[SLP]Add shuffling of extractelements to avoid extra costs/data movement.

[ABataev]

Summary

If the scalar must be extracted and then used in the gather node,
instead we can emit shuffle instruction to avoid those extra
extractelements and vector-to-scalar and back data movement.

Part of D110978

[ABataev]

Rebase

[ABataev]

Rebase, ping!

[ABataev]

Rebase, ping!

[ABataev]

Ping!

[RKSimon]

There's a lot going on in this patch - making it really difficult to understand

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

637

I > E?

6875

we're looking for VF? == VF so would this be better as:

((VF == VF1 || VF == VF2) && GatheredScalars.size() != VF)

9422

((VF == VF1 || VF == Vf2) && GatheredScalars.size() != VF)

[ABataev]

In D141940#4104462, @RKSimon wrote:

There's a lot going on in this patch - making it really difficult to understand

Yeah, I understand, did everything I could. It is a part of another big patch, which improves previous attempt to vectorize the gathered node with extractelement instructions. Original isFixedVectorShuffle is very limited, this patch introduces new function, which tries to handle more cases.

1. Handles extractelements with vectors of different sizes.
2. If we have the extractelements from more than 2 vectors, it selects 2 vector with the most number of uses in the node.
3. If the number of the extracts from one of the vectors is greater than the extracts from 2 vectors with other vector factors, select permutation of single vector, not the very first pair.

Then it reuses the existing isFixedVectorShuffle to build the shuffle for the best subset of instructions.
If it is not succeeded, just exit (later can be extended to try to analyze some other subsets).

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

637

Very rare case, missed it.

[ABataev]

Address comments

[ABataev]

Rebase

[ABataev]

Rebase

[ABataev]

Rebase

[RKSimon]

A couple of minors

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

6852

if (Value *VecBase = AdjustExtractsCost(Cost, ExtractMask)) ?

9402

if (Value *VecBase = AdjustExtracts(E, ExtractMask)) ?

[ABataev]

Address comments

[RKSimon]

LGTM - I'm still worried there's too much in this patch, but I can't see anything wrong.

[dmgreen]

Hi - We noticed some regressions from this. The largest ones appear to be similar to the potential regressions that were reported in https://reviews.llvm.org/D142359#4131731, but with code that includes intrinsics and didn't change the vector cost: https://godbolt.org/z/P1zoxz85f.

Any idea how to improve that case and what might be going wrong? @SjoerdMeijer FYI. Thanks.

[ABataev]

In D141940#4141069, @dmgreen wrote:

Hi - We noticed some regressions from this. The largest ones appear to be similar to the potential regressions that were reported in https://reviews.llvm.org/D142359#4131731, but with code that includes intrinsics and didn't change the vector cost: https://godbolt.org/z/P1zoxz85f.

Any idea how to improve that case and what might be going wrong? @SjoerdMeijer FYI. Thanks.

I have a small fix, will commit it later today (working on a small reproducer)

[ABataev]

In D141940#4141069, @dmgreen wrote:

Hi - We noticed some regressions from this. The largest ones appear to be similar to the potential regressions that were reported in https://reviews.llvm.org/D142359#4131731, but with code that includes intrinsics and didn't change the vector cost: https://godbolt.org/z/P1zoxz85f.

Any idea how to improve that case and what might be going wrong? @SjoerdMeijer FYI. Thanks.

Fixed in cbcdd747e85b8d33b821d94d8114b971f31fd0d2

[dmgreen]

Fixed in cbcdd747e85b8d33b821d94d8114b971f31fd0d2

Yeah the results look great. Thanks!

[eaeltsin]

Heads up: this commits crashes one of our tests, the stack trace is:

...
 #2 0x000055c65640bbd6 CrashRecoverySignalHandler(int) (clang+0x800bbd6)
 #3 0x00007fda88e7b1c0 __restore_rt (/usr/grte/v5/lib64/libpthread.so.0+0x151c0)
 #4 0x000055c655b71964 llvm::slpvectorizer::BoUpSLP::isGatherShuffledEntry(llvm::slpvectorizer::BoUpSLP::TreeEntry const*, llvm::ArrayRef<llvm::Value*>, llvm::SmallVectorImpl<int>&, llvm::SmallVectorImpl<llvm::slpvectorizer::BoUpSLP::TreeEntry const*>&) (clang+0x7771964)
 #5 0x000055c655b6c47c llvm::slpvectorizer::BoUpSLP::getEntryCost(llvm::slpvectorizer::BoUpSLP::TreeEntry const*, llvm::ArrayRef<llvm::Value*>) (clang+0x776c47c)
 #6 0x000055c655b74903 llvm::slpvectorizer::BoUpSLP::getTreeCost(llvm::ArrayRef<llvm::Value*>) (clang+0x7774903)
 #7 0x000055c655b8d3a5 llvm::SLPVectorizerPass::tryToVectorizeList(llvm::ArrayRef<llvm::Value*>, llvm::slpvectorizer::BoUpSLP&, bool) (clang+0x778d3a5)
 #8 0x000055c655b92a0a bool tryToVectorizeSequence<llvm::Value>(llvm::SmallVectorImpl<llvm::Value*>&, llvm::function_ref<unsigned int (llvm::Value*)>, llvm::function_ref<bool (llvm::Value*, llvm::Value*)>, llvm::function_ref<bool (llvm::Value*, llvm::Value*)>, llvm::function_ref<bool (llvm::ArrayRef<llvm::Value*>, bool)>, bool) (clang+0x7792a0a)
 #9 0x000055c655b89681 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (clang+0x7789681)
#10 0x000055c655b87eb5 llvm::SLPVectorizerPass::runImpl(llvm::Function&, llvm::ScalarEvolution*, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo*, llvm::AAResults*, llvm::LoopInfo*, llvm::DominatorTree*, llvm::AssumptionCache*, llvm::DemandedBits*, llvm::OptimizationRemarkEmitter*) (clang+0x7787eb5)
#11 0x000055c655b877e6 llvm::SLPVectorizerPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x77877e6)
#12 0x000055c654ce1e32 llvm::detail::PassModel<llvm::Function, llvm::SLPVectorizerPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x68e1e32)
#13 0x000055c6562b07cb llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x7eb07cb)
#14 0x000055c651f02472 llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x3b02472)
#15 0x000055c6562b2b27 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (clang+0x7eb2b27)
#16 0x000055c651f04512 llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (clang+0x3b04512)
#17 0x000055c6562afcab llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (clang+0x7eafcab)
#18 0x000055c651efa03d (anonymous namespace)::EmitAssemblyHelper::RunOptimizationPipeline(clang::BackendAction, std::__u::unique_ptr<llvm::raw_pwrite_stream, std::__u::default_delete<llvm::raw_pwrite_stream>>&, std::__u::unique_ptr<llvm::ToolOutputFile, std::__u::default_delete<llvm::ToolOutputFile>>&) (clang+0x3afa03d)
#19 0x000055c651ef2d47 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, llvm::IntrusiveRefCntPtr<llvm::vfs::FileSystem>, std::__u::unique_ptr<llvm::raw_pwrite_stream, std::__u::default_delete<llvm::raw_pwrite_stream>>) (clang+0x3af2d47)
#20 0x000055c651eef4ca clang::CodeGenAction::ExecuteAction() (clang+0x3aef4ca)
#21 0x000055c6529aad5a clang::FrontendAction::Execute() (clang+0x45aad5a)
#22 0x000055c652919a04 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (clang+0x4519a04)
#23 0x000055c651b958ce clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (clang+0x37958ce)
#24 0x000055c651b8a4af cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (clang+0x378a4af)
...

[ABataev]

In D141940#4155223, @eaeltsin wrote:

Heads up: this commits crashes one of our tests, the stack trace is:

...
 #2 0x000055c65640bbd6 CrashRecoverySignalHandler(int) (clang+0x800bbd6)
 #3 0x00007fda88e7b1c0 __restore_rt (/usr/grte/v5/lib64/libpthread.so.0+0x151c0)
 #4 0x000055c655b71964 llvm::slpvectorizer::BoUpSLP::isGatherShuffledEntry(llvm::slpvectorizer::BoUpSLP::TreeEntry const*, llvm::ArrayRef<llvm::Value*>, llvm::SmallVectorImpl<int>&, llvm::SmallVectorImpl<llvm::slpvectorizer::BoUpSLP::TreeEntry const*>&) (clang+0x7771964)
 #5 0x000055c655b6c47c llvm::slpvectorizer::BoUpSLP::getEntryCost(llvm::slpvectorizer::BoUpSLP::TreeEntry const*, llvm::ArrayRef<llvm::Value*>) (clang+0x776c47c)
 #6 0x000055c655b74903 llvm::slpvectorizer::BoUpSLP::getTreeCost(llvm::ArrayRef<llvm::Value*>) (clang+0x7774903)
 #7 0x000055c655b8d3a5 llvm::SLPVectorizerPass::tryToVectorizeList(llvm::ArrayRef<llvm::Value*>, llvm::slpvectorizer::BoUpSLP&, bool) (clang+0x778d3a5)
 #8 0x000055c655b92a0a bool tryToVectorizeSequence<llvm::Value>(llvm::SmallVectorImpl<llvm::Value*>&, llvm::function_ref<unsigned int (llvm::Value*)>, llvm::function_ref<bool (llvm::Value*, llvm::Value*)>, llvm::function_ref<bool (llvm::Value*, llvm::Value*)>, llvm::function_ref<bool (llvm::ArrayRef<llvm::Value*>, bool)>, bool) (clang+0x7792a0a)
 #9 0x000055c655b89681 llvm::SLPVectorizerPass::vectorizeChainsInBlock(llvm::BasicBlock*, llvm::slpvectorizer::BoUpSLP&) (clang+0x7789681)
#10 0x000055c655b87eb5 llvm::SLPVectorizerPass::runImpl(llvm::Function&, llvm::ScalarEvolution*, llvm::TargetTransformInfo*, llvm::TargetLibraryInfo*, llvm::AAResults*, llvm::LoopInfo*, llvm::DominatorTree*, llvm::AssumptionCache*, llvm::DemandedBits*, llvm::OptimizationRemarkEmitter*) (clang+0x7787eb5)
#11 0x000055c655b877e6 llvm::SLPVectorizerPass::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x77877e6)
#12 0x000055c654ce1e32 llvm::detail::PassModel<llvm::Function, llvm::SLPVectorizerPass, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x68e1e32)
#13 0x000055c6562b07cb llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x7eb07cb)
#14 0x000055c651f02472 llvm::detail::PassModel<llvm::Function, llvm::PassManager<llvm::Function, llvm::AnalysisManager<llvm::Function>>, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Function>>::run(llvm::Function&, llvm::AnalysisManager<llvm::Function>&) (clang+0x3b02472)
#15 0x000055c6562b2b27 llvm::ModuleToFunctionPassAdaptor::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (clang+0x7eb2b27)
#16 0x000055c651f04512 llvm::detail::PassModel<llvm::Module, llvm::ModuleToFunctionPassAdaptor, llvm::PreservedAnalyses, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (clang+0x3b04512)
#17 0x000055c6562afcab llvm::PassManager<llvm::Module, llvm::AnalysisManager<llvm::Module>>::run(llvm::Module&, llvm::AnalysisManager<llvm::Module>&) (clang+0x7eafcab)
#18 0x000055c651efa03d (anonymous namespace)::EmitAssemblyHelper::RunOptimizationPipeline(clang::BackendAction, std::__u::unique_ptr<llvm::raw_pwrite_stream, std::__u::default_delete<llvm::raw_pwrite_stream>>&, std::__u::unique_ptr<llvm::ToolOutputFile, std::__u::default_delete<llvm::ToolOutputFile>>&) (clang+0x3afa03d)
#19 0x000055c651ef2d47 clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::HeaderSearchOptions const&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module*, clang::BackendAction, llvm::IntrusiveRefCntPtr<llvm::vfs::FileSystem>, std::__u::unique_ptr<llvm::raw_pwrite_stream, std::__u::default_delete<llvm::raw_pwrite_stream>>) (clang+0x3af2d47)
#20 0x000055c651eef4ca clang::CodeGenAction::ExecuteAction() (clang+0x3aef4ca)
#21 0x000055c6529aad5a clang::FrontendAction::Execute() (clang+0x45aad5a)
#22 0x000055c652919a04 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (clang+0x4519a04)
#23 0x000055c651b958ce clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (clang+0x37958ce)
#24 0x000055c651b8a4af cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (clang+0x378a4af)
...

Thanks for the report, please send the reproducer.

[eaeltsin]

Failing assertion is

assert.h assertion failed at ...llvm/include/llvm/ADT/SmallVector.h:298 in const_reference llvm::SmallVectorTemplateCommon<llvm::Value *>::operator[](size_type) const [T = llvm::Value *]: idx < size()

In D141940#4155223, @eaeltsin wrote:

Heads up: this commits crashes one of our tests, the stack trace is:

[ABataev]

In D141940#4155254, @eaeltsin wrote:

Failing assertion is

assert.h assertion failed at ...llvm/include/llvm/ADT/SmallVector.h:298 in const_reference llvm::SmallVectorTemplateCommon<llvm::Value *>::operator[](size_type) const [T = llvm::Value *]: idx < size()

In D141940#4155223, @eaeltsin wrote:

Heads up: this commits crashes one of our tests, the stack trace is:

Need a reproducer, unable to investigate the crash without actual repro.

[eaeltsin]

Need a reproducer, unable to investigate the crash without actual repro.

We are working on it. Unfortunately, reducing and stripping the internal stuff for this particular test is really hard.

[vporpo]

I think this is the cause of the crash: https://reviews.llvm.org/D144895
@ABataev could you take a look ?

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

if (Opcode == Instruction::ExtractElement) {
     return CI->getZExtValue();
   }
   auto *EI = cast<ExtractValueInst>(E);
   if (EI->getNumIndices() != 1)
     return std::nullopt;
   return *EI->idx_begin();
 }
 
+/// Tries to find extractelement instructions with constant indices from fixed
+/// vector type and gather such instructions into a bunch, which highly likely
+/// might be detected as a shuffle of 1 or 2 input vectors. If this attempt was
+/// successful, the matched scalars are replaced by poison values in \p VL for
+/// future analysis.
+static std::optional<TTI::ShuffleKind>
+tryToGatherExtractElements(SmallVectorImpl<Value *> &VL,
+                           SmallVectorImpl<int> &Mask) {
+  // Scan list of gathered scalars for extractelements that can be represented
+  // as shuffles.
+  MapVector<Value *, SmallVector<int>> VectorOpToIdx;
+  SmallVector<int> UndefVectorExtracts;
+  for (int I = 0, E = VL.size(); I < E; ++I) {
+    auto *EI = dyn_cast<ExtractElementInst>(VL[I]);
+    if (!EI)
+      continue;
+    auto *VecTy = dyn_cast<FixedVectorType>(EI->getVectorOperandType());
+    if (!VecTy || !isa<ConstantInt, UndefValue>(EI->getIndexOperand()))
+      continue;
+    std::optional<unsigned> Idx = getExtractIndex(EI);
+    // Undefined index.
+    if (!Idx) {
+      UndefVectorExtracts.push_back(I);
+      continue;
+    }
+    SmallBitVector ExtractMask(VecTy->getNumElements(), true);
+    ExtractMask.reset(*Idx);
+    if (isUndefVector(EI->getVectorOperand(), ExtractMask).all()) {
+      UndefVectorExtracts.push_back(I);
+      continue;
+    }
+    VectorOpToIdx[EI->getVectorOperand()].push_back(I);
+  }
+  // Sort the vector operands by the maximum number of uses in extractelements.
+  MapVector<unsigned, SmallVector<Value *>> VFToVector;
+  for (const auto &Data : VectorOpToIdx)
+    VFToVector[cast<FixedVectorType>(Data.first->getType())->getNumElements()]
+        .push_back(Data.first);
+  for (auto &Data : VFToVector) {
+    stable_sort(Data.second, [&VectorOpToIdx](Value *V1, Value *V2) {
+      return VectorOpToIdx.find(V1)->second.size() >
+             VectorOpToIdx.find(V2)->second.size();
+    });
+  }
+  // Find the best pair of the vectors with the same number of elements or a
+  // single vector.
+  const int UndefSz = UndefVectorExtracts.size();
+  unsigned SingleMax = 0;
+  Value *SingleVec = nullptr;
+  unsigned PairMax = 0;
+  std::pair<Value *, Value *> PairVec(nullptr, nullptr);
+  for (auto &Data : VFToVector) {
+    Value *V1 = Data.second.front();
+    if (SingleMax < VectorOpToIdx[V1].size() + UndefSz) {
+      SingleMax = VectorOpToIdx[V1].size() + UndefSz;
+      SingleVec = V1;
+    }
+    Value *V2 = nullptr;
+    if (Data.second.size() > 1)
+      V2 = *std::next(Data.second.begin());
+    if (V2 && PairMax < VectorOpToIdx[V1].size() + VectorOpToIdx[V2].size() +
+                            UndefSz) {
+      PairMax = VectorOpToIdx[V1].size() + VectorOpToIdx[V2].size() + UndefSz;
+      PairVec = std::make_pair(V1, V2);
+    }
+  }
+  if (SingleMax == 0 && PairMax == 0 && UndefSz == 0)
+    return std::nullopt;
+  // Check if better to perform a shuffle of 2 vectors or just of a single
+  // vector.
+  SmallVector<Value *> SavedVL(VL.begin(), VL.end());
+  SmallVector<Value *> GatheredExtracts(
+      VL.size(), PoisonValue::get(VL.front()->getType()));
+  if (SingleMax >= PairMax && SingleMax) {
+    for (int Idx : VectorOpToIdx[SingleVec])
+      std::swap(GatheredExtracts[Idx], VL[Idx]);
+  } else {
+    for (Value *V : {PairVec.first, PairVec.second})
+      for (int Idx : VectorOpToIdx[V])
+        std::swap(GatheredExtracts[Idx], VL[Idx]);
+  }
+  // Add extracts from undefs too.
+  for (int Idx : UndefVectorExtracts)
+    std::swap(GatheredExtracts[Idx], VL[Idx]);
+  // Check that gather of extractelements can be represented as just a
+  // shuffle of a single/two vectors the scalars are extracted from.
+  std::optional<TTI::ShuffleKind> Res =
+      isFixedVectorShuffle(GatheredExtracts, Mask);
+  if (!Res) {
+    // TODO: try to check other subsets if possible.
+    // Restore the original VL if attempt was not successful.
+    VL.swap(SavedVL);
+    return std::nullopt;
+  }
+  // Restore unused scalars from mask, if some of the extractelements were not
+  // selected for shuffle.
+  for (int I = 0, E = GatheredExtracts.size(); I < E; ++I) {
+    auto *EI = dyn_cast<ExtractElementInst>(VL[I]);
+    if (!EI || !isa<FixedVectorType>(EI->getVectorOperandType()) ||
+        !isa<ConstantInt, UndefValue>(EI->getIndexOperand()) ||
+        is_contained(UndefVectorExtracts, I))
+      continue;
+    if (Mask[I] == UndefMaskElem)
+      std::swap(VL[I], GatheredExtracts[I]);
+  }
+  return Res;
+}
+
 namespace {
 
 /// Main data required for vectorization of instructions.
 struct InstructionsState {
   /// The very first instruction in the list with the main opcode.
   Value *OpValue = nullptr;
 
   /// The main/alternate instruction.
if (MinBWs.count(VL[0]))
     VecTy = FixedVectorType::get(
         IntegerType::get(F->getContext(), MinBWs[VL[0]].first), VL.size());
   unsigned EntryVF = E->getVectorFactor();
   auto *FinalVecTy = FixedVectorType::get(VecTy->getElementType(), EntryVF);
 
   bool NeedToShuffleReuses = !E->ReuseShuffleIndices.empty();
   // FIXME: it tries to fix a problem with MSVC buildbots.
   TargetTransformInfo *TTI = this->TTI;
-  auto AdjustExtractsCost = [=](InstructionCost &Cost) {
+  auto AdjustExtractsCost = [=](InstructionCost &Cost,
+                                ArrayRef<int> Mask) -> Value * {
+    if (Mask.empty())
+      return nullptr;
+    Value *VecBase = nullptr;
     // If the resulting type is scalarized, do not adjust the cost.
     unsigned VecNumParts = TTI->getNumberOfParts(VecTy);
     if (VecNumParts == VecTy->getNumElements())
-      return;
+      return nullptr;
     DenseMap<Value *, int> ExtractVectorsTys;
     SmallPtrSet<Value *, 4> CheckedExtracts;
-    for (auto *V : VL) {
-      if (isa<UndefValue>(V))
+    for (auto [I, V] : enumerate(VL)) {
+      if (isa<UndefValue>(V) || (!Mask.empty() && Mask[I] == UndefMaskElem))
         continue;
       // If all users of instruction are going to be vectorized and this
       // instruction itself is not going to be vectorized, consider this
       // instruction as dead and remove its cost from the final cost of the
       // vectorized tree.
       // Also, avoid adjusting the cost for extractelements with multiple uses
       // in different graph entries.
       const TreeEntry *VE = getTreeEntry(V);
       if (!CheckedExtracts.insert(V).second ||
           !areAllUsersVectorized(cast<Instruction>(V), VectorizedVals) ||
           (VE && VE != E))
         continue;
       auto *EE = cast<ExtractElementInst>(V);
+      VecBase = EE->getVectorOperand();
       std::optional<unsigned> EEIdx = getExtractIndex(EE);
       if (!EEIdx)
         continue;
       unsigned Idx = *EEIdx;
       if (VecNumParts != TTI->getNumberOfParts(EE->getVectorOperandType())) {
         auto It =
             ExtractVectorsTys.try_emplace(EE->getVectorOperand(), Idx).first;
         It->getSecond() = std::min<int>(It->second, Idx);
auto AdjustExtractsCost = [=](InstructionCost &Cost,
     for (const auto &Data : ExtractVectorsTys) {
       auto *EEVTy = cast<FixedVectorType>(Data.first->getType());
       unsigned NumElts = VecTy->getNumElements();
       if (Data.second % NumElts == 0)
         continue;
       if (TTI->getNumberOfParts(EEVTy) > VecNumParts) {
         unsigned Idx = (Data.second / NumElts) * NumElts;
         unsigned EENumElts = EEVTy->getNumElements();
+        if (Idx % NumElts == 0)
+          continue;
         if (Idx + NumElts <= EENumElts) {
           Cost +=
               TTI->getShuffleCost(TargetTransformInfo::SK_ExtractSubvector,
                                   EEVTy, std::nullopt, CostKind, Idx, VecTy);
         } else {
           // Need to round up the subvector type vectorization factor to avoid a
           // crash in cost model functions. Make SubVT so that Idx + VF of SubVT
           // <= EENumElts.
           auto *SubVT =
               FixedVectorType::get(VecTy->getElementType(), EENumElts - Idx);
           Cost +=
               TTI->getShuffleCost(TargetTransformInfo::SK_ExtractSubvector,
                                   EEVTy, std::nullopt, CostKind, Idx, SubVT);
         }
       } else {
         Cost += TTI->getShuffleCost(TargetTransformInfo::SK_InsertSubvector,
                                     VecTy, std::nullopt, CostKind, 0, EEVTy);
       }
     }
+    return VecBase;
   };
   if (E->State == TreeEntry::NeedToGather) {
     if (allConstant(VL))
       return 0;
     if (isa<InsertElementInst>(VL[0]))
       return InstructionCost::getInvalid();
+    unsigned VF = E->getVectorFactor();
+    SmallVector<int> ReuseShuffleIndicies(E->ReuseShuffleIndices.begin(),
+                                          E->ReuseShuffleIndices.end());
     SmallVector<Value *> GatheredScalars(E->Scalars.begin(), E->Scalars.end());
     // Build a mask out of the reorder indices and reorder scalars per this
     // mask.
     SmallVector<int> ReorderMask;
     inversePermutation(E->ReorderIndices, ReorderMask);
     if (!ReorderMask.empty())
       reorderScalars(GatheredScalars, ReorderMask);
     SmallVector<int> Mask;
+    SmallVector<int> ExtractMask;
+    std::optional<TargetTransformInfo::ShuffleKind> ExtractShuffle;
     std::optional<TargetTransformInfo::ShuffleKind> GatherShuffle;
     SmallVector<const TreeEntry *> Entries;
+    Type *ScalarTy = GatheredScalars.front()->getType();
+    // Check for gathered extracts.
+    ExtractShuffle = tryToGatherExtractElements(GatheredScalars, ExtractMask);
+    SmallVector<Value *> IgnoredVals;
+    if (UserIgnoreList)
+      IgnoredVals.assign(UserIgnoreList->begin(), UserIgnoreList->end());
+
+    InstructionCost Cost = 0;
+    bool Resized = false;
+    if (Value *VecBase = AdjustExtractsCost(Cost, ExtractMask))
+      if (auto *VecBaseTy = dyn_cast<FixedVectorType>(VecBase->getType()))
+        if (VF == VecBaseTy->getNumElements() && GatheredScalars.size() != VF) {
+          Resized = true;
+          GatheredScalars.append(VF - GatheredScalars.size(),
+                                 PoisonValue::get(ScalarTy));
+        }
+
     // Do not try to look for reshuffled loads for gathered loads (they will be
     // handled later), for vectorized scalars, and cases, which are definitely
     // not profitable (splats and small gather nodes.)
-    if (E->getOpcode() != Instruction::Load || E->isAltShuffle() ||
+    if (ExtractShuffle || E->getOpcode() != Instruction::Load ||
+        E->isAltShuffle() ||
         all_of(E->Scalars, [this](Value *V) { return getTreeEntry(V); }) ||
         isSplat(E->Scalars) ||
         (E->Scalars != GatheredScalars && GatheredScalars.size() <= 2))
       GatherShuffle = isGatherShuffledEntry(E, GatheredScalars, Mask, Entries);
     if (GatherShuffle) {
+      assert((Entries.size() == 1 || Entries.size() == 2) &&
+             "Expected shuffle of 1 or 2 entries.");
+      if (!Resized) {
+        unsigned VF1 = Entries.front()->getVectorFactor();
+        unsigned VF2 = Entries.back()->getVectorFactor();
+        if ((VF == VF1 || VF == VF2) && GatheredScalars.size() != VF)
+          GatheredScalars.append(VF - GatheredScalars.size(),
+                                 PoisonValue::get(ScalarTy));
+      }
       // Remove shuffled elements from list of gathers.
       for (int I = 0, Sz = Mask.size(); I < Sz; ++I) {
         if (Mask[I] != UndefMaskElem)
           GatheredScalars[I] = PoisonValue::get(ScalarTy);
       }
-      assert((Entries.size() == 1 || Entries.size() == 2) &&
-             "Expected shuffle of 1 or 2 entries.");
       InstructionCost GatherCost = 0;
       int Limit = Mask.size() * 2;
       if (all_of(Mask, [=](int Idx) { return Idx < Limit; }) &&
           ShuffleVectorInst::isIdentityMask(Mask)) {
         // Perfect match in the graph, will reuse the previously vectorized
         // node. Cost is 0.
         LLVM_DEBUG(
             dbgs()
if (GatherShuffle) {
         // than gather.
         ::addMask(Mask, E->ReuseShuffleIndices);
         GatherCost = TTI->getShuffleCost(*GatherShuffle, FinalVecTy, Mask);
       }
       if (!all_of(GatheredScalars, UndefValue::classof))
         GatherCost += getGatherCost(GatheredScalars);
       return GatherCost;
     }
-    if ((E->getOpcode() == Instruction::ExtractElement ||
-         all_of(E->Scalars,
-                [](Value *V) {
-                  return isa<ExtractElementInst, UndefValue>(V);
-                })) &&
-        allSameType(VL)) {
+    if (ExtractShuffle && all_of(GatheredScalars, PoisonValue::classof)) {
       // Check that gather of extractelements can be represented as just a
       // shuffle of a single/two vectors the scalars are extracted from.
-      SmallVector<int> Mask;
-      std::optional<TargetTransformInfo::ShuffleKind> ShuffleKind =
-          isFixedVectorShuffle(VL, Mask);
-      if (ShuffleKind) {
-        // Found the bunch of extractelement instructions that must be gathered
-        // into a vector and can be represented as a permutation elements in a
-        // single input vector or of 2 input vectors.
-        InstructionCost Cost =
-            computeExtractCost(VL, VecTy, *ShuffleKind, Mask, *TTI);
-        AdjustExtractsCost(Cost);
-        if (NeedToShuffleReuses)
-          Cost += TTI->getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc,
-                                      FinalVecTy, E->ReuseShuffleIndices);
-        return Cost;
-      }
-    }
+          computeExtractCost(VL, VecTy, *ExtractShuffle, ExtractMask, *TTI);
+      AdjustExtractsCost(Cost, ExtractMask);
     if (isSplat(VL)) {
       // Found the broadcasting of the single scalar, calculate the cost as the
       // broadcast.
       assert(VecTy == FinalVecTy &&
              "No reused scalars expected for broadcast.");
       const auto *It =
           find_if(VL, [](Value *V) { return !isa<UndefValue>(V); });
       // If all values are undefs - consider cost free.
if (E->Idx > 0) {
       // We are in the middle of a vectorizable chain. We need to gather the
       // scalars from the users.
       Value *Vec = createBuildVector(E);
       E->VectorizedValue = Vec;
       return Vec;
     }
     if (E->getMainOp())
       setInsertPointAfterBundle(E);
+    unsigned VF = E->getVectorFactor();
+    auto AdjustExtracts = [&](const TreeEntry *E, ArrayRef<int> Mask) {
+      Value *VecBase = nullptr;
+      for (int I = 0, Sz = Mask.size(); I < Sz; ++I) {
+        int Idx = Mask[I];
+        if (Idx == UndefMaskElem)
+          continue;
+        auto *EI = cast<ExtractElementInst>(E->Scalars[I]);
+        VecBase = EI->getVectorOperand();
+        // If all users are vectorized - can delete the extractelement
+        // itself.
+        if (any_of(EI->users(),
+                   [&](User *U) { return !ScalarToTreeEntry.count(U); }))
+          continue;
+        eraseInstruction(EI);
+      }
+      return VecBase;
+    };
+    auto CreateShuffle = [&](Value *V1, Value *V2, ArrayRef<int> Mask) {
+      if (V1->getType() != V2->getType()) {
+        unsigned VF1 = cast<FixedVectorType>(V1->getType())->getNumElements();
+        unsigned VF2 = cast<FixedVectorType>(V2->getType())->getNumElements();
+        unsigned VF = std::max(VF1, VF2);
+        SmallVector<int> ExtMask(VF, UndefMaskElem);
+        std::iota(ExtMask.begin(),
+                  std::next(ExtMask.begin(), std::min(VF1, VF2)), 0);
+        if (VF1 < VF2) {
+          V1 = Builder.CreateShuffleVector(V1, ExtMask);
+          if (auto *I = dyn_cast<Instruction>(V1)) {
+            GatherShuffleExtractSeq.insert(I);
+            CSEBlocks.insert(I->getParent());
+          }
+        } else {
+          V2 = Builder.CreateShuffleVector(V2, ExtMask);
+          if (auto *I = dyn_cast<Instruction>(V2)) {
+            GatherShuffleExtractSeq.insert(I);
+            CSEBlocks.insert(I->getParent());
+          }
+        }
+      }
+      Value *Vec = Builder.CreateShuffleVector(V1, V2, Mask);
+      if (auto *I = dyn_cast<Instruction>(Vec)) {
+        GatherShuffleExtractSeq.insert(I);
+        CSEBlocks.insert(I->getParent());
+      }
+      return Vec;
+    };
+
+    SmallVector<int> ReuseShuffleIndicies(E->ReuseShuffleIndices.begin(),
+                                          E->ReuseShuffleIndices.end());
     SmallVector<Value *> GatheredScalars(E->Scalars.begin(), E->Scalars.end());
     // Build a mask out of the reorder indices and reorder scalars per this
     // mask.
     SmallVector<int> ReorderMask;
     inversePermutation(E->ReorderIndices, ReorderMask);
     if (!ReorderMask.empty())
       reorderScalars(GatheredScalars, ReorderMask);
-    Value *Vec;
+    Value *Vec = nullptr;
     SmallVector<int> Mask;
+    SmallVector<int> ExtractMask;
+    std::optional<TargetTransformInfo::ShuffleKind> ExtractShuffle;
+    std::optional<TargetTransformInfo::ShuffleKind> GatherShuffle;
     SmallVector<const TreeEntry *> Entries;
-    std::optional<TargetTransformInfo::ShuffleKind> Shuffle =
-        isGatherShuffledEntry(E, GatheredScalars, Mask, Entries);
-    if (Shuffle) {
+    Type *ScalarTy = GatheredScalars.front()->getType();
+    // Check for gathered extracts.
+    ExtractShuffle = tryToGatherExtractElements(GatheredScalars, ExtractMask);
+    SmallVector<Value *> IgnoredVals;
+    if (UserIgnoreList)
+      IgnoredVals.assign(UserIgnoreList->begin(), UserIgnoreList->end());
+    bool Resized = false;
+    if (Value *VecBase = AdjustExtracts(E, ExtractMask))
+      if (auto *VecBaseTy = dyn_cast<FixedVectorType>(VecBase->getType()))
+        if (VF == VecBaseTy->getNumElements() && GatheredScalars.size() != VF) {
+          Resized = true;
+          GatheredScalars.append(VF - GatheredScalars.size(),
+                                 PoisonValue::get(ScalarTy));
+        }
+    // Gather extracts after we check for full matched gathers only.
+    if (ExtractShuffle || E->getOpcode() != Instruction::Load ||
+        E->isAltShuffle() ||
+        all_of(E->Scalars, [this](Value *V) { return getTreeEntry(V); }) ||
+        isSplat(E->Scalars) ||
+        (E->Scalars != GatheredScalars && GatheredScalars.size() <= 2))
+      GatherShuffle = isGatherShuffledEntry(E, GatheredScalars, Mask, Entries);
+    if (GatherShuffle) {
       assert((Entries.size() == 1 || Entries.size() == 2) &&
              "Expected shuffle of 1 or 2 entries.");
-      Vec = Builder.CreateShuffleVector(Entries.front()->VectorizedValue,
-                                        Entries.back()->VectorizedValue, Mask);
-      if (auto *I = dyn_cast<Instruction>(Vec)) {
-        GatherShuffleExtractSeq.insert(I);
-        CSEBlocks.insert(I->getParent());
+      if (!Resized) {
+        unsigned VF1 = Entries.front()->getVectorFactor();
+        unsigned VF2 = Entries.back()->getVectorFactor();
+        if ((VF == VF1 || VF == VF2) && GatheredScalars.size() != VF)
+          GatheredScalars.append(VF - GatheredScalars.size(),
+                                 PoisonValue::get(ScalarTy));
+      }
+      // Remove shuffled elements from list of gathers.
+      for (int I = 0, Sz = Mask.size(); I < Sz; ++I) {
+        if (Mask[I] != UndefMaskElem)
+          GatheredScalars[I] = PoisonValue::get(ScalarTy);
+      }
+    }
+    if ((ExtractShuffle || GatherShuffle) &&
+        all_of(GatheredScalars, PoisonValue::classof)) {
+      Value *Vec1 = nullptr;
+      if (ExtractShuffle) {
+        // Gather of extractelements can be represented as just a shuffle of
+        // a single/two vectors the scalars are extracted from.
+        // Find input vectors.
+        Value *Vec2 = nullptr;
+        for (unsigned I = 0, Sz = ExtractMask.size(); I < Sz; ++I) {
+          if (ExtractMask[I] == UndefMaskElem ||
+              (!Mask.empty() && Mask[I] != UndefMaskElem)) {
+            ExtractMask[I] = UndefMaskElem;
+            continue;
+          }
+          if (isa<UndefValue>(E->Scalars[I]))
+            continue;
+          auto *EI = cast<ExtractElementInst>(E->Scalars[I]);
+          if (!Vec1) {
+            Vec1 = EI->getVectorOperand();
+          } else if (Vec1 != EI->getVectorOperand()) {
+            assert((!Vec2 || Vec2 == EI->getVectorOperand()) &&
+                   "Expected only 1 or 2 vectors shuffle.");
+            Vec2 = EI->getVectorOperand();
+          }
+        }
+        if (Vec2)
+          Vec1 = CreateShuffle(Vec1, Vec2, ExtractMask);
+        else if (Vec1)
+          Vec1 = CreateShuffle(Vec1, Vec1, ExtractMask);
+      }
+      if (GatherShuffle) {
+        Vec = CreateShuffle(Entries.front()->VectorizedValue,
+        if (Vec1) {
+          // Build final mask.
+          for (auto [I, Idx] : enumerate(Mask)) {
+            if (ExtractMask[I] != UndefMaskElem)
+              Idx = I;
+            else if (Idx != UndefMaskElem)
+              Idx = I + VF;
+          }
+          Vec = CreateShuffle(Vec1, Vec, Mask);
+        }
+      } else {
+        Vec = Vec1;
       }
     } else {
       Vec = gather(E->Scalars);
     }
     Vec = FinalShuffle(Vec, E);
     E->VectorizedValue = Vec;
     return Vec;
   }

llvm/test/Transforms/SLPVectorizer/X86/blending-shuffle-inseltpoison.ll

;
   %1 = add i8 %x0x0, %x3x3
   %2 = add i8 %y1y1, %y2y2
   %3 = sdiv i8 %1, %2
   ret i8 %3
 }
 
 define i8 @k(<4 x i8> %x) {
 ; CHECK-LABEL: @k(
-; CHECK-NEXT:    [[X0:%.*]] = extractelement <4 x i8> [[X:%.*]], i64 0
-; CHECK-NEXT:    [[X3:%.*]] = extractelement <4 x i8> [[X]], i64 3
-; CHECK-NEXT:    [[X1:%.*]] = extractelement <4 x i8> [[X]], i64 1
-; CHECK-NEXT:    [[X2:%.*]] = extractelement <4 x i8> [[X]], i64 2
-; CHECK-NEXT:    [[X0X0:%.*]] = mul i8 [[X0]], [[X0]]
-; CHECK-NEXT:    [[X3X3:%.*]] = mul i8 [[X3]], [[X3]]
-; CHECK-NEXT:    [[X1X1:%.*]] = mul i8 [[X1]], [[X1]]
-; CHECK-NEXT:    [[X2X2:%.*]] = mul i8 [[X2]], [[X2]]
-; CHECK-NEXT:    [[TMP1:%.*]] = add i8 [[X0X0]], [[X3X3]]
-; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[X1X1]], [[X2X2]]
-; CHECK-NEXT:    [[TMP3:%.*]] = sdiv i8 [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    ret i8 [[TMP3]]
+; CHECK-NEXT:    [[TMP1:%.*]] = mul <4 x i8> [[X:%.*]], [[X]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i8> [[X]], [[X]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <2 x i32> <i32 3, i32 2>
+; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i8> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i8> [[TMP5]], i64 0
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP5]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = sdiv i8 [[TMP6]], [[TMP7]]
+; CHECK-NEXT:    ret i8 [[TMP8]]
 ;
   %x0 = extractelement <4 x i8> %x, i32 0
   %x3 = extractelement <4 x i8> %x, i32 3
   %x1 = extractelement <4 x i8> %x, i32 1
   %x2 = extractelement <4 x i8> %x, i32 2
   %x0x0 = mul i8 %x0, %x0
   %x3x3 = mul i8 %x3, %x3
   %x1x1 = mul i8 %x1, %x1
   %x2x2 = mul i8 %x2, %x2
   %1 = add i8 %x0x0, %x3x3
   %2 = add i8 %x1x1, %x2x2
   %3 = sdiv i8 %1, %2
   ret i8 %3
 }
 
 define i8 @k_bb(<4 x i8> %x) {
 ; CHECK-LABEL: @k_bb(
 ; CHECK-NEXT:    br label [[BB1:%.*]]
 ; CHECK:       bb1:
-; CHECK-NEXT:    [[X0:%.*]] = extractelement <4 x i8> [[X:%.*]], i64 0
-; CHECK-NEXT:    [[X3:%.*]] = extractelement <4 x i8> [[X]], i64 3
-; CHECK-NEXT:    [[X1:%.*]] = extractelement <4 x i8> [[X]], i64 1
-; CHECK-NEXT:    [[X2:%.*]] = extractelement <4 x i8> [[X]], i64 2
-; CHECK-NEXT:    [[X0X0:%.*]] = mul i8 [[X0]], [[X0]]
-; CHECK-NEXT:    [[X3X3:%.*]] = mul i8 [[X3]], [[X3]]
-; CHECK-NEXT:    [[X1X1:%.*]] = mul i8 [[X1]], [[X1]]
-; CHECK-NEXT:    [[X2X2:%.*]] = mul i8 [[X2]], [[X2]]
-; CHECK-NEXT:    [[TMP1:%.*]] = add i8 [[X0X0]], [[X3X3]]
-; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[X1X1]], [[X2X2]]
-; CHECK-NEXT:    [[TMP3:%.*]] = sdiv i8 [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    ret i8 [[TMP3]]
+; CHECK-NEXT:    [[TMP1:%.*]] = mul <4 x i8> [[X:%.*]], [[X]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i8> [[X]], [[X]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <2 x i32> <i32 3, i32 2>
+; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i8> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i8> [[TMP5]], i64 0
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP5]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = sdiv i8 [[TMP6]], [[TMP7]]
+; CHECK-NEXT:    ret i8 [[TMP8]]
 ;
   %x0 = extractelement <4 x i8> %x, i32 0
   br label %bb1
 bb1:
   %x3 = extractelement <4 x i8> %x, i32 3
   %x1 = extractelement <4 x i8> %x, i32 1
   %x2 = extractelement <4 x i8> %x, i32 2
   %x0x0 = mul i8 %x0, %x0
   %x3x3 = mul i8 %x3, %x3
   %x1x1 = mul i8 %x1, %x1
   %x2x2 = mul i8 %x2, %x2
   %1 = add i8 %x0x0, %x3x3
   %2 = add i8 %x1x1, %x2x2
   %3 = sdiv i8 %1, %2
   ret i8 %3
 }

llvm/test/Transforms/SLPVectorizer/X86/blending-shuffle.ll

;
   %1 = add i8 %x0x0, %x3x3
   %2 = add i8 %y1y1, %y2y2
   %3 = sdiv i8 %1, %2
   ret i8 %3
 }
 
 define i8 @k(<4 x i8> %x) {
 ; CHECK-LABEL: @k(
-; CHECK-NEXT:    [[X0:%.*]] = extractelement <4 x i8> [[X:%.*]], i64 0
-; CHECK-NEXT:    [[X3:%.*]] = extractelement <4 x i8> [[X]], i64 3
-; CHECK-NEXT:    [[X1:%.*]] = extractelement <4 x i8> [[X]], i64 1
-; CHECK-NEXT:    [[X2:%.*]] = extractelement <4 x i8> [[X]], i64 2
-; CHECK-NEXT:    [[X0X0:%.*]] = mul i8 [[X0]], [[X0]]
-; CHECK-NEXT:    [[X3X3:%.*]] = mul i8 [[X3]], [[X3]]
-; CHECK-NEXT:    [[X1X1:%.*]] = mul i8 [[X1]], [[X1]]
-; CHECK-NEXT:    [[X2X2:%.*]] = mul i8 [[X2]], [[X2]]
-; CHECK-NEXT:    [[TMP1:%.*]] = add i8 [[X0X0]], [[X3X3]]
-; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[X1X1]], [[X2X2]]
-; CHECK-NEXT:    [[TMP3:%.*]] = sdiv i8 [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    ret i8 [[TMP3]]
+; CHECK-NEXT:    [[TMP1:%.*]] = mul <4 x i8> [[X:%.*]], [[X]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i8> [[X]], [[X]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <2 x i32> <i32 3, i32 2>
+; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i8> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i8> [[TMP5]], i64 0
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP5]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = sdiv i8 [[TMP6]], [[TMP7]]
+; CHECK-NEXT:    ret i8 [[TMP8]]
 ;
   %x0 = extractelement <4 x i8> %x, i32 0
   %x3 = extractelement <4 x i8> %x, i32 3
   %x1 = extractelement <4 x i8> %x, i32 1
   %x2 = extractelement <4 x i8> %x, i32 2
   %x0x0 = mul i8 %x0, %x0
   %x3x3 = mul i8 %x3, %x3
   %x1x1 = mul i8 %x1, %x1
   %x2x2 = mul i8 %x2, %x2
   %1 = add i8 %x0x0, %x3x3
   %2 = add i8 %x1x1, %x2x2
   %3 = sdiv i8 %1, %2
   ret i8 %3
 }
 
 define i8 @k_bb(<4 x i8> %x) {
 ; CHECK-LABEL: @k_bb(
 ; CHECK-NEXT:    br label [[BB1:%.*]]
 ; CHECK:       bb1:
-; CHECK-NEXT:    [[X0:%.*]] = extractelement <4 x i8> [[X:%.*]], i64 0
-; CHECK-NEXT:    [[X3:%.*]] = extractelement <4 x i8> [[X]], i64 3
-; CHECK-NEXT:    [[X1:%.*]] = extractelement <4 x i8> [[X]], i64 1
-; CHECK-NEXT:    [[X2:%.*]] = extractelement <4 x i8> [[X]], i64 2
-; CHECK-NEXT:    [[X0X0:%.*]] = mul i8 [[X0]], [[X0]]
-; CHECK-NEXT:    [[X3X3:%.*]] = mul i8 [[X3]], [[X3]]
-; CHECK-NEXT:    [[X1X1:%.*]] = mul i8 [[X1]], [[X1]]
-; CHECK-NEXT:    [[X2X2:%.*]] = mul i8 [[X2]], [[X2]]
-; CHECK-NEXT:    [[TMP1:%.*]] = add i8 [[X0X0]], [[X3X3]]
-; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[X1X1]], [[X2X2]]
-; CHECK-NEXT:    [[TMP3:%.*]] = sdiv i8 [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    ret i8 [[TMP3]]
+; CHECK-NEXT:    [[TMP1:%.*]] = mul <4 x i8> [[X:%.*]], [[X]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i8> [[X]], [[X]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <2 x i32> <i32 3, i32 2>
+; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i8> [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i8> [[TMP5]], i64 0
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP5]], i64 1
+; CHECK-NEXT:    [[TMP8:%.*]] = sdiv i8 [[TMP6]], [[TMP7]]
+; CHECK-NEXT:    ret i8 [[TMP8]]
 ;
   %x0 = extractelement <4 x i8> %x, i32 0
   br label %bb1
 bb1:
   %x3 = extractelement <4 x i8> %x, i32 3
   %x1 = extractelement <4 x i8> %x, i32 1
   %x2 = extractelement <4 x i8> %x, i32 2
   %x0x0 = mul i8 %x0, %x0
   %x3x3 = mul i8 %x3, %x3
   %x1x1 = mul i8 %x1, %x1
   %x2x2 = mul i8 %x2, %x2
   %1 = add i8 %x0x0, %x3x3
   %2 = add i8 %x1x1, %x2x2
   %3 = sdiv i8 %1, %2
   ret i8 %3
 }

llvm/test/Transforms/SLPVectorizer/X86/commutativity.ll

 
 ; Check that we correctly detect a splat/broadcast by leveraging the
 ; commutativity property of `xor`.
 
 define void @splat(i8 %a, i8 %b, i8 %c) {
 ; SSE-LABEL: @splat(
 ; SSE-NEXT:    [[TMP1:%.*]] = insertelement <16 x i8> poison, i8 [[A:%.*]], i32 0
 ; SSE-NEXT:    [[TMP2:%.*]] = insertelement <16 x i8> [[TMP1]], i8 [[B:%.*]], i32 1
-; SSE-NEXT:    [[SHUFFLE:%.*]] = shufflevector <16 x i8> [[TMP2]], <16 x i8> poison, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
-; SSE-NEXT:    [[TMP3:%.*]] = insertelement <16 x i8> poison, i8 [[C:%.*]], i32 0
-; SSE-NEXT:    [[SHUFFLE1:%.*]] = shufflevector <16 x i8> [[TMP3]], <16 x i8> poison, <16 x i32> zeroinitializer
-; SSE-NEXT:    [[TMP4:%.*]] = xor <16 x i8> [[SHUFFLE]], [[SHUFFLE1]]
-; SSE-NEXT:    store <16 x i8> [[TMP4]], ptr @cle, align 16
+; SSE-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i8> [[TMP2]], <16 x i8> poison, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
+; SSE-NEXT:    [[TMP4:%.*]] = insertelement <16 x i8> poison, i8 [[C:%.*]], i32 0
+; SSE-NEXT:    [[TMP5:%.*]] = shufflevector <16 x i8> [[TMP4]], <16 x i8> poison, <16 x i32> zeroinitializer
+; SSE-NEXT:    [[TMP6:%.*]] = xor <16 x i8> [[TMP3]], [[TMP5]]
+; SSE-NEXT:    store <16 x i8> [[TMP6]], ptr @cle, align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @splat(
 ; AVX-NEXT:    [[TMP1:%.*]] = insertelement <16 x i8> poison, i8 [[A:%.*]], i32 0
 ; AVX-NEXT:    [[TMP2:%.*]] = insertelement <16 x i8> [[TMP1]], i8 [[B:%.*]], i32 1
-; AVX-NEXT:    [[SHUFFLE:%.*]] = shufflevector <16 x i8> [[TMP2]], <16 x i8> poison, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
-; AVX-NEXT:    [[TMP3:%.*]] = insertelement <16 x i8> poison, i8 [[C:%.*]], i32 0
-; AVX-NEXT:    [[SHUFFLE1:%.*]] = shufflevector <16 x i8> [[TMP3]], <16 x i8> poison, <16 x i32> zeroinitializer
-; AVX-NEXT:    [[TMP4:%.*]] = xor <16 x i8> [[SHUFFLE]], [[SHUFFLE1]]
-; AVX-NEXT:    store <16 x i8> [[TMP4]], ptr @cle, align 16
+; AVX-NEXT:    [[TMP3:%.*]] = shufflevector <16 x i8> [[TMP2]], <16 x i8> poison, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
+; AVX-NEXT:    [[TMP4:%.*]] = insertelement <16 x i8> poison, i8 [[C:%.*]], i32 0
+; AVX-NEXT:    [[TMP5:%.*]] = shufflevector <16 x i8> [[TMP4]], <16 x i8> poison, <16 x i32> zeroinitializer
+; AVX-NEXT:    [[TMP6:%.*]] = xor <16 x i8> [[TMP3]], [[TMP5]]
+; AVX-NEXT:    store <16 x i8> [[TMP6]], ptr @cle, align 16
 ; AVX-NEXT:    ret void
 ;
   %1 = xor i8 %c, %a
   store i8 %1, ptr @cle, align 16
   %2 = xor i8 %a, %c
   store i8 %2, ptr getelementptr inbounds ([32 x i8], ptr @cle, i64 0, i64 1)
   %3 = xor i8 %a, %c
   store i8 %3, ptr getelementptr inbounds ([32 x i8], ptr @cle, i64 0, i64 2)

 ; SSE-NEXT:    [[TMP3:%.*]] = xor i32 [[C]], [[ADD3]]
 ; SSE-NEXT:    store i32 [[TMP3]], ptr getelementptr inbounds ([32 x i32], ptr @cle32, i64 0, i64 2), align 4
 ; SSE-NEXT:    [[TMP4:%.*]] = xor i32 [[A]], [[ADD4]]
 ; SSE-NEXT:    store i32 [[TMP4]], ptr getelementptr inbounds ([32 x i32], ptr @cle32, i64 0, i64 3), align 4
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @same_opcode_on_one_side(
 ; AVX-NEXT:    [[TMP1:%.*]] = insertelement <4 x i32> poison, i32 [[C:%.*]], i32 0
-; AVX-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer
-; AVX-NEXT:    [[TMP2:%.*]] = insertelement <4 x i32> poison, i32 [[A:%.*]], i32 0
-; AVX-NEXT:    [[SHUFFLE1:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <4 x i32> zeroinitializer
-; AVX-NEXT:    [[TMP3:%.*]] = add <4 x i32> [[SHUFFLE]], [[SHUFFLE1]]
-; AVX-NEXT:    [[TMP4:%.*]] = insertelement <4 x i32> [[TMP2]], i32 [[B:%.*]], i32 1
-; AVX-NEXT:    [[TMP5:%.*]] = insertelement <4 x i32> [[TMP4]], i32 [[C]], i32 2
-; AVX-NEXT:    [[SHUFFLE2:%.*]] = shufflevector <4 x i32> [[TMP5]], <4 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 0>
-; AVX-NEXT:    [[TMP6:%.*]] = xor <4 x i32> [[TMP3]], [[SHUFFLE2]]
-; AVX-NEXT:    store <4 x i32> [[TMP6]], ptr @cle32, align 16
+; AVX-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer
+; AVX-NEXT:    [[TMP3:%.*]] = insertelement <4 x i32> poison, i32 [[A:%.*]], i32 0
+; AVX-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> poison, <4 x i32> zeroinitializer
+; AVX-NEXT:    [[TMP5:%.*]] = add <4 x i32> [[TMP2]], [[TMP4]]
+; AVX-NEXT:    [[TMP6:%.*]] = insertelement <4 x i32> [[TMP3]], i32 [[B:%.*]], i32 1
+; AVX-NEXT:    [[TMP7:%.*]] = insertelement <4 x i32> [[TMP6]], i32 [[C]], i32 2
+; AVX-NEXT:    [[TMP8:%.*]] = shufflevector <4 x i32> [[TMP7]], <4 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 0>
+; AVX-NEXT:    [[TMP9:%.*]] = xor <4 x i32> [[TMP5]], [[TMP8]]
+; AVX-NEXT:    store <4 x i32> [[TMP9]], ptr @cle32, align 16
 ; AVX-NEXT:    ret void
 ;
   %add1 = add i32 %c, %a
   %add2 = add i32 %c, %a
   %add3 = add i32 %a, %c
   %add4 = add i32 %c, %a
   %1 = xor i32 %add1, %a
   store i32 %1, ptr @cle32, align 16
   %2 = xor i32 %b, %add2
   store i32 %2, ptr getelementptr inbounds ([32 x i32], ptr @cle32, i64 0, i64 1)
   %3 = xor i32 %c, %add3
   store i32 %3, ptr getelementptr inbounds ([32 x i32], ptr @cle32, i64 0, i64 2)
   %4 = xor i32 %a, %add4
   store i32 %4, ptr getelementptr inbounds ([32 x i32], ptr @cle32, i64 0, i64 3)
   ret void
 }

llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector-inseltpoison.ll

 ; MINTREESIZE-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
 ; MINTREESIZE-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
 ; MINTREESIZE-NEXT:    [[RA:%.*]] = insertelement <4 x float> poison, float [[S0]], i32 0
 ; MINTREESIZE-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
 ; MINTREESIZE-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
 ; MINTREESIZE-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
 ; MINTREESIZE-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
 ; MINTREESIZE-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
-; MINTREESIZE-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> poison, float [[Q0]], i32 0
-; MINTREESIZE-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> [[TMP5]], float [[Q1]], i32 1
+; MINTREESIZE-NEXT:    [[TMP5:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> [[RD]], <2 x i32> <i32 0, i32 1>
 ; MINTREESIZE-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
 ; MINTREESIZE-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
-; MINTREESIZE-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[Q2]], i32 0
-; MINTREESIZE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[Q3]], i32 1
+; MINTREESIZE-NEXT:    [[TMP6:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> [[RD]], <2 x i32> <i32 2, i32 3>
 ; MINTREESIZE-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
 ; MINTREESIZE-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
-; MINTREESIZE-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0
-; MINTREESIZE-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[Q5]], i32 1
+; MINTREESIZE-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0
+; MINTREESIZE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[Q5]], i32 1
 ; MINTREESIZE-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
 ; MINTREESIZE-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
 ; MINTREESIZE-NEXT:    call void @llvm.assume(i1 [[QI]])
 ; MINTREESIZE-NEXT:    ret <4 x float> undef
 ;
   %c0 = extractelement <4 x i32> %c, i32 0
   %c1 = extractelement <4 x i32> %c, i32 1
   %c2 = extractelement <4 x i32> %c, i32 2

llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll

 ; MINTREESIZE-NEXT:    [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]]
 ; MINTREESIZE-NEXT:    [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]]
 ; MINTREESIZE-NEXT:    [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0
 ; MINTREESIZE-NEXT:    [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1
 ; MINTREESIZE-NEXT:    [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2
 ; MINTREESIZE-NEXT:    [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3
 ; MINTREESIZE-NEXT:    [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0
 ; MINTREESIZE-NEXT:    [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1
-; MINTREESIZE-NEXT:    [[TMP5:%.*]] = insertelement <2 x float> poison, float [[Q0]], i32 0
-; MINTREESIZE-NEXT:    [[TMP6:%.*]] = insertelement <2 x float> [[TMP5]], float [[Q1]], i32 1
+; MINTREESIZE-NEXT:    [[TMP5:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> [[RD]], <2 x i32> <i32 0, i32 1>
 ; MINTREESIZE-NEXT:    [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2
 ; MINTREESIZE-NEXT:    [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3
-; MINTREESIZE-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[Q2]], i32 0
-; MINTREESIZE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[Q3]], i32 1
+; MINTREESIZE-NEXT:    [[TMP6:%.*]] = shufflevector <4 x float> [[RD]], <4 x float> [[RD]], <2 x i32> <i32 2, i32 3>
 ; MINTREESIZE-NEXT:    [[Q4:%.*]] = fadd float [[Q0]], [[Q1]]
 ; MINTREESIZE-NEXT:    [[Q5:%.*]] = fadd float [[Q2]], [[Q3]]
-; MINTREESIZE-NEXT:    [[TMP9:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0
-; MINTREESIZE-NEXT:    [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[Q5]], i32 1
+; MINTREESIZE-NEXT:    [[TMP7:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0
+; MINTREESIZE-NEXT:    [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[Q5]], i32 1
 ; MINTREESIZE-NEXT:    [[Q6:%.*]] = fadd float [[Q4]], [[Q5]]
 ; MINTREESIZE-NEXT:    [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]]
 ; MINTREESIZE-NEXT:    call void @llvm.assume(i1 [[QI]])
 ; MINTREESIZE-NEXT:    ret <4 x float> undef
 ;
   %c0 = extractelement <4 x i32> %c, i32 0
   %c1 = extractelement <4 x i32> %c, i32 1
   %c2 = extractelement <4 x i32> %c, i32 2

llvm/test/Transforms/SLPVectorizer/X86/reduction-transpose.ll

 ;   acc &= v3[0] & v3[1] & v3[2] & v3[3];
 ;   acc &= v4[0] & v4[1] & v4[2] & v4[3];
 ;   return acc;
 ; }
 
 define i32 @reduce_and4(i32 %acc, <4 x i32> %v1, <4 x i32> %v2, <4 x i32> %v3, <4 x i32> %v4) {
 ; SSE2-LABEL: @reduce_and4(
 ; SSE2-NEXT:  entry:
-; SSE2-NEXT:    [[VECEXT:%.*]] = extractelement <4 x i32> [[V1:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT1:%.*]] = extractelement <4 x i32> [[V1]], i64 1
-; SSE2-NEXT:    [[VECEXT2:%.*]] = extractelement <4 x i32> [[V1]], i64 2
-; SSE2-NEXT:    [[VECEXT4:%.*]] = extractelement <4 x i32> [[V1]], i64 3
-; SSE2-NEXT:    [[VECEXT7:%.*]] = extractelement <4 x i32> [[V2:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT8:%.*]] = extractelement <4 x i32> [[V2]], i64 1
-; SSE2-NEXT:    [[VECEXT10:%.*]] = extractelement <4 x i32> [[V2]], i64 2
-; SSE2-NEXT:    [[VECEXT12:%.*]] = extractelement <4 x i32> [[V2]], i64 3
-; SSE2-NEXT:    [[TMP0:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT8]], i32 0
-; SSE2-NEXT:    [[TMP1:%.*]] = insertelement <8 x i32> [[TMP0]], i32 [[VECEXT7]], i32 1
-; SSE2-NEXT:    [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[VECEXT10]], i32 2
-; SSE2-NEXT:    [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[VECEXT12]], i32 3
-; SSE2-NEXT:    [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[VECEXT1]], i32 4
-; SSE2-NEXT:    [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[VECEXT]], i32 5
-; SSE2-NEXT:    [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[VECEXT2]], i32 6
-; SSE2-NEXT:    [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[VECEXT4]], i32 7
-; SSE2-NEXT:    [[VECEXT15:%.*]] = extractelement <4 x i32> [[V3:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT16:%.*]] = extractelement <4 x i32> [[V3]], i64 1
-; SSE2-NEXT:    [[VECEXT18:%.*]] = extractelement <4 x i32> [[V3]], i64 2
-; SSE2-NEXT:    [[VECEXT20:%.*]] = extractelement <4 x i32> [[V3]], i64 3
-; SSE2-NEXT:    [[VECEXT23:%.*]] = extractelement <4 x i32> [[V4:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT24:%.*]] = extractelement <4 x i32> [[V4]], i64 1
-; SSE2-NEXT:    [[VECEXT26:%.*]] = extractelement <4 x i32> [[V4]], i64 2
-; SSE2-NEXT:    [[VECEXT28:%.*]] = extractelement <4 x i32> [[V4]], i64 3
-; SSE2-NEXT:    [[TMP8:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT24]], i32 0
-; SSE2-NEXT:    [[TMP9:%.*]] = insertelement <8 x i32> [[TMP8]], i32 [[VECEXT23]], i32 1
-; SSE2-NEXT:    [[TMP10:%.*]] = insertelement <8 x i32> [[TMP9]], i32 [[VECEXT26]], i32 2
-; SSE2-NEXT:    [[TMP11:%.*]] = insertelement <8 x i32> [[TMP10]], i32 [[VECEXT28]], i32 3
-; SSE2-NEXT:    [[TMP12:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[VECEXT16]], i32 4
-; SSE2-NEXT:    [[TMP13:%.*]] = insertelement <8 x i32> [[TMP12]], i32 [[VECEXT15]], i32 5
-; SSE2-NEXT:    [[TMP14:%.*]] = insertelement <8 x i32> [[TMP13]], i32 [[VECEXT18]], i32 6
-; SSE2-NEXT:    [[TMP15:%.*]] = insertelement <8 x i32> [[TMP14]], i32 [[VECEXT20]], i32 7
-; SSE2-NEXT:    [[TMP16:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP15]])
-; SSE2-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP7]])
-; SSE2-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP16]], [[TMP17]]
+; SSE2-NEXT:    [[TMP0:%.*]] = shufflevector <4 x i32> [[V2:%.*]], <4 x i32> [[V1:%.*]], <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
+; SSE2-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[V4:%.*]], <4 x i32> [[V3:%.*]], <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
+; SSE2-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP1]])
+; SSE2-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP0]])
+; SSE2-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP2]], [[TMP3]]
 ; SSE2-NEXT:    [[OP_RDX1:%.*]] = and i32 [[OP_RDX]], [[ACC:%.*]]
 ; SSE2-NEXT:    ret i32 [[OP_RDX1]]
 ;
 ; SSE42-LABEL: @reduce_and4(
 ; SSE42-NEXT:  entry:
 ; SSE42-NEXT:    [[TMP0:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V4:%.*]])
 ; SSE42-NEXT:    [[TMP1:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V3:%.*]])
 ; SSE42-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP0]], [[TMP1]]
 ; SSE42-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V2:%.*]])
 ; SSE42-NEXT:    [[OP_RDX1:%.*]] = and i32 [[OP_RDX]], [[TMP2]]
 ; SSE42-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V1:%.*]])
 ; SSE42-NEXT:    [[OP_RDX2:%.*]] = and i32 [[OP_RDX1]], [[TMP3]]
 ; SSE42-NEXT:    [[OP_RDX3:%.*]] = and i32 [[OP_RDX2]], [[ACC:%.*]]
 ; SSE42-NEXT:    ret i32 [[OP_RDX3]]
 ;
 ; AVX-LABEL: @reduce_and4(
 ; AVX-NEXT:  entry:
-; AVX-NEXT:    [[VECEXT:%.*]] = extractelement <4 x i32> [[V1:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT1:%.*]] = extractelement <4 x i32> [[V1]], i64 1
-; AVX-NEXT:    [[VECEXT2:%.*]] = extractelement <4 x i32> [[V1]], i64 2
-; AVX-NEXT:    [[VECEXT4:%.*]] = extractelement <4 x i32> [[V1]], i64 3
-; AVX-NEXT:    [[VECEXT7:%.*]] = extractelement <4 x i32> [[V2:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT8:%.*]] = extractelement <4 x i32> [[V2]], i64 1
-; AVX-NEXT:    [[VECEXT10:%.*]] = extractelement <4 x i32> [[V2]], i64 2
-; AVX-NEXT:    [[VECEXT12:%.*]] = extractelement <4 x i32> [[V2]], i64 3
-; AVX-NEXT:    [[TMP0:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT8]], i32 0
-; AVX-NEXT:    [[TMP1:%.*]] = insertelement <8 x i32> [[TMP0]], i32 [[VECEXT7]], i32 1
-; AVX-NEXT:    [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[VECEXT10]], i32 2
-; AVX-NEXT:    [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[VECEXT12]], i32 3
-; AVX-NEXT:    [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[VECEXT1]], i32 4
-; AVX-NEXT:    [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[VECEXT]], i32 5
-; AVX-NEXT:    [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[VECEXT2]], i32 6
-; AVX-NEXT:    [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[VECEXT4]], i32 7
-; AVX-NEXT:    [[VECEXT15:%.*]] = extractelement <4 x i32> [[V3:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT16:%.*]] = extractelement <4 x i32> [[V3]], i64 1
-; AVX-NEXT:    [[VECEXT18:%.*]] = extractelement <4 x i32> [[V3]], i64 2
-; AVX-NEXT:    [[VECEXT20:%.*]] = extractelement <4 x i32> [[V3]], i64 3
-; AVX-NEXT:    [[VECEXT23:%.*]] = extractelement <4 x i32> [[V4:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT24:%.*]] = extractelement <4 x i32> [[V4]], i64 1
-; AVX-NEXT:    [[VECEXT26:%.*]] = extractelement <4 x i32> [[V4]], i64 2
-; AVX-NEXT:    [[VECEXT28:%.*]] = extractelement <4 x i32> [[V4]], i64 3
-; AVX-NEXT:    [[TMP8:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT24]], i32 0
-; AVX-NEXT:    [[TMP9:%.*]] = insertelement <8 x i32> [[TMP8]], i32 [[VECEXT23]], i32 1
-; AVX-NEXT:    [[TMP10:%.*]] = insertelement <8 x i32> [[TMP9]], i32 [[VECEXT26]], i32 2
-; AVX-NEXT:    [[TMP11:%.*]] = insertelement <8 x i32> [[TMP10]], i32 [[VECEXT28]], i32 3
-; AVX-NEXT:    [[TMP12:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[VECEXT16]], i32 4
-; AVX-NEXT:    [[TMP13:%.*]] = insertelement <8 x i32> [[TMP12]], i32 [[VECEXT15]], i32 5
-; AVX-NEXT:    [[TMP14:%.*]] = insertelement <8 x i32> [[TMP13]], i32 [[VECEXT18]], i32 6
-; AVX-NEXT:    [[TMP15:%.*]] = insertelement <8 x i32> [[TMP14]], i32 [[VECEXT20]], i32 7
-; AVX-NEXT:    [[TMP16:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP15]])
-; AVX-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP7]])
-; AVX-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP16]], [[TMP17]]
+; AVX-NEXT:    [[TMP0:%.*]] = shufflevector <4 x i32> [[V2:%.*]], <4 x i32> [[V1:%.*]], <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
+; AVX-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[V4:%.*]], <4 x i32> [[V3:%.*]], <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7>
+; AVX-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP1]])
+; AVX-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP0]])
+; AVX-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP2]], [[TMP3]]
 ; AVX-NEXT:    [[OP_RDX1:%.*]] = and i32 [[OP_RDX]], [[ACC:%.*]]
 ; AVX-NEXT:    ret i32 [[OP_RDX1]]
 ;
 entry:
   %vecext = extractelement <4 x i32> %v1, i64 0
   %vecext1 = extractelement <4 x i32> %v1, i64 1
   %vecext2 = extractelement <4 x i32> %v1, i64 2
   %vecext4 = extractelement <4 x i32> %v1, i64 3

 ;   acc &= v1[1] & v2[1] & v3[1] & v4[1];
 ;   acc &= v1[2] & v2[2] & v3[2] & v4[2];
 ;   acc &= v1[3] & v2[3] & v3[3] & v4[3];
 ;   return acc;
 ; }
 
 define i32 @reduce_and4_transpose(i32 %acc, <4 x i32> %v1, <4 x i32> %v2, <4 x i32> %v3, <4 x i32> %v4) {
 ; SSE2-LABEL: @reduce_and4_transpose(
-; SSE2-NEXT:    [[VECEXT:%.*]] = extractelement <4 x i32> [[V1:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT1:%.*]] = extractelement <4 x i32> [[V2:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT2:%.*]] = extractelement <4 x i32> [[V3:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT4:%.*]] = extractelement <4 x i32> [[V4:%.*]], i64 0
-; SSE2-NEXT:    [[VECEXT7:%.*]] = extractelement <4 x i32> [[V1]], i64 1
-; SSE2-NEXT:    [[VECEXT8:%.*]] = extractelement <4 x i32> [[V2]], i64 1
-; SSE2-NEXT:    [[VECEXT10:%.*]] = extractelement <4 x i32> [[V3]], i64 1
-; SSE2-NEXT:    [[VECEXT12:%.*]] = extractelement <4 x i32> [[V4]], i64 1
-; SSE2-NEXT:    [[VECEXT15:%.*]] = extractelement <4 x i32> [[V1]], i64 2
-; SSE2-NEXT:    [[VECEXT16:%.*]] = extractelement <4 x i32> [[V2]], i64 2
-; SSE2-NEXT:    [[VECEXT18:%.*]] = extractelement <4 x i32> [[V3]], i64 2
-; SSE2-NEXT:    [[VECEXT20:%.*]] = extractelement <4 x i32> [[V4]], i64 2
-; SSE2-NEXT:    [[VECEXT23:%.*]] = extractelement <4 x i32> [[V1]], i64 3
-; SSE2-NEXT:    [[VECEXT24:%.*]] = extractelement <4 x i32> [[V2]], i64 3
-; SSE2-NEXT:    [[TMP1:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT24]], i32 0
-; SSE2-NEXT:    [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[VECEXT16]], i32 1
-; SSE2-NEXT:    [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[VECEXT8]], i32 2
-; SSE2-NEXT:    [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[VECEXT1]], i32 3
-; SSE2-NEXT:    [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[VECEXT23]], i32 4
-; SSE2-NEXT:    [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[VECEXT15]], i32 5
-; SSE2-NEXT:    [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[VECEXT7]], i32 6
-; SSE2-NEXT:    [[TMP8:%.*]] = insertelement <8 x i32> [[TMP7]], i32 [[VECEXT]], i32 7
-; SSE2-NEXT:    [[VECEXT26:%.*]] = extractelement <4 x i32> [[V3]], i64 3
-; SSE2-NEXT:    [[VECEXT28:%.*]] = extractelement <4 x i32> [[V4]], i64 3
-; SSE2-NEXT:    [[TMP9:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT28]], i32 0
-; SSE2-NEXT:    [[TMP10:%.*]] = insertelement <8 x i32> [[TMP9]], i32 [[VECEXT20]], i32 1
-; SSE2-NEXT:    [[TMP11:%.*]] = insertelement <8 x i32> [[TMP10]], i32 [[VECEXT12]], i32 2
-; SSE2-NEXT:    [[TMP12:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[VECEXT4]], i32 3
-; SSE2-NEXT:    [[TMP13:%.*]] = insertelement <8 x i32> [[TMP12]], i32 [[VECEXT26]], i32 4
-; SSE2-NEXT:    [[TMP14:%.*]] = insertelement <8 x i32> [[TMP13]], i32 [[VECEXT18]], i32 5
-; SSE2-NEXT:    [[TMP15:%.*]] = insertelement <8 x i32> [[TMP14]], i32 [[VECEXT10]], i32 6
-; SSE2-NEXT:    [[TMP16:%.*]] = insertelement <8 x i32> [[TMP15]], i32 [[VECEXT2]], i32 7
-; SSE2-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP16]])
-; SSE2-NEXT:    [[TMP18:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP8]])
-; SSE2-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP17]], [[TMP18]]
+; SSE2-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[V2:%.*]], <4 x i32> [[V1:%.*]], <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
+; SSE2-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i32> [[V4:%.*]], <4 x i32> [[V3:%.*]], <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
+; SSE2-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP2]])
+; SSE2-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP1]])
+; SSE2-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP3]], [[TMP4]]
 ; SSE2-NEXT:    [[OP_RDX1:%.*]] = and i32 [[OP_RDX]], [[ACC:%.*]]
 ; SSE2-NEXT:    ret i32 [[OP_RDX1]]
 ;
 ; SSE42-LABEL: @reduce_and4_transpose(
 ; SSE42-NEXT:    [[TMP1:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V4:%.*]])
 ; SSE42-NEXT:    [[TMP2:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V3:%.*]])
 ; SSE42-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP1]], [[TMP2]]
 ; SSE42-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V2:%.*]])
 ; SSE42-NEXT:    [[OP_RDX1:%.*]] = and i32 [[OP_RDX]], [[TMP3]]
 ; SSE42-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[V1:%.*]])
 ; SSE42-NEXT:    [[OP_RDX2:%.*]] = and i32 [[OP_RDX1]], [[TMP4]]
 ; SSE42-NEXT:    [[OP_RDX3:%.*]] = and i32 [[OP_RDX2]], [[ACC:%.*]]
 ; SSE42-NEXT:    ret i32 [[OP_RDX3]]
 ;
 ; AVX-LABEL: @reduce_and4_transpose(
-; AVX-NEXT:    [[VECEXT:%.*]] = extractelement <4 x i32> [[V1:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT1:%.*]] = extractelement <4 x i32> [[V2:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT2:%.*]] = extractelement <4 x i32> [[V3:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT4:%.*]] = extractelement <4 x i32> [[V4:%.*]], i64 0
-; AVX-NEXT:    [[VECEXT7:%.*]] = extractelement <4 x i32> [[V1]], i64 1
-; AVX-NEXT:    [[VECEXT8:%.*]] = extractelement <4 x i32> [[V2]], i64 1
-; AVX-NEXT:    [[VECEXT10:%.*]] = extractelement <4 x i32> [[V3]], i64 1
-; AVX-NEXT:    [[VECEXT12:%.*]] = extractelement <4 x i32> [[V4]], i64 1
-; AVX-NEXT:    [[VECEXT15:%.*]] = extractelement <4 x i32> [[V1]], i64 2
-; AVX-NEXT:    [[VECEXT16:%.*]] = extractelement <4 x i32> [[V2]], i64 2
-; AVX-NEXT:    [[VECEXT18:%.*]] = extractelement <4 x i32> [[V3]], i64 2
-; AVX-NEXT:    [[VECEXT20:%.*]] = extractelement <4 x i32> [[V4]], i64 2
-; AVX-NEXT:    [[VECEXT23:%.*]] = extractelement <4 x i32> [[V1]], i64 3
-; AVX-NEXT:    [[VECEXT24:%.*]] = extractelement <4 x i32> [[V2]], i64 3
-; AVX-NEXT:    [[TMP1:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT24]], i32 0
-; AVX-NEXT:    [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[VECEXT16]], i32 1
-; AVX-NEXT:    [[TMP3:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[VECEXT8]], i32 2
-; AVX-NEXT:    [[TMP4:%.*]] = insertelement <8 x i32> [[TMP3]], i32 [[VECEXT1]], i32 3
-; AVX-NEXT:    [[TMP5:%.*]] = insertelement <8 x i32> [[TMP4]], i32 [[VECEXT23]], i32 4
-; AVX-NEXT:    [[TMP6:%.*]] = insertelement <8 x i32> [[TMP5]], i32 [[VECEXT15]], i32 5
-; AVX-NEXT:    [[TMP7:%.*]] = insertelement <8 x i32> [[TMP6]], i32 [[VECEXT7]], i32 6
-; AVX-NEXT:    [[TMP8:%.*]] = insertelement <8 x i32> [[TMP7]], i32 [[VECEXT]], i32 7
-; AVX-NEXT:    [[VECEXT26:%.*]] = extractelement <4 x i32> [[V3]], i64 3
-; AVX-NEXT:    [[VECEXT28:%.*]] = extractelement <4 x i32> [[V4]], i64 3
-; AVX-NEXT:    [[TMP9:%.*]] = insertelement <8 x i32> poison, i32 [[VECEXT28]], i32 0
-; AVX-NEXT:    [[TMP10:%.*]] = insertelement <8 x i32> [[TMP9]], i32 [[VECEXT20]], i32 1
-; AVX-NEXT:    [[TMP11:%.*]] = insertelement <8 x i32> [[TMP10]], i32 [[VECEXT12]], i32 2
-; AVX-NEXT:    [[TMP12:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[VECEXT4]], i32 3
-; AVX-NEXT:    [[TMP13:%.*]] = insertelement <8 x i32> [[TMP12]], i32 [[VECEXT26]], i32 4
-; AVX-NEXT:    [[TMP14:%.*]] = insertelement <8 x i32> [[TMP13]], i32 [[VECEXT18]], i32 5
-; AVX-NEXT:    [[TMP15:%.*]] = insertelement <8 x i32> [[TMP14]], i32 [[VECEXT10]], i32 6
-; AVX-NEXT:    [[TMP16:%.*]] = insertelement <8 x i32> [[TMP15]], i32 [[VECEXT2]], i32 7
-; AVX-NEXT:    [[TMP17:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP16]])
-; AVX-NEXT:    [[TMP18:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP8]])
-; AVX-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP17]], [[TMP18]]
+; AVX-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[V2:%.*]], <4 x i32> [[V1:%.*]], <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
+; AVX-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i32> [[V4:%.*]], <4 x i32> [[V3:%.*]], <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
+; AVX-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP2]])
+; AVX-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.and.v8i32(<8 x i32> [[TMP1]])
+; AVX-NEXT:    [[OP_RDX:%.*]] = and i32 [[TMP3]], [[TMP4]]
 ; AVX-NEXT:    [[OP_RDX1:%.*]] = and i32 [[OP_RDX]], [[ACC:%.*]]
 ; AVX-NEXT:    ret i32 [[OP_RDX1]]
 ;
   %vecext = extractelement <4 x i32> %v1, i64 0
   %vecext1 = extractelement <4 x i32> %v2, i64 0
   %vecext2 = extractelement <4 x i32> %v3, i64 0
   %vecext4 = extractelement <4 x i32> %v4, i64 0
   %vecext7 = extractelement <4 x i32> %v1, i64 1

llvm/test/Transforms/SLPVectorizer/X86/remark_extract_broadcast.ll

 ; CHECK-NEXT:    store <8 x i16> [[TMP0]], ptr [[PTR:%.*]], align 2
 ; CHECK-NEXT:    ret void
 ;
 ; YAML:      Pass:            slp-vectorizer
 ; YAML-NEXT: Name:            StoresVectorized
 ; YAML-NEXT: Function:        fextr
 ; YAML-NEXT: Args:
 ; YAML-NEXT:   - String:          'Stores SLP vectorized with cost '
-; YAML-NEXT:   - Cost:            '-20'
+; YAML-NEXT:   - Cost:            '-19'
 ; YAML-NEXT:   - String:          ' and with tree size '
 ; YAML-NEXT:   - TreeSize:        '4'
 
 entry:
   %LD = load <8 x i16>, ptr undef
   %V0 = extractelement <8 x i16> %LD, i32 0
   br label %t