1; RUN: opt < %s -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck %s
2
3target datalayout = "e-m:e-i64:64-n32:64"
4target triple = "powerpc64le-unknown-linux-gnu"
5
6; Check delinearization in loop cache analysis can handle fixed-size arrays.
7; The IR is copied from llvm/test/Analysis/DependenceAnalysis/SimpleSIVNoValidityCheckFixedSize.ll
8
9; CHECK-DAG: Loop 'for.body' has cost = 4186116
10; CHECK-DAG: Loop 'for.body4' has cost = 128898
11
12;; #define N 1024
13;; #define M 2048
14;; void t1(int a[N][M]) {
15;;   for (int i = 0; i < N-1; ++i)
16;;     for (int j = 2; j < M; ++j)
17;;       a[i][j] = a[i+1][j-2];
18;; }
19
20define void @t1([2048 x i32]* %a) {
21entry:
22  br label %for.body
23
24for.body:                                         ; preds = %entry, %for.inc11
25  %indvars.iv4 = phi i64 [ 0, %entry ], [ %indvars.iv.next5, %for.inc11 ]
26  br label %for.body4
27
28for.body4:                                        ; preds = %for.body, %for.body4
29  %indvars.iv = phi i64 [ 2, %for.body ], [ %indvars.iv.next, %for.body4 ]
30  %0 = add nuw nsw i64 %indvars.iv4, 1
31  %1 = add nsw i64 %indvars.iv, -2
32  %arrayidx6 = getelementptr inbounds [2048 x i32], [2048 x i32]* %a, i64 %0, i64 %1
33  %2 = load i32, i32* %arrayidx6, align 4
34  %a_gep = getelementptr inbounds [2048 x i32], [2048 x i32]* %a, i64 0
35  %arrayidx10 = getelementptr inbounds [2048 x i32], [2048 x i32]* %a_gep, i64 %indvars.iv4, i64 %indvars.iv
36  store i32 %2, i32* %arrayidx10, align 4
37  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
38  %exitcond = icmp ne i64 %indvars.iv.next, 2048
39  br i1 %exitcond, label %for.body4, label %for.inc11
40
41for.inc11:                                        ; preds = %for.body4
42  %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 1
43  %exitcond7 = icmp ne i64 %indvars.iv.next5, 1023
44  br i1 %exitcond7, label %for.body, label %for.end13
45
46for.end13:                                        ; preds = %for.inc11
47  ret void
48}
49
50
51; CHECK-DAG: Loop 'for.body' has cost = 4186116
52; CHECK-DAG: Loop 'for.body4' has cost = 128898
53
54define void @t2([2048 x i32]* %a) {
55entry:
56  br label %for.body
57
58for.body:                                         ; preds = %entry, %for.inc11
59  %indvars.iv4 = phi i64 [ 0, %entry ], [ %indvars.iv.next5, %for.inc11 ]
60  br label %for.body4
61
62for.body4:                                        ; preds = %for.body, %for.body4
63  %indvars.iv = phi i64 [ 2, %for.body ], [ %indvars.iv.next, %for.body4 ]
64  %0 = add nuw nsw i64 %indvars.iv4, 1
65  %1 = add nsw i64 %indvars.iv, -2
66  %arrayidx6 = getelementptr inbounds [2048 x i32], [2048 x i32]* %a, i64 %0, i64 %1
67  %2 = load i32, i32* %arrayidx6, align 4
68  %call = call [2048 x i32]* @func_with_returned_arg([2048 x i32]* returned %a)
69  %arrayidx10 = getelementptr inbounds [2048 x i32], [2048 x i32]* %call, i64 %indvars.iv4, i64 %indvars.iv
70  store i32 %2, i32* %arrayidx10, align 4
71  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
72  %exitcond = icmp ne i64 %indvars.iv.next, 2048
73  br i1 %exitcond, label %for.body4, label %for.inc11
74
75for.inc11:                                        ; preds = %for.body4
76  %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 1
77  %exitcond7 = icmp ne i64 %indvars.iv.next5, 1023
78  br i1 %exitcond7, label %for.body, label %for.end13
79
80for.end13:                                        ; preds = %for.inc11
81  ret void
82}
83
84declare [2048 x i32]* @func_with_returned_arg([2048 x i32]* returned %arg)
85
86; CHECK-DAG: Loop 'for.body' has cost = 4472886244958208
87; CHECK-DAG: Loop 'for.body4' has cost = 4472886244958208
88; CHECK-DAG: Loop 'for.body8' has cost = 4472886244958208
89; CHECK-DAG: Loop 'for.body12' has cost = 4472886244958208
90; CHECK-DAG: Loop 'for.body16' has cost = 137728168833024
91
92
93;; #define N 1024
94;; #define M 2048
95;; void t3(int a[][N][N][N][M]) {
96;;   for (int i1 = 0; i1 < N-1; ++i1)
97;;     for (int i2 = 2; i2 < N; ++i2)
98;;       for (int i3 = 0; i3 < N; ++i3)
99;;         for (int i4 = 3; i4 < N; ++i4)
100;;           for (int i5 = 0; i5 < M-2; ++i5)
101;;             a[i1][i2][i3][i4][i5] = a[i1+1][i2-2][i3][i4-3][i5+2];
102;; }
103
104define void @t3([1024 x [1024 x [1024 x [2048 x i32]]]]* %a) {
105entry:
106  br label %for.body
107
108for.body:                                         ; preds = %entry, %for.inc46
109  %indvars.iv18 = phi i64 [ 0, %entry ], [ %indvars.iv.next19, %for.inc46 ]
110  br label %for.body4
111
112for.body4:                                        ; preds = %for.body, %for.inc43
113  %indvars.iv14 = phi i64 [ 2, %for.body ], [ %indvars.iv.next15, %for.inc43 ]
114  br label %for.body8
115
116for.body8:                                        ; preds = %for.body4, %for.inc40
117  %indvars.iv11 = phi i64 [ 0, %for.body4 ], [ %indvars.iv.next12, %for.inc40 ]
118  br label %for.body12
119
120for.body12:                                       ; preds = %for.body8, %for.inc37
121  %indvars.iv7 = phi i64 [ 3, %for.body8 ], [ %indvars.iv.next8, %for.inc37 ]
122  br label %for.body16
123
124for.body16:                                       ; preds = %for.body12, %for.body16
125  %indvars.iv = phi i64 [ 0, %for.body12 ], [ %indvars.iv.next, %for.body16 ]
126  %0 = add nuw nsw i64 %indvars.iv18, 1
127  %1 = add nsw i64 %indvars.iv14, -2
128  %2 = add nsw i64 %indvars.iv7, -3
129  %3 = add nuw nsw i64 %indvars.iv, 2
130  %arrayidx26 = getelementptr inbounds [1024 x [1024 x [1024 x [2048 x i32]]]], [1024 x [1024 x [1024 x [2048 x i32]]]]* %a, i64 %0, i64 %1, i64 %indvars.iv11, i64 %2, i64 %3
131  %4 = load i32, i32* %arrayidx26, align 4
132  %arrayidx36 = getelementptr inbounds [1024 x [1024 x [1024 x [2048 x i32]]]], [1024 x [1024 x [1024 x [2048 x i32]]]]* %a, i64 %indvars.iv18, i64 %indvars.iv14, i64 %indvars.iv11, i64 %indvars.iv7, i64 %indvars.iv
133  store i32 %4, i32* %arrayidx36, align 4
134  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
135  %exitcond = icmp ne i64 %indvars.iv.next, 2046
136  br i1 %exitcond, label %for.body16, label %for.inc37
137
138for.inc37:                                        ; preds = %for.body16
139  %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 1
140  %exitcond10 = icmp ne i64 %indvars.iv.next8, 1024
141  br i1 %exitcond10, label %for.body12, label %for.inc40
142
143for.inc40:                                        ; preds = %for.inc37
144  %indvars.iv.next12 = add nuw nsw i64 %indvars.iv11, 1
145  %exitcond13 = icmp ne i64 %indvars.iv.next12, 1024
146  br i1 %exitcond13, label %for.body8, label %for.inc43
147
148for.inc43:                                        ; preds = %for.inc40
149  %indvars.iv.next15 = add nuw nsw i64 %indvars.iv14, 1
150  %exitcond17 = icmp ne i64 %indvars.iv.next15, 1024
151  br i1 %exitcond17, label %for.body4, label %for.inc46
152
153for.inc46:                                        ; preds = %for.inc43
154  %indvars.iv.next19 = add nuw nsw i64 %indvars.iv18, 1
155  %exitcond21 = icmp ne i64 %indvars.iv.next19, 1023
156  br i1 %exitcond21, label %for.body, label %for.end48
157
158for.end48:                                        ; preds = %for.inc46
159  ret void
160}
161