1; RUN: opt < %s  -opaque-pointers -cache-line-size=32 -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck -check-prefix=SMALLER-CACHELINE %s
2; RUN: opt < %s  -opaque-pointers -cache-line-size=256 -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck -check-prefix=LARGER-CACHELINE %s
3
4;; This test is similar to test/Analysis/LoopCacheAnalysis/PowerPC/compute-cost.ll,
5;; with differences that it tests the scenarios where an option for cache line size is
6;; specified with different values.
7
8; Check IndexedReference::computeRefCost can handle type differences between
9; Stride and TripCount
10
11; SMALLER-CACHELINE: Loop 'for.cond' has cost = 256
12; LARGER-CACHELINE: Loop 'for.cond' has cost = 32
13%struct._Handleitem = type { %struct._Handleitem* }
14
15define void @handle_to_ptr(%struct._Handleitem** %blocks) {
16; Preheader:
17entry:
18  br label %for.cond
19
20; Loop:
21for.cond:                                         ; preds = %for.body, %entry
22  %i.0 = phi i32 [ 1, %entry ], [ %inc, %for.body ]
23  %cmp = icmp ult i32 %i.0, 1024
24  br i1 %cmp, label %for.body, label %for.end
25
26for.body:                                         ; preds = %for.cond
27  %idxprom = zext i32 %i.0 to i64
28  %arrayidx = getelementptr inbounds %struct._Handleitem*, %struct._Handleitem** %blocks, i64 %idxprom
29  store %struct._Handleitem* null, %struct._Handleitem** %arrayidx, align 8
30  %inc = add nuw nsw i32 %i.0, 1
31  br label %for.cond
32
33; Exit blocks
34for.end:                                          ; preds = %for.cond
35  ret void
36}
37
38
39
40; Check IndexedReference::computeRefCost can handle negative stride
41
42; SMALLER-CACHELINE: Loop 'for.neg.cond' has cost = 256
43; LARGER-CACHELINE: Loop 'for.neg.cond' has cost = 32
44define void @handle_to_ptr_neg_stride(%struct._Handleitem** %blocks) {
45; Preheader:
46entry:
47  br label %for.neg.cond
48
49; Loop:
50for.neg.cond:                                         ; preds = %for.neg.body, %entry
51  %i.0 = phi i32 [ 1023, %entry ], [ %dec, %for.neg.body ]
52  %cmp = icmp sgt i32 %i.0, 0
53  br i1 %cmp, label %for.neg.body, label %for.neg.end
54
55for.neg.body:                                         ; preds = %for.neg.cond
56  %idxprom = zext i32 %i.0 to i64
57  %arrayidx = getelementptr inbounds %struct._Handleitem*, %struct._Handleitem** %blocks, i64 %idxprom
58  store %struct._Handleitem* null, %struct._Handleitem** %arrayidx, align 8
59  %dec = add nsw i32 %i.0, -1
60  br label %for.neg.cond
61
62; Exit blocks
63for.neg.end:                                          ; preds = %for.neg.cond
64  ret void
65}
66
67
68
69;   for (int i = 40960; i > 0; i--)
70;     B[i] = B[40960 - i];
71
72; FIXME: Currently negative access functions are treated the same as positive
73; access functions. When this is fixed this testcase should have a cost
74; approximately 2x higher.
75
76; SMALLER-CACHELINE: Loop 'for.cond2' has cost = 10240
77; LARGER-CACHELINE: Loop 'for.cond2' has cost = 1280
78define void @Test2(double* %B) {
79entry:
80  br label %for.cond2
81
82for.cond2:                                         ; preds = %for.body, %entry
83  %i.0 = phi i32 [ 40960, %entry ], [ %dec, %for.body ]
84  %cmp = icmp sgt i32 %i.0, 0
85  br i1 %cmp, label %for.body, label %for.end
86
87for.body:                                         ; preds = %for.cond
88  %sub = sub nsw i32 40960, %i.0
89  %idxprom = sext i32 %sub to i64
90  %arrayidx = getelementptr inbounds double, double* %B, i64 %idxprom
91  %0 = load double, double* %arrayidx, align 8
92  %idxprom1 = sext i32 %i.0 to i64
93  %arrayidx2 = getelementptr inbounds double, double* %B, i64 %idxprom1
94  store double %0, double* %arrayidx2, align 8
95  %dec = add nsw i32 %i.0, -1
96  br label %for.cond2
97
98for.end:                                          ; preds = %for.cond
99  ret void
100}
101
102
103
104;   for (i = 40960; i > 0; i--)
105;     C[i] = C[i];
106
107; SMALLER-CACHELINE: Loop 'for.cond3' has cost = 10240
108; LARGER-CACHELINE: Loop 'for.cond3' has cost = 1280
109define void @Test3(double** %C) {
110entry:
111  br label %for.cond3
112
113for.cond3:                                         ; preds = %for.body, %entry
114  %i.0 = phi i32 [ 40960, %entry ], [ %dec, %for.body ]
115  %cmp = icmp sgt i32 %i.0, 0
116  br i1 %cmp, label %for.body, label %for.end
117
118for.body:                                         ; preds = %for.cond
119  %idxprom = sext i32 %i.0 to i64
120  %arrayidx = getelementptr inbounds double*, double** %C, i64 %idxprom
121  %0 = load double*, double** %arrayidx, align 8
122  %idxprom1 = sext i32 %i.0 to i64
123  %arrayidx2 = getelementptr inbounds double*, double** %C, i64 %idxprom1
124  store double* %0, double** %arrayidx2, align 8
125  %dec = add nsw i32 %i.0, -1
126  br label %for.cond3
127
128for.end:                                          ; preds = %for.cond
129  ret void
130}
131
132
133
134;  for (i = 0; i < 40960; i++)
135;     D[i] = D[i];
136
137; SMALLER-CACHELINE: Loop 'for.cond4' has cost = 10240
138; LARGER-CACHELINE: Loop 'for.cond4' has cost = 1280
139define void @Test4(double** %D) {
140entry:
141  br label %for.cond4
142
143for.cond4:                                         ; preds = %for.body, %entry
144  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
145  %cmp = icmp slt i32 %i.0, 40960
146  br i1 %cmp, label %for.body, label %for.end
147
148for.body:                                         ; preds = %for.cond
149  %idxprom = sext i32 %i.0 to i64
150  %arrayidx = getelementptr inbounds double*, double** %D, i64 %idxprom
151  %0 = load double*, double** %arrayidx, align 8
152  %idxprom1 = sext i32 %i.0 to i64
153  %arrayidx2 = getelementptr inbounds double*, double** %D, i64 %idxprom1
154  store double* %0, double** %arrayidx2, align 8
155  %inc = add nsw i32 %i.0, 1
156  br label %for.cond4
157
158for.end:                                          ; preds = %for.cond
159  ret void
160}
161