Repository navigation
Expand file tree
/
Copy pathpycore_atomic.h
More file actions
546 lines (486 loc) · 16.1 KB
/
Copy pathpycore_atomic.h
File metadata and controls
546 lines (486 loc) · 16.1 KB
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
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
#ifndef Py_ATOMIC_H
#define Py_ATOMIC_H
#ifdef __cplusplus
extern "C" {
#endif
#if !defined(Py_BUILD_CORE) && !defined(Py_BUILD_CORE_BUILTIN)
# error "this header requires Py_BUILD_CORE or Py_BUILD_CORE_BUILTIN define"
#endif
#include "dynamic_annotations.h"
#include "pyconfig.h"
#if defined(HAVE_STD_ATOMIC)
#include <stdatomic.h>
#endif
#if defined(_MSC_VER)
#include <intrin.h>
#if defined(_M_IX86) || defined(_M_X64)
# include <immintrin.h>
#endif
#endif
/* This is modeled after the atomics interface from C1x, according to
* the draft at
* http://www.open-std.org/JTC1/SC22/wg14/www/docs/n1425.pdf.
* Operations and types are named the same except with a _Py_ prefix
* and have the same semantics.
*
* Beware, the implementations here are deep magic.
*/
#if defined(HAVE_STD_ATOMIC)
typedef enum _Py_memory_order {
_Py_memory_order_relaxed = memory_order_relaxed,
_Py_memory_order_acquire = memory_order_acquire,
_Py_memory_order_release = memory_order_release,
_Py_memory_order_acq_rel = memory_order_acq_rel,
_Py_memory_order_seq_cst = memory_order_seq_cst
} _Py_memory_order;
typedef struct _Py_atomic_address {
atomic_uintptr_t _value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
atomic_int _value;
} _Py_atomic_int;
#define _Py_atomic_signal_fence(/*memory_order*/ ORDER) \
atomic_signal_fence(ORDER)
#define _Py_atomic_thread_fence(/*memory_order*/ ORDER) \
atomic_thread_fence(ORDER)
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
atomic_store_explicit(&((ATOMIC_VAL)->_value), NEW_VAL, ORDER)
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
atomic_load_explicit(&((ATOMIC_VAL)->_value), ORDER)
/* Use builtin atomic operations in GCC >= 4.7 */
#elif defined(HAVE_BUILTIN_ATOMIC)
typedef enum _Py_memory_order {
_Py_memory_order_relaxed = __ATOMIC_RELAXED,
_Py_memory_order_acquire = __ATOMIC_ACQUIRE,
_Py_memory_order_release = __ATOMIC_RELEASE,
_Py_memory_order_acq_rel = __ATOMIC_ACQ_REL,
_Py_memory_order_seq_cst = __ATOMIC_SEQ_CST
} _Py_memory_order;
typedef struct _Py_atomic_address {
uintptr_t _value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
int _value;
} _Py_atomic_int;
#define _Py_atomic_signal_fence(/*memory_order*/ ORDER) \
__atomic_signal_fence(ORDER)
#define _Py_atomic_thread_fence(/*memory_order*/ ORDER) \
__atomic_thread_fence(ORDER)
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
(assert((ORDER) == __ATOMIC_RELAXED \
|| (ORDER) == __ATOMIC_SEQ_CST \
|| (ORDER) == __ATOMIC_RELEASE), \
__atomic_store_n(&((ATOMIC_VAL)->_value), NEW_VAL, ORDER))
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
(assert((ORDER) == __ATOMIC_RELAXED \
|| (ORDER) == __ATOMIC_SEQ_CST \
|| (ORDER) == __ATOMIC_ACQUIRE \
|| (ORDER) == __ATOMIC_CONSUME), \
__atomic_load_n(&((ATOMIC_VAL)->_value), ORDER))
/* Only support GCC (for expression statements) and x86 (for simple
* atomic semantics) and MSVC x86/x64/ARM */
#elif defined(__GNUC__) && (defined(__i386__) || defined(__amd64))
typedef enum _Py_memory_order {
_Py_memory_order_relaxed,
_Py_memory_order_acquire,
_Py_memory_order_release,
_Py_memory_order_acq_rel,
_Py_memory_order_seq_cst
} _Py_memory_order;
typedef struct _Py_atomic_address {
uintptr_t _value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
int _value;
} _Py_atomic_int;
static __inline__ void
_Py_atomic_signal_fence(_Py_memory_order order)
{
if (order != _Py_memory_order_relaxed)
__asm__ volatile("":::"memory");
}
static __inline__ void
_Py_atomic_thread_fence(_Py_memory_order order)
{
if (order != _Py_memory_order_relaxed)
__asm__ volatile("mfence":::"memory");
}
/* Tell the race checker about this operation's effects. */
static __inline__ void
_Py_ANNOTATE_MEMORY_ORDER(const volatile void *address, _Py_memory_order order)
{
(void)address; /* shut up -Wunused-parameter */
switch(order) {
case _Py_memory_order_release:
case _Py_memory_order_acq_rel:
case _Py_memory_order_seq_cst:
_Py_ANNOTATE_HAPPENS_BEFORE(address);
break;
case _Py_memory_order_relaxed:
case _Py_memory_order_acquire:
break;
}
switch(order) {
case _Py_memory_order_acquire:
case _Py_memory_order_acq_rel:
case _Py_memory_order_seq_cst:
_Py_ANNOTATE_HAPPENS_AFTER(address);
break;
case _Py_memory_order_relaxed:
case _Py_memory_order_release:
break;
}
}
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
__extension__ ({ \
__typeof__(ATOMIC_VAL) atomic_val = ATOMIC_VAL; \
__typeof__(atomic_val->_value) new_val = NEW_VAL;\
volatile __typeof__(new_val) *volatile_data = &atomic_val->_value; \
_Py_memory_order order = ORDER; \
_Py_ANNOTATE_MEMORY_ORDER(atomic_val, order); \
\
/* Perform the operation. */ \
_Py_ANNOTATE_IGNORE_WRITES_BEGIN(); \
switch(order) { \
case _Py_memory_order_release: \
_Py_atomic_signal_fence(_Py_memory_order_release); \
/* fallthrough */ \
case _Py_memory_order_relaxed: \
*volatile_data = new_val; \
break; \
\
case _Py_memory_order_acquire: \
case _Py_memory_order_acq_rel: \
case _Py_memory_order_seq_cst: \
__asm__ volatile("xchg %0, %1" \
: "+r"(new_val) \
: "m"(atomic_val->_value) \
: "memory"); \
break; \
} \
_Py_ANNOTATE_IGNORE_WRITES_END(); \
})
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
__extension__ ({ \
__typeof__(ATOMIC_VAL) atomic_val = ATOMIC_VAL; \
__typeof__(atomic_val->_value) result; \
volatile __typeof__(result) *volatile_data = &atomic_val->_value; \
_Py_memory_order order = ORDER; \
_Py_ANNOTATE_MEMORY_ORDER(atomic_val, order); \
\
/* Perform the operation. */ \
_Py_ANNOTATE_IGNORE_READS_BEGIN(); \
switch(order) { \
case _Py_memory_order_release: \
case _Py_memory_order_acq_rel: \
case _Py_memory_order_seq_cst: \
/* Loads on x86 are not releases by default, so need a */ \
/* thread fence. */ \
_Py_atomic_thread_fence(_Py_memory_order_release); \
break; \
default: \
/* No fence */ \
break; \
} \
result = *volatile_data; \
switch(order) { \
case _Py_memory_order_acquire: \
case _Py_memory_order_acq_rel: \
case _Py_memory_order_seq_cst: \
/* Loads on x86 are automatically acquire operations so */ \
/* can get by with just a compiler fence. */ \
_Py_atomic_signal_fence(_Py_memory_order_acquire); \
break; \
default: \
/* No fence */ \
break; \
} \
_Py_ANNOTATE_IGNORE_READS_END(); \
result; \
})
#elif defined(_MSC_VER)
/* _Interlocked* functions provide a full memory barrier and are therefore
enough for acq_rel and seq_cst. If the HLE variants aren't available
in hardware they will fall back to a full memory barrier as well.
This might affect performance but likely only in some very specific and
hard to meassure scenario.
*/
#if defined(_M_IX86) || defined(_M_X64)
typedef enum _Py_memory_order {
_Py_memory_order_relaxed,
_Py_memory_order_acquire,
_Py_memory_order_release,
_Py_memory_order_acq_rel,
_Py_memory_order_seq_cst
} _Py_memory_order;
typedef struct _Py_atomic_address {
volatile uintptr_t _value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
volatile int _value;
} _Py_atomic_int;
#if defined(_M_X64)
#define _Py_atomic_store_64bit(ATOMIC_VAL, NEW_VAL, ORDER) \
switch (ORDER) { \
case _Py_memory_order_acquire: \
_InterlockedExchange64_HLEAcquire((__int64 volatile*)ATOMIC_VAL, (__int64)NEW_VAL); \
break; \
case _Py_memory_order_release: \
_InterlockedExchange64_HLERelease((__int64 volatile*)ATOMIC_VAL, (__int64)NEW_VAL); \
break; \
default: \
_InterlockedExchange64((__int64 volatile*)ATOMIC_VAL, (__int64)NEW_VAL); \
break; \
}
#else
#define _Py_atomic_store_64bit(ATOMIC_VAL, NEW_VAL, ORDER) ((void)0);
#endif
#define _Py_atomic_store_32bit(ATOMIC_VAL, NEW_VAL, ORDER) \
switch (ORDER) { \
case _Py_memory_order_acquire: \
_InterlockedExchange_HLEAcquire((volatile long*)ATOMIC_VAL, (int)NEW_VAL); \
break; \
case _Py_memory_order_release: \
_InterlockedExchange_HLERelease((volatile long*)ATOMIC_VAL, (int)NEW_VAL); \
break; \
default: \
_InterlockedExchange((volatile long*)ATOMIC_VAL, (int)NEW_VAL); \
break; \
}
#if defined(_M_X64)
/* This has to be an intptr_t for now.
gil_created() uses -1 as a sentinel value, if this returns
a uintptr_t it will do an unsigned compare and crash
*/
inline intptr_t _Py_atomic_load_64bit(volatile uintptr_t* value, int order) {
__int64 old;
switch (order) {
case _Py_memory_order_acquire:
{
do {
old = *value;
} while(_InterlockedCompareExchange64_HLEAcquire((volatile __int64*)value, old, old) != old);
break;
}
case _Py_memory_order_release:
{
do {
old = *value;
} while(_InterlockedCompareExchange64_HLERelease((volatile __int64*)value, old, old) != old);
break;
}
case _Py_memory_order_relaxed:
old = *value;
break;
default:
{
do {
old = *value;
} while(_InterlockedCompareExchange64((volatile __int64*)value, old, old) != old);
break;
}
}
return old;
}
#else
#define _Py_atomic_load_64bit(ATOMIC_VAL, ORDER) *(ATOMIC_VAL)
#endif
inline int _Py_atomic_load_32bit(volatile int* value, int order) {
long old;
switch (order) {
case _Py_memory_order_acquire:
{
do {
old = *value;
} while(_InterlockedCompareExchange_HLEAcquire((volatile long*)value, old, old) != old);
break;
}
case _Py_memory_order_release:
{
do {
old = *value;
} while(_InterlockedCompareExchange_HLERelease((volatile long*)value, old, old) != old);
break;
}
case _Py_memory_order_relaxed:
old = *value;
break;
default:
{
do {
old = *value;
} while(_InterlockedCompareExchange((volatile long*)value, old, old) != old);
break;
}
}
return old;
}
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
if (sizeof((ATOMIC_VAL)->_value) == 8) { \
_Py_atomic_store_64bit((volatile long long*)&((ATOMIC_VAL)->_value), NEW_VAL, ORDER) } else { \
_Py_atomic_store_32bit((volatile long*)&((ATOMIC_VAL)->_value), NEW_VAL, ORDER) }
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
( \
sizeof((ATOMIC_VAL)->_value) == 8 ? \
_Py_atomic_load_64bit((volatile long long*)&((ATOMIC_VAL)->_value), ORDER) : \
_Py_atomic_load_32bit((volatile long*)&((ATOMIC_VAL)->_value), ORDER) \
)
#elif defined(_M_ARM) || defined(_M_ARM64)
typedef enum _Py_memory_order {
_Py_memory_order_relaxed,
_Py_memory_order_acquire,
_Py_memory_order_release,
_Py_memory_order_acq_rel,
_Py_memory_order_seq_cst
} _Py_memory_order;
typedef struct _Py_atomic_address {
volatile uintptr_t _value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
volatile int _value;
} _Py_atomic_int;
#if defined(_M_ARM64)
#define _Py_atomic_store_64bit(ATOMIC_VAL, NEW_VAL, ORDER) \
switch (ORDER) { \
case _Py_memory_order_acquire: \
_InterlockedExchange64_acq((__int64 volatile*)&((ATOMIC_VAL)->_value), (__int64)NEW_VAL); \
break; \
case _Py_memory_order_release: \
_InterlockedExchange64_rel((__int64 volatile*)&((ATOMIC_VAL)->_value), (__int64)NEW_VAL); \
break; \
default: \
_InterlockedExchange64((__int64 volatile*)&((ATOMIC_VAL)->_value), (__int64)NEW_VAL); \
break; \
}
#else
#define _Py_atomic_store_64bit(ATOMIC_VAL, NEW_VAL, ORDER) ((void)0);
#endif
#define _Py_atomic_store_32bit(ATOMIC_VAL, NEW_VAL, ORDER) \
switch (ORDER) { \
case _Py_memory_order_acquire: \
_InterlockedExchange_acq((volatile long*)&((ATOMIC_VAL)->_value), (int)NEW_VAL); \
break; \
case _Py_memory_order_release: \
_InterlockedExchange_rel((volatile long*)&((ATOMIC_VAL)->_value), (int)NEW_VAL); \
break; \
default: \
_InterlockedExchange((volatile long*)&((ATOMIC_VAL)->_value), (int)NEW_VAL); \
break; \
}
#if defined(_M_ARM64)
/* This has to be an intptr_t for now.
gil_created() uses -1 as a sentinel value, if this returns
a uintptr_t it will do an unsigned compare and crash
*/
inline intptr_t _Py_atomic_load_64bit(volatile uintptr_t* value, int order) {
uintptr_t old;
switch (order) {
case _Py_memory_order_acquire:
{
do {
old = *value;
} while(_InterlockedCompareExchange64_acq(value, old, old) != old);
break;
}
case _Py_memory_order_release:
{
do {
old = *value;
} while(_InterlockedCompareExchange64_rel(value, old, old) != old);
break;
}
case _Py_memory_order_relaxed:
old = *value;
break;
default:
{
do {
old = *value;
} while(_InterlockedCompareExchange64(value, old, old) != old);
break;
}
}
return old;
}
#else
#define _Py_atomic_load_64bit(ATOMIC_VAL, ORDER) *(ATOMIC_VAL)
#endif
inline int _Py_atomic_load_32bit(volatile int* value, int order) {
int old;
switch (order) {
case _Py_memory_order_acquire:
{
do {
old = *value;
} while(_InterlockedCompareExchange_acq(value, old, old) != old);
break;
}
case _Py_memory_order_release:
{
do {
old = *value;
} while(_InterlockedCompareExchange_rel(value, old, old) != old);
break;
}
case _Py_memory_order_relaxed:
old = *value;
break;
default:
{
do {
old = *value;
} while(_InterlockedCompareExchange(value, old, old) != old);
break;
}
}
return old;
}
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
if (sizeof((ATOMIC_VAL)->_value) == 8) { \
_Py_atomic_store_64bit(&((ATOMIC_VAL)->_value), NEW_VAL, ORDER) } else { \
_Py_atomic_store_32bit(&((ATOMIC_VAL)->_value), NEW_VAL, ORDER) }
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
( \
sizeof((ATOMIC_VAL)->_value) == 8 ? \
_Py_atomic_load_64bit(&((ATOMIC_VAL)->_value), ORDER) : \
_Py_atomic_load_32bit(&((ATOMIC_VAL)->_value), ORDER) \
)
#endif
#else /* !gcc x86 !_msc_ver */
typedef enum _Py_memory_order {
_Py_memory_order_relaxed,
_Py_memory_order_acquire,
_Py_memory_order_release,
_Py_memory_order_acq_rel,
_Py_memory_order_seq_cst
} _Py_memory_order;
typedef struct _Py_atomic_address {
uintptr_t _value;
} _Py_atomic_address;
typedef struct _Py_atomic_int {
int _value;
} _Py_atomic_int;
/* Fall back to other compilers and processors by assuming that simple
volatile accesses are atomic. This is false, so people should port
this. */
#define _Py_atomic_signal_fence(/*memory_order*/ ORDER) ((void)0)
#define _Py_atomic_thread_fence(/*memory_order*/ ORDER) ((void)0)
#define _Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, ORDER) \
((ATOMIC_VAL)->_value = NEW_VAL)
#define _Py_atomic_load_explicit(ATOMIC_VAL, ORDER) \
((ATOMIC_VAL)->_value)
#endif
/* Standardized shortcuts. */
#define _Py_atomic_store(ATOMIC_VAL, NEW_VAL) \
_Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, _Py_memory_order_seq_cst)
#define _Py_atomic_load(ATOMIC_VAL) \
_Py_atomic_load_explicit(ATOMIC_VAL, _Py_memory_order_seq_cst)
/* Python-local extensions */
#define _Py_atomic_store_relaxed(ATOMIC_VAL, NEW_VAL) \
_Py_atomic_store_explicit(ATOMIC_VAL, NEW_VAL, _Py_memory_order_relaxed)
#define _Py_atomic_load_relaxed(ATOMIC_VAL) \
_Py_atomic_load_explicit(ATOMIC_VAL, _Py_memory_order_relaxed)
#ifdef __cplusplus
}
#endif
#endif /* Py_ATOMIC_H */