1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -basic-aa -loop-distribute -enable-loop-distribute -S -enable-mem-access-versioning=0 < %s | FileCheck %s
3
4target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
5
6; PredicatedScalarEvolution decides it needs to insert a bounds check
7; not based on memory access.
8
9define void @f(i32* noalias %a, i32* noalias %b, i32* noalias %c, i32* noalias %d, i32* noalias %e, i64 %N) {
10; CHECK-LABEL: @f(
11; CHECK-NEXT:  entry:
12; CHECK-NEXT:    [[A5:%.*]] = bitcast i32* [[A:%.*]] to i8*
13; CHECK-NEXT:    br label [[FOR_BODY_LVER_CHECK:%.*]]
14; CHECK:       for.body.lver.check:
15; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N:%.*]], -1
16; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
17; CHECK-NEXT:    [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[TMP1]])
18; CHECK-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
19; CHECK-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
20; CHECK-NEXT:    [[TMP2:%.*]] = add i32 0, [[MUL_RESULT]]
21; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP2]], 0
22; CHECK-NEXT:    [[TMP7:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
23; CHECK-NEXT:    [[TMP8:%.*]] = or i1 [[TMP5]], [[TMP7]]
24; CHECK-NEXT:    [[TMP9:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW]]
25; CHECK-NEXT:    [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 8, i64 [[TMP0]])
26; CHECK-NEXT:    [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0
27; CHECK-NEXT:    [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1
28; CHECK-NEXT:    [[TMP11:%.*]] = sub i64 0, [[MUL_RESULT3]]
29; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr i8, i8* [[A5]], i64 [[MUL_RESULT3]]
30; CHECK-NEXT:    [[TMP15:%.*]] = icmp ult i8* [[TMP12]], [[A5]]
31; CHECK-NEXT:    [[TMP17:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW4]]
32; CHECK-NEXT:    [[TMP18:%.*]] = or i1 [[TMP9]], [[TMP17]]
33; CHECK-NEXT:    br i1 [[TMP18]], label [[FOR_BODY_PH_LVER_ORIG:%.*]], label [[FOR_BODY_PH_LDIST1:%.*]]
34; CHECK:       for.body.ph.lver.orig:
35; CHECK-NEXT:    br label [[FOR_BODY_LVER_ORIG:%.*]]
36; CHECK:       for.body.lver.orig:
37; CHECK-NEXT:    [[IND_LVER_ORIG:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[ADD_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
38; CHECK-NEXT:    [[IND1_LVER_ORIG:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[INC1_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
39; CHECK-NEXT:    [[MUL_LVER_ORIG:%.*]] = mul i32 [[IND1_LVER_ORIG]], 2
40; CHECK-NEXT:    [[MUL_EXT_LVER_ORIG:%.*]] = zext i32 [[MUL_LVER_ORIG]] to i64
41; CHECK-NEXT:    [[ARRAYIDXA_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[MUL_EXT_LVER_ORIG]]
42; CHECK-NEXT:    [[LOADA_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXA_LVER_ORIG]], align 4
43; CHECK-NEXT:    [[ARRAYIDXB_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
44; CHECK-NEXT:    [[LOADB_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXB_LVER_ORIG]], align 4
45; CHECK-NEXT:    [[MULA_LVER_ORIG:%.*]] = mul i32 [[LOADB_LVER_ORIG]], [[LOADA_LVER_ORIG]]
46; CHECK-NEXT:    [[ADD_LVER_ORIG]] = add nuw nsw i64 [[IND_LVER_ORIG]], 1
47; CHECK-NEXT:    [[INC1_LVER_ORIG]] = add i32 [[IND1_LVER_ORIG]], 1
48; CHECK-NEXT:    [[ARRAYIDXA_PLUS_4_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[ADD_LVER_ORIG]]
49; CHECK-NEXT:    store i32 [[MULA_LVER_ORIG]], i32* [[ARRAYIDXA_PLUS_4_LVER_ORIG]], align 4
50; CHECK-NEXT:    [[ARRAYIDXD_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[D:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
51; CHECK-NEXT:    [[LOADD_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXD_LVER_ORIG]], align 4
52; CHECK-NEXT:    [[ARRAYIDXE_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[E:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
53; CHECK-NEXT:    [[LOADE_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXE_LVER_ORIG]], align 4
54; CHECK-NEXT:    [[MULC_LVER_ORIG:%.*]] = mul i32 [[LOADD_LVER_ORIG]], [[LOADE_LVER_ORIG]]
55; CHECK-NEXT:    [[ARRAYIDXC_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
56; CHECK-NEXT:    store i32 [[MULC_LVER_ORIG]], i32* [[ARRAYIDXC_LVER_ORIG]], align 4
57; CHECK-NEXT:    [[EXITCOND_LVER_ORIG:%.*]] = icmp eq i64 [[ADD_LVER_ORIG]], [[N]]
58; CHECK-NEXT:    br i1 [[EXITCOND_LVER_ORIG]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY_LVER_ORIG]]
59; CHECK:       for.body.ph.ldist1:
60; CHECK-NEXT:    br label [[FOR_BODY_LDIST1:%.*]]
61; CHECK:       for.body.ldist1:
62; CHECK-NEXT:    [[IND_LDIST1:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[ADD_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
63; CHECK-NEXT:    [[IND1_LDIST1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[INC1_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
64; CHECK-NEXT:    [[MUL_LDIST1:%.*]] = mul i32 [[IND1_LDIST1]], 2
65; CHECK-NEXT:    [[MUL_EXT_LDIST1:%.*]] = zext i32 [[MUL_LDIST1]] to i64
66; CHECK-NEXT:    [[ARRAYIDXA_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[MUL_EXT_LDIST1]]
67; CHECK-NEXT:    [[LOADA_LDIST1:%.*]] = load i32, i32* [[ARRAYIDXA_LDIST1]], align 4, !alias.scope !0
68; CHECK-NEXT:    [[ARRAYIDXB_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[MUL_EXT_LDIST1]]
69; CHECK-NEXT:    [[LOADB_LDIST1:%.*]] = load i32, i32* [[ARRAYIDXB_LDIST1]], align 4
70; CHECK-NEXT:    [[MULA_LDIST1:%.*]] = mul i32 [[LOADB_LDIST1]], [[LOADA_LDIST1]]
71; CHECK-NEXT:    [[ADD_LDIST1]] = add nuw nsw i64 [[IND_LDIST1]], 1
72; CHECK-NEXT:    [[INC1_LDIST1]] = add i32 [[IND1_LDIST1]], 1
73; CHECK-NEXT:    [[ARRAYIDXA_PLUS_4_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[ADD_LDIST1]]
74; CHECK-NEXT:    store i32 [[MULA_LDIST1]], i32* [[ARRAYIDXA_PLUS_4_LDIST1]], align 4, !alias.scope !3
75; CHECK-NEXT:    [[EXITCOND_LDIST1:%.*]] = icmp eq i64 [[ADD_LDIST1]], [[N]]
76; CHECK-NEXT:    br i1 [[EXITCOND_LDIST1]], label [[FOR_BODY_PH:%.*]], label [[FOR_BODY_LDIST1]]
77; CHECK:       for.body.ph:
78; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
79; CHECK:       for.body:
80; CHECK-NEXT:    [[IND:%.*]] = phi i64 [ 0, [[FOR_BODY_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
81; CHECK-NEXT:    [[IND1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
82; CHECK-NEXT:    [[MUL:%.*]] = mul i32 [[IND1]], 2
83; CHECK-NEXT:    [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
84; CHECK-NEXT:    [[ADD]] = add nuw nsw i64 [[IND]], 1
85; CHECK-NEXT:    [[INC1]] = add i32 [[IND1]], 1
86; CHECK-NEXT:    [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, i32* [[D]], i64 [[MUL_EXT]]
87; CHECK-NEXT:    [[LOADD:%.*]] = load i32, i32* [[ARRAYIDXD]], align 4
88; CHECK-NEXT:    [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, i32* [[E]], i64 [[MUL_EXT]]
89; CHECK-NEXT:    [[LOADE:%.*]] = load i32, i32* [[ARRAYIDXE]], align 4
90; CHECK-NEXT:    [[MULC:%.*]] = mul i32 [[LOADD]], [[LOADE]]
91; CHECK-NEXT:    [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, i32* [[C]], i64 [[MUL_EXT]]
92; CHECK-NEXT:    store i32 [[MULC]], i32* [[ARRAYIDXC]], align 4
93; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], [[N]]
94; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT6:%.*]], label [[FOR_BODY]]
95; CHECK:       for.end.loopexit:
96; CHECK-NEXT:    br label [[FOR_END:%.*]]
97; CHECK:       for.end.loopexit6:
98; CHECK-NEXT:    br label [[FOR_END]]
99; CHECK:       for.end:
100; CHECK-NEXT:    ret void
101;
102entry:
103  br label %for.body
104
105for.body:                                         ; preds = %for.body, %entry
106  %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
107  %ind1 = phi i32 [ 0, %entry ], [ %inc1, %for.body ]
108
109  %mul = mul i32 %ind1, 2
110  %mul_ext = zext i32 %mul to i64
111
112
113  %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %mul_ext
114  %loadA = load i32, i32* %arrayidxA, align 4
115
116  %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %mul_ext
117  %loadB = load i32, i32* %arrayidxB, align 4
118
119  %mulA = mul i32 %loadB, %loadA
120
121  %add = add nuw nsw i64 %ind, 1
122  %inc1 = add i32 %ind1, 1
123
124  %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add
125  store i32 %mulA, i32* %arrayidxA_plus_4, align 4
126
127  %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %mul_ext
128  %loadD = load i32, i32* %arrayidxD, align 4
129
130  %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %mul_ext
131  %loadE = load i32, i32* %arrayidxE, align 4
132
133  %mulC = mul i32 %loadD, %loadE
134
135  %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %mul_ext
136  store i32 %mulC, i32* %arrayidxC, align 4
137
138  %exitcond = icmp eq i64 %add, %N
139  br i1 %exitcond, label %for.end, label %for.body
140
141for.end:                                          ; preds = %for.body
142  ret void
143}
144
145declare void @use64(i64)
146@global_a = common local_unnamed_addr global [8192 x i32] zeroinitializer, align 16
147
148define void @f_with_offset(i32* noalias %b, i32* noalias %c, i32* noalias %d, i32* noalias %e, i64 %N) {
149; CHECK-LABEL: @f_with_offset(
150; CHECK-NEXT:  entry:
151; CHECK-NEXT:    [[A_BASE:%.*]] = getelementptr [8192 x i32], [8192 x i32]* @global_a, i32 0, i32 0
152; CHECK-NEXT:    [[A_INTPTR:%.*]] = ptrtoint i32* [[A_BASE]] to i64
153; CHECK-NEXT:    call void @use64(i64 [[A_INTPTR]])
154; CHECK-NEXT:    [[A:%.*]] = getelementptr i32, i32* [[A_BASE]], i32 42
155; CHECK-NEXT:    br label [[FOR_BODY_LVER_CHECK:%.*]]
156; CHECK:       for.body.lver.check:
157; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N:%.*]], -1
158; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i32
159; CHECK-NEXT:    [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[TMP1]])
160; CHECK-NEXT:    [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
161; CHECK-NEXT:    [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
162; CHECK-NEXT:    [[TMP2:%.*]] = add i32 0, [[MUL_RESULT]]
163; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP2]], 0
164; CHECK-NEXT:    [[TMP7:%.*]] = icmp ugt i64 [[TMP0]], 4294967295
165; CHECK-NEXT:    [[TMP8:%.*]] = or i1 [[TMP5]], [[TMP7]]
166; CHECK-NEXT:    [[TMP9:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW]]
167; CHECK-NEXT:    [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 8, i64 [[TMP0]])
168; CHECK-NEXT:    [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0
169; CHECK-NEXT:    [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1
170; CHECK-NEXT:    [[TMP11:%.*]] = sub i64 0, [[MUL_RESULT3]]
171; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr i8, i8* bitcast (i32* getelementptr inbounds ([8192 x i32], [8192 x i32]* @global_a, i64 0, i64 42) to i8*), i64 [[MUL_RESULT3]]
172; CHECK-NEXT:    [[TMP15:%.*]] = icmp ult i8* [[TMP12]], bitcast (i32* getelementptr inbounds ([8192 x i32], [8192 x i32]* @global_a, i64 0, i64 42) to i8*)
173; CHECK-NEXT:    [[TMP17:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW4]]
174; CHECK-NEXT:    [[TMP18:%.*]] = or i1 [[TMP9]], [[TMP17]]
175; CHECK-NEXT:    br i1 [[TMP18]], label [[FOR_BODY_PH_LVER_ORIG:%.*]], label [[FOR_BODY_PH_LDIST1:%.*]]
176; CHECK:       for.body.ph.lver.orig:
177; CHECK-NEXT:    br label [[FOR_BODY_LVER_ORIG:%.*]]
178; CHECK:       for.body.lver.orig:
179; CHECK-NEXT:    [[IND_LVER_ORIG:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[ADD_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
180; CHECK-NEXT:    [[IND1_LVER_ORIG:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LVER_ORIG]] ], [ [[INC1_LVER_ORIG:%.*]], [[FOR_BODY_LVER_ORIG]] ]
181; CHECK-NEXT:    [[MUL_LVER_ORIG:%.*]] = mul i32 [[IND1_LVER_ORIG]], 2
182; CHECK-NEXT:    [[MUL_EXT_LVER_ORIG:%.*]] = zext i32 [[MUL_LVER_ORIG]] to i64
183; CHECK-NEXT:    [[ARRAYIDXA_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[MUL_EXT_LVER_ORIG]]
184; CHECK-NEXT:    [[LOADA_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXA_LVER_ORIG]], align 4
185; CHECK-NEXT:    [[ARRAYIDXB_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
186; CHECK-NEXT:    [[LOADB_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXB_LVER_ORIG]], align 4
187; CHECK-NEXT:    [[MULA_LVER_ORIG:%.*]] = mul i32 [[LOADB_LVER_ORIG]], [[LOADA_LVER_ORIG]]
188; CHECK-NEXT:    [[ADD_LVER_ORIG]] = add nuw nsw i64 [[IND_LVER_ORIG]], 1
189; CHECK-NEXT:    [[INC1_LVER_ORIG]] = add i32 [[IND1_LVER_ORIG]], 1
190; CHECK-NEXT:    [[ARRAYIDXA_PLUS_4_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[ADD_LVER_ORIG]]
191; CHECK-NEXT:    store i32 [[MULA_LVER_ORIG]], i32* [[ARRAYIDXA_PLUS_4_LVER_ORIG]], align 4
192; CHECK-NEXT:    [[ARRAYIDXD_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[D:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
193; CHECK-NEXT:    [[LOADD_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXD_LVER_ORIG]], align 4
194; CHECK-NEXT:    [[ARRAYIDXE_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[E:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
195; CHECK-NEXT:    [[LOADE_LVER_ORIG:%.*]] = load i32, i32* [[ARRAYIDXE_LVER_ORIG]], align 4
196; CHECK-NEXT:    [[MULC_LVER_ORIG:%.*]] = mul i32 [[LOADD_LVER_ORIG]], [[LOADE_LVER_ORIG]]
197; CHECK-NEXT:    [[ARRAYIDXC_LVER_ORIG:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i64 [[MUL_EXT_LVER_ORIG]]
198; CHECK-NEXT:    store i32 [[MULC_LVER_ORIG]], i32* [[ARRAYIDXC_LVER_ORIG]], align 4
199; CHECK-NEXT:    [[EXITCOND_LVER_ORIG:%.*]] = icmp eq i64 [[ADD_LVER_ORIG]], [[N]]
200; CHECK-NEXT:    br i1 [[EXITCOND_LVER_ORIG]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY_LVER_ORIG]]
201; CHECK:       for.body.ph.ldist1:
202; CHECK-NEXT:    br label [[FOR_BODY_LDIST1:%.*]]
203; CHECK:       for.body.ldist1:
204; CHECK-NEXT:    [[IND_LDIST1:%.*]] = phi i64 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[ADD_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
205; CHECK-NEXT:    [[IND1_LDIST1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH_LDIST1]] ], [ [[INC1_LDIST1:%.*]], [[FOR_BODY_LDIST1]] ]
206; CHECK-NEXT:    [[MUL_LDIST1:%.*]] = mul i32 [[IND1_LDIST1]], 2
207; CHECK-NEXT:    [[MUL_EXT_LDIST1:%.*]] = zext i32 [[MUL_LDIST1]] to i64
208; CHECK-NEXT:    [[ARRAYIDXA_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[MUL_EXT_LDIST1]]
209; CHECK-NEXT:    [[LOADA_LDIST1:%.*]] = load i32, i32* [[ARRAYIDXA_LDIST1]], align 4, !alias.scope !5
210; CHECK-NEXT:    [[ARRAYIDXB_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[MUL_EXT_LDIST1]]
211; CHECK-NEXT:    [[LOADB_LDIST1:%.*]] = load i32, i32* [[ARRAYIDXB_LDIST1]], align 4
212; CHECK-NEXT:    [[MULA_LDIST1:%.*]] = mul i32 [[LOADB_LDIST1]], [[LOADA_LDIST1]]
213; CHECK-NEXT:    [[ADD_LDIST1]] = add nuw nsw i64 [[IND_LDIST1]], 1
214; CHECK-NEXT:    [[INC1_LDIST1]] = add i32 [[IND1_LDIST1]], 1
215; CHECK-NEXT:    [[ARRAYIDXA_PLUS_4_LDIST1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[ADD_LDIST1]]
216; CHECK-NEXT:    store i32 [[MULA_LDIST1]], i32* [[ARRAYIDXA_PLUS_4_LDIST1]], align 4, !alias.scope !8
217; CHECK-NEXT:    [[EXITCOND_LDIST1:%.*]] = icmp eq i64 [[ADD_LDIST1]], [[N]]
218; CHECK-NEXT:    br i1 [[EXITCOND_LDIST1]], label [[FOR_BODY_PH:%.*]], label [[FOR_BODY_LDIST1]]
219; CHECK:       for.body.ph:
220; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
221; CHECK:       for.body:
222; CHECK-NEXT:    [[IND:%.*]] = phi i64 [ 0, [[FOR_BODY_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
223; CHECK-NEXT:    [[IND1:%.*]] = phi i32 [ 0, [[FOR_BODY_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
224; CHECK-NEXT:    [[MUL:%.*]] = mul i32 [[IND1]], 2
225; CHECK-NEXT:    [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
226; CHECK-NEXT:    [[ADD]] = add nuw nsw i64 [[IND]], 1
227; CHECK-NEXT:    [[INC1]] = add i32 [[IND1]], 1
228; CHECK-NEXT:    [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, i32* [[D]], i64 [[MUL_EXT]]
229; CHECK-NEXT:    [[LOADD:%.*]] = load i32, i32* [[ARRAYIDXD]], align 4
230; CHECK-NEXT:    [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, i32* [[E]], i64 [[MUL_EXT]]
231; CHECK-NEXT:    [[LOADE:%.*]] = load i32, i32* [[ARRAYIDXE]], align 4
232; CHECK-NEXT:    [[MULC:%.*]] = mul i32 [[LOADD]], [[LOADE]]
233; CHECK-NEXT:    [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, i32* [[C]], i64 [[MUL_EXT]]
234; CHECK-NEXT:    store i32 [[MULC]], i32* [[ARRAYIDXC]], align 4
235; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], [[N]]
236; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT5:%.*]], label [[FOR_BODY]]
237; CHECK:       for.end.loopexit:
238; CHECK-NEXT:    br label [[FOR_END:%.*]]
239; CHECK:       for.end.loopexit5:
240; CHECK-NEXT:    br label [[FOR_END]]
241; CHECK:       for.end:
242; CHECK-NEXT:    ret void
243;
244entry:
245  %a_base = getelementptr [8192 x i32], [8192 x i32]* @global_a, i32 0, i32 0
246  %a_intptr = ptrtoint i32* %a_base to i64
247  call void @use64(i64 %a_intptr)
248  %a = getelementptr i32, i32* %a_base, i32 42
249  br label %for.body
250
251for.body:                                         ; preds = %for.body, %entry
252  %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
253  %ind1 = phi i32 [ 0, %entry ], [ %inc1, %for.body ]
254
255  %mul = mul i32 %ind1, 2
256  %mul_ext = zext i32 %mul to i64
257
258
259  %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %mul_ext
260  %loadA = load i32, i32* %arrayidxA, align 4
261
262  %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %mul_ext
263  %loadB = load i32, i32* %arrayidxB, align 4
264
265  %mulA = mul i32 %loadB, %loadA
266
267  %add = add nuw nsw i64 %ind, 1
268  %inc1 = add i32 %ind1, 1
269
270  %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add
271  store i32 %mulA, i32* %arrayidxA_plus_4, align 4
272
273  %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %mul_ext
274  %loadD = load i32, i32* %arrayidxD, align 4
275
276  %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %mul_ext
277  %loadE = load i32, i32* %arrayidxE, align 4
278
279  %mulC = mul i32 %loadD, %loadE
280
281  %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %mul_ext
282  store i32 %mulC, i32* %arrayidxC, align 4
283
284  %exitcond = icmp eq i64 %add, %N
285  br i1 %exitcond, label %for.end, label %for.body
286
287for.end:                                          ; preds = %for.body
288  ret void
289}
290
291; Can't add control dependency with convergent in loop body.
292define void @f_with_convergent(i32* noalias %a, i32* noalias %b, i32* noalias %c, i32* noalias %d, i32* noalias %e, i64 %N) #1 {
293; CHECK-LABEL: @f_with_convergent(
294; CHECK-NEXT:  entry:
295; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
296; CHECK:       for.body:
297; CHECK-NEXT:    [[IND:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[ADD:%.*]], [[FOR_BODY]] ]
298; CHECK-NEXT:    [[IND1:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
299; CHECK-NEXT:    [[MUL:%.*]] = mul i32 [[IND1]], 2
300; CHECK-NEXT:    [[MUL_EXT:%.*]] = zext i32 [[MUL]] to i64
301; CHECK-NEXT:    [[ARRAYIDXA:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[MUL_EXT]]
302; CHECK-NEXT:    [[LOADA:%.*]] = load i32, i32* [[ARRAYIDXA]], align 4
303; CHECK-NEXT:    [[ARRAYIDXB:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[MUL_EXT]]
304; CHECK-NEXT:    [[LOADB:%.*]] = load i32, i32* [[ARRAYIDXB]], align 4
305; CHECK-NEXT:    [[MULA:%.*]] = mul i32 [[LOADB]], [[LOADA]]
306; CHECK-NEXT:    [[ADD]] = add nuw nsw i64 [[IND]], 1
307; CHECK-NEXT:    [[INC1]] = add i32 [[IND1]], 1
308; CHECK-NEXT:    [[ARRAYIDXA_PLUS_4:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[ADD]]
309; CHECK-NEXT:    store i32 [[MULA]], i32* [[ARRAYIDXA_PLUS_4]], align 4
310; CHECK-NEXT:    [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, i32* [[D:%.*]], i64 [[MUL_EXT]]
311; CHECK-NEXT:    [[LOADD:%.*]] = load i32, i32* [[ARRAYIDXD]], align 4
312; CHECK-NEXT:    [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, i32* [[E:%.*]], i64 [[MUL_EXT]]
313; CHECK-NEXT:    [[LOADE:%.*]] = load i32, i32* [[ARRAYIDXE]], align 4
314; CHECK-NEXT:    [[CONVERGENTD:%.*]] = call i32 @llvm.convergent(i32 [[LOADD]])
315; CHECK-NEXT:    [[MULC:%.*]] = mul i32 [[CONVERGENTD]], [[LOADE]]
316; CHECK-NEXT:    [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, i32* [[C:%.*]], i64 [[MUL_EXT]]
317; CHECK-NEXT:    store i32 [[MULC]], i32* [[ARRAYIDXC]], align 4
318; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], [[N:%.*]]
319; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
320; CHECK:       for.end:
321; CHECK-NEXT:    ret void
322;
323entry:
324  br label %for.body
325
326for.body:                                         ; preds = %for.body, %entry
327  %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
328  %ind1 = phi i32 [ 0, %entry ], [ %inc1, %for.body ]
329
330  %mul = mul i32 %ind1, 2
331  %mul_ext = zext i32 %mul to i64
332
333
334  %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %mul_ext
335  %loadA = load i32, i32* %arrayidxA, align 4
336
337  %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %mul_ext
338  %loadB = load i32, i32* %arrayidxB, align 4
339
340  %mulA = mul i32 %loadB, %loadA
341
342  %add = add nuw nsw i64 %ind, 1
343  %inc1 = add i32 %ind1, 1
344
345  %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add
346  store i32 %mulA, i32* %arrayidxA_plus_4, align 4
347
348  %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %mul_ext
349  %loadD = load i32, i32* %arrayidxD, align 4
350
351  %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %mul_ext
352  %loadE = load i32, i32* %arrayidxE, align 4
353
354  %convergentD = call i32 @llvm.convergent(i32 %loadD)
355  %mulC = mul i32 %convergentD, %loadE
356
357  %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %mul_ext
358  store i32 %mulC, i32* %arrayidxC, align 4
359
360  %exitcond = icmp eq i64 %add, %N
361  br i1 %exitcond, label %for.end, label %for.body
362
363for.end:                                          ; preds = %for.body
364  ret void
365}
366
367declare i32 @llvm.convergent(i32) #0
368
369attributes #0 = { nounwind readnone convergent }
370attributes #1 = { nounwind convergent }
371