1; Test extensions of f32 to f128.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
4
5; Check register extension.
6define void @f1(fp128 *%dst, float %val) {
7; CHECK-LABEL: f1:
8; CHECK: lxebr %f0, %f0
9; CHECK: std %f0, 0(%r2)
10; CHECK: std %f2, 8(%r2)
11; CHECK: br %r14
12  %res = fpext float %val to fp128
13  store fp128 %res, fp128 *%dst
14  ret void
15}
16
17; Check the low end of the LXEB range.
18define void @f2(fp128 *%dst, float *%ptr) {
19; CHECK-LABEL: f2:
20; CHECK: lxeb %f0, 0(%r3)
21; CHECK: std %f0, 0(%r2)
22; CHECK: std %f2, 8(%r2)
23; CHECK: br %r14
24  %val = load float , float *%ptr
25  %res = fpext float %val to fp128
26  store fp128 %res, fp128 *%dst
27  ret void
28}
29
30; Check the high end of the aligned LXEB range.
31define void @f3(fp128 *%dst, float *%base) {
32; CHECK-LABEL: f3:
33; CHECK: lxeb %f0, 4092(%r3)
34; CHECK: std %f0, 0(%r2)
35; CHECK: std %f2, 8(%r2)
36; CHECK: br %r14
37  %ptr = getelementptr float, float *%base, i64 1023
38  %val = load float , float *%ptr
39  %res = fpext float %val to fp128
40  store fp128 %res, fp128 *%dst
41  ret void
42}
43
44; Check the next word up, which needs separate address logic.
45; Other sequences besides this one would be OK.
46define void @f4(fp128 *%dst, float *%base) {
47; CHECK-LABEL: f4:
48; CHECK: aghi %r3, 4096
49; CHECK: lxeb %f0, 0(%r3)
50; CHECK: std %f0, 0(%r2)
51; CHECK: std %f2, 8(%r2)
52; CHECK: br %r14
53  %ptr = getelementptr float, float *%base, i64 1024
54  %val = load float , float *%ptr
55  %res = fpext float %val to fp128
56  store fp128 %res, fp128 *%dst
57  ret void
58}
59
60; Check negative displacements, which also need separate address logic.
61define void @f5(fp128 *%dst, float *%base) {
62; CHECK-LABEL: f5:
63; CHECK: aghi %r3, -4
64; CHECK: lxeb %f0, 0(%r3)
65; CHECK: std %f0, 0(%r2)
66; CHECK: std %f2, 8(%r2)
67; CHECK: br %r14
68  %ptr = getelementptr float, float *%base, i64 -1
69  %val = load float , float *%ptr
70  %res = fpext float %val to fp128
71  store fp128 %res, fp128 *%dst
72  ret void
73}
74
75; Check that LXEB allows indices.
76define void @f6(fp128 *%dst, float *%base, i64 %index) {
77; CHECK-LABEL: f6:
78; CHECK: sllg %r1, %r4, 2
79; CHECK: lxeb %f0, 400(%r1,%r3)
80; CHECK: std %f0, 0(%r2)
81; CHECK: std %f2, 8(%r2)
82; CHECK: br %r14
83  %ptr1 = getelementptr float, float *%base, i64 %index
84  %ptr2 = getelementptr float, float *%ptr1, i64 100
85  %val = load float , float *%ptr2
86  %res = fpext float %val to fp128
87  store fp128 %res, fp128 *%dst
88  ret void
89}
90
91; Test a case where we spill the source of at least one LXEBR.  We want
92; to use LXEB if possible.
93define void @f7(fp128 *%ptr1, float *%ptr2) {
94; CHECK-LABEL: f7:
95; CHECK: lxeb {{%f[0-9]+}}, 16{{[04]}}(%r15)
96; CHECK: br %r14
97  %val0 = load volatile float , float *%ptr2
98  %val1 = load volatile float , float *%ptr2
99  %val2 = load volatile float , float *%ptr2
100  %val3 = load volatile float , float *%ptr2
101  %val4 = load volatile float , float *%ptr2
102  %val5 = load volatile float , float *%ptr2
103  %val6 = load volatile float , float *%ptr2
104  %val7 = load volatile float , float *%ptr2
105  %val8 = load volatile float , float *%ptr2
106  %val9 = load volatile float , float *%ptr2
107  %val10 = load volatile float , float *%ptr2
108  %val11 = load volatile float , float *%ptr2
109  %val12 = load volatile float , float *%ptr2
110  %val13 = load volatile float , float *%ptr2
111  %val14 = load volatile float , float *%ptr2
112  %val15 = load volatile float , float *%ptr2
113  %val16 = load volatile float , float *%ptr2
114
115  %ext0 = fpext float %val0 to fp128
116  %ext1 = fpext float %val1 to fp128
117  %ext2 = fpext float %val2 to fp128
118  %ext3 = fpext float %val3 to fp128
119  %ext4 = fpext float %val4 to fp128
120  %ext5 = fpext float %val5 to fp128
121  %ext6 = fpext float %val6 to fp128
122  %ext7 = fpext float %val7 to fp128
123  %ext8 = fpext float %val8 to fp128
124  %ext9 = fpext float %val9 to fp128
125  %ext10 = fpext float %val10 to fp128
126  %ext11 = fpext float %val11 to fp128
127  %ext12 = fpext float %val12 to fp128
128  %ext13 = fpext float %val13 to fp128
129  %ext14 = fpext float %val14 to fp128
130  %ext15 = fpext float %val15 to fp128
131  %ext16 = fpext float %val16 to fp128
132
133  store volatile float %val0, float *%ptr2
134  store volatile float %val1, float *%ptr2
135  store volatile float %val2, float *%ptr2
136  store volatile float %val3, float *%ptr2
137  store volatile float %val4, float *%ptr2
138  store volatile float %val5, float *%ptr2
139  store volatile float %val6, float *%ptr2
140  store volatile float %val7, float *%ptr2
141  store volatile float %val8, float *%ptr2
142  store volatile float %val9, float *%ptr2
143  store volatile float %val10, float *%ptr2
144  store volatile float %val11, float *%ptr2
145  store volatile float %val12, float *%ptr2
146  store volatile float %val13, float *%ptr2
147  store volatile float %val14, float *%ptr2
148  store volatile float %val15, float *%ptr2
149  store volatile float %val16, float *%ptr2
150
151  store volatile fp128 %ext0, fp128 *%ptr1
152  store volatile fp128 %ext1, fp128 *%ptr1
153  store volatile fp128 %ext2, fp128 *%ptr1
154  store volatile fp128 %ext3, fp128 *%ptr1
155  store volatile fp128 %ext4, fp128 *%ptr1
156  store volatile fp128 %ext5, fp128 *%ptr1
157  store volatile fp128 %ext6, fp128 *%ptr1
158  store volatile fp128 %ext7, fp128 *%ptr1
159  store volatile fp128 %ext8, fp128 *%ptr1
160  store volatile fp128 %ext9, fp128 *%ptr1
161  store volatile fp128 %ext10, fp128 *%ptr1
162  store volatile fp128 %ext11, fp128 *%ptr1
163  store volatile fp128 %ext12, fp128 *%ptr1
164  store volatile fp128 %ext13, fp128 *%ptr1
165  store volatile fp128 %ext14, fp128 *%ptr1
166  store volatile fp128 %ext15, fp128 *%ptr1
167  store volatile fp128 %ext16, fp128 *%ptr1
168
169  ret void
170}
171