llvm/test/Transforms/LoopVectorize/X86/cost-conditional-branches.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -S | FileCheck %s

target triple = "x86_64-apple-macosx10.8.0"

; Test case for https://github.com/llvm/llvm-project/issues/156091.
define void @test_replicate_call_chain(float %x, ptr noalias %A, ptr noalias %B, ptr align 4 noalias %C, ptr align 4 noalias %D, ptr noalias %E) #0 {
; CHECK-LABEL: @test_replicate_call_chain(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    br 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:    [[TMP0:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <16 x float>, ptr [[TMP1]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = fcmp ogt <16 x float> [[WIDE_LOAD]], zeroinitializer
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[INDEX]]
; CHECK-NEXT:    [[WIDE_LOAD1:%.*]] = load <16 x float>, ptr [[TMP3]], align 4
; CHECK-NEXT:    [[TMP4:%.*]] = fcmp ogt <16 x float> [[WIDE_LOAD1]], zeroinitializer
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr float, ptr [[C:%.*]], i64 [[TMP0]]
; CHECK-NEXT:    [[TMP6:%.*]] = and <16 x i1> [[TMP2]], [[TMP4]]
; CHECK-NEXT:    [[TMP7:%.*]] = xor <16 x i1> [[TMP6]], splat (i1 true)
; CHECK-NEXT:    [[TMP8:%.*]] = shl i64 [[INDEX]], 2
; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr i8, ptr [[D:%.*]], i64 [[TMP8]]
; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <16 x float> @llvm.masked.load.v16f32.p0(ptr align 4 [[TMP9]], <16 x i1> [[TMP7]], <16 x float> poison)
; CHECK-NEXT:    [[TMP10:%.*]] = fmul <16 x float> [[WIDE_MASKED_LOAD]], splat (float 2.000000e+00)
; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <16 x float> [[TMP10]], i64 0
; CHECK-NEXT:    [[TMP12:%.*]] = tail call float @llvm.pow.f32(float [[TMP11]], float [[X:%.*]])
; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <16 x float> [[TMP10]], i64 1
; CHECK-NEXT:    [[TMP14:%.*]] = tail call float @llvm.pow.f32(float [[TMP13]], float [[X]])
; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <16 x float> [[TMP10]], i64 2
; CHECK-NEXT:    [[TMP16:%.*]] = tail call float @llvm.pow.f32(float [[TMP15]], float [[X]])
; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <16 x float> [[TMP10]], i64 3
; CHECK-NEXT:    [[TMP18:%.*]] = tail call float @llvm.pow.f32(float [[TMP17]], float [[X]])
; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <16 x float> [[TMP10]], i64 4
; CHECK-NEXT:    [[TMP20:%.*]] = tail call float @llvm.pow.f32(float [[TMP19]], float [[X]])
; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <16 x float> [[TMP10]], i64 5
; CHECK-NEXT:    [[TMP22:%.*]] = tail call float @llvm.pow.f32(float [[TMP21]], float [[X]])
; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <16 x float> [[TMP10]], i64 6
; CHECK-NEXT:    [[TMP24:%.*]] = tail call float @llvm.pow.f32(float [[TMP23]], float [[X]])
; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <16 x float> [[TMP10]], i64 7
; CHECK-NEXT:    [[TMP26:%.*]] = tail call float @llvm.pow.f32(float [[TMP25]], float [[X]])
; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <16 x float> [[TMP10]], i64 8
; CHECK-NEXT:    [[TMP28:%.*]] = tail call float @llvm.pow.f32(float [[TMP27]], float [[X]])
; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <16 x float> [[TMP10]], i64 9
; CHECK-NEXT:    [[TMP30:%.*]] = tail call float @llvm.pow.f32(float [[TMP29]], float [[X]])
; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <16 x float> [[TMP10]], i64 10
; CHECK-NEXT:    [[TMP32:%.*]] = tail call float @llvm.pow.f32(float [[TMP31]], float [[X]])
; CHECK-NEXT:    [[TMP33:%.*]] = extractelement <16 x float> [[TMP10]], i64 11
; CHECK-NEXT:    [[TMP34:%.*]] = tail call float @llvm.pow.f32(float [[TMP33]], float [[X]])
; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <16 x float> [[TMP10]], i64 12
; CHECK-NEXT:    [[TMP36:%.*]] = tail call float @llvm.pow.f32(float [[TMP35]], float [[X]])
; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <16 x float> [[TMP10]], i64 13
; CHECK-NEXT:    [[TMP38:%.*]] = tail call float @llvm.pow.f32(float [[TMP37]], float [[X]])
; CHECK-NEXT:    [[TMP39:%.*]] = extractelement <16 x float> [[TMP10]], i64 14
; CHECK-NEXT:    [[TMP40:%.*]] = tail call float @llvm.pow.f32(float [[TMP39]], float [[X]])
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <16 x float> [[TMP10]], i64 15
; CHECK-NEXT:    [[TMP42:%.*]] = tail call float @llvm.pow.f32(float [[TMP41]], float [[X]])
; CHECK-NEXT:    [[TMP43:%.*]] = tail call float @llvm.pow.f32(float [[TMP12]], float [[X]])
; CHECK-NEXT:    [[TMP44:%.*]] = tail call float @llvm.pow.f32(float [[TMP14]], float [[X]])
; CHECK-NEXT:    [[TMP45:%.*]] = tail call float @llvm.pow.f32(float [[TMP16]], float [[X]])
; CHECK-NEXT:    [[TMP46:%.*]] = tail call float @llvm.pow.f32(float [[TMP18]], float [[X]])
; CHECK-NEXT:    [[TMP47:%.*]] = tail call float @llvm.pow.f32(float [[TMP20]], float [[X]])
; CHECK-NEXT:    [[TMP48:%.*]] = tail call float @llvm.pow.f32(float [[TMP22]], float [[X]])
; CHECK-NEXT:    [[TMP49:%.*]] = tail call float @llvm.pow.f32(float [[TMP24]], float [[X]])
; CHECK-NEXT:    [[TMP50:%.*]] = tail call float @llvm.pow.f32(float [[TMP26]], float [[X]])
; CHECK-NEXT:    [[TMP51:%.*]] = tail call float @llvm.pow.f32(float [[TMP28]], float [[X]])
; CHECK-NEXT:    [[TMP52:%.*]] = tail call float @llvm.pow.f32(float [[TMP30]], float [[X]])
; CHECK-NEXT:    [[TMP53:%.*]] = tail call float @llvm.pow.f32(float [[TMP32]], float [[X]])
; CHECK-NEXT:    [[TMP54:%.*]] = tail call float @llvm.pow.f32(float [[TMP34]], float [[X]])
; CHECK-NEXT:    [[TMP55:%.*]] = tail call float @llvm.pow.f32(float [[TMP36]], float [[X]])
; CHECK-NEXT:    [[TMP56:%.*]] = tail call float @llvm.pow.f32(float [[TMP38]], float [[X]])
; CHECK-NEXT:    [[TMP57:%.*]] = tail call float @llvm.pow.f32(float [[TMP40]], float [[X]])
; CHECK-NEXT:    [[TMP58:%.*]] = tail call float @llvm.pow.f32(float [[TMP42]], float [[X]])
; CHECK-NEXT:    [[TMP59:%.*]] = insertelement <16 x float> poison, float [[TMP43]], i64 0
; CHECK-NEXT:    [[TMP60:%.*]] = insertelement <16 x float> [[TMP59]], float [[TMP44]], i64 1
; CHECK-NEXT:    [[TMP61:%.*]] = insertelement <16 x float> [[TMP60]], float [[TMP45]], i64 2
; CHECK-NEXT:    [[TMP62:%.*]] = insertelement <16 x float> [[TMP61]], float [[TMP46]], i64 3
; CHECK-NEXT:    [[TMP63:%.*]] = insertelement <16 x float> [[TMP62]], float [[TMP47]], i64 4
; CHECK-NEXT:    [[TMP64:%.*]] = insertelement <16 x float> [[TMP63]], float [[TMP48]], i64 5
; CHECK-NEXT:    [[TMP65:%.*]] = insertelement <16 x float> [[TMP64]], float [[TMP49]], i64 6
; CHECK-NEXT:    [[TMP66:%.*]] = insertelement <16 x float> [[TMP65]], float [[TMP50]], i64 7
; CHECK-NEXT:    [[TMP67:%.*]] = insertelement <16 x float> [[TMP66]], float [[TMP51]], i64 8
; CHECK-NEXT:    [[TMP68:%.*]] = insertelement <16 x float> [[TMP67]], float [[TMP52]], i64 9
; CHECK-NEXT:    [[TMP69:%.*]] = insertelement <16 x float> [[TMP68]], float [[TMP53]], i64 10
; CHECK-NEXT:    [[TMP70:%.*]] = insertelement <16 x float> [[TMP69]], float [[TMP54]], i64 11
; CHECK-NEXT:    [[TMP71:%.*]] = insertelement <16 x float> [[TMP70]], float [[TMP55]], i64 12
; CHECK-NEXT:    [[TMP72:%.*]] = insertelement <16 x float> [[TMP71]], float [[TMP56]], i64 13
; CHECK-NEXT:    [[TMP73:%.*]] = insertelement <16 x float> [[TMP72]], float [[TMP57]], i64 14
; CHECK-NEXT:    [[TMP74:%.*]] = insertelement <16 x float> [[TMP73]], float [[TMP58]], i64 15
; CHECK-NEXT:    call void @llvm.masked.store.v16f32.p0(<16 x float> [[TMP74]], ptr align 4 [[TMP5]], <16 x i1> [[TMP7]])
; CHECK-NEXT:    call void @llvm.masked.store.v16f32.p0(<16 x float> zeroinitializer, ptr align 4 [[TMP5]], <16 x i1> [[TMP6]])
; CHECK-NEXT:    store float 0.000000e+00, ptr [[E:%.*]], align 4
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT:    [[TMP75:%.*]] = icmp eq i64 [[INDEX_NEXT]], 96
; CHECK-NEXT:    br i1 [[TMP75]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    br label [[SCALAR_PH:%.*]]
; CHECK:       scalar.ph:
; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
; CHECK:       loop.header:
; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ 96, [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT:    [[DEC_IV:%.*]] = phi i64 [ 4, [[SCALAR_PH]] ], [ [[DEC_IV_NEXT:%.*]], [[LOOP_LATCH]] ]
; CHECK-NEXT:    [[IV_INC:%.*]] = add i64 [[IV]], 1
; CHECK-NEXT:    [[GEP_A:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]]
; CHECK-NEXT:    [[L_A:%.*]] = load float, ptr [[GEP_A]], align 4
; CHECK-NEXT:    [[C_A:%.*]] = fcmp ogt float [[L_A]], 0.000000e+00
; CHECK-NEXT:    [[GEP_B:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[IV]]
; CHECK-NEXT:    [[L_B:%.*]] = load float, ptr [[GEP_B]], align 4
; CHECK-NEXT:    [[C_B:%.*]] = fcmp ogt float [[L_B]], 0.000000e+00
; CHECK-NEXT:    [[GEP_C:%.*]] = getelementptr float, ptr [[C]], i64 [[IV_INC]]
; CHECK-NEXT:    [[AND:%.*]] = and i1 [[C_A]], [[C_B]]
; CHECK-NEXT:    br i1 [[AND]], label [[THEN:%.*]], label [[ELSE:%.*]]
; CHECK:       then:
; CHECK-NEXT:    store float 0.000000e+00, ptr [[GEP_C]], align 4
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       else:
; CHECK-NEXT:    [[IV_MUL_2:%.*]] = shl i64 [[IV]], 2
; CHECK-NEXT:    [[GEP_D:%.*]] = getelementptr i8, ptr [[D]], i64 [[IV_MUL_2]]
; CHECK-NEXT:    [[L_D:%.*]] = load float, ptr [[GEP_D]], align 4
; CHECK-NEXT:    [[MUL:%.*]] = fmul float [[L_D]], 2.000000e+00
; CHECK-NEXT:    [[POW_1:%.*]] = tail call float @llvm.pow.f32(float [[MUL]], float [[X]])
; CHECK-NEXT:    [[POW_2:%.*]] = tail call float @llvm.pow.f32(float [[POW_1]], float [[X]])
; CHECK-NEXT:    store float [[POW_2]], ptr [[GEP_C]], align 4
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       loop.latch:
; CHECK-NEXT:    store float 0.000000e+00, ptr [[E]], align 4
; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
; CHECK-NEXT:    [[DEC_IV_NEXT]] = add i64 [[DEC_IV]], -1
; CHECK-NEXT:    [[EC:%.*]] = icmp ne i64 [[DEC_IV_NEXT]], 0
; CHECK-NEXT:    br i1 [[EC]], label [[LOOP_HEADER]], label [[EXIT:%.*]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK:       exit:
; CHECK-NEXT:    ret void
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %dec.iv = phi i64 [ 100, %entry ], [ %dec.iv.next, %loop.latch ]
  %iv.inc = add i64 %iv, 1
  %gep.A = getelementptr inbounds float, ptr %A, i64 %iv
  %l.A = load float, ptr %gep.A, align 4
  %c.A = fcmp ogt float %l.A, 0.0
  %gep.B = getelementptr inbounds float, ptr %B, i64 %iv
  %l.B = load float, ptr %gep.B, align 4
  %c.B = fcmp ogt float %l.B, 0.0
  %gep.C = getelementptr float, ptr %C, i64 %iv.inc
  %and = and i1 %c.A, %c.B
  br i1 %and, label %then, label %else

then:
  store float 0.0, ptr %gep.C, align 4
  br label %loop.latch

else:
  %iv.mul.2 = shl i64 %iv, 2
  %gep.D = getelementptr i8, ptr %D, i64 %iv.mul.2
  %l.D = load float, ptr %gep.D, align 4
  %mul = fmul float %l.D, 2.0
  %pow.1 = tail call float @llvm.pow.f32(float %mul, float %x)
  %pow.2 = tail call float @llvm.pow.f32(float %pow.1, float %x)
  store float %pow.2, ptr %gep.C, align 4
  br label %loop.latch

loop.latch:
  store float 0.000000e+00, ptr %E, align 4
  %iv.next = add i64 %iv, 1
  %dec.iv.next = add i64 %dec.iv, -1
  %ec = icmp ne i64 %dec.iv.next, 0
  br i1 %ec, label %loop.header, label %exit

exit:
  ret void
}

define i64 @avx512_cond_load_cost(ptr %src, i32 %a, i64 %b, i32 %c, i32 %d) #1 {
; CHECK-LABEL: @avx512_cond_load_cost(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[C:%.*]], 1
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 8
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[TMP0]], 7
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP0]], [[TMP1]]
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <8 x i32> poison, i32 [[A:%.*]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT1:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT2]], <8 x i32> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <8 x i32> poison, i32 [[C]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT1]], <8 x i32> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <8 x i32> poison, i32 [[D:%.*]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT3]], <8 x i32> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[B:%.*]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <8 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[TMP2:%.*]] = icmp slt <8 x i32> [[VEC_IND]], zeroinitializer
; CHECK-NEXT:    [[TMP34:%.*]] = call <8 x i32> @llvm.masked.urem.v8i32(<8 x i32> [[BROADCAST_SPLAT1]], <8 x i32> [[BROADCAST_SPLAT2]], <8 x i1> [[TMP2]])
; CHECK-NEXT:    [[TMP35:%.*]] = sub <8 x i32> zeroinitializer, [[TMP34]]
; CHECK-NEXT:    [[TMP59:%.*]] = call <8 x i32> @llvm.masked.udiv.v8i32(<8 x i32> [[BROADCAST_SPLAT2]], <8 x i32> [[BROADCAST_SPLAT4]], <8 x i1> [[TMP2]])
; CHECK-NEXT:    [[TMP60:%.*]] = or <8 x i32> [[TMP59]], [[TMP35]]
; CHECK-NEXT:    [[TMP61:%.*]] = sext <8 x i32> [[TMP60]] to <8 x i64>
; CHECK-NEXT:    [[WIDE_GEP:%.*]] = getelementptr { i64, i64, i64 }, ptr [[SRC:%.*]], <8 x i64> [[TMP61]], i32 2
; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <8 x i64> @llvm.masked.gather.v8i64.v8p0(<8 x ptr> align 8 [[WIDE_GEP]], <8 x i1> [[TMP2]], <8 x i64> poison)
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 8
; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <8 x i32> [[VEC_IND]], splat (i32 8)
; CHECK-NEXT:    [[TMP62:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP62]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    [[TMP63:%.*]] = or <8 x i64> [[WIDE_MASKED_GATHER]], [[BROADCAST_SPLAT]]
; CHECK-NEXT:    [[PREDPHI:%.*]] = select <8 x i1> [[TMP2]], <8 x i64> [[TMP63]], <8 x i64> zeroinitializer
; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <8 x i64> [[PREDPHI]], i64 7
; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP0]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK:       scalar.ph:
; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
; CHECK:       loop.header:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT:    [[C_1:%.*]] = icmp slt i32 [[IV]], 0
; CHECK-NEXT:    br i1 [[C_1]], label [[IF_THEN:%.*]], label [[LOOP_LATCH]]
; CHECK:       if.then:
; CHECK-NEXT:    [[TMP65:%.*]] = urem i32 [[A]], [[C]]
; CHECK-NEXT:    [[MUL:%.*]] = sub i32 0, [[TMP65]]
; CHECK-NEXT:    [[DIV:%.*]] = udiv i32 [[C]], [[D]]
; CHECK-NEXT:    [[OR:%.*]] = or i32 [[DIV]], [[MUL]]
; CHECK-NEXT:    [[EXT:%.*]] = sext i32 [[OR]] to i64
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr { i64, i64, i64 }, ptr [[SRC]], i64 [[EXT]], i32 2
; CHECK-NEXT:    [[L:%.*]] = load i64, ptr [[GEP]], align 8
; CHECK-NEXT:    [[OR_2:%.*]] = or i64 [[L]], [[B]]
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       loop.latch:
; CHECK-NEXT:    [[RES:%.*]] = phi i64 [ 0, [[LOOP_HEADER]] ], [ [[OR_2]], [[IF_THEN]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add i32 [[IV]], 1
; CHECK-NEXT:    [[EC:%.*]] = icmp ult i32 [[IV]], [[C]]
; CHECK-NEXT:    br i1 [[EC]], label [[LOOP_HEADER]], label [[EXIT]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK:       exit:
; CHECK-NEXT:    [[RES_LCSSA:%.*]] = phi i64 [ [[RES]], [[LOOP_LATCH]] ], [ [[TMP64]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT:    ret i64 [[RES_LCSSA]]
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %c.1 = icmp slt i32 %iv, 0
  br i1 %c.1, label %if.then, label %loop.latch

if.then:
  %1 = urem i32 %a, %c
  %mul = sub i32 0, %1
  %div = udiv i32 %c, %d
  %or = or i32 %div, %mul
  %ext = sext i32 %or to i64
  %gep = getelementptr { i64, i64, i64 }, ptr %src, i64 %ext, i32 2
  %l = load i64, ptr %gep, align 8
  %or.2 = or i64 %l, %b
  br label %loop.latch

loop.latch:
  %res = phi i64 [ 0, %loop.header ], [ %or.2, %if.then ]
  %iv.next = add i32 %iv, 1
  %ec = icmp ult i32 %iv, %c
  br i1 %ec, label %loop.header, label %exit

exit:
  ret i64 %res
}

define void @cost_duplicate_recipe_for_sinking(ptr %A, i64 %N) #2 {
; CHECK-LABEL: @cost_duplicate_recipe_for_sinking(
; CHECK-NEXT:  iter.check:
; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N:%.*]], 1
; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP0]], 4
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
; CHECK:       vector.main.loop.iter.check:
; CHECK-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ule i64 [[TMP0]], 16
; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    [[TMP1:%.*]] = and i64 [[TMP0]], 15
; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i64 [[TMP1]], 0
; CHECK-NEXT:    [[TMP3:%.*]] = select i1 [[TMP2]], i64 16, i64 [[TMP1]]
; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP0]], [[TMP3]]
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE37:%.*]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = add i64 [[INDEX]], 4
; CHECK-NEXT:    [[TMP5:%.*]] = add i64 [[INDEX]], 8
; CHECK-NEXT:    [[TMP6:%.*]] = add i64 [[INDEX]], 12
; CHECK-NEXT:    [[TMP7:%.*]] = shl nsw i64 [[INDEX]], 2
; CHECK-NEXT:    [[TMP8:%.*]] = shl nsw i64 [[TMP4]], 2
; CHECK-NEXT:    [[TMP9:%.*]] = shl nsw i64 [[TMP5]], 2
; CHECK-NEXT:    [[TMP10:%.*]] = shl nsw i64 [[TMP6]], 2
; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr double, ptr [[A:%.*]], i64 [[TMP7]]
; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP8]]
; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP9]]
; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP10]]
; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x double>, ptr [[TMP11]], align 8
; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x double> [[WIDE_VEC]], <16 x double> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
; CHECK-NEXT:    [[WIDE_VEC2:%.*]] = load <16 x double>, ptr [[TMP12]], align 8
; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <16 x double> [[WIDE_VEC2]], <16 x double> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
; CHECK-NEXT:    [[WIDE_VEC4:%.*]] = load <16 x double>, ptr [[TMP13]], align 8
; CHECK-NEXT:    [[STRIDED_VEC5:%.*]] = shufflevector <16 x double> [[WIDE_VEC4]], <16 x double> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
; CHECK-NEXT:    [[WIDE_VEC6:%.*]] = load <16 x double>, ptr [[TMP14]], align 8
; CHECK-NEXT:    [[STRIDED_VEC7:%.*]] = shufflevector <16 x double> [[WIDE_VEC6]], <16 x double> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
; CHECK-NEXT:    [[TMP15:%.*]] = fcmp oeq <4 x double> [[STRIDED_VEC]], zeroinitializer
; CHECK-NEXT:    [[TMP16:%.*]] = fcmp oeq <4 x double> [[STRIDED_VEC3]], zeroinitializer
; CHECK-NEXT:    [[TMP17:%.*]] = fcmp oeq <4 x double> [[STRIDED_VEC5]], zeroinitializer
; CHECK-NEXT:    [[TMP18:%.*]] = fcmp oeq <4 x double> [[STRIDED_VEC7]], zeroinitializer
; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <4 x i1> [[TMP15]], i64 0
; CHECK-NEXT:    br i1 [[TMP19]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK:       pred.store.if:
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP11]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE]]
; CHECK:       pred.store.continue:
; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <4 x i1> [[TMP15]], i64 1
; CHECK-NEXT:    br i1 [[TMP20]], label [[PRED_STORE_IF8:%.*]], label [[PRED_STORE_CONTINUE9:%.*]]
; CHECK:       pred.store.if8:
; CHECK-NEXT:    [[TMP21:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT:    [[TMP22:%.*]] = shl nsw i64 [[TMP21]], 2
; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP22]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP23]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE9]]
; CHECK:       pred.store.continue9:
; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <4 x i1> [[TMP15]], i64 2
; CHECK-NEXT:    br i1 [[TMP24]], label [[PRED_STORE_IF10:%.*]], label [[PRED_STORE_CONTINUE11:%.*]]
; CHECK:       pred.store.if10:
; CHECK-NEXT:    [[TMP25:%.*]] = add i64 [[INDEX]], 2
; CHECK-NEXT:    [[TMP26:%.*]] = shl nsw i64 [[TMP25]], 2
; CHECK-NEXT:    [[TMP27:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP26]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP27]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE11]]
; CHECK:       pred.store.continue11:
; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <4 x i1> [[TMP15]], i64 3
; CHECK-NEXT:    br i1 [[TMP28]], label [[PRED_STORE_IF12:%.*]], label [[PRED_STORE_CONTINUE13:%.*]]
; CHECK:       pred.store.if12:
; CHECK-NEXT:    [[TMP29:%.*]] = add i64 [[INDEX]], 3
; CHECK-NEXT:    [[TMP30:%.*]] = shl nsw i64 [[TMP29]], 2
; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP30]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP31]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE13]]
; CHECK:       pred.store.continue13:
; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <4 x i1> [[TMP16]], i64 0
; CHECK-NEXT:    br i1 [[TMP32]], label [[PRED_STORE_IF14:%.*]], label [[PRED_STORE_CONTINUE15:%.*]]
; CHECK:       pred.store.if14:
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP12]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE15]]
; CHECK:       pred.store.continue15:
; CHECK-NEXT:    [[TMP33:%.*]] = extractelement <4 x i1> [[TMP16]], i64 1
; CHECK-NEXT:    br i1 [[TMP33]], label [[PRED_STORE_IF16:%.*]], label [[PRED_STORE_CONTINUE17:%.*]]
; CHECK:       pred.store.if16:
; CHECK-NEXT:    [[TMP34:%.*]] = add i64 [[INDEX]], 5
; CHECK-NEXT:    [[TMP35:%.*]] = shl nsw i64 [[TMP34]], 2
; CHECK-NEXT:    [[TMP36:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP35]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP36]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE17]]
; CHECK:       pred.store.continue17:
; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <4 x i1> [[TMP16]], i64 2
; CHECK-NEXT:    br i1 [[TMP37]], label [[PRED_STORE_IF18:%.*]], label [[PRED_STORE_CONTINUE19:%.*]]
; CHECK:       pred.store.if18:
; CHECK-NEXT:    [[TMP38:%.*]] = add i64 [[INDEX]], 6
; CHECK-NEXT:    [[TMP39:%.*]] = shl nsw i64 [[TMP38]], 2
; CHECK-NEXT:    [[TMP40:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP39]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP40]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE19]]
; CHECK:       pred.store.continue19:
; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <4 x i1> [[TMP16]], i64 3
; CHECK-NEXT:    br i1 [[TMP41]], label [[PRED_STORE_IF20:%.*]], label [[PRED_STORE_CONTINUE21:%.*]]
; CHECK:       pred.store.if20:
; CHECK-NEXT:    [[TMP42:%.*]] = add i64 [[INDEX]], 7
; CHECK-NEXT:    [[TMP43:%.*]] = shl nsw i64 [[TMP42]], 2
; CHECK-NEXT:    [[TMP44:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP43]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP44]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE21]]
; CHECK:       pred.store.continue21:
; CHECK-NEXT:    [[TMP45:%.*]] = extractelement <4 x i1> [[TMP17]], i64 0
; CHECK-NEXT:    br i1 [[TMP45]], label [[PRED_STORE_IF22:%.*]], label [[PRED_STORE_CONTINUE23:%.*]]
; CHECK:       pred.store.if22:
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP13]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE23]]
; CHECK:       pred.store.continue23:
; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <4 x i1> [[TMP17]], i64 1
; CHECK-NEXT:    br i1 [[TMP46]], label [[PRED_STORE_IF24:%.*]], label [[PRED_STORE_CONTINUE25:%.*]]
; CHECK:       pred.store.if24:
; CHECK-NEXT:    [[TMP47:%.*]] = add i64 [[INDEX]], 9
; CHECK-NEXT:    [[TMP48:%.*]] = shl nsw i64 [[TMP47]], 2
; CHECK-NEXT:    [[TMP49:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP48]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP49]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE25]]
; CHECK:       pred.store.continue25:
; CHECK-NEXT:    [[TMP50:%.*]] = extractelement <4 x i1> [[TMP17]], i64 2
; CHECK-NEXT:    br i1 [[TMP50]], label [[PRED_STORE_IF26:%.*]], label [[PRED_STORE_CONTINUE27:%.*]]
; CHECK:       pred.store.if26:
; CHECK-NEXT:    [[TMP51:%.*]] = add i64 [[INDEX]], 10
; CHECK-NEXT:    [[TMP52:%.*]] = shl nsw i64 [[TMP51]], 2
; CHECK-NEXT:    [[TMP53:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP52]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP53]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE27]]
; CHECK:       pred.store.continue27:
; CHECK-NEXT:    [[TMP54:%.*]] = extractelement <4 x i1> [[TMP17]], i64 3
; CHECK-NEXT:    br i1 [[TMP54]], label [[PRED_STORE_IF28:%.*]], label [[PRED_STORE_CONTINUE29:%.*]]
; CHECK:       pred.store.if28:
; CHECK-NEXT:    [[TMP55:%.*]] = add i64 [[INDEX]], 11
; CHECK-NEXT:    [[TMP56:%.*]] = shl nsw i64 [[TMP55]], 2
; CHECK-NEXT:    [[TMP57:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP56]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP57]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE29]]
; CHECK:       pred.store.continue29:
; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <4 x i1> [[TMP18]], i64 0
; CHECK-NEXT:    br i1 [[TMP58]], label [[PRED_STORE_IF30:%.*]], label [[PRED_STORE_CONTINUE31:%.*]]
; CHECK:       pred.store.if30:
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP14]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE31]]
; CHECK:       pred.store.continue31:
; CHECK-NEXT:    [[TMP59:%.*]] = extractelement <4 x i1> [[TMP18]], i64 1
; CHECK-NEXT:    br i1 [[TMP59]], label [[PRED_STORE_IF32:%.*]], label [[PRED_STORE_CONTINUE33:%.*]]
; CHECK:       pred.store.if32:
; CHECK-NEXT:    [[TMP60:%.*]] = add i64 [[INDEX]], 13
; CHECK-NEXT:    [[TMP61:%.*]] = shl nsw i64 [[TMP60]], 2
; CHECK-NEXT:    [[TMP62:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP61]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP62]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE33]]
; CHECK:       pred.store.continue33:
; CHECK-NEXT:    [[TMP63:%.*]] = extractelement <4 x i1> [[TMP18]], i64 2
; CHECK-NEXT:    br i1 [[TMP63]], label [[PRED_STORE_IF34:%.*]], label [[PRED_STORE_CONTINUE35:%.*]]
; CHECK:       pred.store.if34:
; CHECK-NEXT:    [[TMP64:%.*]] = add i64 [[INDEX]], 14
; CHECK-NEXT:    [[TMP65:%.*]] = shl nsw i64 [[TMP64]], 2
; CHECK-NEXT:    [[TMP66:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP65]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP66]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE35]]
; CHECK:       pred.store.continue35:
; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <4 x i1> [[TMP18]], i64 3
; CHECK-NEXT:    br i1 [[TMP67]], label [[PRED_STORE_IF36:%.*]], label [[PRED_STORE_CONTINUE37]]
; CHECK:       pred.store.if36:
; CHECK-NEXT:    [[TMP68:%.*]] = add i64 [[INDEX]], 15
; CHECK-NEXT:    [[TMP69:%.*]] = shl nsw i64 [[TMP68]], 2
; CHECK-NEXT:    [[TMP70:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP69]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP70]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE37]]
; CHECK:       pred.store.continue37:
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; CHECK-NEXT:    [[TMP71:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT:    br i1 [[TMP71]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    br label [[VEC_EPILOG_ITER_CHECK:%.*]]
; CHECK:       vec.epilog.iter.check:
; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP3]], 4
; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF7:![0-9]+]]
; CHECK:       vec.epilog.ph:
; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-NEXT:    [[TMP72:%.*]] = and i64 [[TMP0]], 3
; CHECK-NEXT:    [[TMP73:%.*]] = icmp eq i64 [[TMP72]], 0
; CHECK-NEXT:    [[TMP74:%.*]] = select i1 [[TMP73]], i64 4, i64 [[TMP72]]
; CHECK-NEXT:    [[N_VEC38:%.*]] = sub i64 [[TMP0]], [[TMP74]]
; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK:       vec.epilog.vector.body:
; CHECK-NEXT:    [[INDEX39:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT50:%.*]], [[PRED_STORE_CONTINUE49:%.*]] ]
; CHECK-NEXT:    [[TMP75:%.*]] = shl nsw i64 [[INDEX39]], 2
; CHECK-NEXT:    [[TMP76:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP75]]
; CHECK-NEXT:    [[WIDE_VEC40:%.*]] = load <16 x double>, ptr [[TMP76]], align 8
; CHECK-NEXT:    [[STRIDED_VEC41:%.*]] = shufflevector <16 x double> [[WIDE_VEC40]], <16 x double> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
; CHECK-NEXT:    [[TMP77:%.*]] = fcmp oeq <4 x double> [[STRIDED_VEC41]], zeroinitializer
; CHECK-NEXT:    [[TMP78:%.*]] = extractelement <4 x i1> [[TMP77]], i64 0
; CHECK-NEXT:    br i1 [[TMP78]], label [[PRED_STORE_IF42:%.*]], label [[PRED_STORE_CONTINUE43:%.*]]
; CHECK:       pred.store.if42:
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP76]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE43]]
; CHECK:       pred.store.continue43:
; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <4 x i1> [[TMP77]], i64 1
; CHECK-NEXT:    br i1 [[TMP79]], label [[PRED_STORE_IF44:%.*]], label [[PRED_STORE_CONTINUE45:%.*]]
; CHECK:       pred.store.if44:
; CHECK-NEXT:    [[TMP80:%.*]] = add i64 [[INDEX39]], 1
; CHECK-NEXT:    [[TMP81:%.*]] = shl nsw i64 [[TMP80]], 2
; CHECK-NEXT:    [[TMP82:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP81]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP82]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE45]]
; CHECK:       pred.store.continue45:
; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <4 x i1> [[TMP77]], i64 2
; CHECK-NEXT:    br i1 [[TMP83]], label [[PRED_STORE_IF46:%.*]], label [[PRED_STORE_CONTINUE47:%.*]]
; CHECK:       pred.store.if46:
; CHECK-NEXT:    [[TMP84:%.*]] = add i64 [[INDEX39]], 2
; CHECK-NEXT:    [[TMP85:%.*]] = shl nsw i64 [[TMP84]], 2
; CHECK-NEXT:    [[TMP86:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP85]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP86]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE47]]
; CHECK:       pred.store.continue47:
; CHECK-NEXT:    [[TMP87:%.*]] = extractelement <4 x i1> [[TMP77]], i64 3
; CHECK-NEXT:    br i1 [[TMP87]], label [[PRED_STORE_IF48:%.*]], label [[PRED_STORE_CONTINUE49]]
; CHECK:       pred.store.if48:
; CHECK-NEXT:    [[TMP88:%.*]] = add i64 [[INDEX39]], 3
; CHECK-NEXT:    [[TMP89:%.*]] = shl nsw i64 [[TMP88]], 2
; CHECK-NEXT:    [[TMP90:%.*]] = getelementptr double, ptr [[A]], i64 [[TMP89]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[TMP90]], align 8
; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE49]]
; CHECK:       pred.store.continue49:
; CHECK-NEXT:    [[INDEX_NEXT50]] = add nuw i64 [[INDEX39]], 4
; CHECK-NEXT:    [[TMP91:%.*]] = icmp eq i64 [[INDEX_NEXT50]], [[N_VEC38]]
; CHECK-NEXT:    br i1 [[TMP91]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK:       vec.epilog.middle.block:
; CHECK-NEXT:    br label [[VEC_EPILOG_SCALAR_PH]]
; CHECK:       vec.epilog.scalar.ph:
; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC38]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
; CHECK:       loop.header:
; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT:    [[IV_SHL:%.*]] = shl nsw i64 [[IV]], 2
; CHECK-NEXT:    [[GEP_0:%.*]] = getelementptr nusw double, ptr [[A]], i64 [[IV_SHL]]
; CHECK-NEXT:    [[L:%.*]] = load double, ptr [[GEP_0]], align 8
; CHECK-NEXT:    [[C:%.*]] = fcmp oeq double [[L]], 0.000000e+00
; CHECK-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[LOOP_LATCH]]
; CHECK:       if.then:
; CHECK-NEXT:    [[GEP_1:%.*]] = getelementptr double, ptr [[A]], i64 [[IV_SHL]]
; CHECK-NEXT:    store double 0.000000e+00, ptr [[GEP_1]], align 8
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       loop.latch:
; CHECK-NEXT:    [[IV_NEXT]] = add nsw i64 [[IV]], 1
; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], [[N]]
; CHECK-NEXT:    br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP_HEADER]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK:       exit:
; CHECK-NEXT:    ret void
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
  %iv.shl = shl nsw i64 %iv, 2
  %gep.0 = getelementptr nusw double, ptr %A, i64 %iv.shl
  %l = load double, ptr %gep.0, align 8
  %c = fcmp oeq double %l, 0.000000e+00
  br i1 %c, label %if.then, label %loop.latch

if.then:
  %gep.1 = getelementptr double, ptr %A, i64 %iv.shl
  store double 0.000000e+00, ptr %gep.1, align 8
  br label %loop.latch

loop.latch:
  %iv.next = add nsw i64 %iv, 1
  %ec = icmp eq i64 %iv, %N
  br i1 %ec, label %exit, label %loop.header

exit:
  ret void
}

; Test for https://github.com/llvm/llvm-project/issues/129236.
define i32 @cost_ashr_with_op_known_invariant_via_scev(i8 %a) {
; CHECK-LABEL: @cost_ashr_with_op_known_invariant_via_scev(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[CMP_I:%.*]] = icmp eq i16 0, 0
; CHECK-NEXT:    [[CONV_I:%.*]] = sext i16 0 to i32
; CHECK-NEXT:    [[CONV5_I:%.*]] = sext i8 [[A:%.*]] to i32
; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
; CHECK:       loop.header:
; CHECK-NEXT:    [[IV:%.*]] = phi i8 [ 100, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT:    br i1 [[CMP_I]], label [[THEN:%.*]], label [[ELSE:%.*]]
; CHECK:       then:
; CHECK-NEXT:    [[P_1:%.*]] = phi i32 [ [[REM_I:%.*]], [[ELSE]] ], [ 0, [[LOOP_HEADER]] ]
; CHECK-NEXT:    [[SHR_I:%.*]] = ashr i32 [[CONV5_I]], [[P_1]]
; CHECK-NEXT:    [[TOBOOL6_NOT_I:%.*]] = icmp eq i32 [[SHR_I]], 0
; CHECK-NEXT:    [[SEXT_I:%.*]] = shl i32 [[P_1]], 24
; CHECK-NEXT:    [[TMP0:%.*]] = ashr exact i32 [[SEXT_I]], 24
; CHECK-NEXT:    [[TMP1:%.*]] = select i1 [[TOBOOL6_NOT_I]], i32 [[TMP0]], i32 0
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       else:
; CHECK-NEXT:    [[REM_I]] = urem i32 -1, [[CONV_I]]
; CHECK-NEXT:    [[CMP3_I:%.*]] = icmp sgt i32 [[REM_I]], 1
; CHECK-NEXT:    br i1 [[CMP3_I]], label [[LOOP_LATCH]], label [[THEN]]
; CHECK:       loop.latch:
; CHECK-NEXT:    [[P_2:%.*]] = phi i32 [ 0, [[ELSE]] ], [ [[TMP1]], [[THEN]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add i8 [[IV]], -1
; CHECK-NEXT:    [[EC:%.*]] = icmp eq i8 [[IV_NEXT]], 0
; CHECK-NEXT:    br i1 [[EC]], label [[EXIT:%.*]], label [[LOOP_HEADER]]
; CHECK:       exit:
; CHECK-NEXT:    [[P_2_LCSSA:%.*]] = phi i32 [ [[P_2]], [[LOOP_LATCH]] ]
; CHECK-NEXT:    ret i32 [[P_2_LCSSA]]
;
entry:
  %cmp.i = icmp eq i16 0, 0
  %conv.i = sext i16 0 to i32
  %conv5.i = sext i8 %a to i32
  br label %loop.header

loop.header:
  %iv = phi i8 [ 100, %entry ], [ %iv.next, %loop.latch ]
  br i1 %cmp.i, label %then, label %else

then:
  %p.1 = phi i32 [ %rem.i, %else ], [ 0, %loop.header ]
  %shr.i = ashr i32 %conv5.i, %p.1
  %tobool6.not.i = icmp eq i32 %shr.i, 0
  %sext.i = shl i32 %p.1, 24
  %2 = ashr exact i32 %sext.i, 24
  %3 = select i1 %tobool6.not.i, i32 %2, i32 0
  br label %loop.latch

else:
  %rem.i = urem i32 -1, %conv.i
  %cmp3.i = icmp sgt i32 %rem.i, 1
  br i1 %cmp3.i, label %loop.latch, label %then

loop.latch:
  %p.2 = phi i32 [ 0, %else ], [ %3, %then ]
  %iv.next = add i8 %iv, -1
  %ec = icmp eq i8 %iv.next, 0
  br i1 %ec, label %exit, label %loop.header

exit:
  ret i32 %p.2
}

; Test case for https://github.com/llvm/llvm-project/issues/156066.
define void @sdiv_by_zero(ptr noalias %src, ptr noalias %dst, i32 %d) #2 {
; CHECK-LABEL: @sdiv_by_zero(
; CHECK-NEXT:  bb:
; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
; CHECK:       loop.header:
; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ], [ 0, [[BB:%.*]] ]
; CHECK-NEXT:    [[GEP_SRC:%.*]] = getelementptr inbounds i32, ptr [[SRC:%.*]], i64 [[IV]]
; CHECK-NEXT:    [[L:%.*]] = load i32, ptr [[GEP_SRC]], align 4
; CHECK-NEXT:    [[ICMP:%.*]] = icmp eq i32 [[L]], 0
; CHECK-NEXT:    br i1 [[ICMP]], label [[LOOP_LATCH]], label [[THEN:%.*]]
; CHECK:       then:
; CHECK-NEXT:    [[SDIV:%.*]] = sdiv i32 [[L]], 0
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       loop.latch:
; CHECK-NEXT:    [[MERGE:%.*]] = phi i32 [ [[SDIV]], [[THEN]] ], [ 0, [[LOOP_HEADER]] ]
; CHECK-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds i32, ptr [[DST:%.*]], i64 [[IV]]
; CHECK-NEXT:    store i32 [[MERGE]], ptr [[GEP_DST]], align 4
; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
; CHECK-NEXT:    [[EC:%.*]] = icmp ult i64 [[IV]], 16
; CHECK-NEXT:    br i1 [[EC]], label [[LOOP_HEADER]], label [[EXIT:%.*]]
; CHECK:       exit:
; CHECK-NEXT:    ret void
;
bb:
  br label %loop.header

loop.header:
  %iv = phi i64 [ %iv.next, %loop.latch ], [ 0, %bb ]
  %gep.src = getelementptr inbounds i32, ptr %src, i64 %iv
  %l = load i32, ptr %gep.src, align 4
  %icmp = icmp eq i32 %l, 0
  br i1 %icmp, label %loop.latch, label %then

then:
  %sdiv = sdiv i32 %l, 0
  br label %loop.latch

loop.latch:
  %merge = phi i32 [ %sdiv, %then ], [ 0, %loop.header ]
  %gep.dst = getelementptr inbounds i32, ptr %dst, i64 %iv
  store i32 %merge, ptr %gep.dst, align 4
  %iv.next = add i64 %iv, 1
  %ec = icmp ult i64 %iv, 16
  br i1 %ec, label %loop.header, label %exit

exit:
  ret void
}

; Test case for https://github.com/llvm/llvm-project/issues/158660.
define i64 @test_predicated_udiv(i32 %d, i1 %c) #2 {
; CHECK-LABEL: @test_predicated_udiv(
; CHECK-NEXT:  iter.check:
; CHECK-NEXT:    br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
; CHECK:       vector.main.loop.iter.check:
; CHECK-NEXT:    br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK:       vector.ph:
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <32 x i1> poison, i1 [[C:%.*]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT1:%.*]] = shufflevector <32 x i1> [[BROADCAST_SPLATINSERT2]], <32 x i1> poison, <32 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP0:%.*]] = xor <32 x i1> [[BROADCAST_SPLAT1]], splat (i1 true)
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <32 x i32> poison, i32 [[D:%.*]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <32 x i32> [[BROADCAST_SPLATINSERT1]], <32 x i32> poison, <32 x i32> zeroinitializer
; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
; CHECK:       vector.body:
; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <32 x i32> [ <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT:    [[STEP_ADD:%.*]] = add nuw <32 x i32> [[VEC_IND]], splat (i32 32)
; CHECK-NEXT:    [[STEP_ADD_2:%.*]] = add nuw <32 x i32> [[STEP_ADD]], splat (i32 32)
; CHECK-NEXT:    [[STEP_ADD_3:%.*]] = add nuw <32 x i32> [[STEP_ADD_2]], splat (i32 32)
; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 128
; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <32 x i32> [[STEP_ADD_3]], splat (i32 32)
; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[INDEX_NEXT]], 896
; CHECK-NEXT:    br i1 [[TMP1]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK:       middle.block:
; CHECK-NEXT:    [[TMP2:%.*]] = call <32 x i32> @llvm.usub.sat.v32i32(<32 x i32> [[STEP_ADD_3]], <32 x i32> splat (i32 1))
; CHECK-NEXT:    [[TMP3:%.*]] = call <32 x i32> @llvm.masked.udiv.v32i32(<32 x i32> [[TMP2]], <32 x i32> [[BROADCAST_SPLAT2]], <32 x i1> [[TMP0]])
; CHECK-NEXT:    [[TMP4:%.*]] = zext <32 x i32> [[TMP3]] to <32 x i64>
; CHECK-NEXT:    [[PREDPHI:%.*]] = select i1 [[C]], <32 x i64> zeroinitializer, <32 x i64> [[TMP4]]
; CHECK-NEXT:    [[TMP132:%.*]] = extractelement <32 x i64> [[PREDPHI]], i64 31
; CHECK-NEXT:    br i1 false, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
; CHECK:       vec.epilog.iter.check:
; CHECK-NEXT:    br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF11:![0-9]+]]
; CHECK:       vec.epilog.ph:
; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i32 [ 896, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <8 x i1> poison, i1 [[C]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <8 x i1> [[BROADCAST_SPLATINSERT3]], <8 x i1> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    [[TMP6:%.*]] = xor <8 x i1> [[BROADCAST_SPLAT4]], splat (i1 true)
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i32> poison, i32 [[D]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT]], <8 x i32> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    [[BROADCAST_SPLATINSERT7:%.*]] = insertelement <8 x i32> poison, i32 [[VEC_EPILOG_RESUME_VAL]], i64 0
; CHECK-NEXT:    [[BROADCAST_SPLAT8:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT7]], <8 x i32> poison, <8 x i32> zeroinitializer
; CHECK-NEXT:    [[INDUCTION:%.*]] = add <8 x i32> [[BROADCAST_SPLAT8]], <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK:       vec.epilog.vector.body:
; CHECK-NEXT:    [[INDEX63:%.*]] = phi i32 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT81:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
; CHECK-NEXT:    [[VEC_IND64:%.*]] = phi <8 x i32> [ [[INDUCTION]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT82:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
; CHECK-NEXT:    [[INDEX_NEXT81]] = add nuw i32 [[INDEX63]], 8
; CHECK-NEXT:    [[VEC_IND_NEXT82]] = add <8 x i32> [[VEC_IND64]], splat (i32 8)
; CHECK-NEXT:    [[TMP167:%.*]] = icmp eq i32 [[INDEX_NEXT81]], 1000
; CHECK-NEXT:    br i1 [[TMP167]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK:       vec.epilog.middle.block:
; CHECK-NEXT:    [[TMP8:%.*]] = call <8 x i32> @llvm.usub.sat.v8i32(<8 x i32> [[VEC_IND64]], <8 x i32> splat (i32 1))
; CHECK-NEXT:    [[TMP166:%.*]] = call <8 x i32> @llvm.masked.udiv.v8i32(<8 x i32> [[TMP8]], <8 x i32> [[BROADCAST_SPLAT]], <8 x i1> [[TMP6]])
; CHECK-NEXT:    [[TMP168:%.*]] = zext <8 x i32> [[TMP166]] to <8 x i64>
; CHECK-NEXT:    [[PREDPHI83:%.*]] = select i1 [[C]], <8 x i64> zeroinitializer, <8 x i64> [[TMP168]]
; CHECK-NEXT:    [[TMP169:%.*]] = extractelement <8 x i64> [[PREDPHI83]], i64 7
; CHECK-NEXT:    br i1 false, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; CHECK:       vec.epilog.scalar.ph:
; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 896, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
; CHECK:       loop.header:
; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
; CHECK-NEXT:    br i1 [[C]], label [[LOOP_LATCH]], label [[THEN:%.*]]
; CHECK:       then:
; CHECK-NEXT:    [[CALL:%.*]] = tail call i32 @llvm.usub.sat.i32(i32 [[IV]], i32 1)
; CHECK-NEXT:    [[UDIV:%.*]] = udiv i32 [[CALL]], [[D]]
; CHECK-NEXT:    [[ZEXT:%.*]] = zext i32 [[UDIV]] to i64
; CHECK-NEXT:    br label [[LOOP_LATCH]]
; CHECK:       loop.latch:
; CHECK-NEXT:    [[MERGE:%.*]] = phi i64 [ [[ZEXT]], [[THEN]] ], [ 0, [[LOOP_HEADER]] ]
; CHECK-NEXT:    [[IV_NEXT]] = add i32 [[IV]], 1
; CHECK-NEXT:    [[EC:%.*]] = icmp eq i32 [[IV]], 1000
; CHECK-NEXT:    br i1 [[EC]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP13:![0-9]+]]
; CHECK:       exit:
; CHECK-NEXT:    [[MERGE_LCSSA:%.*]] = phi i64 [ [[MERGE]], [[LOOP_LATCH]] ], [ [[TMP132]], [[MIDDLE_BLOCK]] ], [ [[TMP169]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
; CHECK-NEXT:    ret i64 [[MERGE_LCSSA]]
;
entry:
  br label %loop.header

loop.header:
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ]
  br i1 %c, label %loop.latch, label %then

then:
  %call = tail call i32 @llvm.usub.sat.i32(i32 %iv, i32 1)
  %udiv = udiv i32 %call, %d
  %zext = zext i32 %udiv to i64
  br label %loop.latch

loop.latch:
  %merge = phi i64 [ %zext, %then ], [ 0, %loop.header ]
  %iv.next = add i32 %iv, 1
  %ec = icmp eq i32 %iv, 1000
  br i1 %ec, label %exit, label %loop.header

exit:
  ret i64 %merge
}

attributes #0 = { "target-cpu"="znver4" }
attributes #1 = { "target-features"="+avx512bw,+avx512cd,+avx512dq,+avx512f,+avx512vl" }
attributes #2 = { "target-cpu"="znver3" }