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; Check IndexedReference::computeRefCost can handle type differences between
39; Coeff and ElemSize.
40
41; SMALLER-CACHELINE: Loop 'for.cond' has cost = 100000000
42; SMALLER-CACHELINE: Loop 'for.cond1' has cost = 1000000
43; SMALLER-CACHELINE: Loop 'for.cond5' has cost = 120000
44; LARGER-CACHELINE: Loop 'for.cond' has cost = 100000000
45; LARGER-CACHELINE: Loop 'for.cond1' has cost = 1000000
46; LARGER-CACHELINE: Loop 'for.cond5' has cost = 10000
47@data = external dso_local global [2 x [4 x [18 x i32]]], align 1
48
49define dso_local void @handle_to_ptr_2(i1 %b0, i1 %b1, i1 %b2) {
50entry:
51  br label %for.cond
52
53for.cond:
54  %i.0 = phi i16 [ 0, %entry ], [ %inc18, %for.inc17 ]
55  %idxprom = zext i16 %i.0 to i32
56  br i1 %b2, label %for.end19, label %for.cond1
57
58for.cond1:
59  %j.0 = phi i16 [ %inc15, %for.inc14 ], [ 0, %for.cond ]
60  br i1 %b1, label %for.inc17, label %for.cond5.preheader
61
62for.cond5.preheader:
63  %idxprom10 = zext i16 %j.0 to i32
64  br label %for.cond5
65
66for.cond5:
67  %k.0 = phi i16 [ %inc, %for.inc ], [ 0, %for.cond5.preheader ]
68  br i1 %b0, label %for.inc14, label %for.inc
69
70for.inc:
71  %idxprom12 = zext i16 %k.0 to i32
72  %arrayidx13 = getelementptr inbounds [2 x [4 x [18 x i32]]], ptr @data, i32 0, i32 %idxprom, i32 %idxprom10, i32 %idxprom12
73  store i32 7, ptr %arrayidx13, align 1
74  %inc = add nuw nsw i16 %k.0, 1
75  br label %for.cond5
76
77for.inc14:
78  %inc15 = add nuw nsw i16 %j.0, 1
79  br label %for.cond1
80
81for.inc17:
82  %inc18 = add nuw nsw i16 %i.0, 1
83  br label %for.cond
84
85for.end19:
86  ret void
87}
88
89; Check IndexedReference::computeRefCost can handle negative stride
90
91; SMALLER-CACHELINE: Loop 'for.neg.cond' has cost = 256
92; LARGER-CACHELINE: Loop 'for.neg.cond' has cost = 32
93define void @handle_to_ptr_neg_stride(%struct._Handleitem** %blocks) {
94; Preheader:
95entry:
96  br label %for.neg.cond
97
98; Loop:
99for.neg.cond:                                         ; preds = %for.neg.body, %entry
100  %i.0 = phi i32 [ 1023, %entry ], [ %dec, %for.neg.body ]
101  %cmp = icmp sgt i32 %i.0, 0
102  br i1 %cmp, label %for.neg.body, label %for.neg.end
103
104for.neg.body:                                         ; preds = %for.neg.cond
105  %idxprom = zext i32 %i.0 to i64
106  %arrayidx = getelementptr inbounds %struct._Handleitem*, %struct._Handleitem** %blocks, i64 %idxprom
107  store %struct._Handleitem* null, %struct._Handleitem** %arrayidx, align 8
108  %dec = add nsw i32 %i.0, -1
109  br label %for.neg.cond
110
111; Exit blocks
112for.neg.end:                                          ; preds = %for.neg.cond
113  ret void
114}
115
116
117
118;   for (int i = 40960; i > 0; i--)
119;     B[i] = B[40960 - i];
120
121; FIXME: Currently negative access functions are treated the same as positive
122; access functions. When this is fixed this testcase should have a cost
123; approximately 2x higher.
124
125; SMALLER-CACHELINE: Loop 'for.cond2' has cost = 10240
126; LARGER-CACHELINE: Loop 'for.cond2' has cost = 1280
127define void @Test2(double* %B) {
128entry:
129  br label %for.cond2
130
131for.cond2:                                         ; preds = %for.body, %entry
132  %i.0 = phi i32 [ 40960, %entry ], [ %dec, %for.body ]
133  %cmp = icmp sgt i32 %i.0, 0
134  br i1 %cmp, label %for.body, label %for.end
135
136for.body:                                         ; preds = %for.cond
137  %sub = sub nsw i32 40960, %i.0
138  %idxprom = sext i32 %sub to i64
139  %arrayidx = getelementptr inbounds double, double* %B, i64 %idxprom
140  %0 = load double, double* %arrayidx, align 8
141  %idxprom1 = sext i32 %i.0 to i64
142  %arrayidx2 = getelementptr inbounds double, double* %B, i64 %idxprom1
143  store double %0, double* %arrayidx2, align 8
144  %dec = add nsw i32 %i.0, -1
145  br label %for.cond2
146
147for.end:                                          ; preds = %for.cond
148  ret void
149}
150
151
152
153;   for (i = 40960; i > 0; i--)
154;     C[i] = C[i];
155
156; SMALLER-CACHELINE: Loop 'for.cond3' has cost = 10240
157; LARGER-CACHELINE: Loop 'for.cond3' has cost = 1280
158define void @Test3(double** %C) {
159entry:
160  br label %for.cond3
161
162for.cond3:                                         ; preds = %for.body, %entry
163  %i.0 = phi i32 [ 40960, %entry ], [ %dec, %for.body ]
164  %cmp = icmp sgt i32 %i.0, 0
165  br i1 %cmp, label %for.body, label %for.end
166
167for.body:                                         ; preds = %for.cond
168  %idxprom = sext i32 %i.0 to i64
169  %arrayidx = getelementptr inbounds double*, double** %C, i64 %idxprom
170  %0 = load double*, double** %arrayidx, align 8
171  %idxprom1 = sext i32 %i.0 to i64
172  %arrayidx2 = getelementptr inbounds double*, double** %C, i64 %idxprom1
173  store double* %0, double** %arrayidx2, align 8
174  %dec = add nsw i32 %i.0, -1
175  br label %for.cond3
176
177for.end:                                          ; preds = %for.cond
178  ret void
179}
180
181
182
183;  for (i = 0; i < 40960; i++)
184;     D[i] = D[i];
185
186; SMALLER-CACHELINE: Loop 'for.cond4' has cost = 10240
187; LARGER-CACHELINE: Loop 'for.cond4' has cost = 1280
188define void @Test4(double** %D) {
189entry:
190  br label %for.cond4
191
192for.cond4:                                         ; preds = %for.body, %entry
193  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
194  %cmp = icmp slt i32 %i.0, 40960
195  br i1 %cmp, label %for.body, label %for.end
196
197for.body:                                         ; preds = %for.cond
198  %idxprom = sext i32 %i.0 to i64
199  %arrayidx = getelementptr inbounds double*, double** %D, i64 %idxprom
200  %0 = load double*, double** %arrayidx, align 8
201  %idxprom1 = sext i32 %i.0 to i64
202  %arrayidx2 = getelementptr inbounds double*, double** %D, i64 %idxprom1
203  store double* %0, double** %arrayidx2, align 8
204  %inc = add nsw i32 %i.0, 1
205  br label %for.cond4
206
207for.end:                                          ; preds = %for.cond
208  ret void
209}
210