1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
|
; Test 64-bit floating-point signaling comparison. The tests assume a z10
; implementation of select, using conditional branches rather than LOCGR.
;
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 \
; RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-SCALAR %s
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 -verify-machineinstrs\
; RUN: | FileCheck -check-prefix=CHECK -check-prefix=CHECK-VECTOR %s
declare double @foo()
; Check comparison with registers.
define i64 @f1(i64 %a, i64 %b, double %f1, double %f2) #0 {
; CHECK-LABEL: f1:
; CHECK: kdbr %f0, %f2
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check the low end of the KDB range.
define i64 @f2(i64 %a, i64 %b, double %f1, ptr %ptr) #0 {
; CHECK-LABEL: f2:
; CHECK: kdb %f0, 0(%r4)
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%f2 = load double, ptr %ptr
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check the high end of the aligned KDB range.
define i64 @f3(i64 %a, i64 %b, double %f1, ptr %base) #0 {
; CHECK-LABEL: f3:
; CHECK: kdb %f0, 4088(%r4)
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%ptr = getelementptr double, ptr %base, i64 511
%f2 = load double, ptr %ptr
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check the next doubleword up, which needs separate address logic.
; Other sequences besides this one would be OK.
define i64 @f4(i64 %a, i64 %b, double %f1, ptr %base) #0 {
; CHECK-LABEL: f4:
; CHECK: aghi %r4, 4096
; CHECK: kdb %f0, 0(%r4)
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%ptr = getelementptr double, ptr %base, i64 512
%f2 = load double, ptr %ptr
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check negative displacements, which also need separate address logic.
define i64 @f5(i64 %a, i64 %b, double %f1, ptr %base) #0 {
; CHECK-LABEL: f5:
; CHECK: aghi %r4, -8
; CHECK: kdb %f0, 0(%r4)
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%ptr = getelementptr double, ptr %base, i64 -1
%f2 = load double, ptr %ptr
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check that KDB allows indices.
define i64 @f6(i64 %a, i64 %b, double %f1, ptr %base, i64 %index) #0 {
; CHECK-LABEL: f6:
; CHECK: sllg %r1, %r5, 3
; CHECK: kdb %f0, 800(%r1,%r4)
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%ptr1 = getelementptr double, ptr %base, i64 %index
%ptr2 = getelementptr double, ptr %ptr1, i64 100
%f2 = load double, ptr %ptr2
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check that comparisons of spilled values can use KDB rather than KDBR.
define double @f7(ptr %ptr0) #0 {
; CHECK-LABEL: f7:
; CHECK: brasl %r14, foo@PLT
; CHECK-SCALAR: kdb {{%f[0-9]+}}, 160(%r15)
; CHECK: br %r14
%ptr1 = getelementptr double, ptr %ptr0, i64 2
%ptr2 = getelementptr double, ptr %ptr0, i64 4
%ptr3 = getelementptr double, ptr %ptr0, i64 6
%ptr4 = getelementptr double, ptr %ptr0, i64 8
%ptr5 = getelementptr double, ptr %ptr0, i64 10
%ptr6 = getelementptr double, ptr %ptr0, i64 12
%ptr7 = getelementptr double, ptr %ptr0, i64 14
%ptr8 = getelementptr double, ptr %ptr0, i64 16
%ptr9 = getelementptr double, ptr %ptr0, i64 18
%ptr10 = getelementptr double, ptr %ptr0, i64 20
%val0 = load double, ptr %ptr0
%val1 = load double, ptr %ptr1
%val2 = load double, ptr %ptr2
%val3 = load double, ptr %ptr3
%val4 = load double, ptr %ptr4
%val5 = load double, ptr %ptr5
%val6 = load double, ptr %ptr6
%val7 = load double, ptr %ptr7
%val8 = load double, ptr %ptr8
%val9 = load double, ptr %ptr9
%val10 = load double, ptr %ptr10
%ret = call double @foo() #0
%cmp0 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val0,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp1 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val1,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp2 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val2,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp3 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val3,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp4 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val4,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp5 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val5,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp6 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val6,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp7 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val7,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp8 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val8,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp9 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val9,
metadata !"olt",
metadata !"fpexcept.strict") #0
%cmp10 = call i1 @llvm.experimental.constrained.fcmps.f64(
double %ret, double %val10,
metadata !"olt",
metadata !"fpexcept.strict") #0
%sel0 = select i1 %cmp0, double %ret, double 0.0
%sel1 = select i1 %cmp1, double %sel0, double 1.0
%sel2 = select i1 %cmp2, double %sel1, double 2.0
%sel3 = select i1 %cmp3, double %sel2, double 3.0
%sel4 = select i1 %cmp4, double %sel3, double 4.0
%sel5 = select i1 %cmp5, double %sel4, double 5.0
%sel6 = select i1 %cmp6, double %sel5, double 6.0
%sel7 = select i1 %cmp7, double %sel6, double 7.0
%sel8 = select i1 %cmp8, double %sel7, double 8.0
%sel9 = select i1 %cmp9, double %sel8, double 9.0
%sel10 = select i1 %cmp10, double %sel9, double 10.0
ret double %sel10
}
; Check comparison with zero - cannot use LOAD AND TEST.
define i64 @f8(i64 %a, i64 %b, double %f) #0 {
; CHECK-LABEL: f8:
; CHECK: lzdr [[REG:%f[0-9]+]]
; CHECK-NEXT: kdbr %f0, [[REG]]
; CHECK-SCALAR-NEXT: ber %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrne %r2, %r3
; CHECK: br %r14
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f, double 0.0,
metadata !"oeq",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
; Check the comparison can be reversed if that allows KDB to be used,
define i64 @f9(i64 %a, i64 %b, double %f2, ptr %ptr) #0 {
; CHECK-LABEL: f9:
; CHECK: kdb %f0, 0(%r4)
; CHECK-SCALAR-NEXT: blr %r14
; CHECK-SCALAR: lgr %r2, %r3
; CHECK-VECTOR-NEXT: locgrnl %r2, %r3
; CHECK: br %r14
%f1 = load double, ptr %ptr
%cond = call i1 @llvm.experimental.constrained.fcmps.f64(
double %f1, double %f2,
metadata !"ogt",
metadata !"fpexcept.strict") #0
%res = select i1 %cond, i64 %a, i64 %b
ret i64 %res
}
attributes #0 = { strictfp }
declare i1 @llvm.experimental.constrained.fcmps.f64(double, double, metadata, metadata)
|
