1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true -unroll-runtime-other-exit-predictable=true -verify-loop-lcssa -verify-dom-info -verify-loop-info -S | FileCheck %s --check-prefix=ENABLED
3; RUN: opt < %s -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true -unroll-runtime-other-exit-predictable=false -verify-loop-lcssa -verify-dom-info -verify-loop-info -S | FileCheck %s --check-prefix=DISABLED
4
5define i32 @test(i32* nocapture %a, i64 %n) {
6; ENABLED-LABEL: @test(
7; ENABLED-NEXT:  entry:
8; ENABLED-NEXT:    [[TMP0:%.*]] = add i64 [[N:%.*]], -1
9; ENABLED-NEXT:    [[XTRAITER:%.*]] = and i64 [[N]], 7
10; ENABLED-NEXT:    [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
11; ENABLED-NEXT:    br i1 [[TMP1]], label [[FOR_END_UNR_LCSSA:%.*]], label [[ENTRY_NEW:%.*]]
12; ENABLED:       entry.new:
13; ENABLED-NEXT:    [[UNROLL_ITER:%.*]] = sub i64 [[N]], [[XTRAITER]]
14; ENABLED-NEXT:    br label [[HEADER:%.*]]
15; ENABLED:       header:
16; ENABLED-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY_7:%.*]] ]
17; ENABLED-NEXT:    [[SUM_02:%.*]] = phi i32 [ 0, [[ENTRY_NEW]] ], [ [[ADD_7:%.*]], [[FOR_BODY_7]] ]
18; ENABLED-NEXT:    [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[ENTRY_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY_7]] ]
19; ENABLED-NEXT:    [[CMP:%.*]] = icmp eq i64 [[N]], 42
20; ENABLED-NEXT:    br i1 [[CMP]], label [[FOR_EXIT2_LOOPEXIT:%.*]], label [[FOR_BODY:%.*]]
21; ENABLED:       for.body:
22; ENABLED-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDVARS_IV]]
23; ENABLED-NEXT:    [[TMP2:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
24; ENABLED-NEXT:    [[ADD:%.*]] = add nsw i32 [[TMP2]], [[SUM_02]]
25; ENABLED-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
26; ENABLED-NEXT:    [[NITER_NSUB:%.*]] = sub i64 [[NITER]], 1
27; ENABLED-NEXT:    [[CMP_1:%.*]] = icmp eq i64 [[N]], 42
28; ENABLED-NEXT:    br i1 [[CMP_1]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_1:%.*]]
29; ENABLED:       for.end.unr-lcssa.loopexit:
30; ENABLED-NEXT:    [[SUM_0_LCSSA_PH_PH:%.*]] = phi i32 [ [[ADD_7]], [[FOR_BODY_7]] ]
31; ENABLED-NEXT:    [[INDVARS_IV_UNR_PH:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_7]], [[FOR_BODY_7]] ]
32; ENABLED-NEXT:    [[SUM_02_UNR_PH:%.*]] = phi i32 [ [[ADD_7]], [[FOR_BODY_7]] ]
33; ENABLED-NEXT:    br label [[FOR_END_UNR_LCSSA]]
34; ENABLED:       for.end.unr-lcssa:
35; ENABLED-NEXT:    [[SUM_0_LCSSA_PH:%.*]] = phi i32 [ undef, [[ENTRY:%.*]] ], [ [[SUM_0_LCSSA_PH_PH]], [[FOR_END_UNR_LCSSA_LOOPEXIT:%.*]] ]
36; ENABLED-NEXT:    [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[INDVARS_IV_UNR_PH]], [[FOR_END_UNR_LCSSA_LOOPEXIT]] ]
37; ENABLED-NEXT:    [[SUM_02_UNR:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[SUM_02_UNR_PH]], [[FOR_END_UNR_LCSSA_LOOPEXIT]] ]
38; ENABLED-NEXT:    [[LCMP_MOD:%.*]] = icmp ne i64 [[XTRAITER]], 0
39; ENABLED-NEXT:    br i1 [[LCMP_MOD]], label [[HEADER_EPIL_PREHEADER:%.*]], label [[FOR_END:%.*]]
40; ENABLED:       header.epil.preheader:
41; ENABLED-NEXT:    br label [[HEADER_EPIL:%.*]]
42; ENABLED:       header.epil:
43; ENABLED-NEXT:    [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL:%.*]] ], [ [[INDVARS_IV_UNR]], [[HEADER_EPIL_PREHEADER]] ]
44; ENABLED-NEXT:    [[SUM_02_EPIL:%.*]] = phi i32 [ [[ADD_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[SUM_02_UNR]], [[HEADER_EPIL_PREHEADER]] ]
45; ENABLED-NEXT:    [[EPIL_ITER:%.*]] = phi i64 [ [[XTRAITER]], [[HEADER_EPIL_PREHEADER]] ], [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ]
46; ENABLED-NEXT:    [[CMP_EPIL:%.*]] = icmp eq i64 [[N]], 42
47; ENABLED-NEXT:    br i1 [[CMP_EPIL]], label [[FOR_EXIT2_LOOPEXIT2:%.*]], label [[FOR_BODY_EPIL]]
48; ENABLED:       for.body.epil:
49; ENABLED-NEXT:    [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_EPIL]]
50; ENABLED-NEXT:    [[TMP3:%.*]] = load i32, i32* [[ARRAYIDX_EPIL]], align 4
51; ENABLED-NEXT:    [[ADD_EPIL]] = add nsw i32 [[TMP3]], [[SUM_02_EPIL]]
52; ENABLED-NEXT:    [[INDVARS_IV_NEXT_EPIL]] = add i64 [[INDVARS_IV_EPIL]], 1
53; ENABLED-NEXT:    [[EXITCOND_EPIL:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_EPIL]], [[N]]
54; ENABLED-NEXT:    [[EPIL_ITER_SUB]] = sub i64 [[EPIL_ITER]], 1
55; ENABLED-NEXT:    [[EPIL_ITER_CMP:%.*]] = icmp ne i64 [[EPIL_ITER_SUB]], 0
56; ENABLED-NEXT:    br i1 [[EPIL_ITER_CMP]], label [[HEADER_EPIL]], label [[FOR_END_EPILOG_LCSSA:%.*]], [[LOOP0:!llvm.loop !.*]]
57; ENABLED:       for.end.epilog-lcssa:
58; ENABLED-NEXT:    [[SUM_0_LCSSA_PH1:%.*]] = phi i32 [ [[ADD_EPIL]], [[FOR_BODY_EPIL]] ]
59; ENABLED-NEXT:    br label [[FOR_END]]
60; ENABLED:       for.end:
61; ENABLED-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ [[SUM_0_LCSSA_PH]], [[FOR_END_UNR_LCSSA]] ], [ [[SUM_0_LCSSA_PH1]], [[FOR_END_EPILOG_LCSSA]] ]
62; ENABLED-NEXT:    ret i32 [[SUM_0_LCSSA]]
63; ENABLED:       for.exit2.loopexit:
64; ENABLED-NEXT:    [[RETVAL_PH:%.*]] = phi i32 [ [[SUM_02]], [[HEADER]] ], [ [[ADD]], [[FOR_BODY]] ], [ [[ADD_1:%.*]], [[FOR_BODY_1]] ], [ [[ADD_2:%.*]], [[FOR_BODY_2:%.*]] ], [ [[ADD_3:%.*]], [[FOR_BODY_3:%.*]] ], [ [[ADD_4:%.*]], [[FOR_BODY_4:%.*]] ], [ [[ADD_5:%.*]], [[FOR_BODY_5:%.*]] ], [ [[ADD_6:%.*]], [[FOR_BODY_6:%.*]] ]
65; ENABLED-NEXT:    br label [[FOR_EXIT2:%.*]]
66; ENABLED:       for.exit2.loopexit2:
67; ENABLED-NEXT:    [[RETVAL_PH3:%.*]] = phi i32 [ [[SUM_02_EPIL]], [[HEADER_EPIL]] ]
68; ENABLED-NEXT:    br label [[FOR_EXIT2]]
69; ENABLED:       for.exit2:
70; ENABLED-NEXT:    [[RETVAL:%.*]] = phi i32 [ [[RETVAL_PH]], [[FOR_EXIT2_LOOPEXIT]] ], [ [[RETVAL_PH3]], [[FOR_EXIT2_LOOPEXIT2]] ]
71; ENABLED-NEXT:    ret i32 [[RETVAL]]
72; ENABLED:       for.body.1:
73; ENABLED-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT]]
74; ENABLED-NEXT:    [[TMP4:%.*]] = load i32, i32* [[ARRAYIDX_1]], align 4
75; ENABLED-NEXT:    [[ADD_1]] = add nsw i32 [[TMP4]], [[ADD]]
76; ENABLED-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT]], 1
77; ENABLED-NEXT:    [[NITER_NSUB_1:%.*]] = sub i64 [[NITER_NSUB]], 1
78; ENABLED-NEXT:    [[CMP_2:%.*]] = icmp eq i64 [[N]], 42
79; ENABLED-NEXT:    br i1 [[CMP_2]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_2]]
80; ENABLED:       for.body.2:
81; ENABLED-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_1]]
82; ENABLED-NEXT:    [[TMP5:%.*]] = load i32, i32* [[ARRAYIDX_2]], align 4
83; ENABLED-NEXT:    [[ADD_2]] = add nsw i32 [[TMP5]], [[ADD_1]]
84; ENABLED-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_1]], 1
85; ENABLED-NEXT:    [[NITER_NSUB_2:%.*]] = sub i64 [[NITER_NSUB_1]], 1
86; ENABLED-NEXT:    [[CMP_3:%.*]] = icmp eq i64 [[N]], 42
87; ENABLED-NEXT:    br i1 [[CMP_3]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_3]]
88; ENABLED:       for.body.3:
89; ENABLED-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_2]]
90; ENABLED-NEXT:    [[TMP6:%.*]] = load i32, i32* [[ARRAYIDX_3]], align 4
91; ENABLED-NEXT:    [[ADD_3]] = add nsw i32 [[TMP6]], [[ADD_2]]
92; ENABLED-NEXT:    [[INDVARS_IV_NEXT_3:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_2]], 1
93; ENABLED-NEXT:    [[NITER_NSUB_3:%.*]] = sub i64 [[NITER_NSUB_2]], 1
94; ENABLED-NEXT:    [[CMP_4:%.*]] = icmp eq i64 [[N]], 42
95; ENABLED-NEXT:    br i1 [[CMP_4]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_4]]
96; ENABLED:       for.body.4:
97; ENABLED-NEXT:    [[ARRAYIDX_4:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_3]]
98; ENABLED-NEXT:    [[TMP7:%.*]] = load i32, i32* [[ARRAYIDX_4]], align 4
99; ENABLED-NEXT:    [[ADD_4]] = add nsw i32 [[TMP7]], [[ADD_3]]
100; ENABLED-NEXT:    [[INDVARS_IV_NEXT_4:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_3]], 1
101; ENABLED-NEXT:    [[NITER_NSUB_4:%.*]] = sub i64 [[NITER_NSUB_3]], 1
102; ENABLED-NEXT:    [[CMP_5:%.*]] = icmp eq i64 [[N]], 42
103; ENABLED-NEXT:    br i1 [[CMP_5]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_5]]
104; ENABLED:       for.body.5:
105; ENABLED-NEXT:    [[ARRAYIDX_5:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_4]]
106; ENABLED-NEXT:    [[TMP8:%.*]] = load i32, i32* [[ARRAYIDX_5]], align 4
107; ENABLED-NEXT:    [[ADD_5]] = add nsw i32 [[TMP8]], [[ADD_4]]
108; ENABLED-NEXT:    [[INDVARS_IV_NEXT_5:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_4]], 1
109; ENABLED-NEXT:    [[NITER_NSUB_5:%.*]] = sub i64 [[NITER_NSUB_4]], 1
110; ENABLED-NEXT:    [[CMP_6:%.*]] = icmp eq i64 [[N]], 42
111; ENABLED-NEXT:    br i1 [[CMP_6]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_6]]
112; ENABLED:       for.body.6:
113; ENABLED-NEXT:    [[ARRAYIDX_6:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_5]]
114; ENABLED-NEXT:    [[TMP9:%.*]] = load i32, i32* [[ARRAYIDX_6]], align 4
115; ENABLED-NEXT:    [[ADD_6]] = add nsw i32 [[TMP9]], [[ADD_5]]
116; ENABLED-NEXT:    [[INDVARS_IV_NEXT_6:%.*]] = add nuw nsw i64 [[INDVARS_IV_NEXT_5]], 1
117; ENABLED-NEXT:    [[NITER_NSUB_6:%.*]] = sub i64 [[NITER_NSUB_5]], 1
118; ENABLED-NEXT:    [[CMP_7:%.*]] = icmp eq i64 [[N]], 42
119; ENABLED-NEXT:    br i1 [[CMP_7]], label [[FOR_EXIT2_LOOPEXIT]], label [[FOR_BODY_7]]
120; ENABLED:       for.body.7:
121; ENABLED-NEXT:    [[ARRAYIDX_7:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV_NEXT_6]]
122; ENABLED-NEXT:    [[TMP10:%.*]] = load i32, i32* [[ARRAYIDX_7]], align 4
123; ENABLED-NEXT:    [[ADD_7]] = add nsw i32 [[TMP10]], [[ADD_6]]
124; ENABLED-NEXT:    [[INDVARS_IV_NEXT_7]] = add i64 [[INDVARS_IV_NEXT_6]], 1
125; ENABLED-NEXT:    [[NITER_NSUB_7]] = sub i64 [[NITER_NSUB_6]], 1
126; ENABLED-NEXT:    [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
127; ENABLED-NEXT:    br i1 [[NITER_NCMP_7]], label [[FOR_END_UNR_LCSSA_LOOPEXIT]], label [[HEADER]]
128;
129; DISABLED-LABEL: @test(
130; DISABLED-NEXT:  entry:
131; DISABLED-NEXT:    br label [[HEADER:%.*]]
132; DISABLED:       header:
133; DISABLED-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY:%.*]] ], [ 0, [[ENTRY:%.*]] ]
134; DISABLED-NEXT:    [[SUM_02:%.*]] = phi i32 [ [[ADD:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY]] ]
135; DISABLED-NEXT:    [[CMP:%.*]] = icmp eq i64 [[N:%.*]], 42
136; DISABLED-NEXT:    br i1 [[CMP]], label [[FOR_EXIT2:%.*]], label [[FOR_BODY]]
137; DISABLED:       for.body:
138; DISABLED-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[INDVARS_IV]]
139; DISABLED-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
140; DISABLED-NEXT:    [[ADD]] = add nsw i32 [[TMP0]], [[SUM_02]]
141; DISABLED-NEXT:    [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1
142; DISABLED-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N]]
143; DISABLED-NEXT:    br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[HEADER]]
144; DISABLED:       for.end:
145; DISABLED-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_BODY]] ]
146; DISABLED-NEXT:    ret i32 [[SUM_0_LCSSA]]
147; DISABLED:       for.exit2:
148; DISABLED-NEXT:    [[RETVAL:%.*]] = phi i32 [ [[SUM_02]], [[HEADER]] ]
149; DISABLED-NEXT:    ret i32 [[RETVAL]]
150;
151entry:
152  br label %header
153
154header:
155  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
156  %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]
157  %cmp = icmp eq i64 %n, 42
158  br i1 %cmp, label %for.exit2, label %for.body
159
160for.body:
161  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
162  %0 = load i32, i32* %arrayidx, align 4
163  %add = add nsw i32 %0, %sum.02
164  %indvars.iv.next = add i64 %indvars.iv, 1
165  %exitcond = icmp eq i64 %indvars.iv.next, %n
166  br i1 %exitcond, label %for.end, label %header
167
168for.end:
169  %sum.0.lcssa = phi i32 [ %add, %for.body ]
170  ret i32 %sum.0.lcssa
171
172for.exit2:
173  %retval = phi i32 [ %sum.02, %header ]
174  ret i32 %retval
175}
176