1; Test extensions of f32 to f64.
2;
3; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
4
5; Check register extension.
6define double @f1(float %val) {
7; CHECK-LABEL: f1:
8; CHECK: ldebr %f0, %f0
9; CHECK: br %r14
10  %res = fpext float %val to double
11  ret double %res
12}
13
14; Check the low end of the LDEB range.
15define double @f2(float *%ptr) {
16; CHECK-LABEL: f2:
17; CHECK: ldeb %f0, 0(%r2)
18; CHECK: br %r14
19  %val = load float , float *%ptr
20  %res = fpext float %val to double
21  ret double %res
22}
23
24; Check the high end of the aligned LDEB range.
25define double @f3(float *%base) {
26; CHECK-LABEL: f3:
27; CHECK: ldeb %f0, 4092(%r2)
28; CHECK: br %r14
29  %ptr = getelementptr float, float *%base, i64 1023
30  %val = load float , float *%ptr
31  %res = fpext float %val to double
32  ret double %res
33}
34
35; Check the next word up, which needs separate address logic.
36; Other sequences besides this one would be OK.
37define double @f4(float *%base) {
38; CHECK-LABEL: f4:
39; CHECK: aghi %r2, 4096
40; CHECK: ldeb %f0, 0(%r2)
41; CHECK: br %r14
42  %ptr = getelementptr float, float *%base, i64 1024
43  %val = load float , float *%ptr
44  %res = fpext float %val to double
45  ret double %res
46}
47
48; Check negative displacements, which also need separate address logic.
49define double @f5(float *%base) {
50; CHECK-LABEL: f5:
51; CHECK: aghi %r2, -4
52; CHECK: ldeb %f0, 0(%r2)
53; CHECK: br %r14
54  %ptr = getelementptr float, float *%base, i64 -1
55  %val = load float , float *%ptr
56  %res = fpext float %val to double
57  ret double %res
58}
59
60; Check that LDEB allows indices.
61define double @f6(float *%base, i64 %index) {
62; CHECK-LABEL: f6:
63; CHECK: sllg %r1, %r3, 2
64; CHECK: ldeb %f0, 400(%r1,%r2)
65; CHECK: br %r14
66  %ptr1 = getelementptr float, float *%base, i64 %index
67  %ptr2 = getelementptr float, float *%ptr1, i64 100
68  %val = load float , float *%ptr2
69  %res = fpext float %val to double
70  ret double %res
71}
72
73; Test a case where we spill the source of at least one LDEBR.  We want
74; to use LDEB if possible.
75define void @f7(double *%ptr1, float *%ptr2) {
76; CHECK-LABEL: f7:
77; CHECK: ldeb {{%f[0-9]+}}, 16{{[04]}}(%r15)
78; CHECK: br %r14
79  %val0 = load volatile float , float *%ptr2
80  %val1 = load volatile float , float *%ptr2
81  %val2 = load volatile float , float *%ptr2
82  %val3 = load volatile float , float *%ptr2
83  %val4 = load volatile float , float *%ptr2
84  %val5 = load volatile float , float *%ptr2
85  %val6 = load volatile float , float *%ptr2
86  %val7 = load volatile float , float *%ptr2
87  %val8 = load volatile float , float *%ptr2
88  %val9 = load volatile float , float *%ptr2
89  %val10 = load volatile float , float *%ptr2
90  %val11 = load volatile float , float *%ptr2
91  %val12 = load volatile float , float *%ptr2
92  %val13 = load volatile float , float *%ptr2
93  %val14 = load volatile float , float *%ptr2
94  %val15 = load volatile float , float *%ptr2
95  %val16 = load volatile float , float *%ptr2
96
97  %ext0 = fpext float %val0 to double
98  %ext1 = fpext float %val1 to double
99  %ext2 = fpext float %val2 to double
100  %ext3 = fpext float %val3 to double
101  %ext4 = fpext float %val4 to double
102  %ext5 = fpext float %val5 to double
103  %ext6 = fpext float %val6 to double
104  %ext7 = fpext float %val7 to double
105  %ext8 = fpext float %val8 to double
106  %ext9 = fpext float %val9 to double
107  %ext10 = fpext float %val10 to double
108  %ext11 = fpext float %val11 to double
109  %ext12 = fpext float %val12 to double
110  %ext13 = fpext float %val13 to double
111  %ext14 = fpext float %val14 to double
112  %ext15 = fpext float %val15 to double
113  %ext16 = fpext float %val16 to double
114
115  store volatile float %val0, float *%ptr2
116  store volatile float %val1, float *%ptr2
117  store volatile float %val2, float *%ptr2
118  store volatile float %val3, float *%ptr2
119  store volatile float %val4, float *%ptr2
120  store volatile float %val5, float *%ptr2
121  store volatile float %val6, float *%ptr2
122  store volatile float %val7, float *%ptr2
123  store volatile float %val8, float *%ptr2
124  store volatile float %val9, float *%ptr2
125  store volatile float %val10, float *%ptr2
126  store volatile float %val11, float *%ptr2
127  store volatile float %val12, float *%ptr2
128  store volatile float %val13, float *%ptr2
129  store volatile float %val14, float *%ptr2
130  store volatile float %val15, float *%ptr2
131  store volatile float %val16, float *%ptr2
132
133  store volatile double %ext0, double *%ptr1
134  store volatile double %ext1, double *%ptr1
135  store volatile double %ext2, double *%ptr1
136  store volatile double %ext3, double *%ptr1
137  store volatile double %ext4, double *%ptr1
138  store volatile double %ext5, double *%ptr1
139  store volatile double %ext6, double *%ptr1
140  store volatile double %ext7, double *%ptr1
141  store volatile double %ext8, double *%ptr1
142  store volatile double %ext9, double *%ptr1
143  store volatile double %ext10, double *%ptr1
144  store volatile double %ext11, double *%ptr1
145  store volatile double %ext12, double *%ptr1
146  store volatile double %ext13, double *%ptr1
147  store volatile double %ext14, double *%ptr1
148  store volatile double %ext15, double *%ptr1
149  store volatile double %ext16, double *%ptr1
150
151  ret void
152}
153