# RUN: llc -mtriple=s390x-linux-gnu -mcpu=z13 -misched=shuffle -verify-machineinstrs -start-before=simple-register-coalescing -systemz-subreg-liveness %s -o - | FileCheck %s

# -misched=shuffle is under !NDEBUG.
# REQUIRES: asserts

# Check for successful compilation.
# CHECK: lhi %r0, 0

--- |
  target datalayout = "E-m:e-i1:8:16-i8:8:16-i64:64-f128:64-v128:64-a:8:16-n32:64"
  target triple = "s390x-unknown-linux-gnu"

  @g_54 = external dso_local unnamed_addr global i8, align 2
  @g_69 = external dso_local unnamed_addr global i32, align 4
  @g_189 = external dso_local unnamed_addr global i16, align 2
  @g_226 = external dso_local unnamed_addr global i8, align 2
  @g_314 = external dso_local global [10 x i8], align 2
  @g_334 = external dso_local global i32, align 4
  @g_352 = external dso_local unnamed_addr global i64, align 8
  @g_747 = external dso_local unnamed_addr global i1, align 2
  @0 = internal unnamed_addr global i8 74, align 2
  @g_1055 = external dso_local unnamed_addr global i16, align 2
  @g_195 = external dso_local global i64**, align 8

  declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #0

  declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #0

  define dso_local fastcc void @func_32(i8 zeroext %arg, i16 zeroext %arg1) unnamed_addr #1 {
  bb:
    %tmp = alloca i32, align 4
    %tmp2 = alloca [5 x [5 x i32***]], align 8
    %tmp3 = bitcast [5 x [5 x i32***]]* %tmp2 to i8*
    %tmp4 = getelementptr inbounds [5 x [5 x i32***]], [5 x [5 x i32***]]* %tmp2, i64 0, i64 2, i64 2
    %tmp5 = bitcast i32**** %tmp4 to i64***
    br label %bb6

  bb6:                                              ; preds = %bb40, %bb
    %tmp7 = phi i8 [ 0, %bb ], [ %tmp43, %bb40 ]
    %tmp8 = phi i16 [ %arg1, %bb ], [ %tmp41, %bb40 ]
    %tmp9 = phi i8 [ %arg, %bb ], [ 0, %bb40 ]
    %tmp10 = sext i8 %tmp7 to i64
    %tmp11 = add nsw i64 %tmp10, 1
    %tmp12 = getelementptr inbounds [10 x i8], [10 x i8]* @g_314, i64 0, i64 %tmp11
    %tmp13 = load volatile i8, i8* %tmp12, align 1
    br i1 undef, label %bb39, label %bb14

  bb14:                                             ; preds = %bb6
    %tmp15 = load i64**, i64*** @g_195, align 8
    %tmp16 = load volatile i8, i8* %tmp12, align 1
    store i32 7, i32* %tmp, align 4
    call void @llvm.lifetime.start.p0i8(i64 200, i8* nonnull %tmp3)
    store i32 580868341, i32* @g_69, align 4
    %tmp17 = zext i8 %tmp9 to i64
    %tmp18 = load i64, i64* @g_352, align 8
    %tmp19 = and i64 %tmp18, %tmp17
    %tmp20 = icmp ne i64 %tmp19, 1
    %tmp21 = zext i1 %tmp20 to i64
    %tmp22 = load i64*, i64** %tmp15, align 8
    store i64 %tmp21, i64* %tmp22, align 8
    %tmp23 = load i32, i32* @g_334, align 4
    %tmp24 = xor i32 %tmp23, 1
    store i32 %tmp24, i32* @g_334, align 4
    %tmp25 = zext i8 %tmp9 to i16
    %tmp26 = mul i16 %tmp25, 26036
    %tmp27 = load i64**, i64*** %tmp5, align 8
    br label %bb28

  bb28:                                             ; preds = %bb14
    %tmp29 = mul i16 %tmp26, %tmp8
    %tmp30 = zext i16 %tmp29 to i32
    store i32 %tmp30, i32* @g_69, align 4
    store i8 0, i8* @g_226, align 2
    br label %bb32

  bb31:                                             ; preds = %bb35
    call void @llvm.lifetime.end.p0i8(i64 200, i8* nonnull %tmp3)
    br label %bb40

  bb32:                                             ; preds = %bb34, %bb28
    store i16 1, i16* @g_1055, align 2
    store i64 0, i64* @g_352, align 8
    store i32* @g_334, i32** undef, align 8
    %tmp33 = or i64 0, 1
    store i64 %tmp33, i64* @g_352, align 8
    store i32* @g_334, i32** null, align 8
    br label %bb34

  bb34:                                             ; preds = %bb32
    br i1 false, label %bb32, label %bb35

  bb35:                                             ; preds = %bb34
    store i32* %tmp, i32** undef, align 8
    store i8 0, i8* @0, align 2
    store i16 2, i16* @g_189, align 2
    store i8 1, i8* @g_54, align 2
    store i1 true, i1* @g_747, align 2
    store i64 0, i64* undef, align 8
    %tmp36 = load i64*, i64** undef, align 8
    %tmp37 = load i64, i64* %tmp36, align 8
    %tmp38 = load i64*, i64** %tmp27, align 8
    store i64 %tmp37, i64* %tmp38, align 8
    store i16 0, i16* @g_189, align 2
    br label %bb31

  bb39:                                             ; preds = %bb6
    br label %bb40

  bb40:                                             ; preds = %bb39, %bb31
    %tmp41 = phi i16 [ undef, %bb39 ], [ 0, %bb31 ]
    %tmp42 = load volatile i8, i8* %tmp12, align 1
    %tmp43 = add i8 %tmp7, 1
    br i1 false, label %bb6, label %bb44

  bb44:                                             ; preds = %bb40
    unreachable
  }

  attributes #0 = { argmemonly nofree nosync nounwind willreturn }
  attributes #1 = { nounwind }
  attributes #2 = { nofree nosync nounwind willreturn }

...
---
name:            func_32
alignment:       16
tracksRegLiveness: true
liveins:
  - { reg: '$r2d', virtual-reg: '%0' }
  - { reg: '$r3d', virtual-reg: '%1' }
frameInfo:
  maxAlignment:    8
stack:
  - { id: 0, name: tmp, size: 4, alignment: 4 }
  - { id: 1, name: tmp2, size: 200, alignment: 8 }
body:             |
  bb.0.bb:
    liveins: $r2d, $r3d

    %1:gr64bit = COPY killed $r3d
    %0:gr64bit = COPY killed $r2d
    %2:grx32bit = COPY killed %1.subreg_l32
    %3:grx32bit = COPY killed %0.subreg_l32
    %4:addr64bit = LA %stack.1.tmp2, 96, $noreg
    %5:gr32bit = LHIMux 0
    %6:addr64bit = LARL @g_314
    %7:gr32bit = IIFMux 580868341
    %8:addr64bit = LARL @g_352
    %9:gr64bit = LGHI 0
    %10:addr64bit = LARL @g_334
    %11:gr32bit = LHIMux 1
    %12:addr64bit = LARL @g_226
    %13:gr64bit = LGHI 1
    %14:gr64bit = LA %stack.0.tmp, 0, $noreg
    %15:addr64bit = LARL @0
    %16:gr32bit = LHIMux 2
    %17:addr64bit = LARL @g_54
    %18:addr64bit = LARL @g_747
    %19:grx32bit = COPY %5
    %20:gr32bit = COPY killed %2
    %21:grx32bit = COPY killed %3

  bb.1.bb6:
    %22:grx32bit = COPY killed %21
    %23:gr32bit = COPY killed %20
    %24:grx32bit = COPY killed %19
    undef %25.subreg_l32:gr64bit = COPY %24
    %26:addr64bit = LGBR killed %25
    %27:addr64bit = LA %26, 1, %6
    dead %28:grx32bit = LBMux killed %26, 1, %6 :: (volatile load (s8) from %ir.tmp12)
    CHIMux %5, 0, implicit-def $cc
    BRC 14, 6, %bb.7, implicit killed $cc
    J %bb.2

  bb.2.bb14:
    %29:addr64bit = LGRL @g_195 :: (dereferenceable load (s64) from @g_195)
    dead %30:grx32bit = LBMux %27, 0, $noreg :: (volatile load (s8) from %ir.tmp12)
    MVHI %stack.0.tmp, 0, 7 :: (store (s32) into %ir.tmp)
    STRL %7, @g_69 :: (store (s32) into @g_69)
    undef %31.subreg_l32:gr64bit = COPY %22
    %32:gr64bit = LLGC %8, 7, $noreg :: (dereferenceable load (s8) from @g_352 + 7)
    %33:gr64bit = COPY killed %32
    %33:gr64bit = RNSBG %33, killed %31, 0, 63, 0, implicit-def dead $cc
    CGHI killed %33, 1, implicit-def $cc
    %34:gr64bit = COPY %9
    %34:gr64bit = LOCGHI %34, 1, 14, 6, implicit killed $cc
    %35:addr64bit = LG killed %29, 0, $noreg :: (load (s64) from %ir.tmp15)
    STG killed %34, killed %35, 0, $noreg :: (store (s64) into %ir.tmp22)
    %36:gr32bit = COPY %11
    %36:gr32bit = X %36, %10, 0, $noreg, implicit-def dead $cc :: (dereferenceable load (s32) from @g_334)
    STRL killed %36, @g_334 :: (store (s32) into @g_334)
    %37:gr32bit = LLCRMux killed %22
    %38:gr32bit = COPY killed %37
    %38:gr32bit = MHI %38, 26036
    %39:addr64bit = LG %4, 0, $noreg :: (dereferenceable load (s64) from %ir.tmp5)

  bb.3.bb28:
    %40:gr32bit = COPY killed %38
    %40:gr32bit = MSR %40, killed %23
    %41:gr32bit = LLHRMux killed %40
    STRL killed %41, @g_69 :: (store (s32) into @g_69)
    MVI %12, 0, 0 :: (store (s8) into @g_226, align 2)
    J %bb.4

  bb.4.bb32:
    STHRL %11, @g_1055 :: (store (s16) into @g_1055)
    STGRL %9, @g_352 :: (store (s64) into @g_352)
    STG %10, undef %42:addr64bit, 0, $noreg :: (store (s64) into `i32** undef`)
    STGRL %13, @g_352 :: (store (s64) into @g_352)
    STG %10, $noreg, 0, $noreg :: (store (s64) into `i32** null`)

  bb.5.bb34:
    successors: %bb.4(0x7c000000), %bb.6(0x04000000)

    CHIMux %5, 0, implicit-def $cc
    BRC 14, 6, %bb.4, implicit killed $cc
    J %bb.6

  bb.6.bb35:
    STG %14, undef %43:addr64bit, 0, $noreg :: (store (s64) into `i32** undef`)
    MVI %15, 0, 0 :: (store (s8) into @0, align 2)
    STHRL %16, @g_189 :: (store (s16) into @g_189)
    MVI %17, 0, 1 :: (store (s8) into @g_54, align 2)
    MVI %18, 0, 1 :: (store (s8) into @g_747, align 2)
    MVGHI undef %44:addr64bit, 0, 0 :: (store (s64) into `i64* undef`)
    %45:gr64bit = LG $noreg, 0, $noreg :: (load (s64) from %ir.tmp36)
    %46:addr64bit = LG killed %39, 0, $noreg :: (load (s64) from %ir.tmp27)
    STG killed %45, killed %46, 0, $noreg :: (store (s64) into %ir.tmp38)
    STHRL %5, @g_189 :: (store (s16) into @g_189)
    %47:grx32bit = LHIMux 0
    %48:grx32bit = COPY killed %47
    J %bb.8

  bb.7.bb39:
    %48:grx32bit = IMPLICIT_DEF

  bb.8.bb40:
    successors: %bb.1(0x7fffffff), %bb.9(0x00000001)

    %49:grx32bit = COPY killed %48
    dead %50:grx32bit = LBMux killed %27, 0, $noreg :: (volatile load (s8) from %ir.tmp12)
    %51:grx32bit = COPY killed %24
    %51:grx32bit = AHIMux %51, 1, implicit-def dead $cc
    %52:grx32bit = LHIMux 0
    CHIMux %52, 0, implicit-def $cc
    %19:grx32bit = COPY killed %51
    %20:gr32bit = COPY killed %49
    %21:grx32bit = COPY killed %52
    BRC 14, 6, %bb.1, implicit killed $cc
    J %bb.9

  bb.9.bb44:

...