1 ; Test 64-bit floating-point comparison. The tests assume a z10 implementation
2 ; of select, using conditional branches rather than LOCGR.
4 ; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 \
5 ; RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-SCALAR %s
6 ; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 -verify-machineinstrs\
7 ; RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-VECTOR %s
11 ; Check comparison with registers.
12 define i64 @f1(i64 %a, i64 %b, double %f1, double %f2) {
14 ; CHECK: cdbr %f0, %f2
15 ; CHECK-SCALAR-NEXT: ber %r14
16 ; CHECK-SCALAR: lgr %r2, %r3
17 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
19 %cond = fcmp oeq double %f1, %f2
20 %res = select i1 %cond, i64 %a, i64 %b
24 ; Check the low end of the CDB range.
25 define i64 @f2(i64 %a, i64 %b, double %f1, double *%ptr) {
27 ; CHECK: cdb %f0, 0(%r4)
28 ; CHECK-SCALAR-NEXT: ber %r14
29 ; CHECK-SCALAR: lgr %r2, %r3
30 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
32 %f2 = load double, double *%ptr
33 %cond = fcmp oeq double %f1, %f2
34 %res = select i1 %cond, i64 %a, i64 %b
38 ; Check the high end of the aligned CDB range.
39 define i64 @f3(i64 %a, i64 %b, double %f1, double *%base) {
41 ; CHECK: cdb %f0, 4088(%r4)
42 ; CHECK-SCALAR-NEXT: ber %r14
43 ; CHECK-SCALAR: lgr %r2, %r3
44 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
46 %ptr = getelementptr double, double *%base, i64 511
47 %f2 = load double, double *%ptr
48 %cond = fcmp oeq double %f1, %f2
49 %res = select i1 %cond, i64 %a, i64 %b
53 ; Check the next doubleword up, which needs separate address logic.
54 ; Other sequences besides this one would be OK.
55 define i64 @f4(i64 %a, i64 %b, double %f1, double *%base) {
57 ; CHECK: aghi %r4, 4096
58 ; CHECK: cdb %f0, 0(%r4)
59 ; CHECK-SCALAR-NEXT: ber %r14
60 ; CHECK-SCALAR: lgr %r2, %r3
61 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
63 %ptr = getelementptr double, double *%base, i64 512
64 %f2 = load double, double *%ptr
65 %cond = fcmp oeq double %f1, %f2
66 %res = select i1 %cond, i64 %a, i64 %b
70 ; Check negative displacements, which also need separate address logic.
71 define i64 @f5(i64 %a, i64 %b, double %f1, double *%base) {
74 ; CHECK: cdb %f0, 0(%r4)
75 ; CHECK-SCALAR-NEXT: ber %r14
76 ; CHECK-SCALAR: lgr %r2, %r3
77 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
79 %ptr = getelementptr double, double *%base, i64 -1
80 %f2 = load double, double *%ptr
81 %cond = fcmp oeq double %f1, %f2
82 %res = select i1 %cond, i64 %a, i64 %b
86 ; Check that CDB allows indices.
87 define i64 @f6(i64 %a, i64 %b, double %f1, double *%base, i64 %index) {
89 ; CHECK: sllg %r1, %r5, 3
90 ; CHECK: cdb %f0, 800(%r1,%r4)
91 ; CHECK-SCALAR-NEXT: ber %r14
92 ; CHECK-SCALAR: lgr %r2, %r3
93 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
95 %ptr1 = getelementptr double, double *%base, i64 %index
96 %ptr2 = getelementptr double, double *%ptr1, i64 100
97 %f2 = load double, double *%ptr2
98 %cond = fcmp oeq double %f1, %f2
99 %res = select i1 %cond, i64 %a, i64 %b
103 ; Check that comparisons of spilled values can use CDB rather than CDBR.
104 define double @f7(double *%ptr0) {
106 ; CHECK: brasl %r14, foo@PLT
107 ; CHECK-SCALAR: cdb {{%f[0-9]+}}, 160(%r15)
109 %ptr1 = getelementptr double, double *%ptr0, i64 2
110 %ptr2 = getelementptr double, double *%ptr0, i64 4
111 %ptr3 = getelementptr double, double *%ptr0, i64 6
112 %ptr4 = getelementptr double, double *%ptr0, i64 8
113 %ptr5 = getelementptr double, double *%ptr0, i64 10
114 %ptr6 = getelementptr double, double *%ptr0, i64 12
115 %ptr7 = getelementptr double, double *%ptr0, i64 14
116 %ptr8 = getelementptr double, double *%ptr0, i64 16
117 %ptr9 = getelementptr double, double *%ptr0, i64 18
118 %ptr10 = getelementptr double, double *%ptr0, i64 20
120 %val0 = load double, double *%ptr0
121 %val1 = load double, double *%ptr1
122 %val2 = load double, double *%ptr2
123 %val3 = load double, double *%ptr3
124 %val4 = load double, double *%ptr4
125 %val5 = load double, double *%ptr5
126 %val6 = load double, double *%ptr6
127 %val7 = load double, double *%ptr7
128 %val8 = load double, double *%ptr8
129 %val9 = load double, double *%ptr9
130 %val10 = load double, double *%ptr10
132 %ret = call double @foo()
134 %cmp0 = fcmp olt double %ret, %val0
135 %cmp1 = fcmp olt double %ret, %val1
136 %cmp2 = fcmp olt double %ret, %val2
137 %cmp3 = fcmp olt double %ret, %val3
138 %cmp4 = fcmp olt double %ret, %val4
139 %cmp5 = fcmp olt double %ret, %val5
140 %cmp6 = fcmp olt double %ret, %val6
141 %cmp7 = fcmp olt double %ret, %val7
142 %cmp8 = fcmp olt double %ret, %val8
143 %cmp9 = fcmp olt double %ret, %val9
144 %cmp10 = fcmp olt double %ret, %val10
146 %sel0 = select i1 %cmp0, double %ret, double 0.0
147 %sel1 = select i1 %cmp1, double %sel0, double 1.0
148 %sel2 = select i1 %cmp2, double %sel1, double 2.0
149 %sel3 = select i1 %cmp3, double %sel2, double 3.0
150 %sel4 = select i1 %cmp4, double %sel3, double 4.0
151 %sel5 = select i1 %cmp5, double %sel4, double 5.0
152 %sel6 = select i1 %cmp6, double %sel5, double 6.0
153 %sel7 = select i1 %cmp7, double %sel6, double 7.0
154 %sel8 = select i1 %cmp8, double %sel7, double 8.0
155 %sel9 = select i1 %cmp9, double %sel8, double 9.0
156 %sel10 = select i1 %cmp10, double %sel9, double 10.0
161 ; Check comparison with zero.
162 define i64 @f8(i64 %a, i64 %b, double %f) {
164 ; CHECK-SCALAR: ltdbr %f0, %f0
165 ; CHECK-SCALAR-NEXT: ber %r14
166 ; CHECK-SCALAR: lgr %r2, %r3
167 ; CHECK-VECTOR: ltdbr %f0, %f0
168 ; CHECK-VECTOR-NEXT: locgrne %r2, %r3
170 %cond = fcmp oeq double %f, 0.0
171 %res = select i1 %cond, i64 %a, i64 %b
175 ; Check the comparison can be reversed if that allows CDB to be used,
176 define i64 @f9(i64 %a, i64 %b, double %f2, double *%ptr) {
178 ; CHECK: cdb %f0, 0(%r4)
179 ; CHECK-SCALAR-NEXT: blr %r14
180 ; CHECK-SCALAR: lgr %r2, %r3
181 ; CHECK-VECTOR-NEXT: locgrnl %r2, %r3
183 %f1 = load double, double *%ptr
184 %cond = fcmp ogt double %f1, %f2
185 %res = select i1 %cond, i64 %a, i64 %b