, i32 24, i32 17, i32 9, i32 21, i32 16>
+; CHECK-NEXT:    [[T1:%.*]] = shufflevector <16 x i32> [[TMP5]], <16 x i32> [[TMP6]], <16 x i32> <i32 10, i32 1, i32 5, i32 24, i32 19, i32 0, i32 3, i32 0, i32 17, i32 22, i32 0, i32 4, i32 1, i32 20, i32 16, i32 6>
 ; CHECK-NEXT:    [[R:%.*]] = xor <16 x i32> [[T0]], [[T1]]
 ; CHECK-NEXT:    ret <16 x i32> [[R]]
 ;
   %s0 = shufflevector <16 x i32> %l0, <16 x i32> %l6, <16 x i32> <i32 1, i32 20, i32 15, i32 3, i32 1, i32 10, i32 17, i32 25, i32 29, i32 23, i32 20, i32 10, i32 0, i32 20, i32 30, i32 30>
   %s1 = shufflevector <16 x i32> %l1, <16 x i32> %l1, <16 x i32> <i32 11, i32 22, i32 1, i32 7, i32 20, i32 0, i32 2, i32 24, i32 28, i32 10, i32 31, i32 12, i32 22, i32 5, i32 11, i32 4>
   %s2 = shufflevector <16 x i32> %l7, <16 x i32> %l7, <16 x i32> <i32 23, i32 20, i32 29, i32 25, i32 14, i32 21, i32 11, i32 9, i32 2, i32 7, i32 5, i32 15, i32 24, i32 30, i32 26, i32 5>
   %s3 = shufflevector <16 x i32> %l6, <16 x i32> %l7, <16 x i32> <i32 29, i32 19, i32 15, i32 30, i32 13, i32 0, i32 30, i32 23, i32 26, i32 3, i32 15, i32 24, i32 29, i32 8, i32 4, i32 0>
   %add = add <16 x i32> %s1, %s1

 ; CHECK-NEXT:    [[TMP52:%.*]] = shufflevector <4 x i8> [[TMP15]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, 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>
 ; CHECK-NEXT:    [[TMP53:%.*]] = shufflevector <16 x i8> [[TMP51]], <16 x i8> [[TMP52]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 undef, i32 undef, i32 undef, i32 undef>
 ; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, 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>
 ; CHECK-NEXT:    [[TMP55:%.*]] = shufflevector <16 x i8> [[TMP53]], <16 x i8> [[TMP54]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
 ; CHECK-NEXT:    [[TMP56:%.*]] = zext <16 x i8> [[TMP55]] to <16 x i32>
 ; CHECK-NEXT:    [[TMP57:%.*]] = sub nsw <16 x i32> [[TMP48]], [[TMP56]]
 ; CHECK-NEXT:    [[TMP58:%.*]] = shl nsw <16 x i32> [[TMP57]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
 ; CHECK-NEXT:    [[TMP59:%.*]] = add nsw <16 x i32> [[TMP58]], [[TMP40]]
-; CHECK-NEXT:    [[REORDER:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
-; CHECK-NEXT:    [[TMP60:%.*]] = add nsw <16 x i32> [[TMP59]], [[REORDER]]
-; CHECK-NEXT:    [[TMP61:%.*]] = sub nsw <16 x i32> [[TMP59]], [[REORDER]]
-; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 7, i32 3, i32 15, i32 11, i32 18, i32 22, i32 30, i32 26, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 1, i32 5, i32 13, i32 9, i32 16, i32 20, i32 28, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP64:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 5, i32 1, i32 13, i32 9, i32 16, i32 20, i32 28, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP65:%.*]] = shufflevector <16 x i32> [[TMP60]], <16 x i32> [[TMP61]], <16 x i32> <i32 3, i32 7, i32 15, i32 11, i32 18, i32 22, i32 30, i32 26, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP66:%.*]] = add nsw <16 x i32> [[TMP62]], [[TMP64]]
-; CHECK-NEXT:    [[TMP67:%.*]] = sub nsw <16 x i32> [[TMP63]], [[TMP65]]
-; CHECK-NEXT:    [[TMP68:%.*]] = shufflevector <16 x i32> [[TMP66]], <16 x i32> [[TMP67]], <16 x i32> <i32 1, i32 4, i32 16, i32 20, i32 3, i32 7, i32 19, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP69:%.*]] = shufflevector <16 x i32> [[TMP66]], <16 x i32> [[TMP67]], <16 x i32> <i32 0, i32 5, i32 17, i32 21, i32 2, i32 6, i32 18, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <16 x i32> [[TMP66]], <16 x i32> [[TMP67]], <16 x i32> <i32 0, i32 5, i32 17, i32 21, i32 2, i32 6, i32 18, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP71:%.*]] = shufflevector <16 x i32> [[TMP66]], <16 x i32> [[TMP67]], <16 x i32> <i32 1, i32 4, i32 16, i32 20, i32 3, i32 7, i32 19, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP72:%.*]] = add nsw <16 x i32> [[TMP68]], [[TMP70]]
-; CHECK-NEXT:    [[TMP73:%.*]] = sub nsw <16 x i32> [[TMP69]], [[TMP71]]
-; CHECK-NEXT:    [[TMP74:%.*]] = shufflevector <16 x i32> [[TMP72]], <16 x i32> [[TMP73]], <16 x i32> <i32 0, i32 16, i32 17, i32 1, i32 18, i32 2, i32 19, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP75:%.*]] = shufflevector <16 x i32> [[TMP72]], <16 x i32> [[TMP73]], <16 x i32> <i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP76:%.*]] = shufflevector <16 x i32> [[TMP72]], <16 x i32> [[TMP73]], <16 x i32> <i32 4, i32 20, i32 21, i32 5, i32 22, i32 6, i32 23, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP77:%.*]] = shufflevector <16 x i32> [[TMP72]], <16 x i32> [[TMP73]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP78:%.*]] = add nsw <16 x i32> [[TMP74]], [[TMP76]]
-; CHECK-NEXT:    [[TMP79:%.*]] = sub nsw <16 x i32> [[TMP75]], [[TMP77]]
-; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <16 x i32> [[TMP78]], <16 x i32> [[TMP79]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
-; CHECK-NEXT:    [[TMP81:%.*]] = lshr <16 x i32> [[TMP80]], <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
-; CHECK-NEXT:    [[TMP82:%.*]] = and <16 x i32> [[TMP81]], <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
-; CHECK-NEXT:    [[TMP83:%.*]] = mul nuw <16 x i32> [[TMP82]], <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-; CHECK-NEXT:    [[TMP84:%.*]] = add <16 x i32> [[TMP83]], [[TMP80]]
-; CHECK-NEXT:    [[TMP85:%.*]] = xor <16 x i32> [[TMP84]], [[TMP83]]
-; CHECK-NEXT:    [[TMP86:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP85]])
-; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP86]], 65535
-; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP86]], 16
+; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> [[TMP59]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> [[TMP59]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> poison, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP59]], <16 x i32> poison, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP64:%.*]] = add nsw <16 x i32> [[TMP61]], [[TMP63]]
+; CHECK-NEXT:    [[TMP65:%.*]] = sub nsw <16 x i32> [[TMP60]], [[TMP62]]
+; CHECK-NEXT:    [[TMP66:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP67:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP68:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP69:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP70:%.*]] = add nsw <16 x i32> [[TMP67]], [[TMP69]]
+; CHECK-NEXT:    [[TMP71:%.*]] = sub nsw <16 x i32> [[TMP66]], [[TMP68]]
+; CHECK-NEXT:    [[TMP72:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP73:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP74:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP75:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP76:%.*]] = add nsw <16 x i32> [[TMP73]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = sub nsw <16 x i32> [[TMP72]], [[TMP74]]
+; CHECK-NEXT:    [[TMP78:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP79:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP81:%.*]] = shufflevector <16 x i32> [[TMP76]], <16 x i32> [[TMP77]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP82:%.*]] = add nsw <16 x i32> [[TMP79]], [[TMP81]]
+; CHECK-NEXT:    [[TMP83:%.*]] = sub nsw <16 x i32> [[TMP78]], [[TMP80]]
+; CHECK-NEXT:    [[TMP84:%.*]] = shufflevector <16 x i32> [[TMP82]], <16 x i32> [[TMP83]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    [[TMP85:%.*]] = lshr <16 x i32> [[TMP84]], <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
+; CHECK-NEXT:    [[TMP86:%.*]] = and <16 x i32> [[TMP85]], <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
+; CHECK-NEXT:    [[TMP87:%.*]] = mul nuw <16 x i32> [[TMP86]], <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
+; CHECK-NEXT:    [[TMP88:%.*]] = add <16 x i32> [[TMP87]], [[TMP84]]
+; CHECK-NEXT:    [[TMP89:%.*]] = xor <16 x i32> [[TMP88]], [[TMP87]]
+; CHECK-NEXT:    [[TMP90:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP89]])
+; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP90]], 65535
+; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP90]], 16
 ; CHECK-NEXT:    [[ADD119:%.*]] = add nuw nsw i32 [[CONV118]], [[SHR]]
 ; CHECK-NEXT:    [[SHR120:%.*]] = lshr i32 [[ADD119]], 1
 ; CHECK-NEXT:    ret i32 [[SHR120]]
 ;
 entry:
   %idx.ext = sext i32 %st1 to i64
   %idx.ext63 = sext i32 %st2 to i64
   %arrayidx3 = getelementptr inbounds i8, i8* %p1, i64 4

 ; CHECK-NEXT:    [[TMP44:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, 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>
 ; CHECK-NEXT:    [[TMP45:%.*]] = shufflevector <16 x i8> [[TMP43]], <16 x i8> [[TMP44]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, 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>
 ; CHECK-NEXT:    [[TMP47:%.*]] = shufflevector <16 x i8> [[TMP45]], <16 x i8> [[TMP46]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
 ; CHECK-NEXT:    [[TMP48:%.*]] = zext <16 x i8> [[TMP47]] to <16 x i32>
 ; CHECK-NEXT:    [[TMP49:%.*]] = sub nsw <16 x i32> [[TMP39]], [[TMP48]]
 ; CHECK-NEXT:    [[TMP50:%.*]] = shl nsw <16 x i32> [[TMP49]], <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
 ; CHECK-NEXT:    [[TMP51:%.*]] = add nsw <16 x i32> [[TMP50]], [[TMP30]]
-; CHECK-NEXT:    [[REORDER:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
-; CHECK-NEXT:    [[TMP52:%.*]] = add nsw <16 x i32> [[TMP51]], [[REORDER]]
-; CHECK-NEXT:    [[TMP53:%.*]] = sub nsw <16 x i32> [[TMP51]], [[REORDER]]
-; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <16 x i32> [[TMP52]], <16 x i32> [[TMP53]], <16 x i32> <i32 7, i32 3, i32 15, i32 11, i32 18, i32 22, i32 30, i32 26, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP55:%.*]] = shufflevector <16 x i32> [[TMP52]], <16 x i32> [[TMP53]], <16 x i32> <i32 1, i32 5, i32 13, i32 9, i32 16, i32 20, i32 28, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP56:%.*]] = shufflevector <16 x i32> [[TMP52]], <16 x i32> [[TMP53]], <16 x i32> <i32 5, i32 1, i32 13, i32 9, i32 16, i32 20, i32 28, i32 24, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP57:%.*]] = shufflevector <16 x i32> [[TMP52]], <16 x i32> [[TMP53]], <16 x i32> <i32 3, i32 7, i32 15, i32 11, i32 18, i32 22, i32 30, i32 26, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP58:%.*]] = add nsw <16 x i32> [[TMP54]], [[TMP56]]
-; CHECK-NEXT:    [[TMP59:%.*]] = sub nsw <16 x i32> [[TMP55]], [[TMP57]]
-; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <16 x i32> [[TMP58]], <16 x i32> [[TMP59]], <16 x i32> <i32 1, i32 4, i32 16, i32 20, i32 3, i32 7, i32 19, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <16 x i32> [[TMP58]], <16 x i32> [[TMP59]], <16 x i32> <i32 0, i32 5, i32 17, i32 21, i32 2, i32 6, i32 18, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <16 x i32> [[TMP58]], <16 x i32> [[TMP59]], <16 x i32> <i32 0, i32 5, i32 17, i32 21, i32 2, i32 6, i32 18, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP58]], <16 x i32> [[TMP59]], <16 x i32> <i32 1, i32 4, i32 16, i32 20, i32 3, i32 7, i32 19, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP64:%.*]] = add nsw <16 x i32> [[TMP60]], [[TMP62]]
-; CHECK-NEXT:    [[TMP65:%.*]] = sub nsw <16 x i32> [[TMP61]], [[TMP63]]
-; CHECK-NEXT:    [[TMP66:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 0, i32 16, i32 17, i32 1, i32 18, i32 2, i32 19, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP67:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP68:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 4, i32 20, i32 21, i32 5, i32 22, i32 6, i32 23, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP69:%.*]] = shufflevector <16 x i32> [[TMP64]], <16 x i32> [[TMP65]], <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; CHECK-NEXT:    [[TMP70:%.*]] = add nsw <16 x i32> [[TMP66]], [[TMP68]]
-; CHECK-NEXT:    [[TMP71:%.*]] = sub nsw <16 x i32> [[TMP67]], [[TMP69]]
-; CHECK-NEXT:    [[TMP72:%.*]] = shufflevector <16 x i32> [[TMP70]], <16 x i32> [[TMP71]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
-; CHECK-NEXT:    [[TMP73:%.*]] = lshr <16 x i32> [[TMP72]], <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
-; CHECK-NEXT:    [[TMP74:%.*]] = and <16 x i32> [[TMP73]], <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
-; CHECK-NEXT:    [[TMP75:%.*]] = mul nuw <16 x i32> [[TMP74]], <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
-; CHECK-NEXT:    [[TMP76:%.*]] = add <16 x i32> [[TMP75]], [[TMP72]]
-; CHECK-NEXT:    [[TMP77:%.*]] = xor <16 x i32> [[TMP76]], [[TMP75]]
-; CHECK-NEXT:    [[TMP78:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP77]])
-; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP78]], 65535
-; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP78]], 16
+; CHECK-NEXT:    [[TMP52:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> [[TMP51]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP53:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> [[TMP51]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> poison, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP55:%.*]] = shufflevector <16 x i32> [[TMP51]], <16 x i32> poison, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP56:%.*]] = add nsw <16 x i32> [[TMP53]], [[TMP55]]
+; CHECK-NEXT:    [[TMP57:%.*]] = sub nsw <16 x i32> [[TMP52]], [[TMP54]]
+; CHECK-NEXT:    [[TMP58:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP59:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <16 x i32> [[TMP56]], <16 x i32> [[TMP57]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP62:%.*]] = add nsw <16 x i32> [[TMP59]], [[TMP61]]
+; CHECK-NEXT:    [[TMP63:%.*]] = sub nsw <16 x i32> [[TMP58]], [[TMP60]]
+; CHECK-NEXT:    [[TMP64:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP65:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP66:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP67:%.*]] = shufflevector <16 x i32> [[TMP62]], <16 x i32> [[TMP63]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP68:%.*]] = add nsw <16 x i32> [[TMP65]], [[TMP67]]
+; CHECK-NEXT:    [[TMP69:%.*]] = sub nsw <16 x i32> [[TMP64]], [[TMP66]]
+; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP71:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP72:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP73:%.*]] = shufflevector <16 x i32> [[TMP68]], <16 x i32> [[TMP69]], <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP74:%.*]] = add nsw <16 x i32> [[TMP71]], [[TMP73]]
+; CHECK-NEXT:    [[TMP75:%.*]] = sub nsw <16 x i32> [[TMP70]], [[TMP72]]
+; CHECK-NEXT:    [[TMP76:%.*]] = shufflevector <16 x i32> [[TMP74]], <16 x i32> [[TMP75]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    [[TMP77:%.*]] = lshr <16 x i32> [[TMP76]], <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
+; CHECK-NEXT:    [[TMP78:%.*]] = and <16 x i32> [[TMP77]], <i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537, i32 65537>
+; CHECK-NEXT:    [[TMP79:%.*]] = mul nuw <16 x i32> [[TMP78]], <i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535, i32 65535>
+; CHECK-NEXT:    [[TMP80:%.*]] = add <16 x i32> [[TMP79]], [[TMP76]]
+; CHECK-NEXT:    [[TMP81:%.*]] = xor <16 x i32> [[TMP80]], [[TMP79]]
+; CHECK-NEXT:    [[TMP82:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP81]])
+; CHECK-NEXT:    [[CONV118:%.*]] = and i32 [[TMP82]], 65535
+; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[TMP82]], 16
 ; CHECK-NEXT:    [[ADD119:%.*]] = add nuw nsw i32 [[CONV118]], [[SHR]]
 ; CHECK-NEXT:    [[SHR120:%.*]] = lshr i32 [[ADD119]], 1
 ; CHECK-NEXT:    ret i32 [[SHR120]]
 ;
 entry:
   %idx.ext = sext i32 %i_pix1 to i64
   %idx.ext63 = sext i32 %i_pix2 to i64
   %arrayidx3 = getelementptr inbounds i8, ptr %pix1, i64 4