1; Test 32-bit unsigned division and remainder.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
4
5declare i32 @foo()
6
7; Test register division.  The result is in the second of the two registers.
8define void @f1(i32 %dummy, i32 %a, i32 %b, i32 *%dest) {
9; CHECK-LABEL: f1:
10; CHECK-NOT: %r3
11; CHECK: {{llill|lhi}} %r2, 0
12; CHECK-NOT: %r3
13; CHECK: dlr %r2, %r4
14; CHECK: st %r3, 0(%r5)
15; CHECK: br %r14
16  %div = udiv i32 %a, %b
17  store i32 %div, i32 *%dest
18  ret void
19}
20
21; Test register remainder.  The result is in the first of the two registers.
22define void @f2(i32 %dummy, i32 %a, i32 %b, i32 *%dest) {
23; CHECK-LABEL: f2:
24; CHECK-NOT: %r3
25; CHECK: {{llill|lhi}} %r2, 0
26; CHECK-NOT: %r3
27; CHECK: dlr %r2, %r4
28; CHECK: st %r2, 0(%r5)
29; CHECK: br %r14
30  %rem = urem i32 %a, %b
31  store i32 %rem, i32 *%dest
32  ret void
33}
34
35; Test that division and remainder use a single instruction.
36define i32 @f3(i32 %dummy1, i32 %a, i32 %b) {
37; CHECK-LABEL: f3:
38; CHECK-NOT: %r3
39; CHECK: {{llill|lhi}} %r2, 0
40; CHECK-NOT: %r3
41; CHECK: dlr %r2, %r4
42; CHECK-NOT: dlr
43; CHECK: or %r2, %r3
44; CHECK: br %r14
45  %div = udiv i32 %a, %b
46  %rem = urem i32 %a, %b
47  %or = or i32 %rem, %div
48  ret i32 %or
49}
50
51; Test memory division with no displacement.
52define void @f4(i32 %dummy, i32 %a, i32 *%src, i32 *%dest) {
53; CHECK-LABEL: f4:
54; CHECK-NOT: %r3
55; CHECK: {{llill|lhi}} %r2, 0
56; CHECK-NOT: %r3
57; CHECK: dl %r2, 0(%r4)
58; CHECK: st %r3, 0(%r5)
59; CHECK: br %r14
60  %b = load i32 , i32 *%src
61  %div = udiv i32 %a, %b
62  store i32 %div, i32 *%dest
63  ret void
64}
65
66; Test memory remainder with no displacement.
67define void @f5(i32 %dummy, i32 %a, i32 *%src, i32 *%dest) {
68; CHECK-LABEL: f5:
69; CHECK-NOT: %r3
70; CHECK: {{llill|lhi}} %r2, 0
71; CHECK-NOT: %r3
72; CHECK: dl %r2, 0(%r4)
73; CHECK: st %r2, 0(%r5)
74; CHECK: br %r14
75  %b = load i32 , i32 *%src
76  %rem = urem i32 %a, %b
77  store i32 %rem, i32 *%dest
78  ret void
79}
80
81; Test both memory division and memory remainder.
82define i32 @f6(i32 %dummy, i32 %a, i32 *%src) {
83; CHECK-LABEL: f6:
84; CHECK-NOT: %r3
85; CHECK: {{llill|lhi}} %r2, 0
86; CHECK-NOT: %r3
87; CHECK: dl %r2, 0(%r4)
88; CHECK-NOT: {{dl|dlr}}
89; CHECK: or %r2, %r3
90; CHECK: br %r14
91  %b = load i32 , i32 *%src
92  %div = udiv i32 %a, %b
93  %rem = urem i32 %a, %b
94  %or = or i32 %rem, %div
95  ret i32 %or
96}
97
98; Check the high end of the DL range.
99define i32 @f7(i32 %dummy, i32 %a, i32 *%src) {
100; CHECK-LABEL: f7:
101; CHECK: dl %r2, 524284(%r4)
102; CHECK: br %r14
103  %ptr = getelementptr i32, i32 *%src, i64 131071
104  %b = load i32 , i32 *%ptr
105  %rem = urem i32 %a, %b
106  ret i32 %rem
107}
108
109; Check the next word up, which needs separate address logic.
110; Other sequences besides this one would be OK.
111define i32 @f8(i32 %dummy, i32 %a, i32 *%src) {
112; CHECK-LABEL: f8:
113; CHECK: agfi %r4, 524288
114; CHECK: dl %r2, 0(%r4)
115; CHECK: br %r14
116  %ptr = getelementptr i32, i32 *%src, i64 131072
117  %b = load i32 , i32 *%ptr
118  %rem = urem i32 %a, %b
119  ret i32 %rem
120}
121
122; Check the high end of the negative aligned DL range.
123define i32 @f9(i32 %dummy, i32 %a, i32 *%src) {
124; CHECK-LABEL: f9:
125; CHECK: dl %r2, -4(%r4)
126; CHECK: br %r14
127  %ptr = getelementptr i32, i32 *%src, i64 -1
128  %b = load i32 , i32 *%ptr
129  %rem = urem i32 %a, %b
130  ret i32 %rem
131}
132
133; Check the low end of the DL range.
134define i32 @f10(i32 %dummy, i32 %a, i32 *%src) {
135; CHECK-LABEL: f10:
136; CHECK: dl %r2, -524288(%r4)
137; CHECK: br %r14
138  %ptr = getelementptr i32, i32 *%src, i64 -131072
139  %b = load i32 , i32 *%ptr
140  %rem = urem i32 %a, %b
141  ret i32 %rem
142}
143
144; Check the next word down, which needs separate address logic.
145; Other sequences besides this one would be OK.
146define i32 @f11(i32 %dummy, i32 %a, i32 *%src) {
147; CHECK-LABEL: f11:
148; CHECK: agfi %r4, -524292
149; CHECK: dl %r2, 0(%r4)
150; CHECK: br %r14
151  %ptr = getelementptr i32, i32 *%src, i64 -131073
152  %b = load i32 , i32 *%ptr
153  %rem = urem i32 %a, %b
154  ret i32 %rem
155}
156
157; Check that DL allows an index.
158define i32 @f12(i32 %dummy, i32 %a, i64 %src, i64 %index) {
159; CHECK-LABEL: f12:
160; CHECK: dl %r2, 524287(%r5,%r4)
161; CHECK: br %r14
162  %add1 = add i64 %src, %index
163  %add2 = add i64 %add1, 524287
164  %ptr = inttoptr i64 %add2 to i32 *
165  %b = load i32 , i32 *%ptr
166  %rem = urem i32 %a, %b
167  ret i32 %rem
168}
169
170; Check that divisions of spilled values can use DL rather than DLR.
171define i32 @f13(i32 *%ptr0) {
172; CHECK-LABEL: f13:
173; CHECK: brasl %r14, foo@PLT
174; CHECK: dl {{%r[0-9]+}}, 16{{[04]}}(%r15)
175; CHECK: br %r14
176  %ptr1 = getelementptr i32, i32 *%ptr0, i64 2
177  %ptr2 = getelementptr i32, i32 *%ptr0, i64 4
178  %ptr3 = getelementptr i32, i32 *%ptr0, i64 6
179  %ptr4 = getelementptr i32, i32 *%ptr0, i64 8
180  %ptr5 = getelementptr i32, i32 *%ptr0, i64 10
181  %ptr6 = getelementptr i32, i32 *%ptr0, i64 12
182  %ptr7 = getelementptr i32, i32 *%ptr0, i64 14
183  %ptr8 = getelementptr i32, i32 *%ptr0, i64 16
184  %ptr9 = getelementptr i32, i32 *%ptr0, i64 18
185
186  %val0 = load i32 , i32 *%ptr0
187  %val1 = load i32 , i32 *%ptr1
188  %val2 = load i32 , i32 *%ptr2
189  %val3 = load i32 , i32 *%ptr3
190  %val4 = load i32 , i32 *%ptr4
191  %val5 = load i32 , i32 *%ptr5
192  %val6 = load i32 , i32 *%ptr6
193  %val7 = load i32 , i32 *%ptr7
194  %val8 = load i32 , i32 *%ptr8
195  %val9 = load i32 , i32 *%ptr9
196
197  %ret = call i32 @foo()
198
199  %div0 = udiv i32 %ret, %val0
200  %div1 = udiv i32 %div0, %val1
201  %div2 = udiv i32 %div1, %val2
202  %div3 = udiv i32 %div2, %val3
203  %div4 = udiv i32 %div3, %val4
204  %div5 = udiv i32 %div4, %val5
205  %div6 = udiv i32 %div5, %val6
206  %div7 = udiv i32 %div6, %val7
207  %div8 = udiv i32 %div7, %val8
208  %div9 = udiv i32 %div8, %val9
209
210  ret i32 %div9
211}
212