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
|
/**
* Many similar implementations exist. See for example libwsbm
* or the linux kernel include/atomic.h
*
* No copyright claimed on this file.
*
*/
#include "no_extern_c.h"
#ifndef U_ATOMIC_H
#define U_ATOMIC_H
#include <stdbool.h>
#include <stdint.h>
/* Favor OS-provided implementations.
*
* Where no OS-provided implementation is available, fall back to
* locally coded assembly, compiler intrinsic or ultimately a
* mutex-based implementation.
*/
#if defined(__sun)
#define PIPE_ATOMIC_OS_SOLARIS
#elif defined(_MSC_VER)
#define PIPE_ATOMIC_MSVC_INTRINSIC
#elif defined(__GNUC__)
#define PIPE_ATOMIC_GCC_INTRINSIC
#else
#error "Unsupported platform"
#endif
/* Implementation using GCC-provided synchronization intrinsics
*/
#if defined(PIPE_ATOMIC_GCC_INTRINSIC)
#define PIPE_ATOMIC "GCC Sync Intrinsics"
#if defined(USE_GCC_ATOMIC_BUILTINS)
/* The builtins with explicit memory model are available since GCC 4.7. */
#define p_atomic_set(_v, _i) __atomic_store_n((_v), (_i), __ATOMIC_RELEASE)
#define p_atomic_read(_v) __atomic_load_n((_v), __ATOMIC_ACQUIRE)
#define p_atomic_dec_zero(v) (__atomic_sub_fetch((v), 1, __ATOMIC_ACQ_REL) == 0)
#define p_atomic_inc(v) (void) __atomic_add_fetch((v), 1, __ATOMIC_ACQ_REL)
#define p_atomic_dec(v) (void) __atomic_sub_fetch((v), 1, __ATOMIC_ACQ_REL)
#define p_atomic_add(v, i) (void) __atomic_add_fetch((v), (i), __ATOMIC_ACQ_REL)
#define p_atomic_inc_return(v) __atomic_add_fetch((v), 1, __ATOMIC_ACQ_REL)
#define p_atomic_dec_return(v) __atomic_sub_fetch((v), 1, __ATOMIC_ACQ_REL)
#define p_atomic_add_return(v, i) __atomic_add_fetch((v), (i), __ATOMIC_ACQ_REL)
#define p_atomic_xchg(v, i) __atomic_exchange_n((v), (i), __ATOMIC_ACQ_REL)
#define PIPE_NATIVE_ATOMIC_XCHG
#else
#define p_atomic_set(_v, _i) (*(_v) = (_i))
#define p_atomic_read(_v) (*(_v))
#define p_atomic_dec_zero(v) (__sync_sub_and_fetch((v), 1) == 0)
#define p_atomic_inc(v) (void) __sync_add_and_fetch((v), 1)
#define p_atomic_dec(v) (void) __sync_sub_and_fetch((v), 1)
#define p_atomic_add(v, i) (void) __sync_add_and_fetch((v), (i))
#define p_atomic_inc_return(v) __sync_add_and_fetch((v), 1)
#define p_atomic_dec_return(v) __sync_sub_and_fetch((v), 1)
#define p_atomic_add_return(v, i) __sync_add_and_fetch((v), (i))
#endif
/* There is no __atomic_* compare and exchange that returns the current value.
* Also, GCC 5.4 seems unable to optimize a compound statement expression that
* uses an additional stack variable with __atomic_compare_exchange[_n].
*/
#define p_atomic_cmpxchg(v, old, _new) \
__sync_val_compare_and_swap((v), (old), (_new))
#endif
/* Unlocked version for single threaded environments, such as some
* windows kernel modules.
*/
#if defined(PIPE_ATOMIC_OS_UNLOCKED)
#define PIPE_ATOMIC "Unlocked"
#define p_atomic_set(_v, _i) (*(_v) = (_i))
#define p_atomic_read(_v) (*(_v))
#define p_atomic_dec_zero(_v) (p_atomic_dec_return(_v) == 0)
#define p_atomic_inc(_v) ((void) p_atomic_inc_return(_v))
#define p_atomic_dec(_v) ((void) p_atomic_dec_return(_v))
#define p_atomic_add(_v, _i) ((void) p_atomic_add_return((_v), (_i)))
#define p_atomic_inc_return(_v) (++(*(_v)))
#define p_atomic_dec_return(_v) (--(*(_v)))
#define p_atomic_add_return(_v, _i) (*(_v) = *(_v) + (_i))
#define p_atomic_cmpxchg(_v, _old, _new) (*(_v) == (_old) ? (*(_v) = (_new), (_old)) : *(_v))
#endif
#if defined(PIPE_ATOMIC_MSVC_INTRINSIC)
#define PIPE_ATOMIC "MSVC Intrinsics"
/* We use the Windows header's Interlocked*64 functions instead of the
* _Interlocked*64 intrinsics wherever we can, as support for the latter varies
* with target CPU, whereas Windows headers take care of all portability
* issues: using intrinsics where available, falling back to library
* implementations where not.
*/
#ifndef WIN32_LEAN_AND_MEAN
#define WIN32_LEAN_AND_MEAN 1
#endif
#include <windows.h>
#include <intrin.h>
#include <assert.h>
/* MSVC supports decltype keyword, but it's only supported on C++ and doesn't
* quite work here; and if a C++-only solution is worthwhile, then it would be
* better to use templates / function overloading, instead of decltype magic.
* Therefore, we rely on implicit casting to LONGLONG for the functions that return
*/
#define p_atomic_set(_v, _i) (*(_v) = (_i))
#define p_atomic_read(_v) (*(_v))
#define p_atomic_dec_zero(_v) \
(p_atomic_dec_return(_v) == 0)
#define p_atomic_inc(_v) \
((void) p_atomic_inc_return(_v))
#define p_atomic_inc_return(_v) (\
sizeof *(_v) == sizeof(short) ? _InterlockedIncrement16((short *) (_v)) : \
sizeof *(_v) == sizeof(long) ? _InterlockedIncrement ((long *) (_v)) : \
sizeof *(_v) == sizeof(__int64) ? InterlockedIncrement64 ((__int64 *)(_v)) : \
(assert(!"should not get here"), 0))
#define p_atomic_dec(_v) \
((void) p_atomic_dec_return(_v))
#define p_atomic_dec_return(_v) (\
sizeof *(_v) == sizeof(short) ? _InterlockedDecrement16((short *) (_v)) : \
sizeof *(_v) == sizeof(long) ? _InterlockedDecrement ((long *) (_v)) : \
sizeof *(_v) == sizeof(__int64) ? InterlockedDecrement64 ((__int64 *)(_v)) : \
(assert(!"should not get here"), 0))
#define p_atomic_add(_v, _i) \
((void) p_atomic_add_return((_v), (_i)))
#define p_atomic_add_return(_v, _i) (\
sizeof *(_v) == sizeof(char) ? _InterlockedExchangeAdd8 ((char *) (_v), (_i)) : \
sizeof *(_v) == sizeof(short) ? _InterlockedExchangeAdd16((short *) (_v), (_i)) : \
sizeof *(_v) == sizeof(long) ? _InterlockedExchangeAdd ((long *) (_v), (_i)) : \
sizeof *(_v) == sizeof(__int64) ? InterlockedExchangeAdd64((__int64 *)(_v), (_i)) : \
(assert(!"should not get here"), 0))
#define p_atomic_cmpxchg(_v, _old, _new) (\
sizeof *(_v) == sizeof(char) ? _InterlockedCompareExchange8 ((char *) (_v), (char) (_new), (char) (_old)) : \
sizeof *(_v) == sizeof(short) ? _InterlockedCompareExchange16((short *) (_v), (short) (_new), (short) (_old)) : \
sizeof *(_v) == sizeof(long) ? _InterlockedCompareExchange ((long *) (_v), (long) (_new), (long) (_old)) : \
sizeof *(_v) == sizeof(__int64) ? InterlockedCompareExchange64 ((__int64 *)(_v), (__int64)(_new), (__int64)(_old)) : \
(assert(!"should not get here"), 0))
#endif
#if defined(PIPE_ATOMIC_OS_SOLARIS)
#define PIPE_ATOMIC "Solaris OS atomic functions"
#include <atomic.h>
#include <assert.h>
#define p_atomic_set(_v, _i) (*(_v) = (_i))
#define p_atomic_read(_v) (*(_v))
#define p_atomic_dec_zero(v) (\
sizeof(*v) == sizeof(uint8_t) ? atomic_dec_8_nv ((uint8_t *)(v)) == 0 : \
sizeof(*v) == sizeof(uint16_t) ? atomic_dec_16_nv((uint16_t *)(v)) == 0 : \
sizeof(*v) == sizeof(uint32_t) ? atomic_dec_32_nv((uint32_t *)(v)) == 0 : \
sizeof(*v) == sizeof(uint64_t) ? atomic_dec_64_nv((uint64_t *)(v)) == 0 : \
(assert(!"should not get here"), 0))
#define p_atomic_inc(v) (void) (\
sizeof(*v) == sizeof(uint8_t) ? atomic_inc_8 ((uint8_t *)(v)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_inc_16((uint16_t *)(v)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_inc_32((uint32_t *)(v)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_inc_64((uint64_t *)(v)) : \
(assert(!"should not get here"), 0))
#define p_atomic_inc_return(v) (__typeof(*v))( \
sizeof(*v) == sizeof(uint8_t) ? atomic_inc_8_nv ((uint8_t *)(v)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_inc_16_nv((uint16_t *)(v)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_inc_32_nv((uint32_t *)(v)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_inc_64_nv((uint64_t *)(v)) : \
(assert(!"should not get here"), 0))
#define p_atomic_dec(v) (void) ( \
sizeof(*v) == sizeof(uint8_t) ? atomic_dec_8 ((uint8_t *)(v)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_dec_16((uint16_t *)(v)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_dec_32((uint32_t *)(v)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_dec_64((uint64_t *)(v)) : \
(assert(!"should not get here"), 0))
#define p_atomic_dec_return(v) (__typeof(*v))( \
sizeof(*v) == sizeof(uint8_t) ? atomic_dec_8_nv ((uint8_t *)(v)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_dec_16_nv((uint16_t *)(v)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_dec_32_nv((uint32_t *)(v)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_dec_64_nv((uint64_t *)(v)) : \
(assert(!"should not get here"), 0))
#define p_atomic_add(v, i) (void) ( \
sizeof(*v) == sizeof(uint8_t) ? atomic_add_8 ((uint8_t *)(v), (i)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_add_16((uint16_t *)(v), (i)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_add_32((uint32_t *)(v), (i)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_add_64((uint64_t *)(v), (i)) : \
(assert(!"should not get here"), 0))
#define p_atomic_add_return(v, i) (__typeof(*v)) ( \
sizeof(*v) == sizeof(uint8_t) ? atomic_add_8_nv ((uint8_t *)(v), (i)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_add_16_nv((uint16_t *)(v), (i)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_add_32_nv((uint32_t *)(v), (i)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_add_64_nv((uint64_t *)(v), (i)) : \
(assert(!"should not get here"), 0))
#define p_atomic_cmpxchg(v, old, _new) (__typeof(*v))( \
sizeof(*v) == sizeof(uint8_t) ? atomic_cas_8 ((uint8_t *)(v), (uint8_t )(old), (uint8_t )(_new)) : \
sizeof(*v) == sizeof(uint16_t) ? atomic_cas_16((uint16_t *)(v), (uint16_t)(old), (uint16_t)(_new)) : \
sizeof(*v) == sizeof(uint32_t) ? atomic_cas_32((uint32_t *)(v), (uint32_t)(old), (uint32_t)(_new)) : \
sizeof(*v) == sizeof(uint64_t) ? atomic_cas_64((uint64_t *)(v), (uint64_t)(old), (uint64_t)(_new)) : \
(assert(!"should not get here"), 0))
#endif
#ifndef PIPE_ATOMIC
#error "No pipe_atomic implementation selected"
#endif
#ifndef PIPE_NATIVE_ATOMIC_XCHG
static inline uint32_t p_atomic_xchg_32(uint32_t *v, uint32_t i)
{
uint32_t actual = p_atomic_read(v);
uint32_t expected;
do {
expected = actual;
actual = p_atomic_cmpxchg(v, expected, i);
} while (expected != actual);
return actual;
}
static inline uint64_t p_atomic_xchg_64(uint64_t *v, uint64_t i)
{
uint64_t actual = p_atomic_read(v);
uint64_t expected;
do {
expected = actual;
actual = p_atomic_cmpxchg(v, expected, i);
} while (expected != actual);
return actual;
}
#define p_atomic_xchg(v, i) (__typeof(*(v)))( \
sizeof(*(v)) == sizeof(uint32_t) ? p_atomic_xchg_32((uint32_t *)(v), (uint32_t)(i)) : \
sizeof(*(v)) == sizeof(uint64_t) ? p_atomic_xchg_64((uint64_t *)(v), (uint64_t)(i)) : \
(assert(!"should not get here"), 0))
#endif
#endif /* U_ATOMIC_H */
|
