File: atomicops.h

package info (click to toggle)
readerwriterqueue 1.0.3-1
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 388 kB
  • sloc: cpp: 2,876; makefile: 79
file content (679 lines) | stat: -rw-r--r-- 20,046 bytes parent folder | download
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
// ©2013-2016 Cameron Desrochers.
// Distributed under the simplified BSD license (see the license file that
// should have come with this header).
// Uses Jeff Preshing's semaphore implementation (under the terms of its
// separate zlib license, embedded below).

#pragma once

// Provides portable (VC++2010+, Intel ICC 13, GCC 4.7+, and anything C++11 compliant) implementation
// of low-level memory barriers, plus a few semi-portable utility macros (for inlining and alignment).
// Also has a basic atomic type (limited to hardware-supported atomics with no memory ordering guarantees).
// Uses the AE_* prefix for macros (historical reasons), and the "moodycamel" namespace for symbols.

#include <cerrno>
#include <cassert>
#include <type_traits>
#include <cerrno>
#include <cstdint>
#include <ctime>

// Platform detection
#if defined(__INTEL_COMPILER)
#define AE_ICC
#elif defined(_MSC_VER)
#define AE_VCPP
#elif defined(__GNUC__)
#define AE_GCC
#endif

#if defined(_M_IA64) || defined(__ia64__)
#define AE_ARCH_IA64
#elif defined(_WIN64) || defined(__amd64__) || defined(_M_X64) || defined(__x86_64__)
#define AE_ARCH_X64
#elif defined(_M_IX86) || defined(__i386__)
#define AE_ARCH_X86
#elif defined(_M_PPC) || defined(__powerpc__)
#define AE_ARCH_PPC
#else
#define AE_ARCH_UNKNOWN
#endif


// AE_UNUSED
#define AE_UNUSED(x) ((void)x)

// AE_NO_TSAN
#if defined(__has_feature)
#if __has_feature(thread_sanitizer)
#define AE_NO_TSAN __attribute__((no_sanitize("thread")))
#else
#define AE_NO_TSAN
#endif
#else
#define AE_NO_TSAN
#endif


// AE_FORCEINLINE
#if defined(AE_VCPP) || defined(AE_ICC)
#define AE_FORCEINLINE __forceinline
#elif defined(AE_GCC)
//#define AE_FORCEINLINE __attribute__((always_inline)) 
#define AE_FORCEINLINE inline
#else
#define AE_FORCEINLINE inline
#endif


// AE_ALIGN
#if defined(AE_VCPP) || defined(AE_ICC)
#define AE_ALIGN(x) __declspec(align(x))
#elif defined(AE_GCC)
#define AE_ALIGN(x) __attribute__((aligned(x)))
#else
// Assume GCC compliant syntax...
#define AE_ALIGN(x) __attribute__((aligned(x)))
#endif


// Portable atomic fences implemented below:

namespace moodycamel {

enum memory_order {
	memory_order_relaxed,
	memory_order_acquire,
	memory_order_release,
	memory_order_acq_rel,
	memory_order_seq_cst,

	// memory_order_sync: Forces a full sync:
	// #LoadLoad, #LoadStore, #StoreStore, and most significantly, #StoreLoad
	memory_order_sync = memory_order_seq_cst
};

}    // end namespace moodycamel

#if (defined(AE_VCPP) && (_MSC_VER < 1700 || defined(__cplusplus_cli))) || (defined(AE_ICC) && __INTEL_COMPILER < 1600)
// VS2010 and ICC13 don't support std::atomic_*_fence, implement our own fences

#include <intrin.h>

#if defined(AE_ARCH_X64) || defined(AE_ARCH_X86)
#define AeFullSync _mm_mfence
#define AeLiteSync _mm_mfence
#elif defined(AE_ARCH_IA64)
#define AeFullSync __mf
#define AeLiteSync __mf
#elif defined(AE_ARCH_PPC)
#include <ppcintrinsics.h>
#define AeFullSync __sync
#define AeLiteSync __lwsync
#endif


#ifdef AE_VCPP
#pragma warning(push)
#pragma warning(disable: 4365)		// Disable erroneous 'conversion from long to unsigned int, signed/unsigned mismatch' error when using `assert`
#ifdef __cplusplus_cli
#pragma managed(push, off)
#endif
#endif

namespace moodycamel {

AE_FORCEINLINE void compiler_fence(memory_order order) AE_NO_TSAN
{
	switch (order) {
		case memory_order_relaxed: break;
		case memory_order_acquire: _ReadBarrier(); break;
		case memory_order_release: _WriteBarrier(); break;
		case memory_order_acq_rel: _ReadWriteBarrier(); break;
		case memory_order_seq_cst: _ReadWriteBarrier(); break;
		default: assert(false);
	}
}

// x86/x64 have a strong memory model -- all loads and stores have
// acquire and release semantics automatically (so only need compiler
// barriers for those).
#if defined(AE_ARCH_X86) || defined(AE_ARCH_X64)
AE_FORCEINLINE void fence(memory_order order) AE_NO_TSAN
{
	switch (order) {
		case memory_order_relaxed: break;
		case memory_order_acquire: _ReadBarrier(); break;
		case memory_order_release: _WriteBarrier(); break;
		case memory_order_acq_rel: _ReadWriteBarrier(); break;
		case memory_order_seq_cst:
			_ReadWriteBarrier();
			AeFullSync();
			_ReadWriteBarrier();
			break;
		default: assert(false);
	}
}
#else
AE_FORCEINLINE void fence(memory_order order) AE_NO_TSAN
{
	// Non-specialized arch, use heavier memory barriers everywhere just in case :-(
	switch (order) {
		case memory_order_relaxed:
			break;
		case memory_order_acquire:
			_ReadBarrier();
			AeLiteSync();
			_ReadBarrier();
			break;
		case memory_order_release:
			_WriteBarrier();
			AeLiteSync();
			_WriteBarrier();
			break;
		case memory_order_acq_rel:
			_ReadWriteBarrier();
			AeLiteSync();
			_ReadWriteBarrier();
			break;
		case memory_order_seq_cst:
			_ReadWriteBarrier();
			AeFullSync();
			_ReadWriteBarrier();
			break;
		default: assert(false);
	}
}
#endif
}    // end namespace moodycamel
#else
// Use standard library of atomics
#include <atomic>

namespace moodycamel {

AE_FORCEINLINE void compiler_fence(memory_order order) AE_NO_TSAN
{
	switch (order) {
		case memory_order_relaxed: break;
		case memory_order_acquire: std::atomic_signal_fence(std::memory_order_acquire); break;
		case memory_order_release: std::atomic_signal_fence(std::memory_order_release); break;
		case memory_order_acq_rel: std::atomic_signal_fence(std::memory_order_acq_rel); break;
		case memory_order_seq_cst: std::atomic_signal_fence(std::memory_order_seq_cst); break;
		default: assert(false);
	}
}

AE_FORCEINLINE void fence(memory_order order) AE_NO_TSAN
{
	switch (order) {
		case memory_order_relaxed: break;
		case memory_order_acquire: std::atomic_thread_fence(std::memory_order_acquire); break;
		case memory_order_release: std::atomic_thread_fence(std::memory_order_release); break;
		case memory_order_acq_rel: std::atomic_thread_fence(std::memory_order_acq_rel); break;
		case memory_order_seq_cst: std::atomic_thread_fence(std::memory_order_seq_cst); break;
		default: assert(false);
	}
}

}    // end namespace moodycamel

#endif


#if !defined(AE_VCPP) || (_MSC_VER >= 1700 && !defined(__cplusplus_cli))
#define AE_USE_STD_ATOMIC_FOR_WEAK_ATOMIC
#endif

#ifdef AE_USE_STD_ATOMIC_FOR_WEAK_ATOMIC
#include <atomic>
#endif
#include <utility>

// WARNING: *NOT* A REPLACEMENT FOR std::atomic. READ CAREFULLY:
// Provides basic support for atomic variables -- no memory ordering guarantees are provided.
// The guarantee of atomicity is only made for types that already have atomic load and store guarantees
// at the hardware level -- on most platforms this generally means aligned pointers and integers (only).
namespace moodycamel {
template<typename T>
class weak_atomic
{
public:
	AE_NO_TSAN weak_atomic() { }
#ifdef AE_VCPP
#pragma warning(push)
#pragma warning(disable: 4100)		// Get rid of (erroneous) 'unreferenced formal parameter' warning
#endif
	template<typename U> AE_NO_TSAN weak_atomic(U&& x) : value(std::forward<U>(x)) {  }
#ifdef __cplusplus_cli
	// Work around bug with universal reference/nullptr combination that only appears when /clr is on
	AE_NO_TSAN weak_atomic(nullptr_t) : value(nullptr) {  }
#endif
	AE_NO_TSAN weak_atomic(weak_atomic const& other) : value(other.load()) {  }
	AE_NO_TSAN weak_atomic(weak_atomic&& other) : value(std::move(other.load())) {  }
#ifdef AE_VCPP
#pragma warning(pop)
#endif

	AE_FORCEINLINE operator T() const AE_NO_TSAN { return load(); }

	
#ifndef AE_USE_STD_ATOMIC_FOR_WEAK_ATOMIC
	template<typename U> AE_FORCEINLINE weak_atomic const& operator=(U&& x) AE_NO_TSAN { value = std::forward<U>(x); return *this; }
	AE_FORCEINLINE weak_atomic const& operator=(weak_atomic const& other) AE_NO_TSAN { value = other.value; return *this; }
	
	AE_FORCEINLINE T load() const AE_NO_TSAN { return value; }
	
	AE_FORCEINLINE T fetch_add_acquire(T increment) AE_NO_TSAN
	{
#if defined(AE_ARCH_X64) || defined(AE_ARCH_X86)
		if (sizeof(T) == 4) return _InterlockedExchangeAdd((long volatile*)&value, (long)increment);
#if defined(_M_AMD64)
		else if (sizeof(T) == 8) return _InterlockedExchangeAdd64((long long volatile*)&value, (long long)increment);
#endif
#else
#error Unsupported platform
#endif
		assert(false && "T must be either a 32 or 64 bit type");
		return value;
	}
	
	AE_FORCEINLINE T fetch_add_release(T increment) AE_NO_TSAN
	{
#if defined(AE_ARCH_X64) || defined(AE_ARCH_X86)
		if (sizeof(T) == 4) return _InterlockedExchangeAdd((long volatile*)&value, (long)increment);
#if defined(_M_AMD64)
		else if (sizeof(T) == 8) return _InterlockedExchangeAdd64((long long volatile*)&value, (long long)increment);
#endif
#else
#error Unsupported platform
#endif
		assert(false && "T must be either a 32 or 64 bit type");
		return value;
	}
#else
	template<typename U>
	AE_FORCEINLINE weak_atomic const& operator=(U&& x) AE_NO_TSAN
	{
		value.store(std::forward<U>(x), std::memory_order_relaxed);
		return *this;
	}
	
	AE_FORCEINLINE weak_atomic const& operator=(weak_atomic const& other) AE_NO_TSAN
	{
		value.store(other.value.load(std::memory_order_relaxed), std::memory_order_relaxed);
		return *this;
	}

	AE_FORCEINLINE T load() const AE_NO_TSAN { return value.load(std::memory_order_relaxed); }
	
	AE_FORCEINLINE T fetch_add_acquire(T increment) AE_NO_TSAN
	{
		return value.fetch_add(increment, std::memory_order_acquire);
	}
	
	AE_FORCEINLINE T fetch_add_release(T increment) AE_NO_TSAN
	{
		return value.fetch_add(increment, std::memory_order_release);
	}
#endif
	

private:
#ifndef AE_USE_STD_ATOMIC_FOR_WEAK_ATOMIC
	// No std::atomic support, but still need to circumvent compiler optimizations.
	// `volatile` will make memory access slow, but is guaranteed to be reliable.
	volatile T value;
#else
	std::atomic<T> value;
#endif
};

}	// end namespace moodycamel



// Portable single-producer, single-consumer semaphore below:

#if defined(_WIN32)
// Avoid including windows.h in a header; we only need a handful of
// items, so we'll redeclare them here (this is relatively safe since
// the API generally has to remain stable between Windows versions).
// I know this is an ugly hack but it still beats polluting the global
// namespace with thousands of generic names or adding a .cpp for nothing.
extern "C" {
	struct _SECURITY_ATTRIBUTES;
	__declspec(dllimport) void* __stdcall CreateSemaphoreW(_SECURITY_ATTRIBUTES* lpSemaphoreAttributes, long lInitialCount, long lMaximumCount, const wchar_t* lpName);
	__declspec(dllimport) int __stdcall CloseHandle(void* hObject);
	__declspec(dllimport) unsigned long __stdcall WaitForSingleObject(void* hHandle, unsigned long dwMilliseconds);
	__declspec(dllimport) int __stdcall ReleaseSemaphore(void* hSemaphore, long lReleaseCount, long* lpPreviousCount);
}
#elif defined(__MACH__)
#include <mach/mach.h>
#elif defined(__unix__)
#include <semaphore.h>
#endif

namespace moodycamel
{
	// Code in the spsc_sema namespace below is an adaptation of Jeff Preshing's
	// portable + lightweight semaphore implementations, originally from
	// https://github.com/preshing/cpp11-on-multicore/blob/master/common/sema.h
	// LICENSE:
	// Copyright (c) 2015 Jeff Preshing
	//
	// This software is provided 'as-is', without any express or implied
	// warranty. In no event will the authors be held liable for any damages
	// arising from the use of this software.
	//
	// Permission is granted to anyone to use this software for any purpose,
	// including commercial applications, and to alter it and redistribute it
	// freely, subject to the following restrictions:
	//
	// 1. The origin of this software must not be misrepresented; you must not
	//    claim that you wrote the original software. If you use this software
	//    in a product, an acknowledgement in the product documentation would be
	//    appreciated but is not required.
	// 2. Altered source versions must be plainly marked as such, and must not be
	//    misrepresented as being the original software.
	// 3. This notice may not be removed or altered from any source distribution.
	namespace spsc_sema
	{
#if defined(_WIN32)
		class Semaphore
		{
		private:
		    void* m_hSema;
		    
		    Semaphore(const Semaphore& other);
		    Semaphore& operator=(const Semaphore& other);

		public:
		    AE_NO_TSAN Semaphore(int initialCount = 0)
		    {
		        assert(initialCount >= 0);
		        const long maxLong = 0x7fffffff;
		        m_hSema = CreateSemaphoreW(nullptr, initialCount, maxLong, nullptr);
		        assert(m_hSema);
		    }

		    AE_NO_TSAN ~Semaphore()
		    {
		        CloseHandle(m_hSema);
		    }

		    bool wait() AE_NO_TSAN
		    {
		    	const unsigned long infinite = 0xffffffff;
		        return WaitForSingleObject(m_hSema, infinite) == 0;
		    }

			bool try_wait() AE_NO_TSAN
			{
				return WaitForSingleObject(m_hSema, 0) == 0;
			}

			bool timed_wait(std::uint64_t usecs) AE_NO_TSAN
			{
				return WaitForSingleObject(m_hSema, (unsigned long)(usecs / 1000)) == 0;
			}

		    void signal(int count = 1) AE_NO_TSAN
		    {
		        while (!ReleaseSemaphore(m_hSema, count, nullptr));
		    }
		};
#elif defined(__MACH__)
		//---------------------------------------------------------
		// Semaphore (Apple iOS and OSX)
		// Can't use POSIX semaphores due to http://lists.apple.com/archives/darwin-kernel/2009/Apr/msg00010.html
		//---------------------------------------------------------
		class Semaphore
		{
		private:
		    semaphore_t m_sema;

		    Semaphore(const Semaphore& other);
		    Semaphore& operator=(const Semaphore& other);

		public:
		    AE_NO_TSAN Semaphore(int initialCount = 0)
		    {
		        assert(initialCount >= 0);
		        kern_return_t rc = semaphore_create(mach_task_self(), &m_sema, SYNC_POLICY_FIFO, initialCount);
		        assert(rc == KERN_SUCCESS);
		        AE_UNUSED(rc);
		    }

		    AE_NO_TSAN ~Semaphore()
		    {
		        semaphore_destroy(mach_task_self(), m_sema);
		    }

		    bool wait() AE_NO_TSAN
		    {
		        return semaphore_wait(m_sema) == KERN_SUCCESS;
		    }

			bool try_wait() AE_NO_TSAN
			{
				return timed_wait(0);
			}

			bool timed_wait(std::int64_t timeout_usecs) AE_NO_TSAN
			{
				mach_timespec_t ts;
				ts.tv_sec = static_cast<unsigned int>(timeout_usecs / 1000000);
				ts.tv_nsec = (timeout_usecs % 1000000) * 1000;

				// added in OSX 10.10: https://developer.apple.com/library/prerelease/mac/documentation/General/Reference/APIDiffsMacOSX10_10SeedDiff/modules/Darwin.html
				kern_return_t rc = semaphore_timedwait(m_sema, ts);
				return rc == KERN_SUCCESS;
			}

		    void signal() AE_NO_TSAN
		    {
		        while (semaphore_signal(m_sema) != KERN_SUCCESS);
		    }

		    void signal(int count) AE_NO_TSAN
		    {
		        while (count-- > 0)
		        {
		            while (semaphore_signal(m_sema) != KERN_SUCCESS);
		        }
		    }
		};
#elif defined(__unix__)
		//---------------------------------------------------------
		// Semaphore (POSIX, Linux)
		//---------------------------------------------------------
		class Semaphore
		{
		private:
		    sem_t m_sema;

		    Semaphore(const Semaphore& other);
		    Semaphore& operator=(const Semaphore& other);

		public:
		    AE_NO_TSAN Semaphore(int initialCount = 0)
		    {
		        assert(initialCount >= 0);
		        int rc = sem_init(&m_sema, 0, initialCount);
		        assert(rc == 0);
		        AE_UNUSED(rc);
		    }

		    AE_NO_TSAN ~Semaphore()
		    {
		        sem_destroy(&m_sema);
		    }

		    bool wait() AE_NO_TSAN
		    {
		        // http://stackoverflow.com/questions/2013181/gdb-causes-sem-wait-to-fail-with-eintr-error
		        int rc;
		        do
		        {
		            rc = sem_wait(&m_sema);
		        }
		        while (rc == -1 && errno == EINTR);
		        return rc == 0;
		    }

			bool try_wait() AE_NO_TSAN
			{
				int rc;
				do {
					rc = sem_trywait(&m_sema);
				} while (rc == -1 && errno == EINTR);
				return rc == 0;
			}

			bool timed_wait(std::uint64_t usecs) AE_NO_TSAN
			{
				struct timespec ts;
				const int usecs_in_1_sec = 1000000;
				const int nsecs_in_1_sec = 1000000000;
				clock_gettime(CLOCK_REALTIME, &ts);
				ts.tv_sec += (time_t)(usecs / usecs_in_1_sec);
				ts.tv_nsec += (long)(usecs % usecs_in_1_sec) * 1000;
				// sem_timedwait bombs if you have more than 1e9 in tv_nsec
				// so we have to clean things up before passing it in
				if (ts.tv_nsec >= nsecs_in_1_sec) {
					ts.tv_nsec -= nsecs_in_1_sec;
					++ts.tv_sec;
				}

				int rc;
				do {
					rc = sem_timedwait(&m_sema, &ts);
				} while (rc == -1 && errno == EINTR);
				return rc == 0;
			}

		    void signal() AE_NO_TSAN
		    {
		        while (sem_post(&m_sema) == -1);
		    }

		    void signal(int count) AE_NO_TSAN
		    {
		        while (count-- > 0)
		        {
		            while (sem_post(&m_sema) == -1);
		        }
		    }
		};
#else
#error Unsupported platform! (No semaphore wrapper available)
#endif

		//---------------------------------------------------------
		// LightweightSemaphore
		//---------------------------------------------------------
		class LightweightSemaphore
		{
		public:
			typedef std::make_signed<std::size_t>::type ssize_t;
			
		private:
		    weak_atomic<ssize_t> m_count;
		    Semaphore m_sema;

		    bool waitWithPartialSpinning(std::int64_t timeout_usecs = -1) AE_NO_TSAN
		    {
		        ssize_t oldCount;
		        // Is there a better way to set the initial spin count?
		        // If we lower it to 1000, testBenaphore becomes 15x slower on my Core i7-5930K Windows PC,
		        // as threads start hitting the kernel semaphore.
		        int spin = 1024;
		        while (--spin >= 0)
		        {
		            if (m_count.load() > 0)
		            {
		                m_count.fetch_add_acquire(-1);
		                return true;
		            }
		            compiler_fence(memory_order_acquire);     // Prevent the compiler from collapsing the loop.
		        }
		        oldCount = m_count.fetch_add_acquire(-1);
				if (oldCount > 0)
					return true;
		        if (timeout_usecs < 0)
				{
					if (m_sema.wait())
						return true;
				}
				if (timeout_usecs > 0 && m_sema.timed_wait(timeout_usecs))
					return true;
				// At this point, we've timed out waiting for the semaphore, but the
				// count is still decremented indicating we may still be waiting on
				// it. So we have to re-adjust the count, but only if the semaphore
				// wasn't signaled enough times for us too since then. If it was, we
				// need to release the semaphore too.
				while (true)
				{
					oldCount = m_count.fetch_add_release(1);
					if (oldCount < 0)
						return false;    // successfully restored things to the way they were
					// Oh, the producer thread just signaled the semaphore after all. Try again:
					oldCount = m_count.fetch_add_acquire(-1);
					if (oldCount > 0 && m_sema.try_wait())
						return true;
				}
		    }

		public:
		    AE_NO_TSAN LightweightSemaphore(ssize_t initialCount = 0) : m_count(initialCount)
		    {
		        assert(initialCount >= 0);
		    }

		    bool tryWait() AE_NO_TSAN
		    {
		        if (m_count.load() > 0)
		        {
		        	m_count.fetch_add_acquire(-1);
		        	return true;
		        }
		        return false;
		    }

		    bool wait() AE_NO_TSAN
		    {
		        return tryWait() || waitWithPartialSpinning();
		    }

			bool wait(std::int64_t timeout_usecs) AE_NO_TSAN
			{
				return tryWait() || waitWithPartialSpinning(timeout_usecs);
			}

		    void signal(ssize_t count = 1) AE_NO_TSAN
		    {
		    	assert(count >= 0);
		        ssize_t oldCount = m_count.fetch_add_release(count);
		        assert(oldCount >= -1);
		        if (oldCount < 0)
		        {
		            m_sema.signal(1);
		        }
		    }
		    
		    ssize_t availableApprox() const AE_NO_TSAN
		    {
		    	ssize_t count = m_count.load();
		    	return count > 0 ? count : 0;
		    }
		};
	}	// end namespace spsc_sema
}	// end namespace moodycamel

#if defined(AE_VCPP) && (_MSC_VER < 1700 || defined(__cplusplus_cli))
#pragma warning(pop)
#ifdef __cplusplus_cli
#pragma managed(pop)
#endif
#endif