; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=loop-vectorize,dce,instcombine -force-vector-interleave=4 -force-vector-width=4 -prefer-inloop-reductions -S | FileCheck %s

target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"

define i32 @reduction_sum_single(ptr noalias nocapture %A) {
; CHECK-LABEL: @reduction_sum_single(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI3:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 16
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 32
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 48
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4
; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i32>, ptr [[TMP1]], align 4
; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x i32>, ptr [[TMP2]], align 4
; CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4
; CHECK-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[WIDE_LOAD]])
; CHECK-NEXT:    [[TMP5]] = add i32 [[TMP4]], [[VEC_PHI]]
; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[WIDE_LOAD4]])
; CHECK-NEXT:    [[TMP7]] = add i32 [[TMP6]], [[VEC_PHI1]]
; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[WIDE_LOAD5]])
; CHECK-NEXT:    [[TMP9]] = add i32 [[TMP8]], [[VEC_PHI2]]
; CHECK-NEXT:    [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[WIDE_LOAD6]])
; CHECK-NEXT:    [[TMP11]] = add i32 [[TMP10]], [[VEC_PHI3]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    [[BIN_RDX:%.*]] = add i32 [[TMP7]], [[TMP5]]
; CHECK-NEXT:    [[BIN_RDX7:%.*]] = add i32 [[TMP9]], [[BIN_RDX]]
; CHECK-NEXT:    [[BIN_RDX8:%.*]] = add i32 [[TMP11]], [[BIN_RDX7]]
; CHECK-NEXT:    br i1 true, label [[DOT_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
; CHECK:       scalar.ph:
; CHECK-NEXT:    br label [[DOTLR_PH:%.*]]
; CHECK:       .lr.ph:
; CHECK-NEXT:    br i1 poison, label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK:       ._crit_edge:
; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ poison, [[DOTLR_PH]] ], [ [[BIN_RDX8]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT:    ret i32 [[SUM_0_LCSSA]]
;
entry:
  br label %.lr.ph

.lr.ph:                                           ; preds = %entry, %.lr.ph
  %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %entry ]
  %sum.02 = phi i32 [ %l7, %.lr.ph ], [ 0, %entry ]
  %l2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
  %l3 = load i32, ptr %l2, align 4
  %l7 = add i32 %sum.02, %l3
  %indvars.iv.next = add i64 %indvars.iv, 1
  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
  %exitcond = icmp eq i32 %lftr.wideiv, 256
  br i1 %exitcond, label %._crit_edge, label %.lr.ph

._crit_edge:                                      ; preds = %.lr.ph
  %sum.0.lcssa = phi i32 [ %l7, %.lr.ph ]
  ret i32 %sum.0.lcssa
}

; Check that we correctly unroll two reductions chained together.
define i64 @reduction_sum_chain(ptr noalias %p, ptr noalias %q) {
; CHECK-LABEL: @reduction_sum_chain(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI2:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP21:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_PHI3:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i64, ptr [[P:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i64, ptr [[Q:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[TMP0]], i64 32
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 64
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i8, ptr [[TMP0]], i64 96
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP0]], align 8
; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP2]], align 8
; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x i64>, ptr [[TMP3]], align 8
; CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i64>, ptr [[TMP4]], align 8
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[TMP1]], i64 32
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr i8, ptr [[TMP1]], i64 64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr i8, ptr [[TMP1]], i64 96
; CHECK-NEXT:    [[WIDE_LOAD7:%.*]] = load <4 x i64>, ptr [[TMP1]], align 8
; CHECK-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x i64>, ptr [[TMP5]], align 8
; CHECK-NEXT:    [[WIDE_LOAD9:%.*]] = load <4 x i64>, ptr [[TMP6]], align 8
; CHECK-NEXT:    [[WIDE_LOAD10:%.*]] = load <4 x i64>, ptr [[TMP7]], align 8
; CHECK-NEXT:    [[TMP8:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD]])
; CHECK-NEXT:    [[TMP9:%.*]] = add i64 [[TMP8]], [[VEC_PHI]]
; CHECK-NEXT:    [[TMP10:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD4]])
; CHECK-NEXT:    [[TMP11:%.*]] = add i64 [[TMP10]], [[VEC_PHI1]]
; CHECK-NEXT:    [[TMP12:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD5]])
; CHECK-NEXT:    [[TMP13:%.*]] = add i64 [[TMP12]], [[VEC_PHI2]]
; CHECK-NEXT:    [[TMP14:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD6]])
; CHECK-NEXT:    [[TMP15:%.*]] = add i64 [[TMP14]], [[VEC_PHI3]]
; CHECK-NEXT:    [[TMP16:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD7]])
; CHECK-NEXT:    [[TMP17]] = add i64 [[TMP16]], [[TMP9]]
; CHECK-NEXT:    [[TMP18:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD8]])
; CHECK-NEXT:    [[TMP19]] = add i64 [[TMP18]], [[TMP11]]
; CHECK-NEXT:    [[TMP20:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD9]])
; CHECK-NEXT:    [[TMP21]] = add i64 [[TMP20]], [[TMP13]]
; CHECK-NEXT:    [[TMP22:%.*]] = call i64 @llvm.vector.reduce.add.v4i64(<4 x i64> [[WIDE_LOAD10]])
; CHECK-NEXT:    [[TMP23]] = add i64 [[TMP22]], [[TMP15]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
; CHECK-NEXT:    br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    [[BIN_RDX:%.*]] = add i64 [[TMP19]], [[TMP17]]
; CHECK-NEXT:    [[BIN_RDX11:%.*]] = add i64 [[TMP21]], [[BIN_RDX]]
; CHECK-NEXT:    [[BIN_RDX12:%.*]] = add i64 [[TMP23]], [[BIN_RDX11]]
; CHECK-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK:       scalar.ph:
; CHECK-NEXT:    br label [[LOOP:%.*]]
; CHECK:       loop:
; CHECK-NEXT:    br i1 poison, label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK:       exit:
; CHECK-NEXT:    [[ADD2_LCSSA:%.*]] = phi i64 [ poison, [[LOOP]] ], [ [[BIN_RDX12]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT:    ret i64 [[ADD2_LCSSA]]
;
entry:
  br label %loop

loop:
  %iv = phi i64 [0, %entry], [%iv.next, %loop]
  %rdx = phi i64 [0, %entry], [%add2, %loop]

  %p.gep = getelementptr i64, ptr %p, i64 %iv
  %q.gep = getelementptr i64, ptr %q, i64 %iv

  %x = load i64, ptr %p.gep
  %y = load i64, ptr %q.gep

  %add1 = add i64 %rdx, %x
  %add2 = add i64 %add1, %y

  %iv.next = add i64 %iv, 1
  %done = icmp eq i64 %iv.next, 256
  br i1 %done, label %exit, label %loop

exit:
  ret i64 %add2
}

define i32 @predicated(ptr noalias nocapture %A) {
; CHECK-LABEL: @predicated(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE36:%.*]] ]
; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE36]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP101:%.*]], [[PRED_LOAD_CONTINUE36]] ]
; CHECK-NEXT:    [[VEC_PHI4:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP104:%.*]], [[PRED_LOAD_CONTINUE36]] ]
; CHECK-NEXT:    [[VEC_PHI5:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP107:%.*]], [[PRED_LOAD_CONTINUE36]] ]
; CHECK-NEXT:    [[VEC_PHI6:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP110:%.*]], [[PRED_LOAD_CONTINUE36]] ]
; CHECK-NEXT:    [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
; CHECK-NEXT:    [[STEP_ADD_2:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 8)
; CHECK-NEXT:    [[STEP_ADD_3:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 12)
; CHECK-NEXT:    [[TMP0:%.*]] = icmp ult <4 x i64> [[VEC_IND]], splat (i64 257)
; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult <4 x i64> [[STEP_ADD]], splat (i64 257)
; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult <4 x i64> [[STEP_ADD_2]], splat (i64 257)
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <4 x i64> [[STEP_ADD_3]], splat (i64 257)
; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i1> [[TMP0]], i64 0
; CHECK-NEXT:    br i1 [[TMP4]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
; CHECK:       pred.load.if:
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP6:%.*]] = load i32, ptr [[TMP5]], align 4
; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x i32> poison, i32 [[TMP6]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE]]
; CHECK:       pred.load.continue:
; CHECK-NEXT:    [[TMP8:%.*]] = phi <4 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_LOAD_IF]] ]
; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i1> [[TMP0]], i64 1
; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_LOAD_IF7:%.*]], label [[PRED_LOAD_CONTINUE8:%.*]]
; CHECK:       pred.load.if4:
; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr i8, ptr [[TMP10]], i64 4
; CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr [[TMP11]], align 4
; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <4 x i32> [[TMP8]], i32 [[TMP12]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE8]]
; CHECK:       pred.load.continue5:
; CHECK-NEXT:    [[TMP14:%.*]] = phi <4 x i32> [ [[TMP8]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP13]], [[PRED_LOAD_IF7]] ]
; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x i1> [[TMP0]], i64 2
; CHECK-NEXT:    br i1 [[TMP15]], label [[PRED_LOAD_IF9:%.*]], label [[PRED_LOAD_CONTINUE10:%.*]]
; CHECK:       pred.load.if6:
; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr i8, ptr [[TMP16]], i64 8
; CHECK-NEXT:    [[TMP18:%.*]] = load i32, ptr [[TMP17]], align 4
; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i32> [[TMP14]], i32 [[TMP18]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE10]]
; CHECK:       pred.load.continue7:
; CHECK-NEXT:    [[TMP20:%.*]] = phi <4 x i32> [ [[TMP14]], [[PRED_LOAD_CONTINUE8]] ], [ [[TMP19]], [[PRED_LOAD_IF9]] ]
; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <4 x i1> [[TMP0]], i64 3
; CHECK-NEXT:    br i1 [[TMP21]], label [[PRED_LOAD_IF11:%.*]], label [[PRED_LOAD_CONTINUE12:%.*]]
; CHECK:       pred.load.if8:
; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr i8, ptr [[TMP22]], i64 12
; CHECK-NEXT:    [[TMP24:%.*]] = load i32, ptr [[TMP23]], align 4
; CHECK-NEXT:    [[TMP25:%.*]] = insertelement <4 x i32> [[TMP20]], i32 [[TMP24]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE12]]
; CHECK:       pred.load.continue9:
; CHECK-NEXT:    [[TMP26:%.*]] = phi <4 x i32> [ [[TMP20]], [[PRED_LOAD_CONTINUE10]] ], [ [[TMP25]], [[PRED_LOAD_IF11]] ]
; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <4 x i1> [[TMP1]], i64 0
; CHECK-NEXT:    br i1 [[TMP27]], label [[PRED_LOAD_IF13:%.*]], label [[PRED_LOAD_CONTINUE14:%.*]]
; CHECK:       pred.load.if10:
; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP29:%.*]] = getelementptr i8, ptr [[TMP28]], i64 16
; CHECK-NEXT:    [[TMP30:%.*]] = load i32, ptr [[TMP29]], align 4
; CHECK-NEXT:    [[TMP31:%.*]] = insertelement <4 x i32> poison, i32 [[TMP30]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE14]]
; CHECK:       pred.load.continue11:
; CHECK-NEXT:    [[TMP32:%.*]] = phi <4 x i32> [ poison, [[PRED_LOAD_CONTINUE12]] ], [ [[TMP31]], [[PRED_LOAD_IF13]] ]
; CHECK-NEXT:    [[TMP33:%.*]] = extractelement <4 x i1> [[TMP1]], i64 1
; CHECK-NEXT:    br i1 [[TMP33]], label [[PRED_LOAD_IF15:%.*]], label [[PRED_LOAD_CONTINUE16:%.*]]
; CHECK:       pred.load.if12:
; CHECK-NEXT:    [[TMP34:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP35:%.*]] = getelementptr i8, ptr [[TMP34]], i64 20
; CHECK-NEXT:    [[TMP36:%.*]] = load i32, ptr [[TMP35]], align 4
; CHECK-NEXT:    [[TMP37:%.*]] = insertelement <4 x i32> [[TMP32]], i32 [[TMP36]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE16]]
; CHECK:       pred.load.continue13:
; CHECK-NEXT:    [[TMP38:%.*]] = phi <4 x i32> [ [[TMP32]], [[PRED_LOAD_CONTINUE14]] ], [ [[TMP37]], [[PRED_LOAD_IF15]] ]
; CHECK-NEXT:    [[TMP39:%.*]] = extractelement <4 x i1> [[TMP1]], i64 2
; CHECK-NEXT:    br i1 [[TMP39]], label [[PRED_LOAD_IF17:%.*]], label [[PRED_LOAD_CONTINUE18:%.*]]
; CHECK:       pred.load.if14:
; CHECK-NEXT:    [[TMP40:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP41:%.*]] = getelementptr i8, ptr [[TMP40]], i64 24
; CHECK-NEXT:    [[TMP42:%.*]] = load i32, ptr [[TMP41]], align 4
; CHECK-NEXT:    [[TMP43:%.*]] = insertelement <4 x i32> [[TMP38]], i32 [[TMP42]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE18]]
; CHECK:       pred.load.continue15:
; CHECK-NEXT:    [[TMP44:%.*]] = phi <4 x i32> [ [[TMP38]], [[PRED_LOAD_CONTINUE16]] ], [ [[TMP43]], [[PRED_LOAD_IF17]] ]
; CHECK-NEXT:    [[TMP45:%.*]] = extractelement <4 x i1> [[TMP1]], i64 3
; CHECK-NEXT:    br i1 [[TMP45]], label [[PRED_LOAD_IF19:%.*]], label [[PRED_LOAD_CONTINUE20:%.*]]
; CHECK:       pred.load.if16:
; CHECK-NEXT:    [[TMP46:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP47:%.*]] = getelementptr i8, ptr [[TMP46]], i64 28
; CHECK-NEXT:    [[TMP48:%.*]] = load i32, ptr [[TMP47]], align 4
; CHECK-NEXT:    [[TMP49:%.*]] = insertelement <4 x i32> [[TMP44]], i32 [[TMP48]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE20]]
; CHECK:       pred.load.continue17:
; CHECK-NEXT:    [[TMP50:%.*]] = phi <4 x i32> [ [[TMP44]], [[PRED_LOAD_CONTINUE18]] ], [ [[TMP49]], [[PRED_LOAD_IF19]] ]
; CHECK-NEXT:    [[TMP51:%.*]] = extractelement <4 x i1> [[TMP2]], i64 0
; CHECK-NEXT:    br i1 [[TMP51]], label [[PRED_LOAD_IF21:%.*]], label [[PRED_LOAD_CONTINUE22:%.*]]
; CHECK:       pred.load.if18:
; CHECK-NEXT:    [[TMP52:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP53:%.*]] = getelementptr i8, ptr [[TMP52]], i64 32
; CHECK-NEXT:    [[TMP54:%.*]] = load i32, ptr [[TMP53]], align 4
; CHECK-NEXT:    [[TMP55:%.*]] = insertelement <4 x i32> poison, i32 [[TMP54]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE22]]
; CHECK:       pred.load.continue19:
; CHECK-NEXT:    [[TMP56:%.*]] = phi <4 x i32> [ poison, [[PRED_LOAD_CONTINUE20]] ], [ [[TMP55]], [[PRED_LOAD_IF21]] ]
; CHECK-NEXT:    [[TMP57:%.*]] = extractelement <4 x i1> [[TMP2]], i64 1
; CHECK-NEXT:    br i1 [[TMP57]], label [[PRED_LOAD_IF23:%.*]], label [[PRED_LOAD_CONTINUE24:%.*]]
; CHECK:       pred.load.if20:
; CHECK-NEXT:    [[TMP58:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP59:%.*]] = getelementptr i8, ptr [[TMP58]], i64 36
; CHECK-NEXT:    [[TMP60:%.*]] = load i32, ptr [[TMP59]], align 4
; CHECK-NEXT:    [[TMP61:%.*]] = insertelement <4 x i32> [[TMP56]], i32 [[TMP60]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE24]]
; CHECK:       pred.load.continue21:
; CHECK-NEXT:    [[TMP62:%.*]] = phi <4 x i32> [ [[TMP56]], [[PRED_LOAD_CONTINUE22]] ], [ [[TMP61]], [[PRED_LOAD_IF23]] ]
; CHECK-NEXT:    [[TMP63:%.*]] = extractelement <4 x i1> [[TMP2]], i64 2
; CHECK-NEXT:    br i1 [[TMP63]], label [[PRED_LOAD_IF25:%.*]], label [[PRED_LOAD_CONTINUE26:%.*]]
; CHECK:       pred.load.if22:
; CHECK-NEXT:    [[TMP64:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP65:%.*]] = getelementptr i8, ptr [[TMP64]], i64 40
; CHECK-NEXT:    [[TMP66:%.*]] = load i32, ptr [[TMP65]], align 4
; CHECK-NEXT:    [[TMP67:%.*]] = insertelement <4 x i32> [[TMP62]], i32 [[TMP66]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE26]]
; CHECK:       pred.load.continue23:
; CHECK-NEXT:    [[TMP68:%.*]] = phi <4 x i32> [ [[TMP62]], [[PRED_LOAD_CONTINUE24]] ], [ [[TMP67]], [[PRED_LOAD_IF25]] ]
; CHECK-NEXT:    [[TMP69:%.*]] = extractelement <4 x i1> [[TMP2]], i64 3
; CHECK-NEXT:    br i1 [[TMP69]], label [[PRED_LOAD_IF27:%.*]], label [[PRED_LOAD_CONTINUE28:%.*]]
; CHECK:       pred.load.if24:
; CHECK-NEXT:    [[TMP70:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP71:%.*]] = getelementptr i8, ptr [[TMP70]], i64 44
; CHECK-NEXT:    [[TMP72:%.*]] = load i32, ptr [[TMP71]], align 4
; CHECK-NEXT:    [[TMP73:%.*]] = insertelement <4 x i32> [[TMP68]], i32 [[TMP72]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE28]]
; CHECK:       pred.load.continue25:
; CHECK-NEXT:    [[TMP74:%.*]] = phi <4 x i32> [ [[TMP68]], [[PRED_LOAD_CONTINUE26]] ], [ [[TMP73]], [[PRED_LOAD_IF27]] ]
; CHECK-NEXT:    [[TMP75:%.*]] = extractelement <4 x i1> [[TMP3]], i64 0
; CHECK-NEXT:    br i1 [[TMP75]], label [[PRED_LOAD_IF29:%.*]], label [[PRED_LOAD_CONTINUE30:%.*]]
; CHECK:       pred.load.if26:
; CHECK-NEXT:    [[TMP76:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP77:%.*]] = getelementptr i8, ptr [[TMP76]], i64 48
; CHECK-NEXT:    [[TMP78:%.*]] = load i32, ptr [[TMP77]], align 4
; CHECK-NEXT:    [[TMP79:%.*]] = insertelement <4 x i32> poison, i32 [[TMP78]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE30]]
; CHECK:       pred.load.continue27:
; CHECK-NEXT:    [[TMP80:%.*]] = phi <4 x i32> [ poison, [[PRED_LOAD_CONTINUE28]] ], [ [[TMP79]], [[PRED_LOAD_IF29]] ]
; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <4 x i1> [[TMP3]], i64 1
; CHECK-NEXT:    br i1 [[TMP81]], label [[PRED_LOAD_IF31:%.*]], label [[PRED_LOAD_CONTINUE32:%.*]]
; CHECK:       pred.load.if28:
; CHECK-NEXT:    [[TMP82:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP83:%.*]] = getelementptr i8, ptr [[TMP82]], i64 52
; CHECK-NEXT:    [[TMP84:%.*]] = load i32, ptr [[TMP83]], align 4
; CHECK-NEXT:    [[TMP85:%.*]] = insertelement <4 x i32> [[TMP80]], i32 [[TMP84]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE32]]
; CHECK:       pred.load.continue29:
; CHECK-NEXT:    [[TMP86:%.*]] = phi <4 x i32> [ [[TMP80]], [[PRED_LOAD_CONTINUE30]] ], [ [[TMP85]], [[PRED_LOAD_IF31]] ]
; CHECK-NEXT:    [[TMP87:%.*]] = extractelement <4 x i1> [[TMP3]], i64 2
; CHECK-NEXT:    br i1 [[TMP87]], label [[PRED_LOAD_IF33:%.*]], label [[PRED_LOAD_CONTINUE34:%.*]]
; CHECK:       pred.load.if30:
; CHECK-NEXT:    [[TMP88:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP89:%.*]] = getelementptr i8, ptr [[TMP88]], i64 56
; CHECK-NEXT:    [[TMP90:%.*]] = load i32, ptr [[TMP89]], align 4
; CHECK-NEXT:    [[TMP91:%.*]] = insertelement <4 x i32> [[TMP86]], i32 [[TMP90]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE34]]
; CHECK:       pred.load.continue31:
; CHECK-NEXT:    [[TMP92:%.*]] = phi <4 x i32> [ [[TMP86]], [[PRED_LOAD_CONTINUE32]] ], [ [[TMP91]], [[PRED_LOAD_IF33]] ]
; CHECK-NEXT:    [[TMP93:%.*]] = extractelement <4 x i1> [[TMP3]], i64 3
; CHECK-NEXT:    br i1 [[TMP93]], label [[PRED_LOAD_IF35:%.*]], label [[PRED_LOAD_CONTINUE36]]
; CHECK:       pred.load.if32:
; CHECK-NEXT:    [[TMP94:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP95:%.*]] = getelementptr i8, ptr [[TMP94]], i64 60
; CHECK-NEXT:    [[TMP96:%.*]] = load i32, ptr [[TMP95]], align 4
; CHECK-NEXT:    [[TMP97:%.*]] = insertelement <4 x i32> [[TMP92]], i32 [[TMP96]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE36]]
; CHECK:       pred.load.continue33:
; CHECK-NEXT:    [[TMP98:%.*]] = phi <4 x i32> [ [[TMP92]], [[PRED_LOAD_CONTINUE34]] ], [ [[TMP97]], [[PRED_LOAD_IF35]] ]
; CHECK-NEXT:    [[TMP99:%.*]] = select <4 x i1> [[TMP0]], <4 x i32> [[TMP26]], <4 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP100:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP99]])
; CHECK-NEXT:    [[TMP101]] = add i32 [[TMP100]], [[VEC_PHI]]
; CHECK-NEXT:    [[TMP102:%.*]] = select <4 x i1> [[TMP1]], <4 x i32> [[TMP50]], <4 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP103:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP102]])
; CHECK-NEXT:    [[TMP104]] = add i32 [[TMP103]], [[VEC_PHI4]]
; CHECK-NEXT:    [[TMP105:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP74]], <4 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP106:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP105]])
; CHECK-NEXT:    [[TMP107]] = add i32 [[TMP106]], [[VEC_PHI5]]
; CHECK-NEXT:    [[TMP108:%.*]] = select <4 x i1> [[TMP3]], <4 x i32> [[TMP98]], <4 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP109:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP108]])
; CHECK-NEXT:    [[TMP110]] = add i32 [[TMP109]], [[VEC_PHI6]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 16)
; CHECK-NEXT:    [[TMP111:%.*]] = icmp eq i64 [[INDEX_NEXT]], 272
; CHECK-NEXT:    br i1 [[TMP111]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    [[BIN_RDX:%.*]] = add i32 [[TMP104]], [[TMP101]]
; CHECK-NEXT:    [[BIN_RDX37:%.*]] = add i32 [[TMP107]], [[BIN_RDX]]
; CHECK-NEXT:    [[BIN_RDX38:%.*]] = add i32 [[TMP110]], [[BIN_RDX37]]
; CHECK-NEXT:    br label [[DOT_CRIT_EDGE:%.*]]
; CHECK:       scalar.ph:
; CHECK-NEXT:    br label [[DOTLR_PH:%.*]]
; CHECK:       .lr.ph:
; CHECK-NEXT:    br i1 poison, label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]], !llvm.loop [[LOOP7:![0-9]+]]
; CHECK:       ._crit_edge:
; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ poison, [[DOTLR_PH]] ], [ [[BIN_RDX38]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT:    ret i32 [[SUM_0_LCSSA]]
;
entry:
  br label %.lr.ph

.lr.ph:                                           ; preds = %entry, %.lr.ph
  %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 0, %entry ]
  %sum.02 = phi i32 [ %l7, %.lr.ph ], [ 0, %entry ]
  %l2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
  %l3 = load i32, ptr %l2, align 4
  %l7 = add i32 %sum.02, %l3
  %indvars.iv.next = add i64 %indvars.iv, 1
  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
  %exitcond = icmp eq i32 %lftr.wideiv, 257
  br i1 %exitcond, label %._crit_edge, label %.lr.ph, !llvm.loop !6

._crit_edge:                                      ; preds = %.lr.ph
  %sum.0.lcssa = phi i32 [ %l7, %.lr.ph ]
  ret i32 %sum.0.lcssa
}

define i32 @cond_rdx_pred(i32 %cond, ptr noalias %a, i64 %N) {
; CHECK-LABEL: @cond_rdx_pred(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i64 [[N:%.*]], 15
; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[N_RND_UP]], -16
; CHECK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = add i64 [[N]], -1
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <4 x i32> poison, i32 [[COND:%.*]], i64 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt <4 x i32> [[BROADCAST_SPLATINSERT7]], splat (i32 7)
; CHECK-NEXT:    [[BROADCAST_SPLAT8:%.*]] = shufflevector <4 x i1> [[TMP4]], <4 x i1> poison, <4 x i32> zeroinitializer
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TRIP_COUNT_MINUS_1]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE38:%.*]] ]
; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE38]] ]
; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi i32 [ 4, [[VECTOR_PH]] ], [ [[TMP109:%.*]], [[PRED_LOAD_CONTINUE38]] ]
; CHECK-NEXT:    [[VEC_PHI4:%.*]] = phi i32 [ 1, [[VECTOR_PH]] ], [ [[TMP112:%.*]], [[PRED_LOAD_CONTINUE38]] ]
; CHECK-NEXT:    [[VEC_PHI5:%.*]] = phi i32 [ 1, [[VECTOR_PH]] ], [ [[TMP115:%.*]], [[PRED_LOAD_CONTINUE38]] ]
; CHECK-NEXT:    [[VEC_PHI6:%.*]] = phi i32 [ 1, [[VECTOR_PH]] ], [ [[TMP118:%.*]], [[PRED_LOAD_CONTINUE38]] ]
; CHECK-NEXT:    [[STEP_ADD:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
; CHECK-NEXT:    [[STEP_ADD1:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 8)
; CHECK-NEXT:    [[STEP_ADD2:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 12)
; CHECK-NEXT:    [[TMP0:%.*]] = icmp ule <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
; CHECK-NEXT:    [[TMP1:%.*]] = icmp ule <4 x i64> [[STEP_ADD]], [[BROADCAST_SPLAT]]
; CHECK-NEXT:    [[TMP2:%.*]] = icmp ule <4 x i64> [[STEP_ADD1]], [[BROADCAST_SPLAT]]
; CHECK-NEXT:    [[TMP3:%.*]] = icmp ule <4 x i64> [[STEP_ADD2]], [[BROADCAST_SPLAT]]
; CHECK-NEXT:    [[TMP8:%.*]] = select <4 x i1> [[TMP0]], <4 x i1> [[BROADCAST_SPLAT8]], <4 x i1> zeroinitializer
; CHECK-NEXT:    [[TMP9:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[BROADCAST_SPLAT8]], <4 x i1> zeroinitializer
; CHECK-NEXT:    [[TMP10:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[BROADCAST_SPLAT8]], <4 x i1> zeroinitializer
; CHECK-NEXT:    [[TMP11:%.*]] = select <4 x i1> [[TMP3]], <4 x i1> [[BROADCAST_SPLAT8]], <4 x i1> zeroinitializer
; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x i1> [[TMP8]], i64 0
; CHECK-NEXT:    br i1 [[TMP12]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
; CHECK:       pred.load.if:
; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP14:%.*]] = load i32, ptr [[TMP13]], align 4
; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <4 x i32> poison, i32 [[TMP14]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE]]
; CHECK:       pred.load.continue:
; CHECK-NEXT:    [[TMP16:%.*]] = phi <4 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP15]], [[PRED_LOAD_IF]] ]
; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <4 x i1> [[TMP8]], i64 1
; CHECK-NEXT:    br i1 [[TMP17]], label [[PRED_LOAD_IF9:%.*]], label [[PRED_LOAD_CONTINUE10:%.*]]
; CHECK:       pred.load.if6:
; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr i8, ptr [[TMP18]], i64 4
; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[TMP19]], align 4
; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <4 x i32> [[TMP16]], i32 [[TMP20]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE10]]
; CHECK:       pred.load.continue7:
; CHECK-NEXT:    [[TMP22:%.*]] = phi <4 x i32> [ [[TMP16]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP21]], [[PRED_LOAD_IF9]] ]
; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <4 x i1> [[TMP8]], i64 2
; CHECK-NEXT:    br i1 [[TMP23]], label [[PRED_LOAD_IF11:%.*]], label [[PRED_LOAD_CONTINUE12:%.*]]
; CHECK:       pred.load.if8:
; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP25:%.*]] = getelementptr i8, ptr [[TMP24]], i64 8
; CHECK-NEXT:    [[TMP26:%.*]] = load i32, ptr [[TMP25]], align 4
; CHECK-NEXT:    [[TMP27:%.*]] = insertelement <4 x i32> [[TMP22]], i32 [[TMP26]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE12]]
; CHECK:       pred.load.continue9:
; CHECK-NEXT:    [[TMP28:%.*]] = phi <4 x i32> [ [[TMP22]], [[PRED_LOAD_CONTINUE10]] ], [ [[TMP27]], [[PRED_LOAD_IF11]] ]
; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <4 x i1> [[TMP8]], i64 3
; CHECK-NEXT:    br i1 [[TMP29]], label [[PRED_LOAD_IF13:%.*]], label [[PRED_LOAD_CONTINUE14:%.*]]
; CHECK:       pred.load.if10:
; CHECK-NEXT:    [[TMP30:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr i8, ptr [[TMP30]], i64 12
; CHECK-NEXT:    [[TMP32:%.*]] = load i32, ptr [[TMP31]], align 4
; CHECK-NEXT:    [[TMP33:%.*]] = insertelement <4 x i32> [[TMP28]], i32 [[TMP32]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE14]]
; CHECK:       pred.load.continue11:
; CHECK-NEXT:    [[TMP34:%.*]] = phi <4 x i32> [ [[TMP28]], [[PRED_LOAD_CONTINUE12]] ], [ [[TMP33]], [[PRED_LOAD_IF13]] ]
; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <4 x i1> [[TMP9]], i64 0
; CHECK-NEXT:    br i1 [[TMP35]], label [[PRED_LOAD_IF15:%.*]], label [[PRED_LOAD_CONTINUE16:%.*]]
; CHECK:       pred.load.if12:
; CHECK-NEXT:    [[TMP36:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP37:%.*]] = getelementptr i8, ptr [[TMP36]], i64 16
; CHECK-NEXT:    [[TMP38:%.*]] = load i32, ptr [[TMP37]], align 4
; CHECK-NEXT:    [[TMP39:%.*]] = insertelement <4 x i32> poison, i32 [[TMP38]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE16]]
; CHECK:       pred.load.continue13:
; CHECK-NEXT:    [[TMP40:%.*]] = phi <4 x i32> [ poison, [[PRED_LOAD_CONTINUE14]] ], [ [[TMP39]], [[PRED_LOAD_IF15]] ]
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <4 x i1> [[TMP9]], i64 1
; CHECK-NEXT:    br i1 [[TMP41]], label [[PRED_LOAD_IF17:%.*]], label [[PRED_LOAD_CONTINUE18:%.*]]
; CHECK:       pred.load.if14:
; CHECK-NEXT:    [[TMP42:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP43:%.*]] = getelementptr i8, ptr [[TMP42]], i64 20
; CHECK-NEXT:    [[TMP44:%.*]] = load i32, ptr [[TMP43]], align 4
; CHECK-NEXT:    [[TMP45:%.*]] = insertelement <4 x i32> [[TMP40]], i32 [[TMP44]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE18]]
; CHECK:       pred.load.continue15:
; CHECK-NEXT:    [[TMP46:%.*]] = phi <4 x i32> [ [[TMP40]], [[PRED_LOAD_CONTINUE16]] ], [ [[TMP45]], [[PRED_LOAD_IF17]] ]
; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <4 x i1> [[TMP9]], i64 2
; CHECK-NEXT:    br i1 [[TMP47]], label [[PRED_LOAD_IF19:%.*]], label [[PRED_LOAD_CONTINUE20:%.*]]
; CHECK:       pred.load.if16:
; CHECK-NEXT:    [[TMP48:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP49:%.*]] = getelementptr i8, ptr [[TMP48]], i64 24
; CHECK-NEXT:    [[TMP50:%.*]] = load i32, ptr [[TMP49]], align 4
; CHECK-NEXT:    [[TMP51:%.*]] = insertelement <4 x i32> [[TMP46]], i32 [[TMP50]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE20]]
; CHECK:       pred.load.continue17:
; CHECK-NEXT:    [[TMP52:%.*]] = phi <4 x i32> [ [[TMP46]], [[PRED_LOAD_CONTINUE18]] ], [ [[TMP51]], [[PRED_LOAD_IF19]] ]
; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <4 x i1> [[TMP9]], i64 3
; CHECK-NEXT:    br i1 [[TMP53]], label [[PRED_LOAD_IF21:%.*]], label [[PRED_LOAD_CONTINUE22:%.*]]
; CHECK:       pred.load.if18:
; CHECK-NEXT:    [[TMP54:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP55:%.*]] = getelementptr i8, ptr [[TMP54]], i64 28
; CHECK-NEXT:    [[TMP56:%.*]] = load i32, ptr [[TMP55]], align 4
; CHECK-NEXT:    [[TMP57:%.*]] = insertelement <4 x i32> [[TMP52]], i32 [[TMP56]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE22]]
; CHECK:       pred.load.continue19:
; CHECK-NEXT:    [[TMP58:%.*]] = phi <4 x i32> [ [[TMP52]], [[PRED_LOAD_CONTINUE20]] ], [ [[TMP57]], [[PRED_LOAD_IF21]] ]
; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <4 x i1> [[TMP10]], i64 0
; CHECK-NEXT:    br i1 [[TMP59]], label [[PRED_LOAD_IF23:%.*]], label [[PRED_LOAD_CONTINUE24:%.*]]
; CHECK:       pred.load.if20:
; CHECK-NEXT:    [[TMP60:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP61:%.*]] = getelementptr i8, ptr [[TMP60]], i64 32
; CHECK-NEXT:    [[TMP62:%.*]] = load i32, ptr [[TMP61]], align 4
; CHECK-NEXT:    [[TMP63:%.*]] = insertelement <4 x i32> poison, i32 [[TMP62]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE24]]
; CHECK:       pred.load.continue21:
; CHECK-NEXT:    [[TMP64:%.*]] = phi <4 x i32> [ poison, [[PRED_LOAD_CONTINUE22]] ], [ [[TMP63]], [[PRED_LOAD_IF23]] ]
; CHECK-NEXT:    [[TMP65:%.*]] = extractelement <4 x i1> [[TMP10]], i64 1
; CHECK-NEXT:    br i1 [[TMP65]], label [[PRED_LOAD_IF25:%.*]], label [[PRED_LOAD_CONTINUE26:%.*]]
; CHECK:       pred.load.if22:
; CHECK-NEXT:    [[TMP66:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP67:%.*]] = getelementptr i8, ptr [[TMP66]], i64 36
; CHECK-NEXT:    [[TMP68:%.*]] = load i32, ptr [[TMP67]], align 4
; CHECK-NEXT:    [[TMP69:%.*]] = insertelement <4 x i32> [[TMP64]], i32 [[TMP68]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE26]]
; CHECK:       pred.load.continue23:
; CHECK-NEXT:    [[TMP70:%.*]] = phi <4 x i32> [ [[TMP64]], [[PRED_LOAD_CONTINUE24]] ], [ [[TMP69]], [[PRED_LOAD_IF25]] ]
; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <4 x i1> [[TMP10]], i64 2
; CHECK-NEXT:    br i1 [[TMP71]], label [[PRED_LOAD_IF27:%.*]], label [[PRED_LOAD_CONTINUE28:%.*]]
; CHECK:       pred.load.if24:
; CHECK-NEXT:    [[TMP72:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP73:%.*]] = getelementptr i8, ptr [[TMP72]], i64 40
; CHECK-NEXT:    [[TMP74:%.*]] = load i32, ptr [[TMP73]], align 4
; CHECK-NEXT:    [[TMP75:%.*]] = insertelement <4 x i32> [[TMP70]], i32 [[TMP74]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE28]]
; CHECK:       pred.load.continue25:
; CHECK-NEXT:    [[TMP76:%.*]] = phi <4 x i32> [ [[TMP70]], [[PRED_LOAD_CONTINUE26]] ], [ [[TMP75]], [[PRED_LOAD_IF27]] ]
; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <4 x i1> [[TMP10]], i64 3
; CHECK-NEXT:    br i1 [[TMP77]], label [[PRED_LOAD_IF29:%.*]], label [[PRED_LOAD_CONTINUE30:%.*]]
; CHECK:       pred.load.if26:
; CHECK-NEXT:    [[TMP78:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP79:%.*]] = getelementptr i8, ptr [[TMP78]], i64 44
; CHECK-NEXT:    [[TMP80:%.*]] = load i32, ptr [[TMP79]], align 4
; CHECK-NEXT:    [[TMP81:%.*]] = insertelement <4 x i32> [[TMP76]], i32 [[TMP80]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE30]]
; CHECK:       pred.load.continue27:
; CHECK-NEXT:    [[TMP82:%.*]] = phi <4 x i32> [ [[TMP76]], [[PRED_LOAD_CONTINUE28]] ], [ [[TMP81]], [[PRED_LOAD_IF29]] ]
; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <4 x i1> [[TMP11]], i64 0
; CHECK-NEXT:    br i1 [[TMP83]], label [[PRED_LOAD_IF31:%.*]], label [[PRED_LOAD_CONTINUE32:%.*]]
; CHECK:       pred.load.if28:
; CHECK-NEXT:    [[TMP84:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP85:%.*]] = getelementptr i8, ptr [[TMP84]], i64 48
; CHECK-NEXT:    [[TMP86:%.*]] = load i32, ptr [[TMP85]], align 4
; CHECK-NEXT:    [[TMP87:%.*]] = insertelement <4 x i32> poison, i32 [[TMP86]], i64 0
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE32]]
; CHECK:       pred.load.continue29:
; CHECK-NEXT:    [[TMP88:%.*]] = phi <4 x i32> [ poison, [[PRED_LOAD_CONTINUE30]] ], [ [[TMP87]], [[PRED_LOAD_IF31]] ]
; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <4 x i1> [[TMP11]], i64 1
; CHECK-NEXT:    br i1 [[TMP89]], label [[PRED_LOAD_IF33:%.*]], label [[PRED_LOAD_CONTINUE34:%.*]]
; CHECK:       pred.load.if30:
; CHECK-NEXT:    [[TMP90:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP91:%.*]] = getelementptr i8, ptr [[TMP90]], i64 52
; CHECK-NEXT:    [[TMP92:%.*]] = load i32, ptr [[TMP91]], align 4
; CHECK-NEXT:    [[TMP93:%.*]] = insertelement <4 x i32> [[TMP88]], i32 [[TMP92]], i64 1
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE34]]
; CHECK:       pred.load.continue31:
; CHECK-NEXT:    [[TMP94:%.*]] = phi <4 x i32> [ [[TMP88]], [[PRED_LOAD_CONTINUE32]] ], [ [[TMP93]], [[PRED_LOAD_IF33]] ]
; CHECK-NEXT:    [[TMP95:%.*]] = extractelement <4 x i1> [[TMP11]], i64 2
; CHECK-NEXT:    br i1 [[TMP95]], label [[PRED_LOAD_IF35:%.*]], label [[PRED_LOAD_CONTINUE36:%.*]]
; CHECK:       pred.load.if32:
; CHECK-NEXT:    [[TMP96:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP97:%.*]] = getelementptr i8, ptr [[TMP96]], i64 56
; CHECK-NEXT:    [[TMP98:%.*]] = load i32, ptr [[TMP97]], align 4
; CHECK-NEXT:    [[TMP99:%.*]] = insertelement <4 x i32> [[TMP94]], i32 [[TMP98]], i64 2
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE36]]
; CHECK:       pred.load.continue33:
; CHECK-NEXT:    [[TMP100:%.*]] = phi <4 x i32> [ [[TMP94]], [[PRED_LOAD_CONTINUE34]] ], [ [[TMP99]], [[PRED_LOAD_IF35]] ]
; CHECK-NEXT:    [[TMP101:%.*]] = extractelement <4 x i1> [[TMP11]], i64 3
; CHECK-NEXT:    br i1 [[TMP101]], label [[PRED_LOAD_IF37:%.*]], label [[PRED_LOAD_CONTINUE38]]
; CHECK:       pred.load.if34:
; CHECK-NEXT:    [[TMP102:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT:    [[TMP103:%.*]] = getelementptr i8, ptr [[TMP102]], i64 60
; CHECK-NEXT:    [[TMP104:%.*]] = load i32, ptr [[TMP103]], align 4
; CHECK-NEXT:    [[TMP105:%.*]] = insertelement <4 x i32> [[TMP100]], i32 [[TMP104]], i64 3
; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE38]]
; CHECK:       pred.load.continue35:
; CHECK-NEXT:    [[TMP106:%.*]] = phi <4 x i32> [ [[TMP100]], [[PRED_LOAD_CONTINUE36]] ], [ [[TMP105]], [[PRED_LOAD_IF37]] ]
; CHECK-NEXT:    [[TMP107:%.*]] = select <4 x i1> [[TMP8]], <4 x i32> [[TMP34]], <4 x i32> splat (i32 1)
; CHECK-NEXT:    [[TMP108:%.*]] = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> [[TMP107]])
; CHECK-NEXT:    [[TMP109]] = mul i32 [[TMP108]], [[VEC_PHI]]
; CHECK-NEXT:    [[TMP110:%.*]] = select <4 x i1> [[TMP9]], <4 x i32> [[TMP58]], <4 x i32> splat (i32 1)
; CHECK-NEXT:    [[TMP111:%.*]] = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> [[TMP110]])
; CHECK-NEXT:    [[TMP112]] = mul i32 [[TMP111]], [[VEC_PHI4]]
; CHECK-NEXT:    [[TMP113:%.*]] = select <4 x i1> [[TMP10]], <4 x i32> [[TMP82]], <4 x i32> splat (i32 1)
; CHECK-NEXT:    [[TMP114:%.*]] = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> [[TMP113]])
; CHECK-NEXT:    [[TMP115]] = mul i32 [[TMP114]], [[VEC_PHI5]]
; CHECK-NEXT:    [[TMP116:%.*]] = select <4 x i1> [[TMP11]], <4 x i32> [[TMP106]], <4 x i32> splat (i32 1)
; CHECK-NEXT:    [[TMP117:%.*]] = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> [[TMP116]])
; CHECK-NEXT:    [[TMP118]] = mul i32 [[TMP117]], [[VEC_PHI6]]
; CHECK-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], 16
; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 16)
; CHECK-NEXT:    [[TMP119:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP119]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    [[BIN_RDX:%.*]] = mul i32 [[TMP112]], [[TMP109]]
; CHECK-NEXT:    [[BIN_RDX39:%.*]] = mul i32 [[TMP115]], [[BIN_RDX]]
; CHECK-NEXT:    [[BIN_RDX40:%.*]] = mul i32 [[TMP118]], [[BIN_RDX39]]
; CHECK-NEXT:    br label [[FOR_END:%.*]]
; CHECK:       scalar.ph:
; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
; CHECK:       for.body:
; CHECK-NEXT:    br i1 poison, label [[IF_THEN:%.*]], label [[FOR_INC:%.*]]
; CHECK:       if.then:
; CHECK-NEXT:    br label [[FOR_INC]]
; CHECK:       for.inc:
; CHECK-NEXT:    br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK:       for.end:
; CHECK-NEXT:    [[RES_LCSSA:%.*]] = phi i32 [ poison, [[FOR_INC]] ], [ [[BIN_RDX40]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT:    ret i32 [[RES_LCSSA]]
;
entry:
  br label %for.body

for.body:
  %iv = phi i64 [ %inc, %for.inc ], [ 0, %entry ]
  %sum = phi i32 [ %res, %for.inc ], [ 4, %entry ]
  %cmp1 = icmp sgt i32 %cond, 7
  br i1 %cmp1, label %if.then, label %for.inc

if.then:
  %arrayidx = getelementptr inbounds i32, ptr %a, i64 %iv
  %load = load i32, ptr %arrayidx
  %mul = mul nsw i32 %load, %sum
  br label %for.inc

for.inc:
  %res = phi i32 [ %mul, %if.then ], [ %sum, %for.body ]
  %inc = add nuw nsw i64 %iv, 1
  %exitcond.not = icmp eq i64 %inc, %N
  br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !6

for.end:
  ret i32 %res
}

!6 = distinct !{!6, !7, !8}
!7 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}
!8 = !{!"llvm.loop.vectorize.enable", i1 true}