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