[VectorCombine] Combine extract/insert from vector

insert (DstVec, (extract SrcVec, ExtIdx), InsIdx)
--> shuffle (DstVec, SrcVec, Mask)

This commit combines extract/insert on a vector into Shuffle with
vector.

Author: ParkHanbum

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

@@ -106,6 +106,7 @@ class VectorCombine {
                        Instruction &I);
   bool foldExtractExtract(Instruction &I);
   bool foldInsExtFNeg(Instruction &I);
+  bool foldInsExtVectorToShuffle(Instruction &I);
   bool foldBitcastShuffle(Instruction &I);
   bool scalarizeBinopOrCmp(Instruction &I);
   bool scalarizeVPIntrinsic(Instruction &I);
@@ -2678,6 +2679,48 @@ bool VectorCombine::shrinkType(llvm::Instruction &I) {
   return true;
 }
 
+/// insert (DstVec, (extract SrcVec, ExtIdx), InsIdx) -->
+/// shuffle (DstVec, SrcVec, Mask)
+bool VectorCombine::foldInsExtVectorToShuffle(Instruction &I) {
+  Value *DstVec, *SrcVec;
+  uint64_t ExtIdx, InsIdx;
+  if (!match(&I, m_InsertElt(m_Value(DstVec),
+                             m_OneUse(m_ExtractElt(m_Value(SrcVec),
+                                                   m_ConstantInt(ExtIdx))),
+                             m_ConstantInt(InsIdx))))
+    return false;
+
+  auto *VecTy = dyn_cast<FixedVectorType>(I.getType());
+  if (!VecTy || SrcVec->getType() != VecTy)
+    return false;
+
+  unsigned NumElts = VecTy->getNumElements();
+  if (ExtIdx >= NumElts)
+    return false;
+
+  SmallVector<int> Mask(NumElts);
+  std::iota(Mask.begin(), Mask.end(), 0);
+  Mask[InsIdx] = ExtIdx + NumElts;
+  // Cost
+  ExtractElementInst *Ext;
+  if ((Ext = dyn_cast<ExtractElementInst>(I.getOperand(0))) == nullptr)
+    Ext = dyn_cast<ExtractElementInst>(I.getOperand(1));
+
+  TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
+  InstructionCost OldCost =
+      TTI.getVectorInstrCost(*Ext, VecTy, CostKind, ExtIdx);
+  InstructionCost NewCost =
+      TTI.getShuffleCost(TargetTransformInfo::SK_Select, VecTy, Mask);
+
+  if (OldCost < NewCost)
+    return false;
+
+  Value *Shuf = Builder.CreateShuffleVector(DstVec, SrcVec, Mask);
+  replaceValue(I, *Shuf);
+
+  return true;
+}
+
 /// This is the entry point for all transforms. Pass manager differences are
 /// handled in the callers of this function.
 bool VectorCombine::run() {
@@ -2734,6 +2777,7 @@ bool VectorCombine::run() {
       switch (Opcode) {
       case Instruction::InsertElement:
         MadeChange |= foldInsExtFNeg(I);
+        MadeChange |= foldInsExtVectorToShuffle(I);
         break;
       case Instruction::ShuffleVector:
         MadeChange |= foldShuffleOfBinops(I);

llvm/test/Transforms/VectorCombine/X86/extract-binop-inseltpoison.ll

@@ -417,12 +417,18 @@ define float @ext14_ext15_fmul_v16f32(<16 x float> %x) {
 }
 
 define <4 x float> @ins_bo_ext_ext(<4 x float> %a, <4 x float> %b) {
-; CHECK-LABEL: @ins_bo_ext_ext(
-; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[SHIFT]], [[A]]
-; CHECK-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i64 3
-; CHECK-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
-; CHECK-NEXT:    ret <4 x float> [[V3]]
+; SSE-LABEL: @ins_bo_ext_ext(
+; SSE-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; SSE-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[SHIFT]], [[A]]
+; SSE-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i64 3
+; SSE-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
+; SSE-NEXT:    ret <4 x float> [[V3]]
+;
+; AVX-LABEL: @ins_bo_ext_ext(
+; AVX-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; AVX-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[SHIFT]], [[A]]
+; AVX-NEXT:    [[V3:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+; AVX-NEXT:    ret <4 x float> [[V3]]
 ;
   %a2 = extractelement <4 x float> %a, i32 2
   %a3 = extractelement <4 x float> %a, i32 3
@@ -435,13 +441,21 @@ define <4 x float> @ins_bo_ext_ext(<4 x float> %a, <4 x float> %b) {
 ;       but it is likely that extracting from index 3 is the better option.
 
 define <4 x float> @ins_bo_ext_ext_uses(<4 x float> %a, <4 x float> %b) {
-; CHECK-LABEL: @ins_bo_ext_ext_uses(
-; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
-; CHECK-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
-; CHECK-NEXT:    call void @use_f32(float [[A23]])
-; CHECK-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
-; CHECK-NEXT:    ret <4 x float> [[V3]]
+; SSE-LABEL: @ins_bo_ext_ext_uses(
+; SSE-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; SSE-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; SSE-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
+; SSE-NEXT:    call void @use_f32(float [[A23]])
+; SSE-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
+; SSE-NEXT:    ret <4 x float> [[V3]]
+;
+; AVX-LABEL: @ins_bo_ext_ext_uses(
+; AVX-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; AVX-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; AVX-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
+; AVX-NEXT:    call void @use_f32(float [[A23]])
+; AVX-NEXT:    [[V3:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 1, i32 2, i32 6>
+; AVX-NEXT:    ret <4 x float> [[V3]]
 ;
   %a2 = extractelement <4 x float> %a, i32 2
   %a3 = extractelement <4 x float> %a, i32 3
@@ -452,22 +466,37 @@ define <4 x float> @ins_bo_ext_ext_uses(<4 x float> %a, <4 x float> %b) {
 }
 
 define <4 x float> @PR34724(<4 x float> %a, <4 x float> %b) {
-; CHECK-LABEL: @PR34724(
-; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
-; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
-; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
-; CHECK-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
-; CHECK-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP2:%.*]] = fadd <4 x float> [[B]], [[SHIFT1]]
-; CHECK-NEXT:    [[B01:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
-; CHECK-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
-; CHECK-NEXT:    [[TMP3:%.*]] = fadd <4 x float> [[SHIFT2]], [[B]]
-; CHECK-NEXT:    [[B23:%.*]] = extractelement <4 x float> [[TMP3]], i64 3
-; CHECK-NEXT:    [[V1:%.*]] = insertelement <4 x float> poison, float [[A23]], i32 1
-; CHECK-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[B01]], i32 2
-; CHECK-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[V2]], float [[B23]], i32 3
-; CHECK-NEXT:    ret <4 x float> [[V3]]
+; SSE-LABEL: @PR34724(
+; SSE-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
+; SSE-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
+; SSE-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; SSE-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; SSE-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
+; SSE-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
+; SSE-NEXT:    [[TMP2:%.*]] = fadd <4 x float> [[B]], [[SHIFT1]]
+; SSE-NEXT:    [[B01:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; SSE-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; SSE-NEXT:    [[TMP3:%.*]] = fadd <4 x float> [[SHIFT2]], [[B]]
+; SSE-NEXT:    [[B23:%.*]] = extractelement <4 x float> [[TMP3]], i64 3
+; SSE-NEXT:    [[V1:%.*]] = insertelement <4 x float> poison, float [[A23]], i32 1
+; SSE-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[B01]], i32 2
+; SSE-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[V2]], float [[B23]], i32 3
+; SSE-NEXT:    ret <4 x float> [[V3]]
+;
+; AVX-LABEL: @PR34724(
+; AVX-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
+; AVX-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
+; AVX-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; AVX-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; AVX-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT:    [[TMP2:%.*]] = fadd <4 x float> [[B]], [[SHIFT1]]
+; AVX-NEXT:    [[B01:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; AVX-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; AVX-NEXT:    [[TMP3:%.*]] = fadd <4 x float> [[SHIFT2]], [[B]]
+; AVX-NEXT:    [[V1:%.*]] = shufflevector <4 x float> poison, <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 6, i32 2, i32 3>
+; AVX-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[B01]], i32 2
+; AVX-NEXT:    [[V3:%.*]] = shufflevector <4 x float> [[V2]], <4 x float> [[TMP3]], <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+; AVX-NEXT:    ret <4 x float> [[V3]]
 ;
   %a0 = extractelement <4 x float> %a, i32 0
   %a1 = extractelement <4 x float> %a, i32 1

llvm/test/Transforms/VectorCombine/X86/extract-binop.ll

@@ -417,12 +417,18 @@ define float @ext14_ext15_fmul_v16f32(<16 x float> %x) {
 }
 
 define <4 x float> @ins_bo_ext_ext(<4 x float> %a, <4 x float> %b) {
-; CHECK-LABEL: @ins_bo_ext_ext(
-; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[SHIFT]], [[A]]
-; CHECK-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i64 3
-; CHECK-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
-; CHECK-NEXT:    ret <4 x float> [[V3]]
+; SSE-LABEL: @ins_bo_ext_ext(
+; SSE-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; SSE-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[SHIFT]], [[A]]
+; SSE-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i64 3
+; SSE-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
+; SSE-NEXT:    ret <4 x float> [[V3]]
+;
+; AVX-LABEL: @ins_bo_ext_ext(
+; AVX-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; AVX-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[SHIFT]], [[A]]
+; AVX-NEXT:    [[V3:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+; AVX-NEXT:    ret <4 x float> [[V3]]
 ;
   %a2 = extractelement <4 x float> %a, i32 2
   %a3 = extractelement <4 x float> %a, i32 3
@@ -435,13 +441,21 @@ define <4 x float> @ins_bo_ext_ext(<4 x float> %a, <4 x float> %b) {
 ;       but it is likely that extracting from index 3 is the better option.
 
 define <4 x float> @ins_bo_ext_ext_uses(<4 x float> %a, <4 x float> %b) {
-; CHECK-LABEL: @ins_bo_ext_ext_uses(
-; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
-; CHECK-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
-; CHECK-NEXT:    call void @use_f32(float [[A23]])
-; CHECK-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
-; CHECK-NEXT:    ret <4 x float> [[V3]]
+; SSE-LABEL: @ins_bo_ext_ext_uses(
+; SSE-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; SSE-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; SSE-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
+; SSE-NEXT:    call void @use_f32(float [[A23]])
+; SSE-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[B:%.*]], float [[A23]], i32 3
+; SSE-NEXT:    ret <4 x float> [[V3]]
+;
+; AVX-LABEL: @ins_bo_ext_ext_uses(
+; AVX-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A:%.*]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; AVX-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; AVX-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
+; AVX-NEXT:    call void @use_f32(float [[A23]])
+; AVX-NEXT:    [[V3:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 1, i32 2, i32 6>
+; AVX-NEXT:    ret <4 x float> [[V3]]
 ;
   %a2 = extractelement <4 x float> %a, i32 2
   %a3 = extractelement <4 x float> %a, i32 3
@@ -452,22 +466,37 @@ define <4 x float> @ins_bo_ext_ext_uses(<4 x float> %a, <4 x float> %b) {
 }
 
 define <4 x float> @PR34724(<4 x float> %a, <4 x float> %b) {
-; CHECK-LABEL: @PR34724(
-; CHECK-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
-; CHECK-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
-; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
-; CHECK-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
-; CHECK-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
-; CHECK-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP2:%.*]] = fadd <4 x float> [[B]], [[SHIFT1]]
-; CHECK-NEXT:    [[B01:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
-; CHECK-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
-; CHECK-NEXT:    [[TMP3:%.*]] = fadd <4 x float> [[SHIFT2]], [[B]]
-; CHECK-NEXT:    [[B23:%.*]] = extractelement <4 x float> [[TMP3]], i64 3
-; CHECK-NEXT:    [[V1:%.*]] = insertelement <4 x float> undef, float [[A23]], i32 1
-; CHECK-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[B01]], i32 2
-; CHECK-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[V2]], float [[B23]], i32 3
-; CHECK-NEXT:    ret <4 x float> [[V3]]
+; SSE-LABEL: @PR34724(
+; SSE-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
+; SSE-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
+; SSE-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; SSE-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; SSE-NEXT:    [[A23:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
+; SSE-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
+; SSE-NEXT:    [[TMP2:%.*]] = fadd <4 x float> [[B]], [[SHIFT1]]
+; SSE-NEXT:    [[B01:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; SSE-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; SSE-NEXT:    [[TMP3:%.*]] = fadd <4 x float> [[SHIFT2]], [[B]]
+; SSE-NEXT:    [[B23:%.*]] = extractelement <4 x float> [[TMP3]], i64 3
+; SSE-NEXT:    [[V1:%.*]] = insertelement <4 x float> undef, float [[A23]], i32 1
+; SSE-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[B01]], i32 2
+; SSE-NEXT:    [[V3:%.*]] = insertelement <4 x float> [[V2]], float [[B23]], i32 3
+; SSE-NEXT:    ret <4 x float> [[V3]]
+;
+; AVX-LABEL: @PR34724(
+; AVX-NEXT:    [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0
+; AVX-NEXT:    [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1
+; AVX-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x float> [[A]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 3, i32 poison>
+; AVX-NEXT:    [[TMP1:%.*]] = fadd <4 x float> [[A]], [[SHIFT]]
+; AVX-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x float> [[B:%.*]], <4 x float> poison, <4 x i32> <i32 1, i32 poison, i32 poison, i32 poison>
+; AVX-NEXT:    [[TMP2:%.*]] = fadd <4 x float> [[B]], [[SHIFT1]]
+; AVX-NEXT:    [[B01:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; AVX-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x float> [[B]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 poison, i32 2>
+; AVX-NEXT:    [[TMP3:%.*]] = fadd <4 x float> [[SHIFT2]], [[B]]
+; AVX-NEXT:    [[V1:%.*]] = shufflevector <4 x float> undef, <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 6, i32 2, i32 3>
+; AVX-NEXT:    [[V2:%.*]] = insertelement <4 x float> [[V1]], float [[B01]], i32 2
+; AVX-NEXT:    [[V3:%.*]] = shufflevector <4 x float> [[V2]], <4 x float> [[TMP3]], <4 x i32> <i32 0, i32 1, i32 2, i32 7>
+; AVX-NEXT:    ret <4 x float> [[V3]]
 ;
   %a0 = extractelement <4 x float> %a, i32 0
   %a1 = extractelement <4 x float> %a, i32 1