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
|
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple aarch64-unknown-unknown -run-pass=instruction-select -global-isel-abort=1 -verify-machineinstrs %s -o - | FileCheck %s
...
---
name: flip_eq
alignment: 4
legalized: true
regBankSelected: true
body: |
; CHECK-LABEL: name: flip_eq
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.0(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: liveins: $x0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %copy:gpr64all = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr32all = COPY %copy.sub_32
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[COPY]]
; CHECK-NEXT: TBNZW [[COPY1]], 3, %bb.1
; CHECK-NEXT: B %bb.0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: RET_ReallyLR
bb.0:
successors: %bb.0, %bb.1
liveins: $x0
%copy:gpr(s64) = COPY $x0
; Check bit 3.
%bit:gpr(s64) = G_CONSTANT i64 8
%zero:gpr(s64) = G_CONSTANT i64 0
; 8 has the third bit set.
%fold_cst:gpr(s64) = G_CONSTANT i64 8
; This only has the third bit set if %copy does not. So, to walk through
; this, we want to use a TBNZW on %copy.
%fold_me:gpr(s64) = G_XOR %copy, %fold_cst
%and:gpr(s64) = G_AND %fold_me, %bit
%cmp:gpr(s32) = G_ICMP intpred(eq), %and(s64), %zero
G_BRCOND %cmp, %bb.1
G_BR %bb.0
bb.1:
RET_ReallyLR
...
---
name: flip_ne
alignment: 4
legalized: true
regBankSelected: true
body: |
; CHECK-LABEL: name: flip_ne
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.0(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: liveins: $x0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %copy:gpr64all = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr32all = COPY %copy.sub_32
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[COPY]]
; CHECK-NEXT: TBZW [[COPY1]], 3, %bb.1
; CHECK-NEXT: B %bb.0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: RET_ReallyLR
bb.0:
successors: %bb.0, %bb.1
liveins: $x0
; Same as eq case, but we should get a TBZW instead.
%copy:gpr(s64) = COPY $x0
%bit:gpr(s64) = G_CONSTANT i64 8
%zero:gpr(s64) = G_CONSTANT i64 0
%fold_cst:gpr(s64) = G_CONSTANT i64 8
%fold_me:gpr(s64) = G_XOR %copy, %fold_cst
%and:gpr(s64) = G_AND %fold_me, %bit
%cmp:gpr(s32) = G_ICMP intpred(ne), %and(s64), %zero
G_BRCOND %cmp, %bb.1
G_BR %bb.0
bb.1:
RET_ReallyLR
...
---
name: dont_flip_eq
alignment: 4
legalized: true
regBankSelected: true
body: |
; CHECK-LABEL: name: dont_flip_eq
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.0(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: liveins: $x0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %copy:gpr64all = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr32all = COPY %copy.sub_32
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[COPY]]
; CHECK-NEXT: TBZW [[COPY1]], 3, %bb.1
; CHECK-NEXT: B %bb.0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: RET_ReallyLR
bb.0:
successors: %bb.0, %bb.1
liveins: $x0
%copy:gpr(s64) = COPY $x0
; Check bit 3.
%bit:gpr(s64) = G_CONSTANT i64 8
%zero:gpr(s64) = G_CONSTANT i64 0
; 7 does not have the third bit set.
%fold_cst:gpr(s64) = G_CONSTANT i64 7
; This only has the third bit set if %copy does. So, to walk through this,
; we should have a TBZW on %copy.
%fold_me:gpr(s64) = G_XOR %fold_cst, %copy
%and:gpr(s64) = G_AND %fold_me, %bit
%cmp:gpr(s32) = G_ICMP intpred(eq), %and(s64), %zero
G_BRCOND %cmp, %bb.1
G_BR %bb.0
bb.1:
RET_ReallyLR
...
---
name: dont_flip_eq_zext
alignment: 4
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
; CHECK-LABEL: name: dont_flip_eq_zext
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.0(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr32 = COPY $wzr
; CHECK-NEXT: [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, [[COPY]], %subreg.sub_32
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[SUBREG_TO_REG]]
; CHECK-NEXT: TBNZX [[COPY1]], 63, %bb.1
; CHECK-NEXT: B %bb.0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: RET_ReallyLR
bb.0:
successors: %bb.0(0x40000000), %bb.1(0x40000000)
%1:gpr(s32) = G_CONSTANT i32 0
%3:gpr(s32) = G_CONSTANT i32 -1
%4:gpr(s32) = G_XOR %1, %3
%5:gpr(s64) = G_ZEXT %4(s32)
%15:gpr(s64) = G_CONSTANT i64 0
%13:gpr(s32) = G_ICMP intpred(slt), %5(s64), %15
G_BRCOND %13, %bb.1
G_BR %bb.0
bb.1:
RET_ReallyLR
...
---
name: dont_flip_ne
alignment: 4
legalized: true
regBankSelected: true
body: |
; CHECK-LABEL: name: dont_flip_ne
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.0(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: liveins: $x0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %copy:gpr64all = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr32all = COPY %copy.sub_32
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[COPY]]
; CHECK-NEXT: TBNZW [[COPY1]], 3, %bb.1
; CHECK-NEXT: B %bb.0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: RET_ReallyLR
bb.0:
successors: %bb.0, %bb.1
liveins: $x0
; Same as eq case, but we should get a TBNZW instead.
%copy:gpr(s64) = COPY $x0
%bit:gpr(s64) = G_CONSTANT i64 8
%zero:gpr(s64) = G_CONSTANT i64 0
%fold_cst:gpr(s64) = G_CONSTANT i64 7
%fold_me:gpr(s64) = G_XOR %fold_cst, %copy
%and:gpr(s64) = G_AND %fold_me, %bit
%cmp:gpr(s32) = G_ICMP intpred(ne), %and(s64), %zero
G_BRCOND %cmp, %bb.1
G_BR %bb.0
bb.1:
RET_ReallyLR
...
---
name: xor_chain
alignment: 4
legalized: true
regBankSelected: true
body: |
; CHECK-LABEL: name: xor_chain
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.0(0x40000000), %bb.1(0x40000000)
; CHECK-NEXT: liveins: $x0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %copy:gpr64all = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr32all = COPY %copy.sub_32
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[COPY]]
; CHECK-NEXT: TBZW [[COPY1]], 3, %bb.1
; CHECK-NEXT: B %bb.0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: RET_ReallyLR
bb.0:
successors: %bb.0, %bb.1
liveins: $x0
%copy:gpr(s64) = COPY $x0
%bit:gpr(s64) = G_CONSTANT i64 8
%zero:gpr(s64) = G_CONSTANT i64 0
%fold_cst:gpr(s64) = G_CONSTANT i64 8
; The G_XORs cancel each other out, so we should get a TBZW.
%xor1:gpr(s64) = G_XOR %copy, %fold_cst
%xor2:gpr(s64) = G_XOR %xor1, %fold_cst
%and:gpr(s64) = G_AND %xor2, %bit
%cmp:gpr(s32) = G_ICMP intpred(eq), %and(s64), %zero
G_BRCOND %cmp, %bb.1
G_BR %bb.0
bb.1:
RET_ReallyLR
|
