xref: /llvm-project/llvm/test/Transforms/PhaseOrdering/X86/SROA-after-final-loop-unrolling-2.ll (revision 7ec4f6094e54911794c142b5d88496a220d807d6)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -O3 -S                   | FileCheck %s
3; RUN: opt < %s -passes="default<O3>" -S | FileCheck %s
4
5; LoopFullUnrollPass will not preform non-full unrolling,
6; yet until we unroll at least 2x,
7; the loads/stores into `%i3 = alloca %t0, align 4`
8; happen at non-constant indexes.
9
10target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
11target triple = "x86_64-pc-linux-gnu"
12
13%t0 = type { [2 x i32] }
14
15$_ZNSt5arrayIiLm2EEixEm = comdat any
16
17$_ZNSt14__array_traitsIiLm2EE6_S_refERA2_Kim = comdat any
18
19@global = private unnamed_addr constant %t0 { [2 x i32] [i32 24, i32 42] }, align 4
20
21; Function Attrs: mustprogress nounwind uwtable
22define dso_local void @foo(i32 noundef %arg, ptr noundef nonnull align 4 dereferenceable(8) %arg1) #0 {
23; CHECK-LABEL: @foo(
24; CHECK-NEXT:  bb:
25; CHECK-NEXT:    [[I9:%.*]] = sdiv i32 [[ARG:%.*]], 128
26; CHECK-NEXT:    [[I10:%.*]] = shl nsw i32 [[I9]], 7
27; CHECK-NEXT:    [[ARG_OFF:%.*]] = add i32 [[ARG]], 127
28; CHECK-NEXT:    [[TMP0:%.*]] = icmp ult i32 [[ARG_OFF]], 255
29; CHECK-NEXT:    br i1 [[TMP0]], label [[BB12:%.*]], label [[BB13:%.*]]
30; CHECK:       bb12.loopexit:
31; CHECK-NEXT:    [[I3_SROA_8_0_INSERT_EXT:%.*]] = zext i32 [[I21_3:%.*]] to i64
32; CHECK-NEXT:    [[I3_SROA_8_0_INSERT_SHIFT:%.*]] = shl nuw i64 [[I3_SROA_8_0_INSERT_EXT]], 32
33; CHECK-NEXT:    [[I3_SROA_0_0_INSERT_EXT:%.*]] = zext i32 [[I21_2:%.*]] to i64
34; CHECK-NEXT:    [[I3_SROA_0_0_INSERT_INSERT:%.*]] = or disjoint i64 [[I3_SROA_8_0_INSERT_SHIFT]], [[I3_SROA_0_0_INSERT_EXT]]
35; CHECK-NEXT:    br label [[BB12]]
36; CHECK:       bb12:
37; CHECK-NEXT:    [[TMP1:%.*]] = phi i64 [ [[I3_SROA_0_0_INSERT_INSERT]], [[BB12_LOOPEXIT:%.*]] ], [ 180388626456, [[BB:%.*]] ]
38; CHECK-NEXT:    store i64 [[TMP1]], ptr [[ARG1:%.*]], align 4, !tbaa [[TBAA5:![0-9]+]]
39; CHECK-NEXT:    ret void
40; CHECK:       bb13:
41; CHECK-NEXT:    [[I3_SROA_8_0:%.*]] = phi i32 [ [[I21_3]], [[BB13]] ], [ 42, [[BB]] ]
42; CHECK-NEXT:    [[I3_SROA_0_0:%.*]] = phi i32 [ [[I21_2]], [[BB13]] ], [ 24, [[BB]] ]
43; CHECK-NEXT:    [[I4_05:%.*]] = phi i32 [ [[I24_3:%.*]], [[BB13]] ], [ 0, [[BB]] ]
44; CHECK-NEXT:    [[I21:%.*]] = mul nsw i32 [[I3_SROA_0_0]], [[I4_05]]
45; CHECK-NEXT:    [[I24:%.*]] = or disjoint i32 [[I4_05]], 1
46; CHECK-NEXT:    [[I21_1:%.*]] = mul nsw i32 [[I3_SROA_8_0]], [[I24]]
47; CHECK-NEXT:    [[I24_1:%.*]] = or disjoint i32 [[I4_05]], 2
48; CHECK-NEXT:    [[I21_2]] = mul nsw i32 [[I21]], [[I24_1]]
49; CHECK-NEXT:    [[I24_2:%.*]] = or disjoint i32 [[I4_05]], 3
50; CHECK-NEXT:    [[I21_3]] = mul nsw i32 [[I21_1]], [[I24_2]]
51; CHECK-NEXT:    [[I24_3]] = add nuw nsw i32 [[I4_05]], 4
52; CHECK-NEXT:    [[I11_NOT_3:%.*]] = icmp eq i32 [[I24_3]], [[I10]]
53; CHECK-NEXT:    br i1 [[I11_NOT_3]], label [[BB12_LOOPEXIT]], label [[BB13]], !llvm.loop [[LOOP8:![0-9]+]]
54;
55bb:
56  %i = alloca i32, align 4
57  %i2 = alloca ptr, align 8
58  %i3 = alloca %t0, align 4
59  %i4 = alloca i32, align 4
60  %i5 = alloca ptr, align 8
61  store i32 %arg, ptr %i, align 4, !tbaa !5
62  store ptr %arg1, ptr %i2, align 8, !tbaa !9
63  call void @llvm.lifetime.start.p0(i64 8, ptr %i3) #3
64  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %i3, ptr align 4 @global, i64 8, i1 false)
65  call void @llvm.lifetime.start.p0(i64 4, ptr %i4) #3
66  store i32 0, ptr %i4, align 4, !tbaa !5
67  br label %bb6
68
69bb6:                                              ; preds = %bb22, %bb
70  %i7 = load i32, ptr %i4, align 4, !tbaa !5
71  %i8 = load i32, ptr %i, align 4, !tbaa !5
72  %i9 = sdiv i32 %i8, 128
73  %i10 = mul nsw i32 128, %i9
74  %i11 = icmp ne i32 %i7, %i10
75  br i1 %i11, label %bb13, label %bb12
76
77bb12:                                             ; preds = %bb6
78  call void @llvm.lifetime.end.p0(i64 4, ptr %i4) #3
79  br label %bb25
80
81bb13:                                             ; preds = %bb6
82  call void @llvm.lifetime.start.p0(i64 8, ptr %i5) #3
83  %i14 = load i32, ptr %i4, align 4, !tbaa !5
84  %i15 = srem i32 %i14, 2
85  %i16 = sext i32 %i15 to i64
86  %i17 = call noundef nonnull align 4 dereferenceable(4) ptr @wibble(ptr noundef nonnull align 4 dereferenceable(8) %i3, i64 noundef %i16) #3
87  store ptr %i17, ptr %i5, align 8, !tbaa !9
88  %i18 = load i32, ptr %i4, align 4, !tbaa !5
89  %i19 = load ptr, ptr %i5, align 8, !tbaa !9
90  %i20 = load i32, ptr %i19, align 4, !tbaa !5
91  %i21 = mul nsw i32 %i20, %i18
92  store i32 %i21, ptr %i19, align 4, !tbaa !5
93  call void @llvm.lifetime.end.p0(i64 8, ptr %i5) #3
94  br label %bb22
95
96bb22:                                             ; preds = %bb13
97  %i23 = load i32, ptr %i4, align 4, !tbaa !5
98  %i24 = add nsw i32 %i23, 1
99  store i32 %i24, ptr %i4, align 4, !tbaa !5
100  br label %bb6, !llvm.loop !11
101
102bb25:                                             ; preds = %bb12
103  %i26 = load ptr, ptr %i2, align 8, !tbaa !9
104  call void @llvm.memcpy.p0.p0.i64(ptr align 4 %i26, ptr align 4 %i3, i64 8, i1 false), !tbaa.struct !13
105  call void @llvm.lifetime.end.p0(i64 8, ptr %i3) #3
106  ret void
107}
108
109; Function Attrs: argmemonly nocallback nofree nosync nounwind willreturn
110declare void @llvm.lifetime.start.p0(i64 immarg, ptr nocapture) #1
111
112; Function Attrs: argmemonly nocallback nofree nounwind willreturn
113declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg) #2
114
115; Function Attrs: mustprogress nounwind uwtable
116define linkonce_odr dso_local noundef nonnull align 4 dereferenceable(4) ptr @wibble(ptr noundef nonnull align 4 dereferenceable(8) %arg, i64 noundef %arg1) #0 comdat($_ZNSt5arrayIiLm2EEixEm) align 2 {
117bb:
118  %i = alloca ptr, align 8
119  %i2 = alloca i64, align 8
120  store ptr %arg, ptr %i, align 8, !tbaa !9
121  store i64 %arg1, ptr %i2, align 8, !tbaa !15
122  %i3 = load ptr, ptr %i, align 8
123  %i5 = load i64, ptr %i2, align 8, !tbaa !15
124  %i6 = call noundef nonnull align 4 dereferenceable(4) ptr @widget(ptr noundef nonnull align 4 dereferenceable(8) %i3, i64 noundef %i5) #3
125  ret ptr %i6
126}
127
128; Function Attrs: argmemonly nocallback nofree nosync nounwind willreturn
129declare void @llvm.lifetime.end.p0(i64 immarg, ptr nocapture) #1
130
131; Function Attrs: mustprogress nounwind uwtable
132define linkonce_odr dso_local noundef nonnull align 4 dereferenceable(4) ptr @widget(ptr noundef nonnull align 4 dereferenceable(8) %arg, i64 noundef %arg1) #0 comdat($_ZNSt14__array_traitsIiLm2EE6_S_refERA2_Kim) align 2 {
133bb:
134  %i = alloca ptr, align 8
135  %i2 = alloca i64, align 8
136  store ptr %arg, ptr %i, align 8, !tbaa !9
137  store i64 %arg1, ptr %i2, align 8, !tbaa !15
138  %i3 = load ptr, ptr %i, align 8, !tbaa !9
139  %i4 = load i64, ptr %i2, align 8, !tbaa !15
140  %i5 = getelementptr inbounds [2 x i32], ptr %i3, i64 0, i64 %i4
141  ret ptr %i5
142}
143
144attributes #0 = { mustprogress nounwind uwtable "frame-pointer"="none" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "tune-cpu"="generic" }
145attributes #1 = { argmemonly nocallback nofree nosync nounwind willreturn }
146attributes #2 = { argmemonly nocallback nofree nounwind willreturn }
147attributes #3 = { nounwind }
148
149!llvm.module.flags = !{!0, !1, !2, !3}
150!llvm.ident = !{!4}
151
152!0 = !{i32 1, !"wchar_size", i32 4}
153!1 = !{i32 8, !"PIC Level", i32 2}
154!2 = !{i32 7, !"PIE Level", i32 2}
155!3 = !{i32 7, !"uwtable", i32 2}
156!4 = !{!"Debian clang version 16.0.0-++20221024100939+996267d20ea0-1~exp1~20221024221046.844"}
157!5 = !{!6, !6, i64 0}
158!6 = !{!"int", !7, i64 0}
159!7 = !{!"omnipotent char", !8, i64 0}
160!8 = !{!"Simple C++ TBAA"}
161!9 = !{!10, !10, i64 0}
162!10 = !{!"any pointer", !7, i64 0}
163!11 = distinct !{!11, !12}
164!12 = !{!"llvm.loop.mustprogress"}
165!13 = !{i64 0, i64 8, !14}
166!14 = !{!7, !7, i64 0}
167!15 = !{!16, !16, i64 0}
168!16 = !{!"long", !7, i64 0}
169