1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -basic-aa -loop-interchange -verify-dom-info -verify-loop-info -verify-scev -verify-loop-lcssa -S | FileCheck %s
3
4target triple = "powerpc64le-unknown-linux-gnu"
5@b = common dso_local local_unnamed_addr global [200 x [200 x i32]] zeroinitializer, align 4
6@a = common dso_local local_unnamed_addr global i32 0, align 4
7
8;; int a, c, d, e;
9;; int b[200][200];
10;; void fn1() {
11;;   for (c = 0; c < 100; c++) {
12;;     for (d = 5, e = 5; d > 0, e > 0; d--, e--)
13;;       a |= b[d][c + 9];
14;;   }
15;; }
16;
17; There are multiple inner loop indvars and only one
18; of them is used in the loop exit condition at the
19; inner loop latch.
20;
21define void @test1() {
22; CHECK-LABEL: @test1(
23; CHECK-NEXT:  entry:
24; CHECK-NEXT:    [[A:%.*]] = load i32, i32* @a, align 4
25; CHECK-NEXT:    br label [[FOR_BODY3_PREHEADER:%.*]]
26; CHECK:       for.body.preheader:
27; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
28; CHECK:       for.body:
29; CHECK-NEXT:    [[INDVARS_OUTER:%.*]] = phi i64 [ [[INDVARS_OUTER_NEXT:%.*]], [[FOR_INC7:%.*]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
30; CHECK-NEXT:    [[OR_REDUCTION_INNER:%.*]] = phi i32 [ [[OR:%.*]], [[FOR_INC7]] ], [ [[OR_REDUCTION_OUTER:%.*]], [[FOR_BODY_PREHEADER]] ]
31; CHECK-NEXT:    [[INDEX:%.*]] = add nsw i64 [[INDVARS_OUTER]], 9
32; CHECK-NEXT:    br label [[FOR_BODY3_SPLIT1:%.*]]
33; CHECK:       for.body3.preheader:
34; CHECK-NEXT:    br label [[FOR_BODY3:%.*]]
35; CHECK:       for.body3:
36; CHECK-NEXT:    [[INDVAR0:%.*]] = phi i64 [ [[TMP0:%.*]], [[FOR_BODY3_SPLIT:%.*]] ], [ 5, [[FOR_BODY3_PREHEADER]] ]
37; CHECK-NEXT:    [[INDVAR1:%.*]] = phi i32 [ [[TMP1:%.*]], [[FOR_BODY3_SPLIT]] ], [ 5, [[FOR_BODY3_PREHEADER]] ]
38; CHECK-NEXT:    [[OR_REDUCTION_OUTER]] = phi i32 [ [[OR_LCSSA:%.*]], [[FOR_BODY3_SPLIT]] ], [ [[A]], [[FOR_BODY3_PREHEADER]] ]
39; CHECK-NEXT:    br label [[FOR_BODY_PREHEADER]]
40; CHECK:       for.body3.split1:
41; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [200 x [200 x i32]], [200 x [200 x i32]]* @b, i64 0, i64 [[INDVAR0]], i64 [[INDEX]]
42; CHECK-NEXT:    [[LOAD_VAL:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
43; CHECK-NEXT:    [[OR]] = or i32 [[OR_REDUCTION_INNER]], [[LOAD_VAL]]
44; CHECK-NEXT:    [[INDVAR0_NEXT:%.*]] = add nsw i64 [[INDVAR0]], -1
45; CHECK-NEXT:    [[INDVAR1_NEXT:%.*]] = add nsw i32 [[INDVAR1]], -1
46; CHECK-NEXT:    [[TOBOOL2:%.*]] = icmp eq i32 [[INDVAR1_NEXT]], 0
47; CHECK-NEXT:    br label [[FOR_INC7]]
48; CHECK:       for.body3.split:
49; CHECK-NEXT:    [[OR_LCSSA]] = phi i32 [ [[OR]], [[FOR_INC7]] ]
50; CHECK-NEXT:    [[TMP0]] = add nsw i64 [[INDVAR0]], -1
51; CHECK-NEXT:    [[TMP1]] = add nsw i32 [[INDVAR1]], -1
52; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
53; CHECK-NEXT:    br i1 [[TMP2]], label [[FOR_COND_FOR_END8_CRIT_EDGE:%.*]], label [[FOR_BODY3]]
54; CHECK:       for.inc7:
55; CHECK-NEXT:    [[INDVARS_OUTER_NEXT]] = add nsw i64 [[INDVARS_OUTER]], 1
56; CHECK-NEXT:    [[INDVARS_OUTER_NEXT_TRUNC:%.*]] = trunc i64 [[INDVARS_OUTER_NEXT]] to i32
57; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp eq i32 [[INDVARS_OUTER_NEXT_TRUNC]], 100
58; CHECK-NEXT:    br i1 [[TOBOOL]], label [[FOR_BODY3_SPLIT]], label [[FOR_BODY]]
59; CHECK:       for.cond.for.end8_crit_edge:
60; CHECK-NEXT:    [[OR_LCSSA_LCSSA:%.*]] = phi i32 [ [[OR_LCSSA]], [[FOR_BODY3_SPLIT]] ]
61; CHECK-NEXT:    store i32 [[OR_LCSSA_LCSSA]], i32* @a, align 4
62; CHECK-NEXT:    br label [[FOR_END8:%.*]]
63; CHECK:       for.end8:
64; CHECK-NEXT:    ret void
65;
66
67entry:
68  %a = load i32, i32* @a
69  br label %for.body
70
71for.body:                                         ; preds = %for.body.lr.ph, %for.inc7
72  %indvars.outer = phi i64 [ 0, %entry ], [ %indvars.outer.next, %for.inc7 ]
73  %or.reduction.outer = phi i32 [ %a, %entry ], [ %or.lcssa, %for.inc7 ]
74  %index = add nsw i64 %indvars.outer, 9
75  br label %for.body3
76
77for.body3:                                        ; preds = %for.body, %for.body3
78  %or.reduction.inner = phi i32 [ %or.reduction.outer, %for.body ], [ %or, %for.body3 ]
79  %indvar0 = phi i64 [ 5, %for.body ], [ %indvar0.next, %for.body3 ]
80  %indvar1 = phi i32 [ 5, %for.body ], [ %indvar1.next, %for.body3 ]
81  %arrayidx5 = getelementptr inbounds [200 x [200 x i32]], [200 x [200 x i32]]* @b, i64 0, i64 %indvar0, i64 %index
82  %load.val = load i32, i32* %arrayidx5, align 4
83  %or = or i32 %or.reduction.inner, %load.val
84  %indvar0.next = add nsw i64 %indvar0, -1
85  %indvar1.next = add nsw i32 %indvar1, -1
86  %tobool2 = icmp eq i32 %indvar1.next, 0
87  br i1 %tobool2, label %for.inc7, label %for.body3
88
89for.inc7:                                         ; preds = %for.body3
90  %or.lcssa = phi i32 [ %or, %for.body3 ]
91  %indvars.outer.next = add nsw i64 %indvars.outer, 1
92  %indvars.outer.next.trunc = trunc i64 %indvars.outer.next to i32
93  %tobool = icmp eq i32 %indvars.outer.next.trunc, 100
94  br i1 %tobool, label %for.cond.for.end8_crit_edge, label %for.body
95
96for.cond.for.end8_crit_edge:                      ; preds = %for.inc7
97  %or.lcssa.lcssa = phi i32 [ %or.lcssa, %for.inc7 ]
98  store i32 %or.lcssa.lcssa, i32* @a
99  br label %for.end8
100
101for.end8:                                         ; preds = %for.cond.for.end8_crit_edge, %entry
102  ret void
103}
104
105;; int a, c, d, e;
106;; int b[200][200];
107;; void fn1() {
108;;   for (c = 0 ; c < 100; c++) {
109;;     for (d = 5, e = 6; d + e > 0; d--, e = e - 2)
110;;       a |= b[d][c + 9];
111;;   }
112;; }
113;
114; All inner loop indvars are used in the inner latch.
115;
116define void @test2() {
117; CHECK-LABEL: @test2(
118; CHECK-NEXT:  entry:
119; CHECK-NEXT:    [[A:%.*]] = load i32, i32* @a, align 4
120; CHECK-NEXT:    br label [[FOR_BODY3_PREHEADER:%.*]]
121; CHECK:       for.body.preheader:
122; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
123; CHECK:       for.body:
124; CHECK-NEXT:    [[INDVARS_OUTER:%.*]] = phi i64 [ [[INDVARS_OUTER_NEXT:%.*]], [[FOR_INC7:%.*]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
125; CHECK-NEXT:    [[OR_REDUCTION_INNER:%.*]] = phi i32 [ [[OR:%.*]], [[FOR_INC7]] ], [ [[OR_REDUCTION_OUTER:%.*]], [[FOR_BODY_PREHEADER]] ]
126; CHECK-NEXT:    [[INDEX:%.*]] = add nsw i64 [[INDVARS_OUTER]], 9
127; CHECK-NEXT:    br label [[FOR_BODY3_SPLIT1:%.*]]
128; CHECK:       for.body3.preheader:
129; CHECK-NEXT:    br label [[FOR_BODY3:%.*]]
130; CHECK:       for.body3:
131; CHECK-NEXT:    [[INDVAR0:%.*]] = phi i64 [ [[TMP2:%.*]], [[FOR_BODY3_SPLIT:%.*]] ], [ 5, [[FOR_BODY3_PREHEADER]] ]
132; CHECK-NEXT:    [[INDVAR1:%.*]] = phi i32 [ [[TMP0:%.*]], [[FOR_BODY3_SPLIT]] ], [ 6, [[FOR_BODY3_PREHEADER]] ]
133; CHECK-NEXT:    [[OR_REDUCTION_OUTER]] = phi i32 [ [[OR_LCSSA:%.*]], [[FOR_BODY3_SPLIT]] ], [ [[A]], [[FOR_BODY3_PREHEADER]] ]
134; CHECK-NEXT:    br label [[FOR_BODY_PREHEADER]]
135; CHECK:       for.body3.split1:
136; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [200 x [200 x i32]], [200 x [200 x i32]]* @b, i64 0, i64 [[INDVAR0]], i64 [[INDEX]]
137; CHECK-NEXT:    [[LOAD_VAL:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
138; CHECK-NEXT:    [[OR]] = or i32 [[OR_REDUCTION_INNER]], [[LOAD_VAL]]
139; CHECK-NEXT:    [[INDVAR0_NEXT:%.*]] = add nsw i64 [[INDVAR0]], -1
140; CHECK-NEXT:    [[INDVAR1_NEXT:%.*]] = add nsw i32 [[INDVAR1]], -2
141; CHECK-NEXT:    [[INDVAR1_NEXT_EXT:%.*]] = sext i32 [[INDVAR1_NEXT]] to i64
142; CHECK-NEXT:    [[INDVARS_ADD:%.*]] = add nsw i64 [[INDVAR0_NEXT]], [[INDVAR1_NEXT_EXT]]
143; CHECK-NEXT:    [[TOBOOL2:%.*]] = icmp eq i64 [[INDVARS_ADD]], 0
144; CHECK-NEXT:    br label [[FOR_INC7]]
145; CHECK:       for.body3.split:
146; CHECK-NEXT:    [[OR_LCSSA]] = phi i32 [ [[OR]], [[FOR_INC7]] ]
147; CHECK-NEXT:    [[TMP0]] = add nsw i32 [[INDVAR1]], -2
148; CHECK-NEXT:    [[TMP1:%.*]] = sext i32 [[TMP0]] to i64
149; CHECK-NEXT:    [[TMP2]] = add nsw i64 [[INDVAR0]], -1
150; CHECK-NEXT:    [[TMP3:%.*]] = add nsw i64 [[TMP2]], [[TMP1]]
151; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[TMP3]], 0
152; CHECK-NEXT:    br i1 [[TMP4]], label [[FOR_COND_FOR_END8_CRIT_EDGE:%.*]], label [[FOR_BODY3]]
153; CHECK:       for.inc7:
154; CHECK-NEXT:    [[INDVARS_OUTER_NEXT]] = add nsw i64 [[INDVARS_OUTER]], 1
155; CHECK-NEXT:    [[INDVARS_OUTER_NEXT_TRUNC:%.*]] = trunc i64 [[INDVARS_OUTER_NEXT]] to i32
156; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp eq i32 [[INDVARS_OUTER_NEXT_TRUNC]], 100
157; CHECK-NEXT:    br i1 [[TOBOOL]], label [[FOR_BODY3_SPLIT]], label [[FOR_BODY]]
158; CHECK:       for.cond.for.end8_crit_edge:
159; CHECK-NEXT:    [[OR_LCSSA_LCSSA:%.*]] = phi i32 [ [[OR_LCSSA]], [[FOR_BODY3_SPLIT]] ]
160; CHECK-NEXT:    store i32 [[OR_LCSSA_LCSSA]], i32* @a, align 4
161; CHECK-NEXT:    br label [[FOR_END8:%.*]]
162; CHECK:       for.end8:
163; CHECK-NEXT:    ret void
164;
165entry:
166  %a = load i32, i32* @a
167  br label %for.body
168
169for.body:                                         ; preds = %for.body.lr.ph, %for.inc7
170  %indvars.outer = phi i64 [ 0, %entry ], [ %indvars.outer.next, %for.inc7 ]
171  %or.reduction.outer = phi i32 [ %a, %entry ], [ %or.lcssa, %for.inc7 ]
172  %index = add nsw i64 %indvars.outer, 9
173  br label %for.body3
174
175for.body3:                                        ; preds = %for.body, %for.body3
176  %or.reduction.inner = phi i32 [ %or.reduction.outer, %for.body ], [ %or, %for.body3 ]
177  %indvar0 = phi i64 [ 5, %for.body ], [ %indvar0.next, %for.body3 ]
178  %indvar1 = phi i32 [ 6, %for.body ], [ %indvar1.next, %for.body3 ]
179  %arrayidx5 = getelementptr inbounds [200 x [200 x i32]], [200 x [200 x i32]]* @b, i64 0, i64 %indvar0, i64 %index
180  %load.val = load i32, i32* %arrayidx5, align 4
181  %or = or i32 %or.reduction.inner, %load.val
182  %indvar0.next = add nsw i64 %indvar0, -1
183  %indvar1.next = add nsw i32 %indvar1, -2
184  %indvar1.next.ext = sext i32 %indvar1.next to i64
185  %indvars.add = add nsw i64 %indvar0.next, %indvar1.next.ext
186  %tobool2 = icmp eq i64 %indvars.add, 0
187  br i1 %tobool2, label %for.inc7, label %for.body3
188
189for.inc7:                                         ; preds = %for.body3
190  %or.lcssa = phi i32 [ %or, %for.body3 ]
191  %indvars.outer.next = add nsw i64 %indvars.outer, 1
192  %indvars.outer.next.trunc = trunc i64 %indvars.outer.next to i32
193  %tobool = icmp eq i32 %indvars.outer.next.trunc, 100
194  br i1 %tobool, label %for.cond.for.end8_crit_edge, label %for.body
195
196for.cond.for.end8_crit_edge:                      ; preds = %for.inc7
197  %or.lcssa.lcssa = phi i32 [ %or.lcssa, %for.inc7 ]
198  store i32 %or.lcssa.lcssa, i32* @a
199  br label %for.end8
200
201for.end8:                                         ; preds = %for.cond.for.end8_crit_edge, %entry
202  ret void
203}
204
205;; int a, c, d, e;
206;; int b[200][200];
207;; void fn1() {
208;;   for (c = 0 ; c < 100; c++) {
209;;     d = 5;
210;;     e = 49;
211;;     for (; d != e; d++, e--)
212;;       a |= b[d][c + 9];
213;;   }
214;; }
215;
216; Two inner loop indvars are involved in the inner loop exit
217; condition as LHS and RHS.
218define void @test3() {
219; CHECK-LABEL: @test3(
220; CHECK-NEXT:  entry:
221; CHECK-NEXT:    [[A:%.*]] = load i32, i32* @a, align 4
222; CHECK-NEXT:    br label [[FOR_BODY3_PREHEADER:%.*]]
223; CHECK:       for.body.preheader:
224; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
225; CHECK:       for.body:
226; CHECK-NEXT:    [[INDVARS_OUTER:%.*]] = phi i64 [ [[INDVARS_OUTER_NEXT:%.*]], [[FOR_INC7:%.*]] ], [ 0, [[FOR_BODY_PREHEADER:%.*]] ]
227; CHECK-NEXT:    [[OR_REDUCTION_INNER:%.*]] = phi i32 [ [[OR:%.*]], [[FOR_INC7]] ], [ [[OR_REDUCTION_OUTER:%.*]], [[FOR_BODY_PREHEADER]] ]
228; CHECK-NEXT:    [[INDEX:%.*]] = add nsw i64 [[INDVARS_OUTER]], 9
229; CHECK-NEXT:    br label [[FOR_BODY3_SPLIT1:%.*]]
230; CHECK:       for.body3.preheader:
231; CHECK-NEXT:    br label [[FOR_BODY3:%.*]]
232; CHECK:       for.body3:
233; CHECK-NEXT:    [[INDVAR0:%.*]] = phi i32 [ [[TMP1:%.*]], [[FOR_BODY3_SPLIT:%.*]] ], [ 5, [[FOR_BODY3_PREHEADER]] ]
234; CHECK-NEXT:    [[INDVAR1:%.*]] = phi i32 [ [[TMP0:%.*]], [[FOR_BODY3_SPLIT]] ], [ 49, [[FOR_BODY3_PREHEADER]] ]
235; CHECK-NEXT:    [[OR_REDUCTION_OUTER]] = phi i32 [ [[OR_LCSSA:%.*]], [[FOR_BODY3_SPLIT]] ], [ [[A]], [[FOR_BODY3_PREHEADER]] ]
236; CHECK-NEXT:    br label [[FOR_BODY_PREHEADER]]
237; CHECK:       for.body3.split1:
238; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [200 x [200 x i32]], [200 x [200 x i32]]* @b, i64 0, i32 [[INDVAR0]], i64 [[INDEX]]
239; CHECK-NEXT:    [[LOAD_VAL:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
240; CHECK-NEXT:    [[OR]] = or i32 [[OR_REDUCTION_INNER]], [[LOAD_VAL]]
241; CHECK-NEXT:    [[INDVAR0_NEXT:%.*]] = add nsw i32 [[INDVAR0]], 1
242; CHECK-NEXT:    [[INDVAR1_NEXT:%.*]] = add nsw i32 [[INDVAR1]], -1
243; CHECK-NEXT:    [[TOBOOL2:%.*]] = icmp eq i32 [[INDVAR0_NEXT]], [[INDVAR1_NEXT]]
244; CHECK-NEXT:    br label [[FOR_INC7]]
245; CHECK:       for.body3.split:
246; CHECK-NEXT:    [[OR_LCSSA]] = phi i32 [ [[OR]], [[FOR_INC7]] ]
247; CHECK-NEXT:    [[TMP0]] = add nsw i32 [[INDVAR1]], -1
248; CHECK-NEXT:    [[TMP1]] = add nsw i32 [[INDVAR0]], 1
249; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[TMP0]]
250; CHECK-NEXT:    br i1 [[TMP2]], label [[FOR_COND_FOR_END8_CRIT_EDGE:%.*]], label [[FOR_BODY3]]
251; CHECK:       for.inc7:
252; CHECK-NEXT:    [[INDVARS_OUTER_NEXT]] = add nsw i64 [[INDVARS_OUTER]], 1
253; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp eq i64 [[INDVARS_OUTER_NEXT]], 100
254; CHECK-NEXT:    br i1 [[TOBOOL]], label [[FOR_BODY3_SPLIT]], label [[FOR_BODY]]
255; CHECK:       for.cond.for.end8_crit_edge:
256; CHECK-NEXT:    [[OR_LCSSA_LCSSA:%.*]] = phi i32 [ [[OR_LCSSA]], [[FOR_BODY3_SPLIT]] ]
257; CHECK-NEXT:    store i32 [[OR_LCSSA_LCSSA]], i32* @a, align 4
258; CHECK-NEXT:    br label [[FOR_END8:%.*]]
259; CHECK:       for.end8:
260; CHECK-NEXT:    ret void
261;
262
263entry:
264  %a = load i32, i32* @a
265  br label %for.body
266
267for.body:                                         ; preds = %for.body.lr.ph, %for.inc7
268  %indvars.outer = phi i64 [ 0, %entry ], [ %indvars.outer.next, %for.inc7 ]
269  %or.reduction.outer = phi i32 [ %a, %entry ], [ %or.lcssa, %for.inc7 ]
270  %index = add nsw i64 %indvars.outer, 9
271  br label %for.body3
272
273for.body3:                                        ; preds = %for.body, %for.body3
274  %or.reduction.inner = phi i32 [ %or.reduction.outer, %for.body ], [ %or, %for.body3 ]
275  %indvar0 = phi i32 [ 5, %for.body ], [ %indvar0.next, %for.body3 ]
276  %indvar1 = phi i32 [ 49, %for.body ], [ %indvar1.next, %for.body3 ]
277  %arrayidx5 = getelementptr inbounds [200 x [200 x i32]], [200 x [200 x i32]]* @b, i64 0, i32 %indvar0, i64 %index
278  %load.val = load i32, i32* %arrayidx5, align 4
279  %or = or i32 %or.reduction.inner, %load.val
280  %indvar0.next = add nsw i32 %indvar0, 1
281  %indvar1.next = add nsw i32 %indvar1, -1
282  %tobool2 = icmp eq i32 %indvar0.next, %indvar1.next
283  br i1 %tobool2, label %for.inc7, label %for.body3
284
285for.inc7:                                         ; preds = %for.body3
286  %or.lcssa = phi i32 [ %or, %for.body3 ]
287  %indvars.outer.next = add nsw i64 %indvars.outer, 1
288  %tobool = icmp eq i64 %indvars.outer.next, 100
289  br i1 %tobool, label %for.cond.for.end8_crit_edge, label %for.body
290
291for.cond.for.end8_crit_edge:                      ; preds = %for.inc7
292  %or.lcssa.lcssa = phi i32 [ %or.lcssa, %for.inc7 ]
293  store i32 %or.lcssa.lcssa, i32* @a
294  br label %for.end8
295
296for.end8:                                         ; preds = %for.cond.for.end8_crit_edge, %entry
297  ret void
298}
299