forked from cameron314/concurrentqueue
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconcurrentqueue.h
4642 lines (4184 loc) · 236 KB
/
concurrentqueue.h
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
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Provides a C++11 implementation of a multi-producer, multi-consumer lock-free queue.
// An overview, including benchmark results, is provided here:
// http://moodycamel.com/blog/2014/a-fast-general-purpose-lock-free-queue-for-c++
// The full design is also described in excruciating detail at:
// http://moodycamel.com/blog/2014/detailed-design-of-a-lock-free-queue
// Simplified BSD license:
// Copyright (c) 2013-2020, Cameron Desrochers.
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// - Redistributions of source code must retain the above copyright notice, this list of
// conditions and the following disclaimer.
// - Redistributions in binary form must reproduce the above copyright notice, this list of
// conditions and the following disclaimer in the documentation and/or other materials
// provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY
// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
// MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
// THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
// OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
// HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
// EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
// Also dual-licensed under the Boost Software License (see LICENSE.md)
#pragma once
#if defined(__GNUC__) && !defined(__INTEL_COMPILER)
// Disable -Wconversion warnings (spuriously triggered when Traits::size_t and
// Traits::index_t are set to < 32 bits, causing integer promotion, causing warnings
// upon assigning any computed values)
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wconversion"
#ifdef MCDBGQ_USE_RELACY
#pragma GCC diagnostic ignored "-Wint-to-pointer-cast"
#endif
#endif
#if defined(_MSC_VER) && (!defined(_HAS_CXX17) || !_HAS_CXX17)
// VS2019 with /W4 warns about constant conditional expressions but unless /std=c++17 or higher
// does not support `if constexpr`, so we have no choice but to simply disable the warning
#pragma warning(push)
#pragma warning(disable : 4127) // conditional expression is constant
#endif
#if defined(__APPLE__)
#include "TargetConditionals.h"
#endif
#ifdef MCDBGQ_USE_RELACY
#include "relacy/relacy_std.hpp"
#include "relacy_shims.h"
// We only use malloc/free anyway, and the delete macro messes up `= delete` method declarations.
// We'll override the default trait malloc ourselves without a macro.
#undef new
#undef delete
#undef malloc
#undef free
#else
#include <atomic> // Requires C++11. Sorry VS2010.
#include <cassert>
#endif
#include <cstddef> // for max_align_t
#include <cstdint>
#include <cstdlib>
#include <type_traits>
#include <algorithm>
#include <utility>
#include <limits>
#include <climits> // for CHAR_BIT
#include <array>
#include <thread> // partly for __WINPTHREADS_VERSION if on MinGW-w64 w/ POSIX threading
#include <iostream>
// Platform-specific definitions of a numeric thread ID type and an invalid value
namespace moodycamel
{
namespace details
{
template <typename thread_id_t>
struct thread_id_converter
{
typedef thread_id_t thread_id_numeric_size_t;
typedef thread_id_t thread_id_hash_t;
static thread_id_hash_t prehash(thread_id_t const &x) { return x; }
};
}
}
#if defined(MCDBGQ_USE_RELACY)
namespace moodycamel
{
namespace details
{
typedef std::uint32_t thread_id_t;
static const thread_id_t invalid_thread_id = 0xFFFFFFFFU;
static const thread_id_t invalid_thread_id2 = 0xFFFFFFFEU;
static inline thread_id_t thread_id() { return rl::thread_index(); }
}
}
#elif defined(_WIN32) || defined(__WINDOWS__) || defined(__WIN32__)
// No sense pulling in windows.h in a header, we'll manually declare the function
// we use and rely on backwards-compatibility for this not to break
extern "C" __declspec(dllimport) unsigned long __stdcall GetCurrentThreadId(void);
namespace moodycamel
{
namespace details
{
static_assert(sizeof(unsigned long) == sizeof(std::uint32_t), "Expected size of unsigned long to be 32 bits on Windows");
typedef std::uint32_t thread_id_t;
static const thread_id_t invalid_thread_id = 0; // See http://blogs.msdn.com/b/oldnewthing/archive/2004/02/23/78395.aspx
static const thread_id_t invalid_thread_id2 = 0xFFFFFFFFU; // Not technically guaranteed to be invalid, but is never used in practice. Note that all Win32 thread IDs are presently multiples of 4.
static inline thread_id_t thread_id() { return static_cast<thread_id_t>(::GetCurrentThreadId()); }
}
}
#elif defined(__arm__) || defined(_M_ARM) || defined(__aarch64__) || (defined(__APPLE__) && TARGET_OS_IPHONE) || defined(MOODYCAMEL_NO_THREAD_LOCAL)
namespace moodycamel
{
namespace details
{
static_assert(sizeof(std::thread::id) == 4 || sizeof(std::thread::id) == 8, "std::thread::id is expected to be either 4 or 8 bytes");
typedef std::thread::id thread_id_t;
static const thread_id_t invalid_thread_id; // Default ctor creates invalid ID
// Note we don't define a invalid_thread_id2 since std::thread::id doesn't have one; it's
// only used if MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED is defined anyway, which it won't
// be.
static inline thread_id_t thread_id() { return std::this_thread::get_id(); }
template <std::size_t>
struct thread_id_size
{
};
template <>
struct thread_id_size<4>
{
typedef std::uint32_t numeric_t;
};
template <>
struct thread_id_size<8>
{
typedef std::uint64_t numeric_t;
};
template <>
struct thread_id_converter<thread_id_t>
{
typedef thread_id_size<sizeof(thread_id_t)>::numeric_t thread_id_numeric_size_t;
#ifndef __APPLE__
typedef std::size_t thread_id_hash_t;
#else
typedef thread_id_numeric_size_t thread_id_hash_t;
#endif
static thread_id_hash_t prehash(thread_id_t const &x)
{
#ifndef __APPLE__
return std::hash<std::thread::id>()(x);
#else
return *reinterpret_cast<thread_id_hash_t const *>(&x);
#endif
}
};
}
}
#else
// Use a nice trick from this answer: http://stackoverflow.com/a/8438730/21475
// In order to get a numeric thread ID in a platform-independent way, we use a thread-local
// static variable's address as a thread identifier :-)
#if defined(__GNUC__) || defined(__INTEL_COMPILER)
#define MOODYCAMEL_THREADLOCAL __thread
#elif defined(_MSC_VER)
#define MOODYCAMEL_THREADLOCAL __declspec(thread)
#else
// Assume C++11 compliant compiler
#define MOODYCAMEL_THREADLOCAL thread_local
#endif
namespace moodycamel
{
namespace details
{
typedef std::uintptr_t thread_id_t;
static const thread_id_t invalid_thread_id = 0; // Address can't be nullptr
static const thread_id_t invalid_thread_id2 = 1; // Member accesses off a null pointer are also generally invalid. Plus it's not aligned.
inline thread_id_t thread_id()
{
static MOODYCAMEL_THREADLOCAL int x;
return reinterpret_cast<thread_id_t>(&x);
}
}
}
#endif
// Constexpr if
#ifndef MOODYCAMEL_CONSTEXPR_IF
#if (defined(_MSC_VER) && defined(_HAS_CXX17) && _HAS_CXX17) || __cplusplus > 201402L
#define MOODYCAMEL_CONSTEXPR_IF if constexpr
#define MOODYCAMEL_MAYBE_UNUSED [[maybe_unused]]
#else
#define MOODYCAMEL_CONSTEXPR_IF if
#define MOODYCAMEL_MAYBE_UNUSED
#endif
#endif
// Exceptions
#ifndef MOODYCAMEL_EXCEPTIONS_ENABLED
#if (defined(_MSC_VER) && defined(_CPPUNWIND)) || (defined(__GNUC__) && defined(__EXCEPTIONS)) || (!defined(_MSC_VER) && !defined(__GNUC__))
#define MOODYCAMEL_EXCEPTIONS_ENABLED
#endif
#endif
#ifdef MOODYCAMEL_EXCEPTIONS_ENABLED
#define MOODYCAMEL_TRY try
#define MOODYCAMEL_CATCH(...) catch (__VA_ARGS__)
#define MOODYCAMEL_RETHROW throw
#define MOODYCAMEL_THROW(expr) throw(expr)
#else
#define MOODYCAMEL_TRY MOODYCAMEL_CONSTEXPR_IF(true)
#define MOODYCAMEL_CATCH(...) else MOODYCAMEL_CONSTEXPR_IF(false)
#define MOODYCAMEL_RETHROW
#define MOODYCAMEL_THROW(expr)
#endif
#ifndef MOODYCAMEL_NOEXCEPT
#if !defined(MOODYCAMEL_EXCEPTIONS_ENABLED)
#define MOODYCAMEL_NOEXCEPT
#define MOODYCAMEL_NOEXCEPT_CTOR(type, valueType, expr) true
#define MOODYCAMEL_NOEXCEPT_ASSIGN(type, valueType, expr) true
#elif defined(_MSC_VER) && defined(_NOEXCEPT) && _MSC_VER < 1800
// VS2012's std::is_nothrow_[move_]constructible is broken and returns true when it shouldn't :-(
// We have to assume *all* non-trivial constructors may throw on VS2012!
#define MOODYCAMEL_NOEXCEPT _NOEXCEPT
#define MOODYCAMEL_NOEXCEPT_CTOR(type, valueType, expr) (std::is_rvalue_reference<valueType>::value && std::is_move_constructible<type>::value ? std::is_trivially_move_constructible<type>::value : std::is_trivially_copy_constructible<type>::value)
#define MOODYCAMEL_NOEXCEPT_ASSIGN(type, valueType, expr) ((std::is_rvalue_reference<valueType>::value && std::is_move_assignable<type>::value ? std::is_trivially_move_assignable<type>::value || std::is_nothrow_move_assignable<type>::value : std::is_trivially_copy_assignable<type>::value || std::is_nothrow_copy_assignable<type>::value) && MOODYCAMEL_NOEXCEPT_CTOR(type, valueType, expr))
#elif defined(_MSC_VER) && defined(_NOEXCEPT) && _MSC_VER < 1900
#define MOODYCAMEL_NOEXCEPT _NOEXCEPT
#define MOODYCAMEL_NOEXCEPT_CTOR(type, valueType, expr) (std::is_rvalue_reference<valueType>::value && std::is_move_constructible<type>::value ? std::is_trivially_move_constructible<type>::value || std::is_nothrow_move_constructible<type>::value : std::is_trivially_copy_constructible<type>::value || std::is_nothrow_copy_constructible<type>::value)
#define MOODYCAMEL_NOEXCEPT_ASSIGN(type, valueType, expr) ((std::is_rvalue_reference<valueType>::value && std::is_move_assignable<type>::value ? std::is_trivially_move_assignable<type>::value || std::is_nothrow_move_assignable<type>::value : std::is_trivially_copy_assignable<type>::value || std::is_nothrow_copy_assignable<type>::value) && MOODYCAMEL_NOEXCEPT_CTOR(type, valueType, expr))
#else
#define MOODYCAMEL_NOEXCEPT noexcept
#define MOODYCAMEL_NOEXCEPT_CTOR(type, valueType, expr) noexcept(expr)
#define MOODYCAMEL_NOEXCEPT_ASSIGN(type, valueType, expr) noexcept(expr)
#endif
#endif
#ifndef MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED
#ifdef MCDBGQ_USE_RELACY
#define MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED
#else
// VS2013 doesn't support `thread_local`, and MinGW-w64 w/ POSIX threading has a crippling bug: http://sourceforge.net/p/mingw-w64/bugs/445
// g++ <=4.7 doesn't support thread_local either.
// Finally, iOS/ARM doesn't have support for it either, and g++/ARM allows it to compile but it's unconfirmed to actually work
#if (!defined(_MSC_VER) || _MSC_VER >= 1900) && (!defined(__MINGW32__) && !defined(__MINGW64__) || !defined(__WINPTHREADS_VERSION)) && (!defined(__GNUC__) || __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) && (!defined(__APPLE__) || !TARGET_OS_IPHONE) && !defined(__arm__) && !defined(_M_ARM) && !defined(__aarch64__)
// Assume `thread_local` is fully supported in all other C++11 compilers/platforms
//#define MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED // always disabled for now since several users report having problems with it on
#endif
#endif
#endif
// VS2012 doesn't support deleted functions.
// In this case, we declare the function normally but don't define it. A link error will be generated if the function is called.
#ifndef MOODYCAMEL_DELETE_FUNCTION
#if defined(_MSC_VER) && _MSC_VER < 1800
#define MOODYCAMEL_DELETE_FUNCTION
#else
#define MOODYCAMEL_DELETE_FUNCTION = delete
#endif
#endif
namespace moodycamel
{
namespace details
{
#ifndef MOODYCAMEL_ALIGNAS
// VS2013 doesn't support alignas or alignof, and align() requires a constant literal
#if defined(_MSC_VER) && _MSC_VER <= 1800
#define MOODYCAMEL_ALIGNAS(alignment) __declspec(align(alignment))
#define MOODYCAMEL_ALIGNOF(obj) __alignof(obj)
#define MOODYCAMEL_ALIGNED_TYPE_LIKE(T, obj) typename details::Vs2013Aligned<std::alignment_of<obj>::value, T>::type
template <int Align, typename T>
struct Vs2013Aligned
{
}; // default, unsupported alignment
template <typename T>
struct Vs2013Aligned<1, T>
{
typedef __declspec(align(1)) T type;
};
template <typename T>
struct Vs2013Aligned<2, T>
{
typedef __declspec(align(2)) T type;
};
template <typename T>
struct Vs2013Aligned<4, T>
{
typedef __declspec(align(4)) T type;
};
template <typename T>
struct Vs2013Aligned<8, T>
{
typedef __declspec(align(8)) T type;
};
template <typename T>
struct Vs2013Aligned<16, T>
{
typedef __declspec(align(16)) T type;
};
template <typename T>
struct Vs2013Aligned<32, T>
{
typedef __declspec(align(32)) T type;
};
template <typename T>
struct Vs2013Aligned<64, T>
{
typedef __declspec(align(64)) T type;
};
template <typename T>
struct Vs2013Aligned<128, T>
{
typedef __declspec(align(128)) T type;
};
template <typename T>
struct Vs2013Aligned<256, T>
{
typedef __declspec(align(256)) T type;
};
#else
template <typename T>
struct identity
{
typedef T type;
};
#define MOODYCAMEL_ALIGNAS(alignment) alignas(alignment)
#define MOODYCAMEL_ALIGNOF(obj) alignof(obj)
#define MOODYCAMEL_ALIGNED_TYPE_LIKE(T, obj) alignas(alignof(obj)) typename details::identity<T>::type
#endif
#endif
}
}
// TSAN can false report races in lock-free code. To enable TSAN to be used from projects that use this one,
// we can apply per-function compile-time suppression.
// See https://clang.llvm.org/docs/ThreadSanitizer.html#has-feature-thread-sanitizer
#define MOODYCAMEL_NO_TSAN
#if defined(__has_feature)
#if __has_feature(thread_sanitizer)
#undef MOODYCAMEL_NO_TSAN
#define MOODYCAMEL_NO_TSAN __attribute__((no_sanitize("thread")))
#endif // TSAN
#endif // TSAN
// Compiler-specific likely/unlikely hints
namespace moodycamel
{
namespace details
{
#if defined(__GNUC__)
static inline bool(likely)(bool x)
{
return __builtin_expect((x), true);
}
static inline bool(unlikely)(bool x) { return __builtin_expect((x), false); }
#else
static inline bool(likely)(bool x)
{
return x;
}
static inline bool(unlikely)(bool x) { return x; }
#endif
}
}
#ifdef MOODYCAMEL_QUEUE_INTERNAL_DEBUG
#include "internal/concurrentqueue_internal_debug.h"
#endif
namespace moodycamel
{
namespace details
{
template <typename T>
struct const_numeric_max
{
static_assert(std::is_integral<T>::value, "const_numeric_max can only be used with integers");
static const T value = std::numeric_limits<T>::is_signed
? (static_cast<T>(1) << (sizeof(T) * CHAR_BIT - 1)) - static_cast<T>(1)
: static_cast<T>(-1);
};
#if defined(__GLIBCXX__)
typedef ::max_align_t std_max_align_t; // libstdc++ forgot to add it to std:: for a while
#else
typedef std::max_align_t std_max_align_t; // Others (e.g. MSVC) insist it can *only* be accessed via std::
#endif
// Some platforms have incorrectly set max_align_t to a type with <8 bytes alignment even while supporting
// 8-byte aligned scalar values (*cough* 32-bit iOS). Work around this with our own union. See issue #64.
typedef union
{
std_max_align_t x;
long long y;
void *z;
} max_align_t;
}
// Default traits for the ConcurrentQueue. To change some of the
// traits without re-implementing all of them, inherit from this
// struct and shadow the declarations you wish to be different;
// since the traits are used as a template type parameter, the
// shadowed declarations will be used where defined, and the defaults
// otherwise.
struct ConcurrentQueueDefaultTraits
{
// General-purpose size type. std::size_t is strongly recommended.
typedef std::size_t size_t;
// The type used for the enqueue and dequeue indices. Must be at least as
// large as size_t. Should be significantly larger than the number of elements
// you expect to hold at once, especially if you have a high turnover rate;
// for example, on 32-bit x86, if you expect to have over a hundred million
// elements or pump several million elements through your queue in a very
// short space of time, using a 32-bit type *may* trigger a race condition.
// A 64-bit int type is recommended in that case, and in practice will
// prevent a race condition no matter the usage of the queue. Note that
// whether the queue is lock-free with a 64-int type depends on the whether
// std::atomic<std::uint64_t> is lock-free, which is platform-specific.
/* *
* 这个类型被用作入队和出队的索引。至少要和 size_t 一样大
* 如果你希望队列有一个很大的吞吐量，则需要将这个值设置的比队列的容量大得多
* 注意，64-int类型的队列是否无锁取决于std::atomic<std::uint64_t>是否无锁，这是特定于平台的
* */
typedef std::size_t index_t;
// Internally, all elements are enqueued and dequeued from multi-element
// blocks; this is the smallest controllable unit. If you expect few elements
// but many producers, a smaller block size should be favoured. For few producers
// and/or many elements, a larger block size is preferred. A sane default
// is provided. Must be a power of 2.
/* *
* 每个 producer 都有一个 SPMC 队列，每个队列用由多个 block 组成。
* 如果元素少而 producer 多，则可以将 block 的大小设置小一些
* 如果 producer 少但是元素多，可以将 block 的大小调大
* */
static const size_t BLOCK_SIZE = 32;
// For explicit producers (i.e. when using a producer token), the block is
// checked for being empty by iterating through a list of flags, one per element.
// For large block sizes, this is too inefficient, and switching to an atomic
// counter-based approach is faster. The switch is made for block sizes strictly
// larger than this threshold.
/* *
* 检查 block 是否为空的阈值：
* 如果 block size 小于这个阈值，则通过遍历每一个元素的 flag 来判断是否 block 为空
* 如果大于这个阈值,则使用基于 atomic 的计数器来判断
* */
static const size_t EXPLICIT_BLOCK_EMPTY_COUNTER_THRESHOLD = 32;
// How many full blocks can be expected for a single explicit producer? This should
// reflect that number's maximum for optimal performance. Must be a power of 2.
// 一个显式生产者可以有多少完整的 block。这是能够发挥最佳性能的最大数值
static const size_t EXPLICIT_INITIAL_INDEX_SIZE = 32;
// How many full blocks can be expected for a single implicit producer? This should
// reflect that number's maximum for optimal performance. Must be a power of 2.
// 一个隐式生产者最多可以拥有的 block 的数量（每个 block index 中初始 index 的大小，即每个 block index 中可以最多拥有的 block 的数量）
// 隐式生产者用循环缓冲区来管理 block index（缓冲区的每个元素都是一个指向 block 的索引），所以也可以理解成一个隐式生产者的循环缓冲区的大小
static const size_t IMPLICIT_INITIAL_INDEX_SIZE = 32;
// The initial size of the hash table mapping thread IDs to implicit producers.
// Note that the hash is resized every time it becomes half full.
// Must be a power of two, and either 0 or at least 1. If 0, implicit production
// (using the enqueue methods without an explicit producer token) is disabled.
// hash 表的初始大小（将线程 ID 映射到 隐式生产者队列的 hash 表）
// 每次哈希表为半满时将会调整哈希表的大小
// 必须是2的倍数，至少是0或者1
// 如果是0，则禁用隐式生产者
// 隐式生产者使用队列方法而不是用显示的生产者令牌
static const size_t INITIAL_IMPLICIT_PRODUCER_HASH_SIZE = 32;
// Controls the number of items that an explicit consumer (i.e. one with a token)
// must consume before it causes all consumers to rotate and move on to the next
// internal queue.
// 控制显式消费者（即带有令牌的消费者）在导致所有消费者旋转并移动到下一个内部队列之前必须消费的数量
static const std::uint32_t EXPLICIT_CONSUMER_CONSUMPTION_QUOTA_BEFORE_ROTATE = 256;
// The maximum number of elements (inclusive) that can be enqueued to a sub-queue.
// Enqueue operations that would cause this limit to be surpassed will fail. Note
// that this limit is enforced at the block level (for performance reasons), i.e.
// it's rounded up to the nearest block size.
/*
* 子队列中可以容纳的最大数量
* 超过此限制的队列操作将失败
* 注意，这个限制是在 block 级别执行的(出于性能原因)，也就是说，它被舍入到最接近的块大小。
**/
static const size_t MAX_SUBQUEUE_SIZE = details::const_numeric_max<size_t>::value;
// The number of times to spin before sleeping when waiting on a semaphore.
// Recommended values are on the order of 1000-10000 unless the number of
// consumer threads exceeds the number of idle cores (in which case try 0-100).
// Only affects instances of the BlockingConcurrentQueue.
/* 等待信号量时自旋的次数
* 建议的值是1000-10000，除非消费线程的数量超过空闲核的数量(在这种情况下尝试0-100)。
* 仅影响 BlockingConcurrentQueue 实例
* */
static const int MAX_SEMA_SPINS = 10000;
#ifndef MCDBGQ_USE_RELACY
// Memory allocation can be customized if needed.
// malloc should return nullptr on failure, and handle alignment like std::malloc.
#if defined(malloc) || defined(free)
// Gah, this is 2015, stop defining macros that break standard code already!
// Work around malloc/free being special macros:
static inline void *WORKAROUND_malloc(size_t size) { return malloc(size); }
static inline void WORKAROUND_free(void *ptr) { return free(ptr); }
static inline void *(malloc)(size_t size) { return WORKAROUND_malloc(size); }
static inline void(free)(void *ptr) { return WORKAROUND_free(ptr); }
#else
static inline void *malloc(size_t size)
{
return std::malloc(size);
}
static inline void free(void *ptr) { return std::free(ptr); }
#endif
#else
// Debug versions when running under the Relacy race detector (ignore
// these in user code)
static inline void *malloc(size_t size) { return rl::rl_malloc(size, $); }
static inline void free(void *ptr) { return rl::rl_free(ptr, $); }
#endif
};
// When producing or consuming many elements, the most efficient way is to:
// 1) Use one of the bulk-operation methods of the queue with a token
// 2) Failing that, use the bulk-operation methods without a token
// 3) Failing that, create a token and use that with the single-item methods
// 4) Failing that, use the single-parameter methods of the queue
// Having said that, don't create tokens willy-nilly -- ideally there should be
// a maximum of one token per thread (of each kind).
struct ProducerToken;
struct ConsumerToken;
template <typename T, typename Traits>
class ConcurrentQueue;
template <typename T, typename Traits>
class BlockingConcurrentQueue;
class ConcurrentQueueTests;
namespace details
{
struct ConcurrentQueueProducerTypelessBase
{
ConcurrentQueueProducerTypelessBase *next;
std::atomic<bool> inactive;
ProducerToken *token;
ConcurrentQueueProducerTypelessBase()
: next(nullptr), inactive(false), token(nullptr)
{
}
};
template <bool use32>
struct _hash_32_or_64
{
static inline std::uint32_t hash(std::uint32_t h)
{
// MurmurHash3 finalizer -- see https://code.google.com/p/smhasher/source/browse/trunk/MurmurHash3.cpp
// Since the thread ID is already unique, all we really want to do is propagate that
// uniqueness evenly across all the bits, so that we can use a subset of the bits while
// reducing collisions significantly
h ^= h >> 16;
h *= 0x85ebca6b;
h ^= h >> 13;
h *= 0xc2b2ae35;
return h ^ (h >> 16);
}
};
template <>
struct _hash_32_or_64<1>
{
static inline std::uint64_t hash(std::uint64_t h)
{
h ^= h >> 33;
h *= 0xff51afd7ed558ccd;
h ^= h >> 33;
h *= 0xc4ceb9fe1a85ec53;
return h ^ (h >> 33);
}
};
template <std::size_t size>
struct hash_32_or_64 : public _hash_32_or_64<(size > 4)>
{
};
static inline size_t hash_thread_id(thread_id_t id)
{
static_assert(sizeof(thread_id_t) <= 8, "Expected a platform where thread IDs are at most 64-bit values");
return static_cast<size_t>(hash_32_or_64<sizeof(thread_id_converter<thread_id_t>::thread_id_hash_t)>::hash(
thread_id_converter<thread_id_t>::prehash(id)));
}
template <typename T>
static inline bool circular_less_than(T a, T b)
{
#ifdef _MSC_VER
#pragma warning(push)
#pragma warning(disable : 4554)
#endif
static_assert(std::is_integral<T>::value && !std::numeric_limits<T>::is_signed, "circular_less_than is intended to be used only with unsigned integer types");
// CHAR_BIT 用来表示一个字符对象所占用的 BIT 数（实际取决于编译器体系结构或者库的实现）
// std::cout << "static_cast<T>(a - b) = " << static_cast<T>(a - b) << std::endl;
// std::cout << "sizeof(T) = " << sizeof(T) << std::endl;
// std::cout << "CHAR_BIT = " << CHAR_BIT << std::endl;
return static_cast<T>(a - b) > static_cast<T>(static_cast<T>(1) << static_cast<T>(sizeof(T) * CHAR_BIT - 1));
#ifdef _MSC_VER
#pragma warning(pop)
#endif
}
template <typename U>
static inline char *align_for(char *ptr)
{
const std::size_t alignment = std::alignment_of<U>::value;
return ptr + (alignment - (reinterpret_cast<std::uintptr_t>(ptr) % alignment)) % alignment;
}
template <typename T>
static inline T ceil_to_pow_2(T x)
{
static_assert(std::is_integral<T>::value && !std::numeric_limits<T>::is_signed, "ceil_to_pow_2 is intended to be used only with unsigned integer types");
// Adapted from http://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2
--x;
x |= x >> 1;
x |= x >> 2;
x |= x >> 4;
for (std::size_t i = 1; i < sizeof(T); i <<= 1)
{
x |= x >> (i << 3);
}
++x;
return x;
}
template <typename T>
static inline void swap_relaxed(std::atomic<T> &left, std::atomic<T> &right)
{
T temp = std::move(left.load(std::memory_order_relaxed));
left.store(std::move(right.load(std::memory_order_relaxed)), std::memory_order_relaxed);
right.store(std::move(temp), std::memory_order_relaxed);
}
template <typename T>
static inline T const &nomove(T const &x)
{
return x;
}
template <bool Enable>
struct nomove_if
{
template <typename T>
static inline T const &eval(T const &x)
{
return x;
}
};
template <>
struct nomove_if<false>
{
template <typename U>
static inline auto eval(U &&x)
-> decltype(std::forward<U>(x))
{
return std::forward<U>(x);
}
};
template <typename It>
static inline auto deref_noexcept(It &it) MOODYCAMEL_NOEXCEPT -> decltype(*it)
{
return *it;
}
#if defined(__clang__) || !defined(__GNUC__) || __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)
template <typename T>
struct is_trivially_destructible : std::is_trivially_destructible<T>
{
};
#else
template <typename T>
struct is_trivially_destructible : std::has_trivial_destructor<T>
{
};
#endif
#ifdef MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED
#ifdef MCDBGQ_USE_RELACY
typedef RelacyThreadExitListener ThreadExitListener;
typedef RelacyThreadExitNotifier ThreadExitNotifier;
#else
struct ThreadExitListener
{
typedef void (*callback_t)(void *);
callback_t callback;
void *userData;
ThreadExitListener *next; // reserved for use by the ThreadExitNotifier
};
class ThreadExitNotifier
{
public:
static void subscribe(ThreadExitListener *listener)
{
auto &tlsInst = instance();
listener->next = tlsInst.tail;
tlsInst.tail = listener;
}
static void unsubscribe(ThreadExitListener *listener)
{
auto &tlsInst = instance();
ThreadExitListener **prev = &tlsInst.tail;
for (auto ptr = tlsInst.tail; ptr != nullptr; ptr = ptr->next)
{
if (ptr == listener)
{
*prev = ptr->next;
break;
}
prev = &ptr->next;
}
}
private:
ThreadExitNotifier() : tail(nullptr) {}
ThreadExitNotifier(ThreadExitNotifier const &) MOODYCAMEL_DELETE_FUNCTION;
ThreadExitNotifier &operator=(ThreadExitNotifier const &) MOODYCAMEL_DELETE_FUNCTION;
~ThreadExitNotifier()
{
// This thread is about to exit, let everyone know!
assert(this == &instance() && "If this assert fails, you likely have a buggy compiler! Change the preprocessor conditions such that MOODYCAMEL_CPP11_THREAD_LOCAL_SUPPORTED is no longer defined.");
for (auto ptr = tail; ptr != nullptr; ptr = ptr->next)
{
ptr->callback(ptr->userData);
}
}
// Thread-local
static inline ThreadExitNotifier &instance()
{
static thread_local ThreadExitNotifier notifier;
return notifier;
}
private:
ThreadExitListener *tail;
};
#endif
#endif
template <typename T>
struct static_is_lock_free_num
{
enum
{
value = 0
};
};
template <>
struct static_is_lock_free_num<signed char>
{
enum
{
value = ATOMIC_CHAR_LOCK_FREE
};
};
template <>
struct static_is_lock_free_num<short>
{
enum
{
value = ATOMIC_SHORT_LOCK_FREE
};
};
template <>
struct static_is_lock_free_num<int>
{
enum
{
value = ATOMIC_INT_LOCK_FREE
};
};
template <>
struct static_is_lock_free_num<long>
{
enum
{
value = ATOMIC_LONG_LOCK_FREE
};
};
template <>
struct static_is_lock_free_num<long long>
{
enum
{
value = ATOMIC_LLONG_LOCK_FREE
};
};
template <typename T>
struct static_is_lock_free : static_is_lock_free_num<typename std::make_signed<T>::type>
{
};
template <>
struct static_is_lock_free<bool>
{
enum
{
value = ATOMIC_BOOL_LOCK_FREE
};
};
template <typename U>
struct static_is_lock_free<U *>
{
enum
{
value = ATOMIC_POINTER_LOCK_FREE
};
};
}
struct ProducerToken
{
template <typename T, typename Traits>
explicit ProducerToken(ConcurrentQueue<T, Traits> &queue);
template <typename T, typename Traits>
explicit ProducerToken(BlockingConcurrentQueue<T, Traits> &queue);
ProducerToken(ProducerToken &&other) MOODYCAMEL_NOEXCEPT
: producer(other.producer)
{
other.producer = nullptr;
if (producer != nullptr)
{
producer->token = this;
}
}
inline ProducerToken &operator=(ProducerToken &&other) MOODYCAMEL_NOEXCEPT
{
swap(other);
return *this;
}
void swap(ProducerToken &other) MOODYCAMEL_NOEXCEPT
{
std::swap(producer, other.producer);
if (producer != nullptr)
{
producer->token = this;
}
if (other.producer != nullptr)
{
other.producer->token = &other;
}
}
// A token is always valid unless:
// 1) Memory allocation failed during construction
// 2) It was moved via the move constructor
// (Note: assignment does a swap, leaving both potentially valid)
// 3) The associated queue was destroyed
// Note that if valid() returns true, that only indicates
// that the token is valid for use with a specific queue,
// but not which one; that's up to the user to track.
inline bool valid() const { return producer != nullptr; }
~ProducerToken()
{
if (producer != nullptr)
{
producer->token = nullptr;
producer->inactive.store(true, std::memory_order_release);
}
}
// Disable copying and assignment
ProducerToken(ProducerToken const &) MOODYCAMEL_DELETE_FUNCTION;
ProducerToken &operator=(ProducerToken const &) MOODYCAMEL_DELETE_FUNCTION;
private:
template <typename T, typename Traits>
friend class ConcurrentQueue;
friend class ConcurrentQueueTests;
protected:
details::ConcurrentQueueProducerTypelessBase *producer;
};
struct ConsumerToken
{
template <typename T, typename Traits>
explicit ConsumerToken(ConcurrentQueue<T, Traits> &q);
template <typename T, typename Traits>
explicit ConsumerToken(BlockingConcurrentQueue<T, Traits> &q);
ConsumerToken(ConsumerToken &&other) MOODYCAMEL_NOEXCEPT
: initialOffset(other.initialOffset),
lastKnownGlobalOffset(other.lastKnownGlobalOffset),
itemsConsumedFromCurrent(other.itemsConsumedFromCurrent),
currentProducer(other.currentProducer),
desiredProducer(other.desiredProducer)
{
}
inline ConsumerToken &operator=(ConsumerToken &&other) MOODYCAMEL_NOEXCEPT
{
swap(other);
return *this;
}
void swap(ConsumerToken &other) MOODYCAMEL_NOEXCEPT
{
std::swap(initialOffset, other.initialOffset);
std::swap(lastKnownGlobalOffset, other.lastKnownGlobalOffset);
std::swap(itemsConsumedFromCurrent, other.itemsConsumedFromCurrent);
std::swap(currentProducer, other.currentProducer);
std::swap(desiredProducer, other.desiredProducer);
}
// Disable copying and assignment
ConsumerToken(ConsumerToken const &) MOODYCAMEL_DELETE_FUNCTION;
ConsumerToken &operator=(ConsumerToken const &) MOODYCAMEL_DELETE_FUNCTION;
private:
template <typename T, typename Traits>
friend class ConcurrentQueue;
friend class ConcurrentQueueTests;
private: // but shared with ConcurrentQueue
std::uint32_t initialOffset;
std::uint32_t lastKnownGlobalOffset;
std::uint32_t itemsConsumedFromCurrent;
details::ConcurrentQueueProducerTypelessBase *currentProducer;
details::ConcurrentQueueProducerTypelessBase *desiredProducer;
};
// Need to forward-declare this swap because it's in a namespace.
// See http://stackoverflow.com/questions/4492062/why-does-a-c-friend-class-need-a-forward-declaration-only-in-other-namespaces
template <typename T, typename Traits>
inline void swap(typename ConcurrentQueue<T, Traits>::ImplicitProducerKVP &a, typename ConcurrentQueue<T, Traits>::ImplicitProducerKVP &b) MOODYCAMEL_NOEXCEPT;
/* design:
* 每个生产者都需要一些线程本地数据，线程本地数据也可以用来提高消费者的速度。
* 线程本地数据可以与用户分配的令牌相关联，
* 如果用户没有为生产者提供令牌，则使用一个无锁哈希表(键值为当前线程ID)来查找线程本地生产者队列，
* 为每个显式分配的生产者令牌创建一个显式队列，并为生产者而不提供令牌的每个线程创建另一个隐式队列。
* 由于令牌包含的数据相当于特定于线程的数据，因此不应该在多个线程中同时使用它们(尽管可以将令牌的所有权转移到另一个线程; 特别是，这允许在线程池任务中使用令牌，即使运行任务的线程中途发生了更改)。
*
* 所有的生产者队列都将自己链接到一个无锁链表中。当一个显示生产者不再需要添加元素时（即它的令牌被销毁），它会被标记为与任何生产者无关，但是它仍然被保存在列表中，并不会进行内存释放。下一个新的生产者将
* 重用旧的生产者内存（无锁生产者列表只能通过这种方式增加）。隐式生产者永远不会被销毁（直到高级队列本身被销毁），因为无法知道是否有线程正在使用数据结构。
* 注意，最坏的情况下，退出队列的速度取决于有多少生产者队列，即时他们都是空的
*
* 显式队列和隐式队列的生命周期的根本区别在于：显式队列的生命周期和令牌的生命周期相关联；隐式队列的生命周期和高级队列的生命周期相关联
* 使用了两种略有不同的 SPMC 算法，以最大化速度和内存使用量。
* 显式生产者队列设计的稍微快一些，内存占用稍微大一些；隐式生产者队列设计的稍微慢一些，但是会将更多的内存回收到高级队列的内存池中
*
* 任何分配到的内存只有在高级队列被销毁时才会被释放。内存分配可以提前完成，如果没有足够的内存，操作就会失败。各种默认大小参数以及队列使用的内存分配函数可以由用户重写
* **/
/* 显示队列和隐式队列的共享设计：
* 1. 生产者队列由 block 组成，可以抽象认为是一个无限数组，有很多 slot。维护着两个变量：tail index 和 head index
* 1.1 tail index 表示下一个即将入队的 slot。tail index 总是增加（除非溢出或者环绕）。入队操作只有一个线程在更新对应的变量
* 1.2 head index 即将出队的 slot。出队操作可能是并发的，由多个消费者进行原子性的递增。
* 2.tail 和 head 的索引计数最终会溢出（预期之内的）
*/
/* block pool design:
* 这里使用了两个不同的 block pool:
* 1. 首先，有预分配块的初始数组。一旦消耗殆尽，这个池子将永远空着。简化了 wait-free 的实现， fetch-and-add（获取空闲块的下一个索引）和检查(确保该索引在范围内)
* 2. 其次，有一个 lock-free 的全局 free-list（这里的全局指的是对高级队列的全局）。它是 lock-free 的，但不是 wait-free 的。
* free-list 元素是一些准备被重用的已用 block。具体实现是一个 lock-free 的单链表，头指针初始化为 null。
* free-list 的 add 操作：block 的 next 指针被设置为 头指针，然后头指针在头没有改变的情况下使用 CAS 操作更新为指向 block，
* free-list 的 remove 操作：读取 head block 的 next 指针，然后将 head 设置为 next（使用 CAS），前提条件是 head 在此期间没有发生改变。
*
* free-list 避免 ABA 问题：https://moodycamel.com/blog/2014/solving-the-aba-problem-for-lock-free-free-lists.htm
*/