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
|
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
; RUN: llc < %s -mtriple=arm -mattr=+vfp4d16sp,+fullfp16,-bf16 -stop-after=finalize-isel | FileCheck %s --check-prefixes=CHECK-NOBF16
; Check that the output instructions have the same fast math flags as the input
; fadd, even when bf16 is legalized to f32.
; FIXME: We should also test with +bf16, but it currently fails in instruction
; selection.
define bfloat @normal_fadd(bfloat %x, bfloat %y) {
; CHECK-NOBF16-LABEL: name: normal_fadd
; CHECK-NOBF16: bb.0.entry:
; CHECK-NOBF16-NEXT: liveins: $r0, $r1
; CHECK-NOBF16-NEXT: {{ $}}
; CHECK-NOBF16-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $r1
; CHECK-NOBF16-NEXT: [[COPY1:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi:%[0-9]+]]:gpr = MOVsi [[COPY]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR:%[0-9]+]]:spr = VMOVSR killed [[MOVsi]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi1:%[0-9]+]]:gpr = MOVsi [[COPY1]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR1:%[0-9]+]]:spr = VMOVSR killed [[MOVsi1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS:%[0-9]+]]:spr = VADDS killed [[VMOVSR1]], killed [[VMOVSR]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS:%[0-9]+]]:gpr = VMOVRS killed [[VADDS]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY2:%[0-9]+]]:rgpr = COPY $r0
; CHECK-NOBF16-NEXT: [[VMOVHR:%[0-9]+]]:hpr = VMOVHR [[COPY2]], 14, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRH:%[0-9]+]]:rgpr = VMOVRH killed [[VMOVHR]], 14, $noreg
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRH]]
; CHECK-NOBF16-NEXT: MOVPCLR 14 /* CC::al */, $noreg, implicit $r0
entry:
%add = fadd bfloat %x, %y
ret bfloat %add
}
define bfloat @fast_fadd(bfloat %x, bfloat %y) {
; CHECK-NOBF16-LABEL: name: fast_fadd
; CHECK-NOBF16: bb.0.entry:
; CHECK-NOBF16-NEXT: liveins: $r0, $r1
; CHECK-NOBF16-NEXT: {{ $}}
; CHECK-NOBF16-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $r1
; CHECK-NOBF16-NEXT: [[COPY1:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi:%[0-9]+]]:gpr = MOVsi [[COPY]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR:%[0-9]+]]:spr = VMOVSR killed [[MOVsi]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi1:%[0-9]+]]:gpr = MOVsi [[COPY1]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR1:%[0-9]+]]:spr = VMOVSR killed [[MOVsi1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS:%[0-9]+]]:spr = nnan ninf nsz arcp contract afn reassoc VADDS killed [[VMOVSR1]], killed [[VMOVSR]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS:%[0-9]+]]:gpr = VMOVRS killed [[VADDS]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY2:%[0-9]+]]:rgpr = COPY $r0
; CHECK-NOBF16-NEXT: [[VMOVHR:%[0-9]+]]:hpr = VMOVHR [[COPY2]], 14, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRH:%[0-9]+]]:rgpr = VMOVRH killed [[VMOVHR]], 14, $noreg
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRH]]
; CHECK-NOBF16-NEXT: MOVPCLR 14 /* CC::al */, $noreg, implicit $r0
entry:
%add = fadd fast bfloat %x, %y
ret bfloat %add
}
define bfloat @ninf_fadd(bfloat %x, bfloat %y) {
; CHECK-NOBF16-LABEL: name: ninf_fadd
; CHECK-NOBF16: bb.0.entry:
; CHECK-NOBF16-NEXT: liveins: $r0, $r1
; CHECK-NOBF16-NEXT: {{ $}}
; CHECK-NOBF16-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $r1
; CHECK-NOBF16-NEXT: [[COPY1:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi:%[0-9]+]]:gpr = MOVsi [[COPY]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR:%[0-9]+]]:spr = VMOVSR killed [[MOVsi]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi1:%[0-9]+]]:gpr = MOVsi [[COPY1]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR1:%[0-9]+]]:spr = VMOVSR killed [[MOVsi1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS:%[0-9]+]]:spr = ninf VADDS killed [[VMOVSR1]], killed [[VMOVSR]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS:%[0-9]+]]:gpr = VMOVRS killed [[VADDS]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY2:%[0-9]+]]:rgpr = COPY $r0
; CHECK-NOBF16-NEXT: [[VMOVHR:%[0-9]+]]:hpr = VMOVHR [[COPY2]], 14, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRH:%[0-9]+]]:rgpr = VMOVRH killed [[VMOVHR]], 14, $noreg
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRH]]
; CHECK-NOBF16-NEXT: MOVPCLR 14 /* CC::al */, $noreg, implicit $r0
entry:
%add = fadd ninf bfloat %x, %y
ret bfloat %add
}
; Check that when we have the right fast math flags the converts in between the
; two fadds are removed.
define bfloat @normal_fadd_sequence(bfloat %x, bfloat %y, bfloat %z) {
; CHECK-NOBF16-LABEL: name: normal_fadd_sequence
; CHECK-NOBF16: bb.0.entry:
; CHECK-NOBF16-NEXT: liveins: $r0, $r1, $r2
; CHECK-NOBF16-NEXT: {{ $}}
; CHECK-NOBF16-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $r2
; CHECK-NOBF16-NEXT: [[COPY1:%[0-9]+]]:gpr = COPY $r1
; CHECK-NOBF16-NEXT: [[COPY2:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi:%[0-9]+]]:gpr = MOVsi [[COPY1]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR:%[0-9]+]]:spr = VMOVSR killed [[MOVsi]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi1:%[0-9]+]]:gpr = MOVsi [[COPY2]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR1:%[0-9]+]]:spr = VMOVSR killed [[MOVsi1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS:%[0-9]+]]:spr = VADDS killed [[VMOVSR1]], killed [[VMOVSR]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS:%[0-9]+]]:gpr = VMOVRS killed [[VADDS]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY3:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi2:%[0-9]+]]:gpr = MOVsi [[COPY]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR2:%[0-9]+]]:spr = VMOVSR killed [[MOVsi2]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi3:%[0-9]+]]:gpr = MOVsi [[COPY3]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR3:%[0-9]+]]:spr = VMOVSR killed [[MOVsi3]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS1:%[0-9]+]]:spr = VADDS killed [[VMOVSR3]], killed [[VMOVSR2]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS1:%[0-9]+]]:gpr = VMOVRS killed [[VADDS1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS1]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY4:%[0-9]+]]:rgpr = COPY $r0
; CHECK-NOBF16-NEXT: [[VMOVHR:%[0-9]+]]:hpr = VMOVHR [[COPY4]], 14, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRH:%[0-9]+]]:rgpr = VMOVRH killed [[VMOVHR]], 14, $noreg
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRH]]
; CHECK-NOBF16-NEXT: MOVPCLR 14 /* CC::al */, $noreg, implicit $r0
entry:
%add1 = fadd bfloat %x, %y
%add2 = fadd bfloat %add1, %z
ret bfloat %add2
}
define bfloat @nnan_ninf_contract_fadd_sequence(bfloat %x, bfloat %y, bfloat %z) {
; CHECK-NOBF16-LABEL: name: nnan_ninf_contract_fadd_sequence
; CHECK-NOBF16: bb.0.entry:
; CHECK-NOBF16-NEXT: liveins: $r0, $r1, $r2
; CHECK-NOBF16-NEXT: {{ $}}
; CHECK-NOBF16-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $r2
; CHECK-NOBF16-NEXT: [[COPY1:%[0-9]+]]:gpr = COPY $r1
; CHECK-NOBF16-NEXT: [[COPY2:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi:%[0-9]+]]:gpr = MOVsi [[COPY1]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR:%[0-9]+]]:spr = VMOVSR killed [[MOVsi]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi1:%[0-9]+]]:gpr = MOVsi [[COPY2]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR1:%[0-9]+]]:spr = VMOVSR killed [[MOVsi1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS:%[0-9]+]]:spr = nnan ninf contract VADDS killed [[VMOVSR1]], killed [[VMOVSR]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi2:%[0-9]+]]:gpr = MOVsi [[COPY]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR2:%[0-9]+]]:spr = VMOVSR killed [[MOVsi2]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS1:%[0-9]+]]:spr = nnan ninf contract VADDS killed [[VADDS]], killed [[VMOVSR2]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS:%[0-9]+]]:gpr = VMOVRS killed [[VADDS1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY3:%[0-9]+]]:rgpr = COPY $r0
; CHECK-NOBF16-NEXT: [[VMOVHR:%[0-9]+]]:hpr = VMOVHR [[COPY3]], 14, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRH:%[0-9]+]]:rgpr = VMOVRH killed [[VMOVHR]], 14, $noreg
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRH]]
; CHECK-NOBF16-NEXT: MOVPCLR 14 /* CC::al */, $noreg, implicit $r0
entry:
%add1 = fadd nnan ninf contract bfloat %x, %y
%add2 = fadd nnan ninf contract bfloat %add1, %z
ret bfloat %add2
}
define bfloat @ninf_fadd_sequence(bfloat %x, bfloat %y, bfloat %z) {
; CHECK-NOBF16-LABEL: name: ninf_fadd_sequence
; CHECK-NOBF16: bb.0.entry:
; CHECK-NOBF16-NEXT: liveins: $r0, $r1, $r2
; CHECK-NOBF16-NEXT: {{ $}}
; CHECK-NOBF16-NEXT: [[COPY:%[0-9]+]]:gpr = COPY $r2
; CHECK-NOBF16-NEXT: [[COPY1:%[0-9]+]]:gpr = COPY $r1
; CHECK-NOBF16-NEXT: [[COPY2:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi:%[0-9]+]]:gpr = MOVsi [[COPY1]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR:%[0-9]+]]:spr = VMOVSR killed [[MOVsi]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi1:%[0-9]+]]:gpr = MOVsi [[COPY2]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR1:%[0-9]+]]:spr = VMOVSR killed [[MOVsi1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS:%[0-9]+]]:spr = ninf VADDS killed [[VMOVSR1]], killed [[VMOVSR]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS:%[0-9]+]]:gpr = VMOVRS killed [[VADDS]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY3:%[0-9]+]]:gpr = COPY $r0
; CHECK-NOBF16-NEXT: [[MOVsi2:%[0-9]+]]:gpr = MOVsi [[COPY]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR2:%[0-9]+]]:spr = VMOVSR killed [[MOVsi2]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[MOVsi3:%[0-9]+]]:gpr = MOVsi [[COPY3]], 130, 14 /* CC::al */, $noreg, $noreg
; CHECK-NOBF16-NEXT: [[VMOVSR3:%[0-9]+]]:spr = VMOVSR killed [[MOVsi3]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VADDS1:%[0-9]+]]:spr = ninf VADDS killed [[VMOVSR3]], killed [[VMOVSR2]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRS1:%[0-9]+]]:gpr = VMOVRS killed [[VADDS1]], 14 /* CC::al */, $noreg
; CHECK-NOBF16-NEXT: ADJCALLSTACKDOWN 0, 0, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRS1]]
; CHECK-NOBF16-NEXT: BL &__truncsfbf2, csr_aapcs, implicit-def dead $lr, implicit $sp, implicit $r0, implicit-def $sp, implicit-def $r0
; CHECK-NOBF16-NEXT: ADJCALLSTACKUP 0, -1, 14 /* CC::al */, $noreg, implicit-def dead $sp, implicit $sp
; CHECK-NOBF16-NEXT: [[COPY4:%[0-9]+]]:rgpr = COPY $r0
; CHECK-NOBF16-NEXT: [[VMOVHR:%[0-9]+]]:hpr = VMOVHR [[COPY4]], 14, $noreg
; CHECK-NOBF16-NEXT: [[VMOVRH:%[0-9]+]]:rgpr = VMOVRH killed [[VMOVHR]], 14, $noreg
; CHECK-NOBF16-NEXT: $r0 = COPY [[VMOVRH]]
; CHECK-NOBF16-NEXT: MOVPCLR 14 /* CC::al */, $noreg, implicit $r0
entry:
%add1 = fadd ninf bfloat %x, %y
%add2 = fadd ninf bfloat %add1, %z
ret bfloat %add2
}
|
