1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -basicaa -loop-interchange -verify-dom-info -verify-loop-info -verify-scev -verify-loop-lcssa -S | FileCheck %s
3
4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
5target triple = "x86_64-unknown-linux-gnu"
6
7@A = common global [100 x [100 x i64]] zeroinitializer
8@B = common global [100 x i64] zeroinitializer
9
10;;  for(int i=0;i<100;i++)
11;;    for(int j=0;j<100;j++)
12;;      A[j][i] = A[j][i]+k;
13
14define void @interchange_01(i64 %k, i64 %N) {
15; CHECK-LABEL: @interchange_01(
16; CHECK-NEXT:  entry:
17; CHECK-NEXT:    br label [[FOR2_PREHEADER:%.*]]
18; CHECK:       for1.header.preheader:
19; CHECK-NEXT:    br label [[FOR1_HEADER:%.*]]
20; CHECK:       for1.header:
21; CHECK-NEXT:    [[J23:%.*]] = phi i64 [ [[J_NEXT24:%.*]], [[FOR1_INC10:%.*]] ], [ 0, [[FOR1_HEADER_PREHEADER:%.*]] ]
22; CHECK-NEXT:    br label [[FOR2_SPLIT1:%.*]]
23; CHECK:       for2.preheader:
24; CHECK-NEXT:    br label [[FOR2:%.*]]
25; CHECK:       for2:
26; CHECK-NEXT:    [[J:%.*]] = phi i64 [ [[TMP0:%.*]], [[FOR2_SPLIT:%.*]] ], [ 0, [[FOR2_PREHEADER]] ]
27; CHECK-NEXT:    br label [[FOR1_HEADER_PREHEADER]]
28; CHECK:       for2.split1:
29; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 [[J]], i64 [[J23]]
30; CHECK-NEXT:    [[LV:%.*]] = load i64, i64* [[ARRAYIDX5]]
31; CHECK-NEXT:    [[ADD:%.*]] = add nsw i64 [[LV]], [[K:%.*]]
32; CHECK-NEXT:    store i64 [[ADD]], i64* [[ARRAYIDX5]]
33; CHECK-NEXT:    [[J_NEXT:%.*]] = add nuw nsw i64 [[J]], 1
34; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[J]], 99
35; CHECK-NEXT:    br label [[FOR1_INC10]]
36; CHECK:       for2.split:
37; CHECK-NEXT:    [[TMP0]] = add nuw nsw i64 [[J]], 1
38; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i64 [[J]], 99
39; CHECK-NEXT:    br i1 [[TMP1]], label [[FOR_END12:%.*]], label [[FOR2]]
40; CHECK:       for1.inc10:
41; CHECK-NEXT:    [[J_NEXT24]] = add nuw nsw i64 [[J23]], 1
42; CHECK-NEXT:    [[EXITCOND26:%.*]] = icmp eq i64 [[J23]], 99
43; CHECK-NEXT:    br i1 [[EXITCOND26]], label [[FOR2_SPLIT]], label [[FOR1_HEADER]]
44; CHECK:       for.end12:
45; CHECK-NEXT:    ret void
46;
47entry:
48  br label %for1.header
49
50for1.header:
51  %j23 = phi i64 [ 0, %entry ], [ %j.next24, %for1.inc10 ]
52  br label %for2
53
54for2:
55  %j = phi i64 [ %j.next, %for2 ], [ 0, %for1.header ]
56  %arrayidx5 = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 %j, i64 %j23
57  %lv = load i64, i64* %arrayidx5
58  %add = add nsw i64 %lv, %k
59  store i64 %add, i64* %arrayidx5
60  %j.next = add nuw nsw i64 %j, 1
61  %exitcond = icmp eq i64 %j, 99
62  br i1 %exitcond, label %for1.inc10, label %for2
63
64for1.inc10:
65  %j.next24 = add nuw nsw i64 %j23, 1
66  %exitcond26 = icmp eq i64 %j23, 99
67  br i1 %exitcond26, label %for.end12, label %for1.header
68
69for.end12:
70  ret void
71}
72
73;; for(int i=0;i<100;i++)
74;;   for(int j=100;j>=0;j--)
75;;     A[j][i] = A[j][i]+k;
76
77define void @interchange_02(i64 %k) {
78; CHECK-LABEL: @interchange_02(
79; CHECK-NEXT:  entry:
80; CHECK-NEXT:    br label [[FOR3_PREHEADER:%.*]]
81; CHECK:       for1.header.preheader:
82; CHECK-NEXT:    br label [[FOR1_HEADER:%.*]]
83; CHECK:       for1.header:
84; CHECK-NEXT:    [[J19:%.*]] = phi i64 [ [[J_NEXT20:%.*]], [[FOR1_INC10:%.*]] ], [ 0, [[FOR1_HEADER_PREHEADER:%.*]] ]
85; CHECK-NEXT:    br label [[FOR3_SPLIT1:%.*]]
86; CHECK:       for3.preheader:
87; CHECK-NEXT:    br label [[FOR3:%.*]]
88; CHECK:       for3:
89; CHECK-NEXT:    [[J:%.*]] = phi i64 [ [[TMP1:%.*]], [[FOR3_SPLIT:%.*]] ], [ 100, [[FOR3_PREHEADER]] ]
90; CHECK-NEXT:    br label [[FOR1_HEADER_PREHEADER]]
91; CHECK:       for3.split1:
92; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 [[J]], i64 [[J19]]
93; CHECK-NEXT:    [[TMP0:%.*]] = load i64, i64* [[ARRAYIDX5]]
94; CHECK-NEXT:    [[ADD:%.*]] = add nsw i64 [[TMP0]], [[K:%.*]]
95; CHECK-NEXT:    store i64 [[ADD]], i64* [[ARRAYIDX5]]
96; CHECK-NEXT:    [[J_NEXT:%.*]] = add nsw i64 [[J]], -1
97; CHECK-NEXT:    [[CMP2:%.*]] = icmp sgt i64 [[J]], 0
98; CHECK-NEXT:    br label [[FOR1_INC10]]
99; CHECK:       for3.split:
100; CHECK-NEXT:    [[TMP1]] = add nsw i64 [[J]], -1
101; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i64 [[J]], 0
102; CHECK-NEXT:    br i1 [[TMP2]], label [[FOR3]], label [[FOR_END11:%.*]]
103; CHECK:       for1.inc10:
104; CHECK-NEXT:    [[J_NEXT20]] = add nuw nsw i64 [[J19]], 1
105; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[J_NEXT20]], 100
106; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR3_SPLIT]], label [[FOR1_HEADER]]
107; CHECK:       for.end11:
108; CHECK-NEXT:    ret void
109;
110entry:
111  br label %for1.header
112
113for1.header:
114  %j19 = phi i64 [ 0, %entry ], [ %j.next20, %for1.inc10 ]
115  br label %for3
116
117for3:
118  %j = phi i64 [ 100, %for1.header ], [ %j.next, %for3 ]
119  %arrayidx5 = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 %j, i64 %j19
120  %0 = load i64, i64* %arrayidx5
121  %add = add nsw i64 %0, %k
122  store i64 %add, i64* %arrayidx5
123  %j.next = add nsw i64 %j, -1
124  %cmp2 = icmp sgt i64 %j, 0
125  br i1 %cmp2, label %for3, label %for1.inc10
126
127for1.inc10:
128  %j.next20 = add nuw nsw i64 %j19, 1
129  %exitcond = icmp eq i64 %j.next20, 100
130  br i1 %exitcond, label %for.end11, label %for1.header
131
132for.end11:
133  ret void
134}
135
136;; Test to make sure we can handle output dependencies.
137;;
138;;  for (int i = 1; i < 100; ++i)
139;;    for(int j = 1; j < 99; ++j) {
140;;      A[j][i] = i;
141;;      A[j][i+1] = j;
142;;    }
143;; FIXME: DA misses this case after D35430
144
145define void @interchange_10() {
146; CHECK-LABEL: @interchange_10(
147; CHECK-NEXT:  entry:
148; CHECK-NEXT:    br label [[FOR1_HEADER:%.*]]
149; CHECK:       for1.header:
150; CHECK-NEXT:    [[J23:%.*]] = phi i64 [ 1, [[ENTRY:%.*]] ], [ [[J_NEXT24:%.*]], [[FOR1_INC10:%.*]] ]
151; CHECK-NEXT:    [[J_NEXT24]] = add nuw nsw i64 [[J23]], 1
152; CHECK-NEXT:    br label [[FOR2:%.*]]
153; CHECK:       for2:
154; CHECK-NEXT:    [[J:%.*]] = phi i64 [ [[J_NEXT:%.*]], [[FOR2]] ], [ 1, [[FOR1_HEADER]] ]
155; CHECK-NEXT:    [[J_NEXT]] = add nuw nsw i64 [[J]], 1
156; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 [[J]], i64 [[J23]]
157; CHECK-NEXT:    store i64 [[J]], i64* [[ARRAYIDX5]]
158; CHECK-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 [[J]], i64 [[J_NEXT24]]
159; CHECK-NEXT:    store i64 [[J23]], i64* [[ARRAYIDX10]]
160; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[J]], 99
161; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR1_INC10]], label [[FOR2]]
162; CHECK:       for1.inc10:
163; CHECK-NEXT:    [[EXITCOND26:%.*]] = icmp eq i64 [[J23]], 98
164; CHECK-NEXT:    br i1 [[EXITCOND26]], label [[FOR_END12:%.*]], label [[FOR1_HEADER]]
165; CHECK:       for.end12:
166; CHECK-NEXT:    ret void
167;
168entry:
169  br label %for1.header
170
171for1.header:
172  %j23 = phi i64 [ 1, %entry ], [ %j.next24, %for1.inc10 ]
173  %j.next24 = add nuw nsw i64 %j23, 1
174  br label %for2
175
176for2:
177  %j = phi i64 [ %j.next, %for2 ], [ 1, %for1.header ]
178  %j.next = add nuw nsw i64 %j, 1
179  %arrayidx5 = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 %j, i64 %j23
180  store i64 %j, i64* %arrayidx5
181  %arrayidx10 = getelementptr inbounds [100 x [100 x i64]], [100 x [100 x i64]]* @A, i64 0, i64 %j, i64 %j.next24
182  store i64 %j23, i64* %arrayidx10
183  %exitcond = icmp eq i64 %j, 99
184  br i1 %exitcond, label %for1.inc10, label %for2
185
186for1.inc10:
187  %exitcond26 = icmp eq i64 %j23, 98
188  br i1 %exitcond26, label %for.end12, label %for1.header
189
190for.end12:
191  ret void
192
193}
194