1; Test 32-bit subtraction.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s
4; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z196 | FileCheck %s
5
6declare i32 @foo()
7
8; Check SR.
9define i32 @f1(i32 %a, i32 %b) {
10; CHECK-LABEL: f1:
11; CHECK: sr %r2, %r3
12; CHECK: br %r14
13  %sub = sub i32 %a, %b
14  ret i32 %sub
15}
16
17; Check the low end of the S range.
18define i32 @f2(i32 %a, i32 *%src) {
19; CHECK-LABEL: f2:
20; CHECK: s %r2, 0(%r3)
21; CHECK: br %r14
22  %b = load i32, i32 *%src
23  %sub = sub i32 %a, %b
24  ret i32 %sub
25}
26
27; Check the high end of the aligned S range.
28define i32 @f3(i32 %a, i32 *%src) {
29; CHECK-LABEL: f3:
30; CHECK: s %r2, 4092(%r3)
31; CHECK: br %r14
32  %ptr = getelementptr i32, i32 *%src, i64 1023
33  %b = load i32, i32 *%ptr
34  %sub = sub i32 %a, %b
35  ret i32 %sub
36}
37
38; Check the next word up, which should use SY instead of S.
39define i32 @f4(i32 %a, i32 *%src) {
40; CHECK-LABEL: f4:
41; CHECK: sy %r2, 4096(%r3)
42; CHECK: br %r14
43  %ptr = getelementptr i32, i32 *%src, i64 1024
44  %b = load i32, i32 *%ptr
45  %sub = sub i32 %a, %b
46  ret i32 %sub
47}
48
49; Check the high end of the aligned SY range.
50define i32 @f5(i32 %a, i32 *%src) {
51; CHECK-LABEL: f5:
52; CHECK: sy %r2, 524284(%r3)
53; CHECK: br %r14
54  %ptr = getelementptr i32, i32 *%src, i64 131071
55  %b = load i32, i32 *%ptr
56  %sub = sub i32 %a, %b
57  ret i32 %sub
58}
59
60; Check the next word up, which needs separate address logic.
61; Other sequences besides this one would be OK.
62define i32 @f6(i32 %a, i32 *%src) {
63; CHECK-LABEL: f6:
64; CHECK: agfi %r3, 524288
65; CHECK: s %r2, 0(%r3)
66; CHECK: br %r14
67  %ptr = getelementptr i32, i32 *%src, i64 131072
68  %b = load i32, i32 *%ptr
69  %sub = sub i32 %a, %b
70  ret i32 %sub
71}
72
73; Check the high end of the negative aligned SY range.
74define i32 @f7(i32 %a, i32 *%src) {
75; CHECK-LABEL: f7:
76; CHECK: sy %r2, -4(%r3)
77; CHECK: br %r14
78  %ptr = getelementptr i32, i32 *%src, i64 -1
79  %b = load i32, i32 *%ptr
80  %sub = sub i32 %a, %b
81  ret i32 %sub
82}
83
84; Check the low end of the SY range.
85define i32 @f8(i32 %a, i32 *%src) {
86; CHECK-LABEL: f8:
87; CHECK: sy %r2, -524288(%r3)
88; CHECK: br %r14
89  %ptr = getelementptr i32, i32 *%src, i64 -131072
90  %b = load i32, i32 *%ptr
91  %sub = sub i32 %a, %b
92  ret i32 %sub
93}
94
95; Check the next word down, which needs separate address logic.
96; Other sequences besides this one would be OK.
97define i32 @f9(i32 %a, i32 *%src) {
98; CHECK-LABEL: f9:
99; CHECK: agfi %r3, -524292
100; CHECK: s %r2, 0(%r3)
101; CHECK: br %r14
102  %ptr = getelementptr i32, i32 *%src, i64 -131073
103  %b = load i32, i32 *%ptr
104  %sub = sub i32 %a, %b
105  ret i32 %sub
106}
107
108; Check that S allows an index.
109define i32 @f10(i32 %a, i64 %src, i64 %index) {
110; CHECK-LABEL: f10:
111; CHECK: s %r2, 4092({{%r4,%r3|%r3,%r4}})
112; CHECK: br %r14
113  %add1 = add i64 %src, %index
114  %add2 = add i64 %add1, 4092
115  %ptr = inttoptr i64 %add2 to i32 *
116  %b = load i32, i32 *%ptr
117  %sub = sub i32 %a, %b
118  ret i32 %sub
119}
120
121; Check that SY allows an index.
122define i32 @f11(i32 %a, i64 %src, i64 %index) {
123; CHECK-LABEL: f11:
124; CHECK: sy %r2, 4096({{%r4,%r3|%r3,%r4}})
125; CHECK: br %r14
126  %add1 = add i64 %src, %index
127  %add2 = add i64 %add1, 4096
128  %ptr = inttoptr i64 %add2 to i32 *
129  %b = load i32, i32 *%ptr
130  %sub = sub i32 %a, %b
131  ret i32 %sub
132}
133
134; Check that subtractions of spilled values can use S rather than SR.
135define i32 @f12(i32 *%ptr0) {
136; CHECK-LABEL: f12:
137; CHECK: brasl %r14, foo@PLT
138; CHECK: s %r2, 16{{[04]}}(%r15)
139; CHECK: br %r14
140  %ptr1 = getelementptr i32, i32 *%ptr0, i64 2
141  %ptr2 = getelementptr i32, i32 *%ptr0, i64 4
142  %ptr3 = getelementptr i32, i32 *%ptr0, i64 6
143  %ptr4 = getelementptr i32, i32 *%ptr0, i64 8
144  %ptr5 = getelementptr i32, i32 *%ptr0, i64 10
145  %ptr6 = getelementptr i32, i32 *%ptr0, i64 12
146  %ptr7 = getelementptr i32, i32 *%ptr0, i64 14
147  %ptr8 = getelementptr i32, i32 *%ptr0, i64 16
148  %ptr9 = getelementptr i32, i32 *%ptr0, i64 18
149
150  %val0 = load i32, i32 *%ptr0
151  %val1 = load i32, i32 *%ptr1
152  %val2 = load i32, i32 *%ptr2
153  %val3 = load i32, i32 *%ptr3
154  %val4 = load i32, i32 *%ptr4
155  %val5 = load i32, i32 *%ptr5
156  %val6 = load i32, i32 *%ptr6
157  %val7 = load i32, i32 *%ptr7
158  %val8 = load i32, i32 *%ptr8
159  %val9 = load i32, i32 *%ptr9
160
161  %ret = call i32 @foo()
162
163  %sub0 = sub i32 %ret, %val0
164  %sub1 = sub i32 %sub0, %val1
165  %sub2 = sub i32 %sub1, %val2
166  %sub3 = sub i32 %sub2, %val3
167  %sub4 = sub i32 %sub3, %val4
168  %sub5 = sub i32 %sub4, %val5
169  %sub6 = sub i32 %sub5, %val6
170  %sub7 = sub i32 %sub6, %val7
171  %sub8 = sub i32 %sub7, %val8
172  %sub9 = sub i32 %sub8, %val9
173
174  ret i32 %sub9
175}
176