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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
| ; Test strict multiplication of two f32s, producing an f64 result.
; FIXME: we do not have a strict version of fpext yet
;
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
declare float @foo()
declare double @llvm.experimental.constrained.fmul.f64(double, double, metadata, metadata)
; Check register multiplication.
define double @f1(float %f1, float %f2) #0 {
; CHECK-LABEL: f1:
; CHECK: mdebr %f0, %f2
; CHECK: br %r14
%f1x = fpext float %f1 to double
%f2x = fpext float %f2 to double
%res = call double @llvm.experimental.constrained.fmul.f64(
double %f1x, double %f2x,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
ret double %res
}
; Check the low end of the MDEB range.
define double @f2(float %f1, float *%ptr) #0 {
; CHECK-LABEL: f2:
; CHECK: mdeb %f0, 0(%r2)
; CHECK: br %r14
%f2 = load float, float *%ptr
%f1x = fpext float %f1 to double
%f2x = fpext float %f2 to double
%res = call double @llvm.experimental.constrained.fmul.f64(
double %f1x, double %f2x,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
ret double %res
}
; Check the high end of the aligned MDEB range.
define double @f3(float %f1, float *%base) #0 {
; CHECK-LABEL: f3:
; CHECK: mdeb %f0, 4092(%r2)
; CHECK: br %r14
%ptr = getelementptr float, float *%base, i64 1023
%f2 = load float, float *%ptr
%f1x = fpext float %f1 to double
%f2x = fpext float %f2 to double
%res = call double @llvm.experimental.constrained.fmul.f64(
double %f1x, double %f2x,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
ret double %res
}
; Check the next word up, which needs separate address logic.
; Other sequences besides this one would be OK.
define double @f4(float %f1, float *%base) #0 {
; CHECK-LABEL: f4:
; CHECK: aghi %r2, 4096
; CHECK: mdeb %f0, 0(%r2)
; CHECK: br %r14
%ptr = getelementptr float, float *%base, i64 1024
%f2 = load float, float *%ptr
%f1x = fpext float %f1 to double
%f2x = fpext float %f2 to double
%res = call double @llvm.experimental.constrained.fmul.f64(
double %f1x, double %f2x,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
ret double %res
}
; Check negative displacements, which also need separate address logic.
define double @f5(float %f1, float *%base) #0 {
; CHECK-LABEL: f5:
; CHECK: aghi %r2, -4
; CHECK: mdeb %f0, 0(%r2)
; CHECK: br %r14
%ptr = getelementptr float, float *%base, i64 -1
%f2 = load float, float *%ptr
%f1x = fpext float %f1 to double
%f2x = fpext float %f2 to double
%res = call double @llvm.experimental.constrained.fmul.f64(
double %f1x, double %f2x,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
ret double %res
}
; Check that MDEB allows indices.
define double @f6(float %f1, float *%base, i64 %index) #0 {
; CHECK-LABEL: f6:
; CHECK: sllg %r1, %r3, 2
; CHECK: mdeb %f0, 400(%r1,%r2)
; CHECK: br %r14
%ptr1 = getelementptr float, float *%base, i64 %index
%ptr2 = getelementptr float, float *%ptr1, i64 100
%f2 = load float, float *%ptr2
%f1x = fpext float %f1 to double
%f2x = fpext float %f2 to double
%res = call double @llvm.experimental.constrained.fmul.f64(
double %f1x, double %f2x,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
ret double %res
}
; Check that multiplications of spilled values can use MDEB rather than MDEBR.
define float @f7(float *%ptr0) #0 {
; CHECK-LABEL: f7:
; CHECK: brasl %r14, foo@PLT
; CHECK: mdeb %f0, 16{{[04]}}(%r15)
; CHECK: br %r14
%ptr1 = getelementptr float, float *%ptr0, i64 2
%ptr2 = getelementptr float, float *%ptr0, i64 4
%ptr3 = getelementptr float, float *%ptr0, i64 6
%ptr4 = getelementptr float, float *%ptr0, i64 8
%ptr5 = getelementptr float, float *%ptr0, i64 10
%ptr6 = getelementptr float, float *%ptr0, i64 12
%ptr7 = getelementptr float, float *%ptr0, i64 14
%ptr8 = getelementptr float, float *%ptr0, i64 16
%ptr9 = getelementptr float, float *%ptr0, i64 18
%ptr10 = getelementptr float, float *%ptr0, i64 20
%val0 = load float, float *%ptr0
%val1 = load float, float *%ptr1
%val2 = load float, float *%ptr2
%val3 = load float, float *%ptr3
%val4 = load float, float *%ptr4
%val5 = load float, float *%ptr5
%val6 = load float, float *%ptr6
%val7 = load float, float *%ptr7
%val8 = load float, float *%ptr8
%val9 = load float, float *%ptr9
%val10 = load float, float *%ptr10
%frob0 = fadd float %val0, %val0
%frob1 = fadd float %val1, %val1
%frob2 = fadd float %val2, %val2
%frob3 = fadd float %val3, %val3
%frob4 = fadd float %val4, %val4
%frob5 = fadd float %val5, %val5
%frob6 = fadd float %val6, %val6
%frob7 = fadd float %val7, %val7
%frob8 = fadd float %val8, %val8
%frob9 = fadd float %val9, %val9
%frob10 = fadd float %val9, %val10
store float %frob0, float *%ptr0
store float %frob1, float *%ptr1
store float %frob2, float *%ptr2
store float %frob3, float *%ptr3
store float %frob4, float *%ptr4
store float %frob5, float *%ptr5
store float %frob6, float *%ptr6
store float %frob7, float *%ptr7
store float %frob8, float *%ptr8
store float %frob9, float *%ptr9
store float %frob10, float *%ptr10
%ret = call float @foo() #0
%accext0 = fpext float %ret to double
%ext0 = fpext float %frob0 to double
%mul0 = call double @llvm.experimental.constrained.fmul.f64(
double %accext0, double %ext0,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra0 = call double @llvm.experimental.constrained.fmul.f64(
double %mul0, double 1.01,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc0 = fptrunc double %extra0 to float
%accext1 = fpext float %trunc0 to double
%ext1 = fpext float %frob1 to double
%mul1 = call double @llvm.experimental.constrained.fmul.f64(
double %accext1, double %ext1,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra1 = call double @llvm.experimental.constrained.fmul.f64(
double %mul1, double 1.11,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc1 = fptrunc double %extra1 to float
%accext2 = fpext float %trunc1 to double
%ext2 = fpext float %frob2 to double
%mul2 = call double @llvm.experimental.constrained.fmul.f64(
double %accext2, double %ext2,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra2 = call double @llvm.experimental.constrained.fmul.f64(
double %mul2, double 1.21,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc2 = fptrunc double %extra2 to float
%accext3 = fpext float %trunc2 to double
%ext3 = fpext float %frob3 to double
%mul3 = call double @llvm.experimental.constrained.fmul.f64(
double %accext3, double %ext3,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra3 = call double @llvm.experimental.constrained.fmul.f64(
double %mul3, double 1.31,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc3 = fptrunc double %extra3 to float
%accext4 = fpext float %trunc3 to double
%ext4 = fpext float %frob4 to double
%mul4 = call double @llvm.experimental.constrained.fmul.f64(
double %accext4, double %ext4,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra4 = call double @llvm.experimental.constrained.fmul.f64(
double %mul4, double 1.41,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc4 = fptrunc double %extra4 to float
%accext5 = fpext float %trunc4 to double
%ext5 = fpext float %frob5 to double
%mul5 = call double @llvm.experimental.constrained.fmul.f64(
double %accext5, double %ext5,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra5 = call double @llvm.experimental.constrained.fmul.f64(
double %mul5, double 1.51,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc5 = fptrunc double %extra5 to float
%accext6 = fpext float %trunc5 to double
%ext6 = fpext float %frob6 to double
%mul6 = call double @llvm.experimental.constrained.fmul.f64(
double %accext6, double %ext6,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra6 = call double @llvm.experimental.constrained.fmul.f64(
double %mul6, double 1.61,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc6 = fptrunc double %extra6 to float
%accext7 = fpext float %trunc6 to double
%ext7 = fpext float %frob7 to double
%mul7 = call double @llvm.experimental.constrained.fmul.f64(
double %accext7, double %ext7,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra7 = call double @llvm.experimental.constrained.fmul.f64(
double %mul7, double 1.71,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc7 = fptrunc double %extra7 to float
%accext8 = fpext float %trunc7 to double
%ext8 = fpext float %frob8 to double
%mul8 = call double @llvm.experimental.constrained.fmul.f64(
double %accext8, double %ext8,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra8 = call double @llvm.experimental.constrained.fmul.f64(
double %mul8, double 1.81,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc8 = fptrunc double %extra8 to float
%accext9 = fpext float %trunc8 to double
%ext9 = fpext float %frob9 to double
%mul9 = call double @llvm.experimental.constrained.fmul.f64(
double %accext9, double %ext9,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%extra9 = call double @llvm.experimental.constrained.fmul.f64(
double %mul9, double 1.91,
metadata !"round.dynamic",
metadata !"fpexcept.strict") #0
%trunc9 = fptrunc double %extra9 to float
ret float %trunc9
}
attributes #0 = { strictfp }
|