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_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -verify-machineinstrs | FileCheck %s --check-prefix=X64
; RUN: llc < %s -mtriple=i686-unknown-unknown -verify-machineinstrs | FileCheck %s --check-prefix=X86
; Make sure that flags are properly preserved despite atomic optimizations.
define i32 @atomic_and_flags_1(i8* %p, i32 %a, i32 %b) {
; X64-LABEL: atomic_and_flags_1:
; X64: # %bb.0:
; X64-NEXT: cmpl %edx, %esi
; X64-NEXT: jne .LBB0_3
; X64-NEXT: # %bb.1: # %L1
; X64-NEXT: incb (%rdi)
; X64-NEXT: cmpl %edx, %esi
; X64-NEXT: jne .LBB0_2
; X64-NEXT: # %bb.4: # %L3
; X64-NEXT: movl $3, %eax
; X64-NEXT: retq
; X64-NEXT: .LBB0_3: # %L2
; X64-NEXT: movl $2, %eax
; X64-NEXT: retq
; X64-NEXT: .LBB0_2: # %L4
; X64-NEXT: movl $4, %eax
; X64-NEXT: retq
;
; X86-LABEL: atomic_and_flags_1:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: cmpl %eax, %ecx
; X86-NEXT: jne .LBB0_3
; X86-NEXT: # %bb.1: # %L1
; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: incb (%edx)
; X86-NEXT: cmpl %eax, %ecx
; X86-NEXT: jne .LBB0_2
; X86-NEXT: # %bb.4: # %L3
; X86-NEXT: movl $3, %eax
; X86-NEXT: retl
; X86-NEXT: .LBB0_3: # %L2
; X86-NEXT: movl $2, %eax
; X86-NEXT: retl
; X86-NEXT: .LBB0_2: # %L4
; X86-NEXT: movl $4, %eax
; X86-NEXT: retl
; Generate flags value, and use it.
%cmp = icmp eq i32 %a, %b
br i1 %cmp, label %L1, label %L2
L1:
; The following pattern will get folded.
%1 = load atomic i8, i8* %p seq_cst, align 1
%2 = add i8 %1, 1 ; This forces the INC instruction to be generated.
store atomic i8 %2, i8* %p release, align 1
; Use the comparison result again. We need to rematerialize the comparison
; somehow. This test checks that cmpl gets emitted again, but any
; rematerialization would work (the optimizer used to clobber the flags with
; the add).
br i1 %cmp, label %L3, label %L4
L2:
ret i32 2
L3:
ret i32 3
L4:
ret i32 4
}
; Same as above, but using 2 as immediate to avoid the INC instruction.
define i32 @atomic_and_flags_2(i8* %p, i32 %a, i32 %b) {
; X64-LABEL: atomic_and_flags_2:
; X64: # %bb.0:
; X64-NEXT: cmpl %edx, %esi
; X64-NEXT: jne .LBB1_3
; X64-NEXT: # %bb.1: # %L1
; X64-NEXT: addb $2, (%rdi)
; X64-NEXT: cmpl %edx, %esi
; X64-NEXT: jne .LBB1_2
; X64-NEXT: # %bb.4: # %L3
; X64-NEXT: movl $3, %eax
; X64-NEXT: retq
; X64-NEXT: .LBB1_3: # %L2
; X64-NEXT: movl $2, %eax
; X64-NEXT: retq
; X64-NEXT: .LBB1_2: # %L4
; X64-NEXT: movl $4, %eax
; X64-NEXT: retq
;
; X86-LABEL: atomic_and_flags_2:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: cmpl %eax, %ecx
; X86-NEXT: jne .LBB1_3
; X86-NEXT: # %bb.1: # %L1
; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
; X86-NEXT: addb $2, (%edx)
; X86-NEXT: cmpl %eax, %ecx
; X86-NEXT: jne .LBB1_2
; X86-NEXT: # %bb.4: # %L3
; X86-NEXT: movl $3, %eax
; X86-NEXT: retl
; X86-NEXT: .LBB1_3: # %L2
; X86-NEXT: movl $2, %eax
; X86-NEXT: retl
; X86-NEXT: .LBB1_2: # %L4
; X86-NEXT: movl $4, %eax
; X86-NEXT: retl
%cmp = icmp eq i32 %a, %b
br i1 %cmp, label %L1, label %L2
L1:
%1 = load atomic i8, i8* %p seq_cst, align 1
%2 = add i8 %1, 2
store atomic i8 %2, i8* %p release, align 1
br i1 %cmp, label %L3, label %L4
L2:
ret i32 2
L3:
ret i32 3
L4:
ret i32 4
}
; PR20841 - ensure we don't reuse the ZF flag from XADD for compares with zero,
; the flags are set for the result of the add result (the value stored to memory),
; not the value returned by the atomicrmw add.
define zeroext i1 @xadd_cmp0_i64(i64* %x) nounwind {
; X64-LABEL: xadd_cmp0_i64:
; X64: # %bb.0:
; X64-NEXT: movl $1, %eax
; X64-NEXT: lock xaddq %rax, (%rdi)
; X64-NEXT: testq %rax, %rax
; X64-NEXT: sete %al
; X64-NEXT: retq
;
; X86-LABEL: xadd_cmp0_i64:
; X86: # %bb.0:
; X86-NEXT: pushl %ebx
; X86-NEXT: pushl %esi
; X86-NEXT: movl {{[0-9]+}}(%esp), %esi
; X86-NEXT: movl (%esi), %eax
; X86-NEXT: movl 4(%esi), %edx
; X86-NEXT: .p2align 4, 0x90
; X86-NEXT: .LBB2_1: # %atomicrmw.start
; X86-NEXT: # =>This Inner Loop Header: Depth=1
; X86-NEXT: movl %eax, %ebx
; X86-NEXT: addl $1, %ebx
; X86-NEXT: movl %edx, %ecx
; X86-NEXT: adcl $0, %ecx
; X86-NEXT: lock cmpxchg8b (%esi)
; X86-NEXT: jne .LBB2_1
; X86-NEXT: # %bb.2: # %atomicrmw.end
; X86-NEXT: orl %edx, %eax
; X86-NEXT: sete %al
; X86-NEXT: popl %esi
; X86-NEXT: popl %ebx
; X86-NEXT: retl
%add = atomicrmw add i64* %x, i64 1 seq_cst
%cmp = icmp eq i64 %add, 0
ret i1 %cmp
}
define zeroext i1 @xadd_cmp0_i32(i32* %x) nounwind {
; X64-LABEL: xadd_cmp0_i32:
; X64: # %bb.0:
; X64-NEXT: movl $1, %eax
; X64-NEXT: lock xaddl %eax, (%rdi)
; X64-NEXT: testl %eax, %eax
; X64-NEXT: setne %al
; X64-NEXT: retq
;
; X86-LABEL: xadd_cmp0_i32:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl $1, %ecx
; X86-NEXT: lock xaddl %ecx, (%eax)
; X86-NEXT: testl %ecx, %ecx
; X86-NEXT: setne %al
; X86-NEXT: retl
%add = atomicrmw add i32* %x, i32 1 seq_cst
%cmp = icmp ne i32 %add, 0
ret i1 %cmp
}
define zeroext i1 @xadd_cmp0_i16(i16* %x) nounwind {
; X64-LABEL: xadd_cmp0_i16:
; X64: # %bb.0:
; X64-NEXT: movw $1, %ax
; X64-NEXT: lock xaddw %ax, (%rdi)
; X64-NEXT: testw %ax, %ax
; X64-NEXT: sete %al
; X64-NEXT: retq
;
; X86-LABEL: xadd_cmp0_i16:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movw $1, %cx
; X86-NEXT: lock xaddw %cx, (%eax)
; X86-NEXT: testw %cx, %cx
; X86-NEXT: sete %al
; X86-NEXT: retl
%add = atomicrmw add i16* %x, i16 1 seq_cst
%cmp = icmp eq i16 %add, 0
ret i1 %cmp
}
define zeroext i1 @xadd_cmp0_i8(i8* %x) nounwind {
; X64-LABEL: xadd_cmp0_i8:
; X64: # %bb.0:
; X64-NEXT: movb $1, %al
; X64-NEXT: lock xaddb %al, (%rdi)
; X64-NEXT: testb %al, %al
; X64-NEXT: setne %al
; X64-NEXT: retq
;
; X86-LABEL: xadd_cmp0_i8:
; X86: # %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movb $1, %cl
; X86-NEXT: lock xaddb %cl, (%eax)
; X86-NEXT: testb %cl, %cl
; X86-NEXT: setne %al
; X86-NEXT: retl
%add = atomicrmw add i8* %x, i8 1 seq_cst
%cmp = icmp ne i8 %add, 0
ret i1 %cmp
}
|
