1; Test 32-bit multiplication in which the second operand is variable.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
4
5declare i32 @foo()
6
7; Check MSR.
8define i32 @f1(i32 %a, i32 %b) {
9; CHECK-LABEL: f1:
10; CHECK: msr %r2, %r3
11; CHECK: br %r14
12  %mul = mul i32 %a, %b
13  ret i32 %mul
14}
15
16; Check the low end of the MS range.
17define i32 @f2(i32 %a, i32 *%src) {
18; CHECK-LABEL: f2:
19; CHECK: ms %r2, 0(%r3)
20; CHECK: br %r14
21  %b = load i32 , i32 *%src
22  %mul = mul i32 %a, %b
23  ret i32 %mul
24}
25
26; Check the high end of the aligned MS range.
27define i32 @f3(i32 %a, i32 *%src) {
28; CHECK-LABEL: f3:
29; CHECK: ms %r2, 4092(%r3)
30; CHECK: br %r14
31  %ptr = getelementptr i32, i32 *%src, i64 1023
32  %b = load i32 , i32 *%ptr
33  %mul = mul i32 %a, %b
34  ret i32 %mul
35}
36
37; Check the next word up, which should use MSY instead of MS.
38define i32 @f4(i32 %a, i32 *%src) {
39; CHECK-LABEL: f4:
40; CHECK: msy %r2, 4096(%r3)
41; CHECK: br %r14
42  %ptr = getelementptr i32, i32 *%src, i64 1024
43  %b = load i32 , i32 *%ptr
44  %mul = mul i32 %a, %b
45  ret i32 %mul
46}
47
48; Check the high end of the aligned MSY range.
49define i32 @f5(i32 %a, i32 *%src) {
50; CHECK-LABEL: f5:
51; CHECK: msy %r2, 524284(%r3)
52; CHECK: br %r14
53  %ptr = getelementptr i32, i32 *%src, i64 131071
54  %b = load i32 , i32 *%ptr
55  %mul = mul i32 %a, %b
56  ret i32 %mul
57}
58
59; Check the next word up, which needs separate address logic.
60; Other sequences besides this one would be OK.
61define i32 @f6(i32 %a, i32 *%src) {
62; CHECK-LABEL: f6:
63; CHECK: agfi %r3, 524288
64; CHECK: ms %r2, 0(%r3)
65; CHECK: br %r14
66  %ptr = getelementptr i32, i32 *%src, i64 131072
67  %b = load i32 , i32 *%ptr
68  %mul = mul i32 %a, %b
69  ret i32 %mul
70}
71
72; Check the high end of the negative aligned MSY range.
73define i32 @f7(i32 %a, i32 *%src) {
74; CHECK-LABEL: f7:
75; CHECK: msy %r2, -4(%r3)
76; CHECK: br %r14
77  %ptr = getelementptr i32, i32 *%src, i64 -1
78  %b = load i32 , i32 *%ptr
79  %mul = mul i32 %a, %b
80  ret i32 %mul
81}
82
83; Check the low end of the MSY range.
84define i32 @f8(i32 %a, i32 *%src) {
85; CHECK-LABEL: f8:
86; CHECK: msy %r2, -524288(%r3)
87; CHECK: br %r14
88  %ptr = getelementptr i32, i32 *%src, i64 -131072
89  %b = load i32 , i32 *%ptr
90  %mul = mul i32 %a, %b
91  ret i32 %mul
92}
93
94; Check the next word down, which needs separate address logic.
95; Other sequences besides this one would be OK.
96define i32 @f9(i32 %a, i32 *%src) {
97; CHECK-LABEL: f9:
98; CHECK: agfi %r3, -524292
99; CHECK: ms %r2, 0(%r3)
100; CHECK: br %r14
101  %ptr = getelementptr i32, i32 *%src, i64 -131073
102  %b = load i32 , i32 *%ptr
103  %mul = mul i32 %a, %b
104  ret i32 %mul
105}
106
107; Check that MS allows an index.
108define i32 @f10(i32 %a, i64 %src, i64 %index) {
109; CHECK-LABEL: f10:
110; CHECK: ms %r2, 4092({{%r4,%r3|%r3,%r4}})
111; CHECK: br %r14
112  %add1 = add i64 %src, %index
113  %add2 = add i64 %add1, 4092
114  %ptr = inttoptr i64 %add2 to i32 *
115  %b = load i32 , i32 *%ptr
116  %mul = mul i32 %a, %b
117  ret i32 %mul
118}
119
120; Check that MSY allows an index.
121define i32 @f11(i32 %a, i64 %src, i64 %index) {
122; CHECK-LABEL: f11:
123; CHECK: msy %r2, 4096({{%r4,%r3|%r3,%r4}})
124; CHECK: br %r14
125  %add1 = add i64 %src, %index
126  %add2 = add i64 %add1, 4096
127  %ptr = inttoptr i64 %add2 to i32 *
128  %b = load i32 , i32 *%ptr
129  %mul = mul i32 %a, %b
130  ret i32 %mul
131}
132
133; Check that multiplications of spilled values can use MS rather than MSR.
134define i32 @f12(i32 *%ptr0) {
135; CHECK-LABEL: f12:
136; CHECK: brasl %r14, foo@PLT
137; CHECK: ms %r2, 16{{[04]}}(%r15)
138; CHECK: br %r14
139  %ptr1 = getelementptr i32, i32 *%ptr0, i64 2
140  %ptr2 = getelementptr i32, i32 *%ptr0, i64 4
141  %ptr3 = getelementptr i32, i32 *%ptr0, i64 6
142  %ptr4 = getelementptr i32, i32 *%ptr0, i64 8
143  %ptr5 = getelementptr i32, i32 *%ptr0, i64 10
144  %ptr6 = getelementptr i32, i32 *%ptr0, i64 12
145  %ptr7 = getelementptr i32, i32 *%ptr0, i64 14
146  %ptr8 = getelementptr i32, i32 *%ptr0, i64 16
147  %ptr9 = getelementptr i32, i32 *%ptr0, i64 18
148
149  %val0 = load i32 , i32 *%ptr0
150  %val1 = load i32 , i32 *%ptr1
151  %val2 = load i32 , i32 *%ptr2
152  %val3 = load i32 , i32 *%ptr3
153  %val4 = load i32 , i32 *%ptr4
154  %val5 = load i32 , i32 *%ptr5
155  %val6 = load i32 , i32 *%ptr6
156  %val7 = load i32 , i32 *%ptr7
157  %val8 = load i32 , i32 *%ptr8
158  %val9 = load i32 , i32 *%ptr9
159
160  %ret = call i32 @foo()
161
162  %mul0 = mul i32 %ret, %val0
163  %mul1 = mul i32 %mul0, %val1
164  %mul2 = mul i32 %mul1, %val2
165  %mul3 = mul i32 %mul2, %val3
166  %mul4 = mul i32 %mul3, %val4
167  %mul5 = mul i32 %mul4, %val5
168  %mul6 = mul i32 %mul5, %val6
169  %mul7 = mul i32 %mul6, %val7
170  %mul8 = mul i32 %mul7, %val8
171  %mul9 = mul i32 %mul8, %val9
172
173  ret i32 %mul9
174}
175