[SLP] Fix PR35777: Incorrect handling of aggregate values.

[ABataev]

Summary

Fixes the bug with incorrect handling of InsertValue|InsertElement
instrucions in SLP vectorizer. Currently, we may use incorrect
ExtractElement instructions as the operands of the original
InsertValue|InsertElement instructions.

[RKSimon]

LGTM

llvm/trunk/include/llvm/Transforms/Vectorize/SLPVectorizer.h

private:
   /// TODO: We can further reduce this cost if we flush the chain creation
   ///       every time we run into a memory barrier.
   void collectSeedInstructions(BasicBlock *BB);
 
   /// \brief Try to vectorize a chain that starts at two arithmetic instrs.
   bool tryToVectorizePair(Value *A, Value *B, slpvectorizer::BoUpSLP &R);
 
   /// \brief Try to vectorize a list of operands.
-  /// \@param BuildVector A list of users to ignore for the purpose of
-  ///                     scheduling and cost estimation when NeedExtraction
-  ///                     is false.
   /// \returns true if a value was vectorized.
   bool tryToVectorizeList(ArrayRef<Value *> VL, slpvectorizer::BoUpSLP &R,
-                          ArrayRef<Value *> BuildVector = None,
-                          bool AllowReorder = false,
-                          bool NeedExtraction = false);
+                          bool AllowReorder = false);
 
   /// \brief Try to vectorize a chain that may start at the operands of \p I.
   bool tryToVectorize(Instruction *I, slpvectorizer::BoUpSLP &R);
 
   /// \brief Vectorize the store instructions collected in Stores.
   bool vectorizeStoreChains(slpvectorizer::BoUpSLP &R);
 
   /// \brief Vectorize the index computations of the getelementptr instructions

llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp

for (Instruction &I : *BB) {
     }
   }
 }
 
 bool SLPVectorizerPass::tryToVectorizePair(Value *A, Value *B, BoUpSLP &R) {
   if (!A || !B)
     return false;
   Value *VL[] = { A, B };
-  return tryToVectorizeList(VL, R, None, true);
+  return tryToVectorizeList(VL, R, true);
 }
 
 bool SLPVectorizerPass::tryToVectorizeList(ArrayRef<Value *> VL, BoUpSLP &R,
-                                           ArrayRef<Value *> BuildVector,
-                                           bool AllowReorder,
-                                           bool NeedExtraction) {
+                                           bool AllowReorder) {
   if (VL.size() < 2)
     return false;
 
   DEBUG(dbgs() << "SLP: Trying to vectorize a list of length = " << VL.size()
                << ".\n");
 
   // Check that all of the parts are scalar instructions of the same type.
   Instruction *I0 = dyn_cast<Instruction>(VL[0]);
for (unsigned I = NextInst; I < MaxInst; ++I) {
       // Check that a previous iteration of this loop did not delete the Value.
       if (hasValueBeenRAUWed(VL, TrackValues, I, OpsWidth))
         continue;
 
       DEBUG(dbgs() << "SLP: Analyzing " << OpsWidth << " operations "
                    << "\n");
       ArrayRef<Value *> Ops = VL.slice(I, OpsWidth);
 
-      ArrayRef<Value *> EmptyArray;
-      ArrayRef<Value *> BuildVectorSlice;
-      if (!BuildVector.empty())
-        BuildVectorSlice = BuildVector.slice(I, OpsWidth);
-
-      R.buildTree(Ops, NeedExtraction ? EmptyArray : BuildVectorSlice);
+      R.buildTree(Ops);
       // TODO: check if we can allow reordering for more cases.
       if (AllowReorder && R.shouldReorder()) {
         // Conceptually, there is nothing actually preventing us from trying to
         // reorder a larger list. In fact, we do exactly this when vectorizing
         // reductions. However, at this point, we only expect to get here when
         // there are exactly two operations.
         assert(Ops.size() == 2);
-        assert(BuildVectorSlice.empty());
         Value *ReorderedOps[] = {Ops[1], Ops[0]};
         R.buildTree(ReorderedOps, None);
       }
       if (R.isTreeTinyAndNotFullyVectorizable())
         continue;
 
       R.computeMinimumValueSizes();
       int Cost = R.getTreeCost();
       CandidateFound = true;
       MinCost = std::min(MinCost, Cost);
 
       if (Cost < -SLPCostThreshold) {
         DEBUG(dbgs() << "SLP: Vectorizing list at cost:" << Cost << ".\n");
         R.getORE()->emit(OptimizationRemark(SV_NAME, "VectorizedList",
                                                     cast<Instruction>(Ops[0]))
                                  << "SLP vectorized with cost " << ore::NV("Cost", Cost)
                                  << " and with tree size "
                                  << ore::NV("TreeSize", R.getTreeSize()));
 
-        Value *VectorizedRoot = R.vectorizeTree();
-
-        // Reconstruct the build vector by extracting the vectorized root. This
-        // way we handle the case where some elements of the vector are
-        // undefined.
-        //  (return (inserelt <4 xi32> (insertelt undef (opd0) 0) (opd1) 2))
-        if (!BuildVectorSlice.empty()) {
-          // The insert point is the last build vector instruction. The
-          // vectorized root will precede it. This guarantees that we get an
-          // instruction. The vectorized tree could have been constant folded.
-          Instruction *InsertAfter = cast<Instruction>(BuildVectorSlice.back());
-          unsigned VecIdx = 0;
-          for (auto &V : BuildVectorSlice) {
-            IRBuilder<NoFolder> Builder(InsertAfter->getParent(),
-                                        ++BasicBlock::iterator(InsertAfter));
-            Instruction *I = cast<Instruction>(V);
-            assert(isa<InsertElementInst>(I) || isa<InsertValueInst>(I));
-            Instruction *Extract =
-                cast<Instruction>(Builder.CreateExtractElement(
-                    VectorizedRoot, Builder.getInt32(VecIdx++)));
-            I->setOperand(1, Extract);
-            I->moveAfter(Extract);
-            InsertAfter = I;
-          }
-        }
+        R.vectorizeTree();
         // Move to the next bundle.
         I += VF - 1;
         NextInst = I + 1;
         Changed = true;
       }
     }
   }
 

 ///  %ra = insertelement <4 x float> undef, float %s0, i32 0
 ///  %rb = insertelement <4 x float> %ra, float %s1, i32 1
 ///  %rc = insertelement <4 x float> %rb, float %s2, i32 2
 ///  %rd = insertelement <4 x float> %rc, float %s3, i32 3
 ///  starting from the last insertelement instruction.
 ///
 /// Returns true if it matches
 static bool findBuildVector(InsertElementInst *LastInsertElem,
-                            SmallVectorImpl<Value *> &BuildVector,
                             SmallVectorImpl<Value *> &BuildVectorOpds) {
   Value *V = nullptr;
   do {
-    BuildVector.push_back(LastInsertElem);
     BuildVectorOpds.push_back(LastInsertElem->getOperand(1));
     V = LastInsertElem->getOperand(0);
     if (isa<UndefValue>(V))
       break;
     LastInsertElem = dyn_cast<InsertElementInst>(V);
     if (!LastInsertElem || !LastInsertElem->hasOneUse())
       return false;
   } while (true);
-  std::reverse(BuildVector.begin(), BuildVector.end());
   std::reverse(BuildVectorOpds.begin(), BuildVectorOpds.end());
   return true;
 }
 
 /// \brief Like findBuildVector, but looks for construction of aggregate.
 ///
 /// \return true if it matches.
 static bool findBuildAggregate(InsertValueInst *IV,
-                               SmallVectorImpl<Value *> &BuildVector,
                                SmallVectorImpl<Value *> &BuildVectorOpds) {
   Value *V;
   do {
-    BuildVector.push_back(IV);
     BuildVectorOpds.push_back(IV->getInsertedValueOperand());
     V = IV->getAggregateOperand();
     if (isa<UndefValue>(V))
       break;
     IV = dyn_cast<InsertValueInst>(V);
     if (!IV || !IV->hasOneUse())
       return false;
   } while (true);
-  std::reverse(BuildVector.begin(), BuildVector.end());
   std::reverse(BuildVectorOpds.begin(), BuildVectorOpds.end());
   return true;
 }
 
 static bool PhiTypeSorterFunc(Value *V, Value *V2) {
   return V->getType() < V2->getType();
 }
 

 }
 
 bool SLPVectorizerPass::vectorizeInsertValueInst(InsertValueInst *IVI,
                                                  BasicBlock *BB, BoUpSLP &R) {
   const DataLayout &DL = BB->getModule()->getDataLayout();
   if (!R.canMapToVector(IVI->getType(), DL))
     return false;
 
-  SmallVector<Value *, 16> BuildVector;
   SmallVector<Value *, 16> BuildVectorOpds;
-  if (!findBuildAggregate(IVI, BuildVector, BuildVectorOpds))
+  if (!findBuildAggregate(IVI, BuildVectorOpds))
     return false;
 
   DEBUG(dbgs() << "SLP: array mappable to vector: " << *IVI << "\n");
   // Aggregate value is unlikely to be processed in vector register, we need to
   // extract scalars into scalar registers, so NeedExtraction is set true.
-  return tryToVectorizeList(BuildVectorOpds, R, BuildVector, false, true);
+  return tryToVectorizeList(BuildVectorOpds, R);
 }
 
 bool SLPVectorizerPass::vectorizeInsertElementInst(InsertElementInst *IEI,
                                                    BasicBlock *BB, BoUpSLP &R) {
-  SmallVector<Value *, 16> BuildVector;
   SmallVector<Value *, 16> BuildVectorOpds;
-  if (!findBuildVector(IEI, BuildVector, BuildVectorOpds))
+  if (!findBuildVector(IEI, BuildVectorOpds))
     return false;
 
   // Vectorize starting with the build vector operands ignoring the BuildVector
   // instructions for the purpose of scheduling and user extraction.
-  return tryToVectorizeList(BuildVectorOpds, R, BuildVector);
+  return tryToVectorizeList(BuildVectorOpds, R);
 }
 
 bool SLPVectorizerPass::vectorizeCmpInst(CmpInst *CI, BasicBlock *BB,
                                          BoUpSLP &R) {
   if (tryToVectorizePair(CI->getOperand(0), CI->getOperand(1), R))
     return true;
 
   bool OpsChanged = false;
for (SmallVector<Value *, 4>::iterator IncIt = Incoming.begin(),
       // Try to vectorize them.
       unsigned NumElts = (SameTypeIt - IncIt);
       DEBUG(errs() << "SLP: Trying to vectorize starting at PHIs (" << NumElts << ")\n");
       // The order in which the phi nodes appear in the program does not matter.
       // So allow tryToVectorizeList to reorder them if it is beneficial. This
       // is done when there are exactly two elements since tryToVectorizeList
       // asserts that there are only two values when AllowReorder is true.
       bool AllowReorder = NumElts == 2;
-      if (NumElts > 1 && tryToVectorizeList(makeArrayRef(IncIt, NumElts), R,
-                                            None, AllowReorder)) {
+      if (NumElts > 1 &&
+          tryToVectorizeList(makeArrayRef(IncIt, NumElts), R, AllowReorder)) {
         // Success start over because instructions might have been changed.
         HaveVectorizedPhiNodes = true;
         Changed = true;
         break;
       }
 
       // Start over at the next instruction of a different type (or the end).
       IncIt = SameTypeIt;

llvm/trunk/test/Transforms/SLPVectorizer/X86/PR35777.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -verify -slp-vectorizer -o - -S -mtriple=x86_64-apple-macosx10.13.0 | FileCheck %s
+
+@global = local_unnamed_addr global [6 x double] zeroinitializer, align 16
+
+define { i64, i64 } @patatino(double %arg) {
+; CHECK-LABEL: @patatino(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    [[TMP0:%.*]] = load <2 x double>, <2 x double>* bitcast ([6 x double]* @global to <2 x double>*), align 16
+; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 2) to <2 x double>*), align 16
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> undef, double [[ARG:%.*]], i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> [[TMP2]], double [[ARG]], i32 1
+; CHECK-NEXT:    [[TMP4:%.*]] = fmul <2 x double> [[TMP3]], [[TMP1]]
+; CHECK-NEXT:    [[TMP5:%.*]] = fadd <2 x double> [[TMP0]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = load <2 x double>, <2 x double>* bitcast (double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 4) to <2 x double>*), align 16
+; CHECK-NEXT:    [[TMP7:%.*]] = fadd <2 x double> [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP8:%.*]] = fptosi <2 x double> [[TMP7]] to <2 x i32>
+; CHECK-NEXT:    [[TMP9:%.*]] = sext <2 x i32> [[TMP8]] to <2 x i64>
+; CHECK-NEXT:    [[TMP10:%.*]] = trunc <2 x i64> [[TMP9]] to <2 x i32>
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i32> [[TMP10]], i32 0
+; CHECK-NEXT:    [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
+; CHECK-NEXT:    [[TMP16:%.*]] = insertvalue { i64, i64 } undef, i64 [[TMP12]], 0
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i32> [[TMP10]], i32 1
+; CHECK-NEXT:    [[TMP14:%.*]] = sext i32 [[TMP13]] to i64
+; CHECK-NEXT:    [[TMP17:%.*]] = insertvalue { i64, i64 } [[TMP16]], i64 [[TMP14]], 1
+; CHECK-NEXT:    ret { i64, i64 } [[TMP17]]
+;
+bb:
+  %tmp = load double, double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 0), align 16
+  %tmp1 = load double, double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 2), align 16
+  %tmp2 = fmul double %tmp1, %arg
+  %tmp3 = fadd double %tmp, %tmp2
+  %tmp4 = load double, double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 4), align 16
+  %tmp5 = fadd double %tmp4, %tmp3
+  %tmp6 = fptosi double %tmp5 to i32
+  %tmp7 = sext i32 %tmp6 to i64
+  %tmp8 = load double, double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 1), align 8
+  %tmp9 = load double, double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 3), align 8
+  %tmp10 = fmul double %tmp9, %arg
+  %tmp11 = fadd double %tmp8, %tmp10
+  %tmp12 = load double, double* getelementptr inbounds ([6 x double], [6 x double]* @global, i64 0, i64 5), align 8
+  %tmp13 = fadd double %tmp12, %tmp11
+  %tmp14 = fptosi double %tmp13 to i32
+  %tmp15 = sext i32 %tmp14 to i64
+  %tmp16 = insertvalue { i64, i64 } undef, i64 %tmp7, 0
+  %tmp17 = insertvalue { i64, i64 } %tmp16, i64 %tmp15, 1
+  ret { i64, i64 } %tmp17
+}

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

;
   %rd = insertelement <4 x float> %rc, float %s3, i32 3
   ret <4 x float> %rd
 }
 
 ; Make sure infinite loop doesn't happen which I ran into when trying
 ; to do this backwards this backwards
 define <4 x i32> @reconstruct(<4 x i32> %c) #0 {
 ; CHECK-LABEL: @reconstruct(
-; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
-; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[TMP1]], i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x i32> [[C]], i32 1
-; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[TMP2]], i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[C]], i32 2
-; CHECK-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[TMP3]], i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[C]], i32 3
-; CHECK-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[TMP4]], i32 3
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 3
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x i32> [[C]], i32 2
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[C]], i32 1
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[C]], i32 0
+; CHECK-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[TMP4]], i32 0
+; CHECK-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[TMP3]], i32 1
+; CHECK-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[TMP2]], i32 2
+; CHECK-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[TMP1]], i32 3
 ; CHECK-NEXT:    ret <4 x i32> [[RD]]
 ;
 ; ZEROTHRESH-LABEL: @reconstruct(
-; ZEROTHRESH-NEXT:    [[TMP1:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
-; ZEROTHRESH-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[TMP1]], i32 0
-; ZEROTHRESH-NEXT:    [[TMP2:%.*]] = extractelement <4 x i32> [[C]], i32 1
-; ZEROTHRESH-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[TMP2]], i32 1
-; ZEROTHRESH-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[C]], i32 2
-; ZEROTHRESH-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[TMP3]], i32 2
-; ZEROTHRESH-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[C]], i32 3
-; ZEROTHRESH-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[TMP4]], i32 3
+; ZEROTHRESH-NEXT:    [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0
+; ZEROTHRESH-NEXT:    [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1
+; ZEROTHRESH-NEXT:    [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2
+; ZEROTHRESH-NEXT:    [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3
+; ZEROTHRESH-NEXT:    [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[C0]], i32 0
+; ZEROTHRESH-NEXT:    [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[C1]], i32 1
+; ZEROTHRESH-NEXT:    [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[C2]], i32 2
+; ZEROTHRESH-NEXT:    [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[C3]], i32 3
 ; ZEROTHRESH-NEXT:    ret <4 x i32> [[RD]]
 ;
   %c0 = extractelement <4 x i32> %c, i32 0
   %c1 = extractelement <4 x i32> %c, i32 1
   %c2 = extractelement <4 x i32> %c, i32 2
   %c3 = extractelement <4 x i32> %c, i32 3
   %ra = insertelement <4 x i32> undef, i32 %c0, i32 0
   %rb = insertelement <4 x i32> %ra, i32 %c1, i32 1

llvm/trunk/test/Transforms/SLPVectorizer/X86/insertvalue.ll

 ; CHECK-NEXT:    [[TMP7:%.*]] = load <2 x double>, <2 x double>* [[TMP6]], align 4
 ; CHECK-NEXT:    [[TMP8:%.*]] = fmul <2 x double> [[TMP5]], [[TMP7]]
 ; CHECK-NEXT:    [[PZ0:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP1:%.*]], i64 0, i64 0
 ; CHECK-NEXT:    [[PZ1:%.*]] = getelementptr inbounds [2 x double], [2 x double]* [[TMP1]], i64 0, i64 1
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast double* [[PZ0]] to <2 x double>*
 ; CHECK-NEXT:    [[TMP10:%.*]] = load <2 x double>, <2 x double>* [[TMP9]], align 4
 ; CHECK-NEXT:    [[TMP11:%.*]] = fadd <2 x double> [[TMP8]], [[TMP10]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x double> [[TMP11]], i32 0
-; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x double> [[TMP11]], i32 1
 ; CHECK-NEXT:    [[I0:%.*]] = insertvalue [2 x double] undef, double [[TMP12]], 0
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x double> [[TMP11]], i32 1
 ; CHECK-NEXT:    [[I1:%.*]] = insertvalue [2 x double] [[I0]], double [[TMP13]], 1
 ; CHECK-NEXT:    store [2 x double] [[I1]], [2 x double]* [[TMP0:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 top:
   %px0 = getelementptr inbounds [2 x double], [2 x double]* %2, i64 0, i64 0
   %x0 = load double, double* %px0, align 4
   %py0 = getelementptr inbounds [2 x double], [2 x double]* %3, i64 0, i64 0

 ; CHECK-NEXT:    [[PZ0:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1:%.*]], i64 0, i64 0
 ; CHECK-NEXT:    [[PZ1:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1]], i64 0, i64 1
 ; CHECK-NEXT:    [[PZ2:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1]], i64 0, i64 2
 ; CHECK-NEXT:    [[PZ3:%.*]] = getelementptr inbounds [4 x float], [4 x float]* [[TMP1]], i64 0, i64 3
 ; CHECK-NEXT:    [[TMP9:%.*]] = bitcast float* [[PZ0]] to <4 x float>*
 ; CHECK-NEXT:    [[TMP10:%.*]] = load <4 x float>, <4 x float>* [[TMP9]], align 4
 ; CHECK-NEXT:    [[TMP11:%.*]] = fadd <4 x float> [[TMP8]], [[TMP10]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x float> [[TMP11]], i32 0
-; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <4 x float> [[TMP11]], i32 1
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x float> [[TMP11]], i32 2
-; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x float> [[TMP11]], i32 3
 ; CHECK-NEXT:    [[I0:%.*]] = insertvalue [4 x float] undef, float [[TMP12]], 0
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <4 x float> [[TMP11]], i32 1
 ; CHECK-NEXT:    [[I1:%.*]] = insertvalue [4 x float] [[I0]], float [[TMP13]], 1
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x float> [[TMP11]], i32 2
 ; CHECK-NEXT:    [[I2:%.*]] = insertvalue [4 x float] [[I1]], float [[TMP14]], 2
+; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x float> [[TMP11]], i32 3
 ; CHECK-NEXT:    [[I3:%.*]] = insertvalue [4 x float] [[I2]], float [[TMP15]], 3
 ; CHECK-NEXT:    store [4 x float] [[I3]], [4 x float]* [[TMP0:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 top:
   %px0 = getelementptr inbounds [4 x float], [4 x float]* %2, i64 0, i64 0
   %x0 = load float, float* %px0, align 4
   %py0 = getelementptr inbounds [4 x float], [4 x float]* %3, i64 0, i64 0

 ; CHECK-NEXT:    [[TMP0:%.*]] = bitcast [4 x float]* [[A:%.*]] to <4 x float>*
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
 ; CHECK-NEXT:    [[A_ARR:%.*]] = load [4 x float], [4 x float]* [[A]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [4 x float]* [[B:%.*]] to <4 x float>*
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
 ; CHECK-NEXT:    [[B_ARR:%.*]] = load [4 x float], [4 x float]* [[B]], align 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = fsub <4 x float> [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP4]], i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP4]], i32 1
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP4]], i32 2
-; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[TMP4]], i32 3
 ; CHECK-NEXT:    [[C_ARR0:%.*]] = insertvalue [4 x float] undef, float [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP4]], i32 1
 ; CHECK-NEXT:    [[C_ARR1:%.*]] = insertvalue [4 x float] [[C_ARR0]], float [[TMP6]], 1
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP4]], i32 2
 ; CHECK-NEXT:    [[C_ARR2:%.*]] = insertvalue [4 x float] [[C_ARR1]], float [[TMP7]], 2
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[TMP4]], i32 3
 ; CHECK-NEXT:    [[C_ARR3:%.*]] = insertvalue [4 x float] [[C_ARR2]], float [[TMP8]], 3
 ; CHECK-NEXT:    store [4 x float] [[C_ARR3]], [4 x float]* [[C:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 top:
   %a_arr = load [4 x float], [4 x float]* %a, align 4
   %a0 = extractvalue [4 x float] %a_arr, 0
   %a2 = extractvalue [4 x float] %a_arr, 2

 ; CHECK-NEXT:    [[TMP0:%.*]] = bitcast [4 x i32]* [[A:%.*]] to <4 x i32>*
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
 ; CHECK-NEXT:    [[A_ARR:%.*]] = load [4 x i32], [4 x i32]* [[A]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast [4 x i32]* [[B:%.*]] to <4 x i32>*
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
 ; CHECK-NEXT:    [[B_ARR:%.*]] = load [4 x i32], [4 x i32]* [[B]], align 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = sub <4 x i32> [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i32> [[TMP4]], i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x i32> [[TMP4]], i32 1
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP4]], i32 2
-; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x i32> [[TMP4]], i32 3
 ; CHECK-NEXT:    [[C_ARR0:%.*]] = insertvalue [4 x i32] undef, i32 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x i32> [[TMP4]], i32 1
 ; CHECK-NEXT:    [[C_ARR1:%.*]] = insertvalue [4 x i32] [[C_ARR0]], i32 [[TMP6]], 1
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[TMP4]], i32 2
 ; CHECK-NEXT:    [[C_ARR2:%.*]] = insertvalue [4 x i32] [[C_ARR1]], i32 [[TMP7]], 2
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x i32> [[TMP4]], i32 3
 ; CHECK-NEXT:    [[C_ARR3:%.*]] = insertvalue [4 x i32] [[C_ARR2]], i32 [[TMP8]], 3
 ; CHECK-NEXT:    store [4 x i32] [[C_ARR3]], [4 x i32]* [[C:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 top:
   %a_arr = load [4 x i32], [4 x i32]* %a, align 4
   %a0 = extractvalue [4 x i32] %a_arr, 0
   %a2 = extractvalue [4 x i32] %a_arr, 2

 ; CHECK-NEXT:    [[TMP0:%.*]] = bitcast %pseudovec* [[A:%.*]] to <4 x float>*
 ; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
 ; CHECK-NEXT:    [[A_STRUCT:%.*]] = load [[PSEUDOVEC:%.*]], %pseudovec* [[A]], align 4
 ; CHECK-NEXT:    [[TMP2:%.*]] = bitcast %pseudovec* [[B:%.*]] to <4 x float>*
 ; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
 ; CHECK-NEXT:    [[B_STRUCT:%.*]] = load [[PSEUDOVEC]], %pseudovec* [[B]], align 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = fsub <4 x float> [[TMP1]], [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[TMP4]], i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP4]], i32 1
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP4]], i32 2
-; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[TMP4]], i32 3
 ; CHECK-NEXT:    [[C_STRUCT0:%.*]] = insertvalue [[PSEUDOVEC]] undef, float [[TMP5]], 0
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP4]], i32 1
 ; CHECK-NEXT:    [[C_STRUCT1:%.*]] = insertvalue [[PSEUDOVEC]] %c_struct0, float [[TMP6]], 1
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP4]], i32 2
 ; CHECK-NEXT:    [[C_STRUCT2:%.*]] = insertvalue [[PSEUDOVEC]] %c_struct1, float [[TMP7]], 2
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <4 x float> [[TMP4]], i32 3
 ; CHECK-NEXT:    [[C_STRUCT3:%.*]] = insertvalue [[PSEUDOVEC]] %c_struct2, float [[TMP8]], 3
 ; CHECK-NEXT:    store [[PSEUDOVEC]] %c_struct3, %pseudovec* [[C:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 top:
   %a_struct = load %pseudovec, %pseudovec* %a, align 4
   %a0 = extractvalue %pseudovec %a_struct, 0
   %a1 = extractvalue %pseudovec %a_struct, 1

llvm/trunk/test/Transforms/SLPVectorizer/X86/value-bug.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -slp-vectorizer < %s -S -mtriple="x86_64-grtev3-linux-gnu" -mcpu=corei7-avx | FileCheck %s
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 
 ; We used to crash on this example because we were building a constant
 ; expression during vectorization and the vectorizer expects instructions
 ; as elements of the vectorized tree.
 ; PR19621
 
 define void @test() {
 ; CHECK-LABEL: @test(
 ; CHECK-NEXT:  bb279:
 ; CHECK-NEXT:    br label [[BB283:%.*]]
 ; CHECK:       bb283:
 ; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x float> [ undef, [[BB279:%.*]] ], [ [[TMP11:%.*]], [[EXIT:%.*]] ]
-; CHECK-NEXT:    [[TMP1:%.*]] = phi <2 x float> [ undef, [[BB279]] ], [ [[TMP15:%.*]], [[EXIT]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = phi <2 x float> [ undef, [[BB279]] ], [ [[TMP13:%.*]], [[EXIT]] ]
 ; CHECK-NEXT:    br label [[BB284:%.*]]
 ; CHECK:       bb284:
 ; CHECK-NEXT:    [[TMP2:%.*]] = fpext <2 x float> [[TMP0]] to <2 x double>
 ; CHECK-NEXT:    [[TMP3:%.*]] = fsub <2 x double> [[TMP2]], undef
 ; CHECK-NEXT:    [[TMP4:%.*]] = fsub <2 x double> [[TMP3]], undef
 ; CHECK-NEXT:    br label [[BB21_I:%.*]]
 ; CHECK:       bb21.i:
 ; CHECK-NEXT:    br i1 undef, label [[BB22_I:%.*]], label [[EXIT]]
 ; CHECK:       bb22.i:
 ; CHECK-NEXT:    [[TMP5:%.*]] = fadd <2 x double> undef, [[TMP4]]
 ; CHECK-NEXT:    br label [[BB32_I:%.*]]
 ; CHECK:       bb32.i:
 ; CHECK-NEXT:    [[TMP6:%.*]] = phi <2 x double> [ [[TMP5]], [[BB22_I]] ], [ zeroinitializer, [[BB32_I]] ]
 ; CHECK-NEXT:    br i1 undef, label [[BB32_I]], label [[BB21_I]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    [[TMP7:%.*]] = fpext <2 x float> [[TMP1]] to <2 x double>
 ; CHECK-NEXT:    [[TMP8:%.*]] = fmul <2 x double> <double undef, double 0.000000e+00>, [[TMP7]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = fadd <2 x double> undef, [[TMP8]]
 ; CHECK-NEXT:    [[TMP10:%.*]] = fadd <2 x double> undef, [[TMP9]]
 ; CHECK-NEXT:    [[TMP11]] = fptrunc <2 x double> [[TMP10]] to <2 x float>
-; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x float> undef, i32 0
-; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x float> undef, float [[TMP12]], i32 0
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x float> undef, i32 1
-; CHECK-NEXT:    [[TMP15]] = insertelement <2 x float> [[TMP13]], float [[TMP14]], i32 1
+; CHECK-NEXT:    [[TMP317:%.*]] = fptrunc double undef to float
+; CHECK-NEXT:    [[TMP319:%.*]] = fptrunc double undef to float
+; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x float> undef, float [[TMP317]], i32 0
+; CHECK-NEXT:    [[TMP13]] = insertelement <2 x float> [[TMP12]], float [[TMP319]], i32 1
 ; CHECK-NEXT:    br label [[BB283]]
 ;
 bb279:
   br label %bb283
 
 bb283:
   %Av.sroa.8.0 = phi float [ undef, %bb279 ], [ %tmp315, %exit ]
   %Av.sroa.5.0 = phi float [ undef, %bb279 ], [ %tmp319, %exit ]

 }
 
 ; Make sure that we probably handle constant folded vectorized trees. The
 ; vectorizer starts at the type (%t2, %t3) and wil constant fold the tree.
 ; The code that handles insertelement instructions must handle this.
 define <4 x double> @constant_folding() {
 ; CHECK-LABEL: @constant_folding(
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = extractelement <2 x double> <double 1.000000e+00, double 2.000000e+00>, i32 0
-; CHECK-NEXT:    [[I1:%.*]] = insertelement <4 x double> undef, double [[TMP0]], i32 1
-; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> <double 1.000000e+00, double 2.000000e+00>, i32 1
-; CHECK-NEXT:    [[I2:%.*]] = insertelement <4 x double> [[I1]], double [[TMP1]], i32 0
+; CHECK-NEXT:    [[I1:%.*]] = insertelement <4 x double> undef, double 1.000000e+00, i32 1
+; CHECK-NEXT:    [[I2:%.*]] = insertelement <4 x double> [[I1]], double 2.000000e+00, i32 0
 ; CHECK-NEXT:    ret <4 x double> [[I2]]
 ;
 entry:
   %t0 = fadd double 1.000000e+00 , 0.000000e+00
   %t1 = fadd double 1.000000e+00 , 1.000000e+00
   %t2 = fmul double %t0, 1.000000e+00
   %i1 = insertelement <4 x double> undef, double %t2, i32 1
   %t3 = fmul double %t1, 1.000000e+00
   %i2 = insertelement <4 x double> %i1, double %t3, i32 0
   ret <4 x double> %i2
 }