diff --git a/.gitmodules b/.gitmodules index 8403d544..6eef7fb4 100644 --- a/.gitmodules +++ b/.gitmodules @@ -6,15 +6,3 @@ path = external/benchmark url = https://github.com/google/benchmark shallow = true -[submodule "external/readerwriterqueue"] - path = external/readerwriterqueue - url = https://github.com/cameron314/readerwriterqueue.git - shallow = true -[submodule "external/cnpy"] - path = external/cnpy - url = https://github.com/paulfd/cnpy.git - shallow = true -[submodule "external/Catch2"] - path = external/Catch2 - url = https://github.com/catchorg/Catch2.git - shallow = true diff --git a/CMakeLists.txt b/CMakeLists.txt index d42c1a5e..b9b5c6f3 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -66,10 +66,6 @@ if (WIN32) target_include_directories(sndfile INTERFACE "${WIN_SNDFILE_PATH}/include") endif() -# Export MoodyCamel's queue as a library -add_library(readerwriterqueue INTERFACE) -target_include_directories(readerwriterqueue INTERFACE "external/readerwriterqueue") - add_subdirectory(sfizz) if (SFIZZ_CLIENTS) @@ -77,9 +73,6 @@ add_subdirectory(clients) endif() if (SFIZZ_TESTS) -# Tests -add_subdirectory(external/Catch2 EXCLUDE_FROM_ALL) -add_subdirectory(external/cnpy EXCLUDE_FROM_ALL) add_subdirectory(tests) endif() diff --git a/external/Catch2 b/external/Catch2 deleted file mode 160000 index af8b2538..00000000 --- a/external/Catch2 +++ /dev/null @@ -1 +0,0 @@ -Subproject commit af8b2538a62db34b3cec10ceb74e3d29bd06e37f diff --git a/external/abseil-cpp b/external/abseil-cpp index 97c1664b..ac78ffc3 160000 --- a/external/abseil-cpp +++ b/external/abseil-cpp @@ -1 +1 @@ -Subproject commit 97c1664b4bbab5f78fac2b151ab02656268fb34b +Subproject commit ac78ffc3bc0a8b295cab9a03817760fd460df2a1 diff --git a/external/benchmark b/external/benchmark index 7ee72863..bf4f2ea0 160000 --- a/external/benchmark +++ b/external/benchmark @@ -1 +1 @@ -Subproject commit 7ee72863fdb1ccb2af5a011250b56af3f49b7511 +Subproject commit bf4f2ea0bd1180b34718ac26eb79b170a4f6290e diff --git a/external/cnpy b/external/cnpy deleted file mode 160000 index 34693799..00000000 --- a/external/cnpy +++ /dev/null @@ -1 +0,0 @@ -Subproject commit 34693799ada6cb84be3f588aabbe75f24d010e2f diff --git a/external/readerwriterqueue b/external/readerwriterqueue deleted file mode 160000 index 2ae710de..00000000 --- a/external/readerwriterqueue +++ /dev/null @@ -1 +0,0 @@ -Subproject commit 2ae710de996a1d02bbc7696b2cdff2c6078e76f8 diff --git a/sfizz/CMakeLists.txt b/sfizz/CMakeLists.txt index aae522ec..b040cc2a 100644 --- a/sfizz/CMakeLists.txt +++ b/sfizz/CMakeLists.txt @@ -51,7 +51,7 @@ if(UNIX) target_link_libraries(sfizz PUBLIC stdc++fs atomic) target_compile_options(sfizz PRIVATE -fno-rtti -fno-exceptions) endif(UNIX) -target_link_libraries(sfizz PUBLIC readerwriterqueue absl::strings) +target_link_libraries(sfizz PUBLIC absl::strings) target_link_libraries(sfizz PRIVATE sndfile absl::flat_hash_map) add_library(sfizz::parser ALIAS sfizz_parser) diff --git a/sfizz/atomicops.h b/sfizz/atomicops.h new file mode 100644 index 00000000..4fd17482 --- /dev/null +++ b/sfizz/atomicops.h @@ -0,0 +1,676 @@ +// ©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 +#include +#include +#include +#include + +// 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 + +#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 +#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 + +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 +#endif +#include + +// 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 +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 AE_NO_TSAN weak_atomic(U&& x) : value(std::forward(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 AE_FORCEINLINE weak_atomic const& operator=(U&& x) AE_NO_TSAN { value = std::forward(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 + AE_FORCEINLINE weak_atomic const& operator=(U&& x) AE_NO_TSAN + { + value.store(std::forward(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 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 +#elif defined(__unix__) +#include +#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); + } + + AE_NO_TSAN ~Semaphore() + { + CloseHandle(m_hSema); + } + + void wait() AE_NO_TSAN + { + const unsigned long infinite = 0xffffffff; + WaitForSingleObject(m_hSema, infinite); + } + + bool try_wait() AE_NO_TSAN + { + const unsigned long RC_WAIT_TIMEOUT = 0x00000102; + return WaitForSingleObject(m_hSema, 0) != RC_WAIT_TIMEOUT; + } + + bool timed_wait(std::uint64_t usecs) AE_NO_TSAN + { + const unsigned long RC_WAIT_TIMEOUT = 0x00000102; + return WaitForSingleObject(m_hSema, (unsigned long)(usecs / 1000)) != RC_WAIT_TIMEOUT; + } + + void signal(int count = 1) AE_NO_TSAN + { + 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); + semaphore_create(mach_task_self(), &m_sema, SYNC_POLICY_FIFO, initialCount); + } + + AE_NO_TSAN ~Semaphore() + { + semaphore_destroy(mach_task_self(), m_sema); + } + + void wait() AE_NO_TSAN + { + semaphore_wait(m_sema); + } + + 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(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_OPERATION_TIMED_OUT && rc != KERN_ABORTED; + } + + void signal() AE_NO_TSAN + { + semaphore_signal(m_sema); + } + + void signal(int count) AE_NO_TSAN + { + while (count-- > 0) + { + semaphore_signal(m_sema); + } + } + }; +#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); + sem_init(&m_sema, 0, initialCount); + } + + AE_NO_TSAN ~Semaphore() + { + sem_destroy(&m_sema); + } + + void 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); + } + + bool try_wait() AE_NO_TSAN + { + int rc; + do { + rc = sem_trywait(&m_sema); + } while (rc == -1 && errno == EINTR); + return !(rc == -1 && errno == EAGAIN); + } + + 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 += usecs / usecs_in_1_sec; + ts.tv_nsec += (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 == -1 && errno == ETIMEDOUT); + } + + void signal() AE_NO_TSAN + { + sem_post(&m_sema); + } + + void signal(int count) AE_NO_TSAN + { + while (count-- > 0) + { + sem_post(&m_sema); + } + } + }; +#else +#error Unsupported platform! (No semaphore wrapper available) +#endif + + //--------------------------------------------------------- + // LightweightSemaphore + //--------------------------------------------------------- + class LightweightSemaphore + { + public: + typedef std::make_signed::type ssize_t; + + private: + weak_atomic 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 = 10000; + 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) + { + m_sema.wait(); + return true; + } + if (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; + } + + void wait() AE_NO_TSAN + { + if (!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 diff --git a/sfizz/readerwriterqueue.h b/sfizz/readerwriterqueue.h new file mode 100644 index 00000000..071147c3 --- /dev/null +++ b/sfizz/readerwriterqueue.h @@ -0,0 +1,906 @@ +// ©2013-2016 Cameron Desrochers. +// Distributed under the simplified BSD license (see the license file that +// should have come with this header). + +#pragma once + +#include "atomicops.h" +#include +#include +#include +#include +#include +#include +#include // For malloc/free/abort & size_t +#include +#if __cplusplus > 199711L || _MSC_VER >= 1700 // C++11 or VS2012 +#include +#endif + + +// A lock-free queue for a single-consumer, single-producer architecture. +// The queue is also wait-free in the common path (except if more memory +// needs to be allocated, in which case malloc is called). +// Allocates memory sparingly (O(lg(n) times, amortized), and only once if +// the original maximum size estimate is never exceeded. +// Tested on x86/x64 processors, but semantics should be correct for all +// architectures (given the right implementations in atomicops.h), provided +// that aligned integer and pointer accesses are naturally atomic. +// Note that there should only be one consumer thread and producer thread; +// Switching roles of the threads, or using multiple consecutive threads for +// one role, is not safe unless properly synchronized. +// Using the queue exclusively from one thread is fine, though a bit silly. + +#ifndef MOODYCAMEL_CACHE_LINE_SIZE +#define MOODYCAMEL_CACHE_LINE_SIZE 64 +#endif + +#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 + +#ifndef MOODYCAMEL_HAS_EMPLACE +#if !defined(_MSC_VER) || _MSC_VER >= 1800 // variadic templates: either a non-MS compiler or VS >= 2013 +#define MOODYCAMEL_HAS_EMPLACE 1 +#endif +#endif + +#ifdef AE_VCPP +#pragma warning(push) +#pragma warning(disable: 4324) // structure was padded due to __declspec(align()) +#pragma warning(disable: 4820) // padding was added +#pragma warning(disable: 4127) // conditional expression is constant +#endif + +namespace moodycamel { + +template +class ReaderWriterQueue +{ + // Design: Based on a queue-of-queues. The low-level queues are just + // circular buffers with front and tail indices indicating where the + // next element to dequeue is and where the next element can be enqueued, + // respectively. Each low-level queue is called a "block". Each block + // wastes exactly one element's worth of space to keep the design simple + // (if front == tail then the queue is empty, and can't be full). + // The high-level queue is a circular linked list of blocks; again there + // is a front and tail, but this time they are pointers to the blocks. + // The front block is where the next element to be dequeued is, provided + // the block is not empty. The back block is where elements are to be + // enqueued, provided the block is not full. + // The producer thread owns all the tail indices/pointers. The consumer + // thread owns all the front indices/pointers. Both threads read each + // other's variables, but only the owning thread updates them. E.g. After + // the consumer reads the producer's tail, the tail may change before the + // consumer is done dequeuing an object, but the consumer knows the tail + // will never go backwards, only forwards. + // If there is no room to enqueue an object, an additional block (of + // equal size to the last block) is added. Blocks are never removed. + +public: + typedef T value_type; + + // Constructs a queue that can hold maxSize elements without further + // allocations. If more than MAX_BLOCK_SIZE elements are requested, + // then several blocks of MAX_BLOCK_SIZE each are reserved (including + // at least one extra buffer block). + AE_NO_TSAN explicit ReaderWriterQueue(size_t maxSize = 15) +#ifndef NDEBUG + : enqueuing(false) + ,dequeuing(false) +#endif + { + assert(maxSize > 0); + assert(MAX_BLOCK_SIZE == ceilToPow2(MAX_BLOCK_SIZE) && "MAX_BLOCK_SIZE must be a power of 2"); + assert(MAX_BLOCK_SIZE >= 2 && "MAX_BLOCK_SIZE must be at least 2"); + + Block* firstBlock = nullptr; + + largestBlockSize = ceilToPow2(maxSize + 1); // We need a spare slot to fit maxSize elements in the block + if (largestBlockSize > MAX_BLOCK_SIZE * 2) { + // We need a spare block in case the producer is writing to a different block the consumer is reading from, and + // wants to enqueue the maximum number of elements. We also need a spare element in each block to avoid the ambiguity + // between front == tail meaning "empty" and "full". + // So the effective number of slots that are guaranteed to be usable at any time is the block size - 1 times the + // number of blocks - 1. Solving for maxSize and applying a ceiling to the division gives us (after simplifying): + size_t initialBlockCount = (maxSize + MAX_BLOCK_SIZE * 2 - 3) / (MAX_BLOCK_SIZE - 1); + largestBlockSize = MAX_BLOCK_SIZE; + Block* lastBlock = nullptr; + for (size_t i = 0; i != initialBlockCount; ++i) { + auto block = make_block(largestBlockSize); + if (block == nullptr) { +#ifdef MOODYCAMEL_EXCEPTIONS_ENABLED + throw std::bad_alloc(); +#else + abort(); +#endif + } + if (firstBlock == nullptr) { + firstBlock = block; + } + else { + lastBlock->next = block; + } + lastBlock = block; + block->next = firstBlock; + } + } + else { + firstBlock = make_block(largestBlockSize); + if (firstBlock == nullptr) { +#ifdef MOODYCAMEL_EXCEPTIONS_ENABLED + throw std::bad_alloc(); +#else + abort(); +#endif + } + firstBlock->next = firstBlock; + } + frontBlock = firstBlock; + tailBlock = firstBlock; + + // Make sure the reader/writer threads will have the initialized memory setup above: + fence(memory_order_sync); + } + + // Note: The queue should not be accessed concurrently while it's + // being moved. It's up to the user to synchronize this. + AE_NO_TSAN ReaderWriterQueue(ReaderWriterQueue&& other) + : frontBlock(other.frontBlock.load()), + tailBlock(other.tailBlock.load()), + largestBlockSize(other.largestBlockSize) +#ifndef NDEBUG + ,enqueuing(false) + ,dequeuing(false) +#endif + { + other.largestBlockSize = 32; + Block* b = other.make_block(other.largestBlockSize); + if (b == nullptr) { +#ifdef MOODYCAMEL_EXCEPTIONS_ENABLED + throw std::bad_alloc(); +#else + abort(); +#endif + } + b->next = b; + other.frontBlock = b; + other.tailBlock = b; + } + + // Note: The queue should not be accessed concurrently while it's + // being moved. It's up to the user to synchronize this. + ReaderWriterQueue& operator=(ReaderWriterQueue&& other) AE_NO_TSAN + { + Block* b = frontBlock.load(); + frontBlock = other.frontBlock.load(); + other.frontBlock = b; + b = tailBlock.load(); + tailBlock = other.tailBlock.load(); + other.tailBlock = b; + std::swap(largestBlockSize, other.largestBlockSize); + return *this; + } + + // Note: The queue should not be accessed concurrently while it's + // being deleted. It's up to the user to synchronize this. + AE_NO_TSAN ~ReaderWriterQueue() + { + // Make sure we get the latest version of all variables from other CPUs: + fence(memory_order_sync); + + // Destroy any remaining objects in queue and free memory + Block* frontBlock_ = frontBlock; + Block* block = frontBlock_; + do { + Block* nextBlock = block->next; + size_t blockFront = block->front; + size_t blockTail = block->tail; + + for (size_t i = blockFront; i != blockTail; i = (i + 1) & block->sizeMask) { + auto element = reinterpret_cast(block->data + i * sizeof(T)); + element->~T(); + (void)element; + } + + auto rawBlock = block->rawThis; + block->~Block(); + std::free(rawBlock); + block = nextBlock; + } while (block != frontBlock_); + } + + + // Enqueues a copy of element if there is room in the queue. + // Returns true if the element was enqueued, false otherwise. + // Does not allocate memory. + AE_FORCEINLINE bool try_enqueue(T const& element) AE_NO_TSAN + { + return inner_enqueue(element); + } + + // Enqueues a moved copy of element if there is room in the queue. + // Returns true if the element was enqueued, false otherwise. + // Does not allocate memory. + AE_FORCEINLINE bool try_enqueue(T&& element) AE_NO_TSAN + { + return inner_enqueue(std::forward(element)); + } + +#if MOODYCAMEL_HAS_EMPLACE + // Like try_enqueue() but with emplace semantics (i.e. construct-in-place). + template + AE_FORCEINLINE bool try_emplace(Args&&... args) AE_NO_TSAN + { + return inner_enqueue(std::forward(args)...); + } +#endif + + // Enqueues a copy of element on the queue. + // Allocates an additional block of memory if needed. + // Only fails (returns false) if memory allocation fails. + AE_FORCEINLINE bool enqueue(T const& element) AE_NO_TSAN + { + return inner_enqueue(element); + } + + // Enqueues a moved copy of element on the queue. + // Allocates an additional block of memory if needed. + // Only fails (returns false) if memory allocation fails. + AE_FORCEINLINE bool enqueue(T&& element) AE_NO_TSAN + { + return inner_enqueue(std::forward(element)); + } + +#if MOODYCAMEL_HAS_EMPLACE + // Like enqueue() but with emplace semantics (i.e. construct-in-place). + template + AE_FORCEINLINE bool emplace(Args&&... args) AE_NO_TSAN + { + return inner_enqueue(std::forward(args)...); + } +#endif + + // Attempts to dequeue an element; if the queue is empty, + // returns false instead. If the queue has at least one element, + // moves front to result using operator=, then returns true. + template + bool try_dequeue(U& result) AE_NO_TSAN + { +#ifndef NDEBUG + ReentrantGuard guard(this->dequeuing); +#endif + + // High-level pseudocode: + // Remember where the tail block is + // If the front block has an element in it, dequeue it + // Else + // If front block was the tail block when we entered the function, return false + // Else advance to next block and dequeue the item there + + // Note that we have to use the value of the tail block from before we check if the front + // block is full or not, in case the front block is empty and then, before we check if the + // tail block is at the front block or not, the producer fills up the front block *and + // moves on*, which would make us skip a filled block. Seems unlikely, but was consistently + // reproducible in practice. + // In order to avoid overhead in the common case, though, we do a double-checked pattern + // where we have the fast path if the front block is not empty, then read the tail block, + // then re-read the front block and check if it's not empty again, then check if the tail + // block has advanced. + + Block* frontBlock_ = frontBlock.load(); + size_t blockTail = frontBlock_->localTail; + size_t blockFront = frontBlock_->front.load(); + + if (blockFront != blockTail || blockFront != (frontBlock_->localTail = frontBlock_->tail.load())) { + fence(memory_order_acquire); + + non_empty_front_block: + // Front block not empty, dequeue from here + auto element = reinterpret_cast(frontBlock_->data + blockFront * sizeof(T)); + result = std::move(*element); + element->~T(); + + blockFront = (blockFront + 1) & frontBlock_->sizeMask; + + fence(memory_order_release); + frontBlock_->front = blockFront; + } + else if (frontBlock_ != tailBlock.load()) { + fence(memory_order_acquire); + + frontBlock_ = frontBlock.load(); + blockTail = frontBlock_->localTail = frontBlock_->tail.load(); + blockFront = frontBlock_->front.load(); + fence(memory_order_acquire); + + if (blockFront != blockTail) { + // Oh look, the front block isn't empty after all + goto non_empty_front_block; + } + + // Front block is empty but there's another block ahead, advance to it + Block* nextBlock = frontBlock_->next; + // Don't need an acquire fence here since next can only ever be set on the tailBlock, + // and we're not the tailBlock, and we did an acquire earlier after reading tailBlock which + // ensures next is up-to-date on this CPU in case we recently were at tailBlock. + + size_t nextBlockFront = nextBlock->front.load(); + size_t nextBlockTail = nextBlock->localTail = nextBlock->tail.load(); + fence(memory_order_acquire); + + // Since the tailBlock is only ever advanced after being written to, + // we know there's for sure an element to dequeue on it + assert(nextBlockFront != nextBlockTail); + AE_UNUSED(nextBlockTail); + + // We're done with this block, let the producer use it if it needs + fence(memory_order_release); // Expose possibly pending changes to frontBlock->front from last dequeue + frontBlock = frontBlock_ = nextBlock; + + compiler_fence(memory_order_release); // Not strictly needed + + auto element = reinterpret_cast(frontBlock_->data + nextBlockFront * sizeof(T)); + + result = std::move(*element); + element->~T(); + + nextBlockFront = (nextBlockFront + 1) & frontBlock_->sizeMask; + + fence(memory_order_release); + frontBlock_->front = nextBlockFront; + } + else { + // No elements in current block and no other block to advance to + return false; + } + + return true; + } + + + // Returns a pointer to the front element in the queue (the one that + // would be removed next by a call to `try_dequeue` or `pop`). If the + // queue appears empty at the time the method is called, nullptr is + // returned instead. + // Must be called only from the consumer thread. + T* peek() AE_NO_TSAN + { +#ifndef NDEBUG + ReentrantGuard guard(this->dequeuing); +#endif + // See try_dequeue() for reasoning + + Block* frontBlock_ = frontBlock.load(); + size_t blockTail = frontBlock_->localTail; + size_t blockFront = frontBlock_->front.load(); + + if (blockFront != blockTail || blockFront != (frontBlock_->localTail = frontBlock_->tail.load())) { + fence(memory_order_acquire); + non_empty_front_block: + return reinterpret_cast(frontBlock_->data + blockFront * sizeof(T)); + } + else if (frontBlock_ != tailBlock.load()) { + fence(memory_order_acquire); + frontBlock_ = frontBlock.load(); + blockTail = frontBlock_->localTail = frontBlock_->tail.load(); + blockFront = frontBlock_->front.load(); + fence(memory_order_acquire); + + if (blockFront != blockTail) { + goto non_empty_front_block; + } + + Block* nextBlock = frontBlock_->next; + + size_t nextBlockFront = nextBlock->front.load(); + fence(memory_order_acquire); + + assert(nextBlockFront != nextBlock->tail.load()); + return reinterpret_cast(nextBlock->data + nextBlockFront * sizeof(T)); + } + + return nullptr; + } + + // Removes the front element from the queue, if any, without returning it. + // Returns true on success, or false if the queue appeared empty at the time + // `pop` was called. + bool pop() AE_NO_TSAN + { +#ifndef NDEBUG + ReentrantGuard guard(this->dequeuing); +#endif + // See try_dequeue() for reasoning + + Block* frontBlock_ = frontBlock.load(); + size_t blockTail = frontBlock_->localTail; + size_t blockFront = frontBlock_->front.load(); + + if (blockFront != blockTail || blockFront != (frontBlock_->localTail = frontBlock_->tail.load())) { + fence(memory_order_acquire); + + non_empty_front_block: + auto element = reinterpret_cast(frontBlock_->data + blockFront * sizeof(T)); + element->~T(); + + blockFront = (blockFront + 1) & frontBlock_->sizeMask; + + fence(memory_order_release); + frontBlock_->front = blockFront; + } + else if (frontBlock_ != tailBlock.load()) { + fence(memory_order_acquire); + frontBlock_ = frontBlock.load(); + blockTail = frontBlock_->localTail = frontBlock_->tail.load(); + blockFront = frontBlock_->front.load(); + fence(memory_order_acquire); + + if (blockFront != blockTail) { + goto non_empty_front_block; + } + + // Front block is empty but there's another block ahead, advance to it + Block* nextBlock = frontBlock_->next; + + size_t nextBlockFront = nextBlock->front.load(); + size_t nextBlockTail = nextBlock->localTail = nextBlock->tail.load(); + fence(memory_order_acquire); + + assert(nextBlockFront != nextBlockTail); + AE_UNUSED(nextBlockTail); + + fence(memory_order_release); + frontBlock = frontBlock_ = nextBlock; + + compiler_fence(memory_order_release); + + auto element = reinterpret_cast(frontBlock_->data + nextBlockFront * sizeof(T)); + element->~T(); + + nextBlockFront = (nextBlockFront + 1) & frontBlock_->sizeMask; + + fence(memory_order_release); + frontBlock_->front = nextBlockFront; + } + else { + // No elements in current block and no other block to advance to + return false; + } + + return true; + } + + // Returns the approximate number of items currently in the queue. + // Safe to call from both the producer and consumer threads. + inline size_t size_approx() const AE_NO_TSAN + { + size_t result = 0; + Block* frontBlock_ = frontBlock.load(); + Block* block = frontBlock_; + do { + fence(memory_order_acquire); + size_t blockFront = block->front.load(); + size_t blockTail = block->tail.load(); + result += (blockTail - blockFront) & block->sizeMask; + block = block->next.load(); + } while (block != frontBlock_); + return result; + } + + +private: + enum AllocationMode { CanAlloc, CannotAlloc }; + +#if MOODYCAMEL_HAS_EMPLACE + template + bool inner_enqueue(Args&&... args) AE_NO_TSAN +#else + template + bool inner_enqueue(U&& element) AE_NO_TSAN +#endif + { +#ifndef NDEBUG + ReentrantGuard guard(this->enqueuing); +#endif + + // High-level pseudocode (assuming we're allowed to alloc a new block): + // If room in tail block, add to tail + // Else check next block + // If next block is not the head block, enqueue on next block + // Else create a new block and enqueue there + // Advance tail to the block we just enqueued to + + Block* tailBlock_ = tailBlock.load(); + size_t blockFront = tailBlock_->localFront; + size_t blockTail = tailBlock_->tail.load(); + + size_t nextBlockTail = (blockTail + 1) & tailBlock_->sizeMask; + if (nextBlockTail != blockFront || nextBlockTail != (tailBlock_->localFront = tailBlock_->front.load())) { + fence(memory_order_acquire); + // This block has room for at least one more element + char* location = tailBlock_->data + blockTail * sizeof(T); +#if MOODYCAMEL_HAS_EMPLACE + new (location) T(std::forward(args)...); +#else + new (location) T(std::forward(element)); +#endif + + fence(memory_order_release); + tailBlock_->tail = nextBlockTail; + } + else { + fence(memory_order_acquire); + if (tailBlock_->next.load() != frontBlock) { + // Note that the reason we can't advance to the frontBlock and start adding new entries there + // is because if we did, then dequeue would stay in that block, eventually reading the new values, + // instead of advancing to the next full block (whose values were enqueued first and so should be + // consumed first). + + fence(memory_order_acquire); // Ensure we get latest writes if we got the latest frontBlock + + // tailBlock is full, but there's a free block ahead, use it + Block* tailBlockNext = tailBlock_->next.load(); + size_t nextBlockFront = tailBlockNext->localFront = tailBlockNext->front.load(); + nextBlockTail = tailBlockNext->tail.load(); + fence(memory_order_acquire); + + // This block must be empty since it's not the head block and we + // go through the blocks in a circle + assert(nextBlockFront == nextBlockTail); + tailBlockNext->localFront = nextBlockFront; + + char* location = tailBlockNext->data + nextBlockTail * sizeof(T); +#if MOODYCAMEL_HAS_EMPLACE + new (location) T(std::forward(args)...); +#else + new (location) T(std::forward(element)); +#endif + + tailBlockNext->tail = (nextBlockTail + 1) & tailBlockNext->sizeMask; + + fence(memory_order_release); + tailBlock = tailBlockNext; + } + else if (canAlloc == CanAlloc) { + // tailBlock is full and there's no free block ahead; create a new block + auto newBlockSize = largestBlockSize >= MAX_BLOCK_SIZE ? largestBlockSize : largestBlockSize * 2; + auto newBlock = make_block(newBlockSize); + if (newBlock == nullptr) { + // Could not allocate a block! + return false; + } + largestBlockSize = newBlockSize; + +#if MOODYCAMEL_HAS_EMPLACE + new (newBlock->data) T(std::forward(args)...); +#else + new (newBlock->data) T(std::forward(element)); +#endif + assert(newBlock->front == 0); + newBlock->tail = newBlock->localTail = 1; + + newBlock->next = tailBlock_->next.load(); + tailBlock_->next = newBlock; + + // Might be possible for the dequeue thread to see the new tailBlock->next + // *without* seeing the new tailBlock value, but this is OK since it can't + // advance to the next block until tailBlock is set anyway (because the only + // case where it could try to read the next is if it's already at the tailBlock, + // and it won't advance past tailBlock in any circumstance). + + fence(memory_order_release); + tailBlock = newBlock; + } + else if (canAlloc == CannotAlloc) { + // Would have had to allocate a new block to enqueue, but not allowed + return false; + } + else { + assert(false && "Should be unreachable code"); + return false; + } + } + + return true; + } + + + // Disable copying + ReaderWriterQueue(ReaderWriterQueue const&) { } + + // Disable assignment + ReaderWriterQueue& operator=(ReaderWriterQueue const&) { } + + + + AE_FORCEINLINE static size_t ceilToPow2(size_t x) + { + // From http://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2 + --x; + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + for (size_t i = 1; i < sizeof(size_t); i <<= 1) { + x |= x >> (i << 3); + } + ++x; + return x; + } + + template + static AE_FORCEINLINE char* align_for(char* ptr) AE_NO_TSAN + { + const std::size_t alignment = std::alignment_of::value; + return ptr + (alignment - (reinterpret_cast(ptr) % alignment)) % alignment; + } +private: +#ifndef NDEBUG + struct ReentrantGuard + { + AE_NO_TSAN ReentrantGuard(bool& _inSection) + : inSection(_inSection) + { + assert(!inSection && "Concurrent (or re-entrant) enqueue or dequeue operation detected (only one thread at a time may hold the producer or consumer role)"); + inSection = true; + } + + AE_NO_TSAN ~ReentrantGuard() { inSection = false; } + + private: + ReentrantGuard& operator=(ReentrantGuard const&); + + private: + bool& inSection; + }; +#endif + + struct Block + { + // Avoid false-sharing by putting highly contended variables on their own cache lines + weak_atomic front; // (Atomic) Elements are read from here + size_t localTail; // An uncontended shadow copy of tail, owned by the consumer + + char cachelineFiller0[MOODYCAMEL_CACHE_LINE_SIZE - sizeof(weak_atomic) - sizeof(size_t)]; + weak_atomic tail; // (Atomic) Elements are enqueued here + size_t localFront; + + char cachelineFiller1[MOODYCAMEL_CACHE_LINE_SIZE - sizeof(weak_atomic) - sizeof(size_t)]; // next isn't very contended, but we don't want it on the same cache line as tail (which is) + weak_atomic next; // (Atomic) + + char* data; // Contents (on heap) are aligned to T's alignment + + const size_t sizeMask; + + + // size must be a power of two (and greater than 0) + AE_NO_TSAN Block(size_t const& _size, char* _rawThis, char* _data) + : front(0), localTail(0), tail(0), localFront(0), next(nullptr), data(_data), sizeMask(_size - 1), rawThis(_rawThis) + { + } + + private: + // C4512 - Assignment operator could not be generated + Block& operator=(Block const&); + + public: + char* rawThis; + }; + + + static Block* make_block(size_t capacity) AE_NO_TSAN + { + // Allocate enough memory for the block itself, as well as all the elements it will contain + auto size = sizeof(Block) + std::alignment_of::value - 1; + size += sizeof(T) * capacity + std::alignment_of::value - 1; + auto newBlockRaw = static_cast(std::malloc(size)); + if (newBlockRaw == nullptr) { + return nullptr; + } + + auto newBlockAligned = align_for(newBlockRaw); + auto newBlockData = align_for(newBlockAligned + sizeof(Block)); + return new (newBlockAligned) Block(capacity, newBlockRaw, newBlockData); + } + +private: + weak_atomic frontBlock; // (Atomic) Elements are enqueued to this block + + char cachelineFiller[MOODYCAMEL_CACHE_LINE_SIZE - sizeof(weak_atomic)]; + weak_atomic tailBlock; // (Atomic) Elements are dequeued from this block + + size_t largestBlockSize; + +#ifndef NDEBUG + bool enqueuing; + bool dequeuing; +#endif +}; + +// Like ReaderWriterQueue, but also providees blocking operations +template +class BlockingReaderWriterQueue +{ +private: + typedef ::moodycamel::ReaderWriterQueue ReaderWriterQueue; + +public: + explicit BlockingReaderWriterQueue(size_t maxSize = 15) AE_NO_TSAN + : inner(maxSize), sema(new spsc_sema::LightweightSemaphore()) + { } + + BlockingReaderWriterQueue(BlockingReaderWriterQueue&& other) AE_NO_TSAN + : inner(std::move(other.inner)), sema(std::move(other.sema)) + { } + + BlockingReaderWriterQueue& operator=(BlockingReaderWriterQueue&& other) AE_NO_TSAN + { + std::swap(sema, other.sema); + std::swap(inner, other.inner); + return *this; + } + + + // Enqueues a copy of element if there is room in the queue. + // Returns true if the element was enqueued, false otherwise. + // Does not allocate memory. + AE_FORCEINLINE bool try_enqueue(T const& element) AE_NO_TSAN + { + if (inner.try_enqueue(element)) { + sema->signal(); + return true; + } + return false; + } + + // Enqueues a moved copy of element if there is room in the queue. + // Returns true if the element was enqueued, false otherwise. + // Does not allocate memory. + AE_FORCEINLINE bool try_enqueue(T&& element) AE_NO_TSAN + { + if (inner.try_enqueue(std::forward(element))) { + sema->signal(); + return true; + } + return false; + } + + + // Enqueues a copy of element on the queue. + // Allocates an additional block of memory if needed. + // Only fails (returns false) if memory allocation fails. + AE_FORCEINLINE bool enqueue(T const& element) AE_NO_TSAN + { + if (inner.enqueue(element)) { + sema->signal(); + return true; + } + return false; + } + + // Enqueues a moved copy of element on the queue. + // Allocates an additional block of memory if needed. + // Only fails (returns false) if memory allocation fails. + AE_FORCEINLINE bool enqueue(T&& element) AE_NO_TSAN + { + if (inner.enqueue(std::forward(element))) { + sema->signal(); + return true; + } + return false; + } + + + // Attempts to dequeue an element; if the queue is empty, + // returns false instead. If the queue has at least one element, + // moves front to result using operator=, then returns true. + template + bool try_dequeue(U& result) AE_NO_TSAN + { + if (sema->tryWait()) { + bool success = inner.try_dequeue(result); + assert(success); + AE_UNUSED(success); + return true; + } + return false; + } + + + // Attempts to dequeue an element; if the queue is empty, + // waits until an element is available, then dequeues it. + template + void wait_dequeue(U& result) AE_NO_TSAN + { + sema->wait(); + bool success = inner.try_dequeue(result); + AE_UNUSED(result); + assert(success); + AE_UNUSED(success); + } + + + // Attempts to dequeue an element; if the queue is empty, + // waits until an element is available up to the specified timeout, + // then dequeues it and returns true, or returns false if the timeout + // expires before an element can be dequeued. + // Using a negative timeout indicates an indefinite timeout, + // and is thus functionally equivalent to calling wait_dequeue. + template + bool wait_dequeue_timed(U& result, std::int64_t timeout_usecs) AE_NO_TSAN + { + if (!sema->wait(timeout_usecs)) { + return false; + } + bool success = inner.try_dequeue(result); + AE_UNUSED(result); + assert(success); + AE_UNUSED(success); + return true; + } + + +#if __cplusplus > 199711L || _MSC_VER >= 1700 + // Attempts to dequeue an element; if the queue is empty, + // waits until an element is available up to the specified timeout, + // then dequeues it and returns true, or returns false if the timeout + // expires before an element can be dequeued. + // Using a negative timeout indicates an indefinite timeout, + // and is thus functionally equivalent to calling wait_dequeue. + template + inline bool wait_dequeue_timed(U& result, std::chrono::duration const& timeout) AE_NO_TSAN + { + return wait_dequeue_timed(result, std::chrono::duration_cast(timeout).count()); + } +#endif + + + // Returns a pointer to the front element in the queue (the one that + // would be removed next by a call to `try_dequeue` or `pop`). If the + // queue appears empty at the time the method is called, nullptr is + // returned instead. + // Must be called only from the consumer thread. + AE_FORCEINLINE T* peek() AE_NO_TSAN + { + return inner.peek(); + } + + // Removes the front element from the queue, if any, without returning it. + // Returns true on success, or false if the queue appeared empty at the time + // `pop` was called. + AE_FORCEINLINE bool pop() AE_NO_TSAN + { + if (sema->tryWait()) { + bool result = inner.pop(); + assert(result); + AE_UNUSED(result); + return true; + } + return false; + } + + // Returns the approximate number of items currently in the queue. + // Safe to call from both the producer and consumer threads. + AE_FORCEINLINE size_t size_approx() const AE_NO_TSAN + { + return sema->availableApprox(); + } + + +private: + // Disable copying & assignment + BlockingReaderWriterQueue(BlockingReaderWriterQueue const&) { } + BlockingReaderWriterQueue& operator=(BlockingReaderWriterQueue const&) { } + +private: + ReaderWriterQueue inner; + std::unique_ptr sema; +}; + +} // end namespace moodycamel + +#ifdef AE_VCPP +#pragma warning(pop) +#endif diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt index 6c901419..c43d5d7d 100644 --- a/tests/CMakeLists.txt +++ b/tests/CMakeLists.txt @@ -24,6 +24,10 @@ set(SFIZZ_TEST_SOURCES RegionTriggersT.cpp ) +find_package(ZLIB REQUIRED) +add_library(cnpy cnpy.cpp) +target_link_libraries(cnpy PRIVATE ZLIB::ZLIB) + add_executable(sfizz_tests ${SFIZZ_TEST_SOURCES}) target_link_libraries(sfizz_tests PRIVATE sfizz) @@ -31,7 +35,7 @@ target_link_libraries(sfizz_tests PRIVATE sfizz) if(UNIX) target_link_libraries(sfizz_tests PRIVATE stdc++fs atomic) endif(UNIX) -target_link_libraries(sfizz_tests PRIVATE Catch2::Catch2 absl::strings absl::str_format absl::flat_hash_map sndfile readerwriterqueue cnpy-static absl::span absl::algorithm) +target_link_libraries(sfizz_tests PRIVATE absl::strings absl::str_format absl::flat_hash_map sndfile cnpy absl::span absl::algorithm) target_include_directories(sfizz_tests SYSTEM PRIVATE sources) file(COPY "." DESTINATION ${CMAKE_BINARY_DIR}/tests) \ No newline at end of file diff --git a/tests/catch2/catch.hpp b/tests/catch2/catch.hpp new file mode 100644 index 00000000..5feb2a4b --- /dev/null +++ b/tests/catch2/catch.hpp @@ -0,0 +1,17075 @@ +/* + * Catch v2.9.2 + * Generated: 2019-08-08 13:35:12.279703 + * ---------------------------------------------------------- + * This file has been merged from multiple headers. Please don't edit it directly + * Copyright (c) 2019 Two Blue Cubes Ltd. All rights reserved. + * + * Distributed under the Boost Software License, Version 1.0. (See accompanying + * file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt) + */ +#ifndef TWOBLUECUBES_SINGLE_INCLUDE_CATCH_HPP_INCLUDED +#define TWOBLUECUBES_SINGLE_INCLUDE_CATCH_HPP_INCLUDED +// start catch.hpp + + +#define CATCH_VERSION_MAJOR 2 +#define CATCH_VERSION_MINOR 9 +#define CATCH_VERSION_PATCH 2 + +#ifdef __clang__ +# pragma clang system_header +#elif defined __GNUC__ +# pragma GCC system_header +#endif + +// start catch_suppress_warnings.h + +#ifdef __clang__ +# ifdef __ICC // icpc defines the __clang__ macro +# pragma warning(push) +# pragma warning(disable: 161 1682) +# else // __ICC +# pragma clang diagnostic push +# pragma clang diagnostic ignored "-Wpadded" +# pragma clang diagnostic ignored "-Wswitch-enum" +# pragma clang diagnostic ignored "-Wcovered-switch-default" +# endif +#elif defined __GNUC__ + // Because REQUIREs trigger GCC's -Wparentheses, and because still + // supported version of g++ have only buggy support for _Pragmas, + // Wparentheses have to be suppressed globally. +# pragma GCC diagnostic ignored "-Wparentheses" // See #674 for details + +# pragma GCC diagnostic push +# pragma GCC diagnostic ignored "-Wunused-variable" +# pragma GCC diagnostic ignored "-Wpadded" +#endif +// end catch_suppress_warnings.h +#if defined(CATCH_CONFIG_MAIN) || defined(CATCH_CONFIG_RUNNER) +# define CATCH_IMPL +# define CATCH_CONFIG_ALL_PARTS +#endif + +// In the impl file, we want to have access to all parts of the headers +// Can also be used to sanely support PCHs +#if defined(CATCH_CONFIG_ALL_PARTS) +# define CATCH_CONFIG_EXTERNAL_INTERFACES +# if defined(CATCH_CONFIG_DISABLE_MATCHERS) +# undef CATCH_CONFIG_DISABLE_MATCHERS +# endif +# if !defined(CATCH_CONFIG_ENABLE_CHRONO_STRINGMAKER) +# define CATCH_CONFIG_ENABLE_CHRONO_STRINGMAKER +# endif +#endif + +#if !defined(CATCH_CONFIG_IMPL_ONLY) +// start catch_platform.h + +#ifdef __APPLE__ +# include +# if TARGET_OS_OSX == 1 +# define CATCH_PLATFORM_MAC +# elif TARGET_OS_IPHONE == 1 +# define CATCH_PLATFORM_IPHONE +# endif + +#elif defined(linux) || defined(__linux) || defined(__linux__) +# define CATCH_PLATFORM_LINUX + +#elif defined(WIN32) || defined(__WIN32__) || defined(_WIN32) || defined(_MSC_VER) || defined(__MINGW32__) +# define CATCH_PLATFORM_WINDOWS +#endif + +// end catch_platform.h + +#ifdef CATCH_IMPL +# ifndef CLARA_CONFIG_MAIN +# define CLARA_CONFIG_MAIN_NOT_DEFINED +# define CLARA_CONFIG_MAIN +# endif +#endif + +// start catch_user_interfaces.h + +namespace Catch { + unsigned int rngSeed(); +} + +// end catch_user_interfaces.h +// start catch_tag_alias_autoregistrar.h + +// start catch_common.h + +// start catch_compiler_capabilities.h + +// Detect a number of compiler features - by compiler +// The following features are defined: +// +// CATCH_CONFIG_COUNTER : is the __COUNTER__ macro supported? +// CATCH_CONFIG_WINDOWS_SEH : is Windows SEH supported? +// CATCH_CONFIG_POSIX_SIGNALS : are POSIX signals supported? +// CATCH_CONFIG_DISABLE_EXCEPTIONS : Are exceptions enabled? +// **************** +// Note to maintainers: if new toggles are added please document them +// in configuration.md, too +// **************** + +// In general each macro has a _NO_ form +// (e.g. CATCH_CONFIG_NO_POSIX_SIGNALS) which disables the feature. +// Many features, at point of detection, define an _INTERNAL_ macro, so they +// can be combined, en-mass, with the _NO_ forms later. + +#ifdef __cplusplus + +# if (__cplusplus >= 201402L) || (defined(_MSVC_LANG) && _MSVC_LANG >= 201402L) +# define CATCH_CPP14_OR_GREATER +# endif + +# if (__cplusplus >= 201703L) || (defined(_MSVC_LANG) && _MSVC_LANG >= 201703L) +# define CATCH_CPP17_OR_GREATER +# endif + +#endif + +#if defined(CATCH_CPP17_OR_GREATER) +# define CATCH_INTERNAL_CONFIG_CPP17_UNCAUGHT_EXCEPTIONS +#endif + +#ifdef __clang__ + +# define CATCH_INTERNAL_SUPPRESS_GLOBALS_WARNINGS \ + _Pragma( "clang diagnostic push" ) \ + _Pragma( "clang diagnostic ignored \"-Wexit-time-destructors\"" ) \ + _Pragma( "clang diagnostic ignored \"-Wglobal-constructors\"") +# define CATCH_INTERNAL_UNSUPPRESS_GLOBALS_WARNINGS \ + _Pragma( "clang diagnostic pop" ) + +# define CATCH_INTERNAL_SUPPRESS_PARENTHESES_WARNINGS \ + _Pragma( "clang diagnostic push" ) \ + _Pragma( "clang diagnostic ignored \"-Wparentheses\"" ) +# define CATCH_INTERNAL_UNSUPPRESS_PARENTHESES_WARNINGS \ + _Pragma( "clang diagnostic pop" ) + +# define CATCH_INTERNAL_SUPPRESS_UNUSED_WARNINGS \ + _Pragma( "clang diagnostic push" ) \ + _Pragma( "clang diagnostic ignored \"-Wunused-variable\"" ) +# define CATCH_INTERNAL_UNSUPPRESS_UNUSED_WARNINGS \ + _Pragma( "clang diagnostic pop" ) + +# define CATCH_INTERNAL_SUPPRESS_ZERO_VARIADIC_WARNINGS \ + _Pragma( "clang diagnostic push" ) \ + _Pragma( "clang diagnostic ignored \"-Wgnu-zero-variadic-macro-arguments\"" ) +# define CATCH_INTERNAL_UNSUPPRESS_ZERO_VARIADIC_WARNINGS \ + _Pragma( "clang diagnostic pop" ) + +#endif // __clang__ + +//////////////////////////////////////////////////////////////////////////////// +// Assume that non-Windows platforms support posix signals by default +#if !defined(CATCH_PLATFORM_WINDOWS) + #define CATCH_INTERNAL_CONFIG_POSIX_SIGNALS +#endif + +//////////////////////////////////////////////////////////////////////////////// +// We know some environments not to support full POSIX signals +#if defined(__CYGWIN__) || defined(__QNX__) || defined(__EMSCRIPTEN__) || defined(__DJGPP__) + #define CATCH_INTERNAL_CONFIG_NO_POSIX_SIGNALS +#endif + +#ifdef __OS400__ +# define CATCH_INTERNAL_CONFIG_NO_POSIX_SIGNALS +# define CATCH_CONFIG_COLOUR_NONE +#endif + +//////////////////////////////////////////////////////////////////////////////// +// Android somehow still does not support std::to_string +#if defined(__ANDROID__) +# define CATCH_INTERNAL_CONFIG_NO_CPP11_TO_STRING +#endif + +//////////////////////////////////////////////////////////////////////////////// +// Not all Windows environments support SEH properly +#if defined(__MINGW32__) +# define CATCH_INTERNAL_CONFIG_NO_WINDOWS_SEH +#endif + +//////////////////////////////////////////////////////////////////////////////// +// PS4 +#if defined(__ORBIS__) +# define CATCH_INTERNAL_CONFIG_NO_NEW_CAPTURE +#endif + +//////////////////////////////////////////////////////////////////////////////// +// Cygwin +#ifdef __CYGWIN__ + +// Required for some versions of Cygwin to declare gettimeofday +// see: http://stackoverflow.com/questions/36901803/gettimeofday-not-declared-in-this-scope-cygwin +# define _BSD_SOURCE +// some versions of cygwin (most) do not support std::to_string. Use the libstd check. +// https://gcc.gnu.org/onlinedocs/gcc-4.8.2/libstdc++/api/a01053_source.html line 2812-2813 +# if !((__cplusplus >= 201103L) && defined(_GLIBCXX_USE_C99) \ + && !defined(_GLIBCXX_HAVE_BROKEN_VSWPRINTF)) + +# define CATCH_INTERNAL_CONFIG_NO_CPP11_TO_STRING + +# endif +#endif // __CYGWIN__ + +//////////////////////////////////////////////////////////////////////////////// +// Visual C++ +#ifdef _MSC_VER + +# if _MSC_VER >= 1900 // Visual Studio 2015 or newer +# define CATCH_INTERNAL_CONFIG_CPP17_UNCAUGHT_EXCEPTIONS +# endif + +// Universal Windows platform does not support SEH +// Or console colours (or console at all...) +# if defined(WINAPI_FAMILY) && (WINAPI_FAMILY == WINAPI_FAMILY_APP) +# define CATCH_CONFIG_COLOUR_NONE +# else +# define CATCH_INTERNAL_CONFIG_WINDOWS_SEH +# endif + +// MSVC traditional preprocessor needs some workaround for __VA_ARGS__ +// _MSVC_TRADITIONAL == 0 means new conformant preprocessor +// _MSVC_TRADITIONAL == 1 means old traditional non-conformant preprocessor +# if !defined(_MSVC_TRADITIONAL) || (defined(_MSVC_TRADITIONAL) && _MSVC_TRADITIONAL) +# define CATCH_INTERNAL_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR +# endif +#endif // _MSC_VER + +#if defined(_REENTRANT) || defined(_MSC_VER) +// Enable async processing, as -pthread is specified or no additional linking is required +# define CATCH_INTERNAL_CONFIG_USE_ASYNC +#endif // _MSC_VER + +//////////////////////////////////////////////////////////////////////////////// +// Check if we are compiled with -fno-exceptions or equivalent +#if defined(__EXCEPTIONS) || defined(__cpp_exceptions) || defined(_CPPUNWIND) +# define CATCH_INTERNAL_CONFIG_EXCEPTIONS_ENABLED +#endif + +//////////////////////////////////////////////////////////////////////////////// +// DJGPP +#ifdef __DJGPP__ +# define CATCH_INTERNAL_CONFIG_NO_WCHAR +#endif // __DJGPP__ + +//////////////////////////////////////////////////////////////////////////////// +// Embarcadero C++Build +#if defined(__BORLANDC__) + #define CATCH_INTERNAL_CONFIG_POLYFILL_ISNAN +#endif + +//////////////////////////////////////////////////////////////////////////////// + +// Use of __COUNTER__ is suppressed during code analysis in +// CLion/AppCode 2017.2.x and former, because __COUNTER__ is not properly +// handled by it. +// Otherwise all supported compilers support COUNTER macro, +// but user still might want to turn it off +#if ( !defined(__JETBRAINS_IDE__) || __JETBRAINS_IDE__ >= 20170300L ) + #define CATCH_INTERNAL_CONFIG_COUNTER +#endif + +//////////////////////////////////////////////////////////////////////////////// + +// RTX is a special version of Windows that is real time. +// This means that it is detected as Windows, but does not provide +// the same set of capabilities as real Windows does. +#if defined(UNDER_RTSS) || defined(RTX64_BUILD) + #define CATCH_INTERNAL_CONFIG_NO_WINDOWS_SEH + #define CATCH_INTERNAL_CONFIG_NO_ASYNC + #define CATCH_CONFIG_COLOUR_NONE +#endif + +//////////////////////////////////////////////////////////////////////////////// +// Check if string_view is available and usable +// The check is split apart to work around v140 (VS2015) preprocessor issue... +#if defined(__has_include) +#if __has_include() && defined(CATCH_CPP17_OR_GREATER) +# define CATCH_INTERNAL_CONFIG_CPP17_STRING_VIEW +#endif +#endif + +//////////////////////////////////////////////////////////////////////////////// +// Check if optional is available and usable +#if defined(__has_include) +# if __has_include() && defined(CATCH_CPP17_OR_GREATER) +# define CATCH_INTERNAL_CONFIG_CPP17_OPTIONAL +# endif // __has_include() && defined(CATCH_CPP17_OR_GREATER) +#endif // __has_include + +//////////////////////////////////////////////////////////////////////////////// +// Check if byte is available and usable +#if defined(__has_include) +# if __has_include() && defined(CATCH_CPP17_OR_GREATER) +# define CATCH_INTERNAL_CONFIG_CPP17_BYTE +# endif // __has_include() && defined(CATCH_CPP17_OR_GREATER) +#endif // __has_include + +//////////////////////////////////////////////////////////////////////////////// +// Check if variant is available and usable +#if defined(__has_include) +# if __has_include() && defined(CATCH_CPP17_OR_GREATER) +# if defined(__clang__) && (__clang_major__ < 8) + // work around clang bug with libstdc++ https://bugs.llvm.org/show_bug.cgi?id=31852 + // fix should be in clang 8, workaround in libstdc++ 8.2 +# include +# if defined(__GLIBCXX__) && defined(_GLIBCXX_RELEASE) && (_GLIBCXX_RELEASE < 9) +# define CATCH_CONFIG_NO_CPP17_VARIANT +# else +# define CATCH_INTERNAL_CONFIG_CPP17_VARIANT +# endif // defined(__GLIBCXX__) && defined(_GLIBCXX_RELEASE) && (_GLIBCXX_RELEASE < 9) +# else +# define CATCH_INTERNAL_CONFIG_CPP17_VARIANT +# endif // defined(__clang__) && (__clang_major__ < 8) +# endif // __has_include() && defined(CATCH_CPP17_OR_GREATER) +#endif // __has_include + +#if defined(CATCH_INTERNAL_CONFIG_COUNTER) && !defined(CATCH_CONFIG_NO_COUNTER) && !defined(CATCH_CONFIG_COUNTER) +# define CATCH_CONFIG_COUNTER +#endif +#if defined(CATCH_INTERNAL_CONFIG_WINDOWS_SEH) && !defined(CATCH_CONFIG_NO_WINDOWS_SEH) && !defined(CATCH_CONFIG_WINDOWS_SEH) && !defined(CATCH_INTERNAL_CONFIG_NO_WINDOWS_SEH) +# define CATCH_CONFIG_WINDOWS_SEH +#endif +// This is set by default, because we assume that unix compilers are posix-signal-compatible by default. +#if defined(CATCH_INTERNAL_CONFIG_POSIX_SIGNALS) && !defined(CATCH_INTERNAL_CONFIG_NO_POSIX_SIGNALS) && !defined(CATCH_CONFIG_NO_POSIX_SIGNALS) && !defined(CATCH_CONFIG_POSIX_SIGNALS) +# define CATCH_CONFIG_POSIX_SIGNALS +#endif +// This is set by default, because we assume that compilers with no wchar_t support are just rare exceptions. +#if !defined(CATCH_INTERNAL_CONFIG_NO_WCHAR) && !defined(CATCH_CONFIG_NO_WCHAR) && !defined(CATCH_CONFIG_WCHAR) +# define CATCH_CONFIG_WCHAR +#endif + +#if !defined(CATCH_INTERNAL_CONFIG_NO_CPP11_TO_STRING) && !defined(CATCH_CONFIG_NO_CPP11_TO_STRING) && !defined(CATCH_CONFIG_CPP11_TO_STRING) +# define CATCH_CONFIG_CPP11_TO_STRING +#endif + +#if defined(CATCH_INTERNAL_CONFIG_CPP17_OPTIONAL) && !defined(CATCH_CONFIG_NO_CPP17_OPTIONAL) && !defined(CATCH_CONFIG_CPP17_OPTIONAL) +# define CATCH_CONFIG_CPP17_OPTIONAL +#endif + +#if defined(CATCH_INTERNAL_CONFIG_CPP17_UNCAUGHT_EXCEPTIONS) && !defined(CATCH_CONFIG_NO_CPP17_UNCAUGHT_EXCEPTIONS) && !defined(CATCH_CONFIG_CPP17_UNCAUGHT_EXCEPTIONS) +# define CATCH_CONFIG_CPP17_UNCAUGHT_EXCEPTIONS +#endif + +#if defined(CATCH_INTERNAL_CONFIG_CPP17_STRING_VIEW) && !defined(CATCH_CONFIG_NO_CPP17_STRING_VIEW) && !defined(CATCH_CONFIG_CPP17_STRING_VIEW) +# define CATCH_CONFIG_CPP17_STRING_VIEW +#endif + +#if defined(CATCH_INTERNAL_CONFIG_CPP17_VARIANT) && !defined(CATCH_CONFIG_NO_CPP17_VARIANT) && !defined(CATCH_CONFIG_CPP17_VARIANT) +# define CATCH_CONFIG_CPP17_VARIANT +#endif + +#if defined(CATCH_INTERNAL_CONFIG_CPP17_BYTE) && !defined(CATCH_CONFIG_NO_CPP17_BYTE) && !defined(CATCH_CONFIG_CPP17_BYTE) +# define CATCH_CONFIG_CPP17_BYTE +#endif + +#if defined(CATCH_CONFIG_EXPERIMENTAL_REDIRECT) +# define CATCH_INTERNAL_CONFIG_NEW_CAPTURE +#endif + +#if defined(CATCH_INTERNAL_CONFIG_NEW_CAPTURE) && !defined(CATCH_INTERNAL_CONFIG_NO_NEW_CAPTURE) && !defined(CATCH_CONFIG_NO_NEW_CAPTURE) && !defined(CATCH_CONFIG_NEW_CAPTURE) +# define CATCH_CONFIG_NEW_CAPTURE +#endif + +#if !defined(CATCH_INTERNAL_CONFIG_EXCEPTIONS_ENABLED) && !defined(CATCH_CONFIG_DISABLE_EXCEPTIONS) +# define CATCH_CONFIG_DISABLE_EXCEPTIONS +#endif + +#if defined(CATCH_INTERNAL_CONFIG_POLYFILL_ISNAN) && !defined(CATCH_CONFIG_NO_POLYFILL_ISNAN) && !defined(CATCH_CONFIG_POLYFILL_ISNAN) +# define CATCH_CONFIG_POLYFILL_ISNAN +#endif + +#if defined(CATCH_INTERNAL_CONFIG_USE_ASYNC) && !defined(CATCH_INTERNAL_CONFIG_NO_ASYNC) && !defined(CATCH_CONFIG_NO_USE_ASYNC) && !defined(CATCH_CONFIG_USE_ASYNC) +# define CATCH_CONFIG_USE_ASYNC +#endif + +#if !defined(CATCH_INTERNAL_SUPPRESS_PARENTHESES_WARNINGS) +# define CATCH_INTERNAL_SUPPRESS_PARENTHESES_WARNINGS +# define CATCH_INTERNAL_UNSUPPRESS_PARENTHESES_WARNINGS +#endif +#if !defined(CATCH_INTERNAL_SUPPRESS_GLOBALS_WARNINGS) +# define CATCH_INTERNAL_SUPPRESS_GLOBALS_WARNINGS +# define CATCH_INTERNAL_UNSUPPRESS_GLOBALS_WARNINGS +#endif +#if !defined(CATCH_INTERNAL_SUPPRESS_UNUSED_WARNINGS) +# define CATCH_INTERNAL_SUPPRESS_UNUSED_WARNINGS +# define CATCH_INTERNAL_UNSUPPRESS_UNUSED_WARNINGS +#endif +#if !defined(CATCH_INTERNAL_SUPPRESS_ZERO_VARIADIC_WARNINGS) +# define CATCH_INTERNAL_SUPPRESS_ZERO_VARIADIC_WARNINGS +# define CATCH_INTERNAL_UNSUPPRESS_ZERO_VARIADIC_WARNINGS +#endif + +#if defined(CATCH_CONFIG_DISABLE_EXCEPTIONS) +#define CATCH_TRY if ((true)) +#define CATCH_CATCH_ALL if ((false)) +#define CATCH_CATCH_ANON(type) if ((false)) +#else +#define CATCH_TRY try +#define CATCH_CATCH_ALL catch (...) +#define CATCH_CATCH_ANON(type) catch (type) +#endif + +#if defined(CATCH_INTERNAL_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR) && !defined(CATCH_CONFIG_NO_TRADITIONAL_MSVC_PREPROCESSOR) && !defined(CATCH_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR) +#define CATCH_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR +#endif + +// end catch_compiler_capabilities.h +#define INTERNAL_CATCH_UNIQUE_NAME_LINE2( name, line ) name##line +#define INTERNAL_CATCH_UNIQUE_NAME_LINE( name, line ) INTERNAL_CATCH_UNIQUE_NAME_LINE2( name, line ) +#ifdef CATCH_CONFIG_COUNTER +# define INTERNAL_CATCH_UNIQUE_NAME( name ) INTERNAL_CATCH_UNIQUE_NAME_LINE( name, __COUNTER__ ) +#else +# define INTERNAL_CATCH_UNIQUE_NAME( name ) INTERNAL_CATCH_UNIQUE_NAME_LINE( name, __LINE__ ) +#endif + +#include +#include +#include + +// We need a dummy global operator<< so we can bring it into Catch namespace later +struct Catch_global_namespace_dummy {}; +std::ostream& operator<<(std::ostream&, Catch_global_namespace_dummy); + +namespace Catch { + + struct CaseSensitive { enum Choice { + Yes, + No + }; }; + + class NonCopyable { + NonCopyable( NonCopyable const& ) = delete; + NonCopyable( NonCopyable && ) = delete; + NonCopyable& operator = ( NonCopyable const& ) = delete; + NonCopyable& operator = ( NonCopyable && ) = delete; + + protected: + NonCopyable(); + virtual ~NonCopyable(); + }; + + struct SourceLineInfo { + + SourceLineInfo() = delete; + SourceLineInfo( char const* _file, std::size_t _line ) noexcept + : file( _file ), + line( _line ) + {} + + SourceLineInfo( SourceLineInfo const& other ) = default; + SourceLineInfo& operator = ( SourceLineInfo const& ) = default; + SourceLineInfo( SourceLineInfo&& ) noexcept = default; + SourceLineInfo& operator = ( SourceLineInfo&& ) noexcept = default; + + bool empty() const noexcept; + bool operator == ( SourceLineInfo const& other ) const noexcept; + bool operator < ( SourceLineInfo const& other ) const noexcept; + + char const* file; + std::size_t line; + }; + + std::ostream& operator << ( std::ostream& os, SourceLineInfo const& info ); + + // Bring in operator<< from global namespace into Catch namespace + // This is necessary because the overload of operator<< above makes + // lookup stop at namespace Catch + using ::operator<<; + + // Use this in variadic streaming macros to allow + // >> +StreamEndStop + // as well as + // >> stuff +StreamEndStop + struct StreamEndStop { + std::string operator+() const; + }; + template + T const& operator + ( T const& value, StreamEndStop ) { + return value; + } +} + +#define CATCH_INTERNAL_LINEINFO \ + ::Catch::SourceLineInfo( __FILE__, static_cast( __LINE__ ) ) + +// end catch_common.h +namespace Catch { + + struct RegistrarForTagAliases { + RegistrarForTagAliases( char const* alias, char const* tag, SourceLineInfo const& lineInfo ); + }; + +} // end namespace Catch + +#define CATCH_REGISTER_TAG_ALIAS( alias, spec ) \ + CATCH_INTERNAL_SUPPRESS_GLOBALS_WARNINGS \ + namespace{ Catch::RegistrarForTagAliases INTERNAL_CATCH_UNIQUE_NAME( AutoRegisterTagAlias )( alias, spec, CATCH_INTERNAL_LINEINFO ); } \ + CATCH_INTERNAL_UNSUPPRESS_GLOBALS_WARNINGS + +// end catch_tag_alias_autoregistrar.h +// start catch_test_registry.h + +// start catch_interfaces_testcase.h + +#include + +namespace Catch { + + class TestSpec; + + struct ITestInvoker { + virtual void invoke () const = 0; + virtual ~ITestInvoker(); + }; + + class TestCase; + struct IConfig; + + struct ITestCaseRegistry { + virtual ~ITestCaseRegistry(); + virtual std::vector const& getAllTests() const = 0; + virtual std::vector const& getAllTestsSorted( IConfig const& config ) const = 0; + }; + + bool isThrowSafe( TestCase const& testCase, IConfig const& config ); + bool matchTest( TestCase const& testCase, TestSpec const& testSpec, IConfig const& config ); + std::vector filterTests( std::vector const& testCases, TestSpec const& testSpec, IConfig const& config ); + std::vector const& getAllTestCasesSorted( IConfig const& config ); + +} + +// end catch_interfaces_testcase.h +// start catch_stringref.h + +#include +#include +#include + +namespace Catch { + + /// A non-owning string class (similar to the forthcoming std::string_view) + /// Note that, because a StringRef may be a substring of another string, + /// it may not be null terminated. c_str() must return a null terminated + /// string, however, and so the StringRef will internally take ownership + /// (taking a copy), if necessary. In theory this ownership is not externally + /// visible - but it does mean (substring) StringRefs should not be shared between + /// threads. + class StringRef { + public: + using size_type = std::size_t; + + private: + friend struct StringRefTestAccess; + + char const* m_start; + size_type m_size; + + char* m_data = nullptr; + + void takeOwnership(); + + static constexpr char const* const s_empty = ""; + + public: // construction/ assignment + StringRef() noexcept + : StringRef( s_empty, 0 ) + {} + + StringRef( StringRef const& other ) noexcept + : m_start( other.m_start ), + m_size( other.m_size ) + {} + + StringRef( StringRef&& other ) noexcept + : m_start( other.m_start ), + m_size( other.m_size ), + m_data( other.m_data ) + { + other.m_data = nullptr; + } + + StringRef( char const* rawChars ) noexcept; + + StringRef( char const* rawChars, size_type size ) noexcept + : m_start( rawChars ), + m_size( size ) + {} + + StringRef( std::string const& stdString ) noexcept + : m_start( stdString.c_str() ), + m_size( stdString.size() ) + {} + + ~StringRef() noexcept { + delete[] m_data; + } + + auto operator = ( StringRef const &other ) noexcept -> StringRef& { + delete[] m_data; + m_data = nullptr; + m_start = other.m_start; + m_size = other.m_size; + return *this; + } + + operator std::string() const; + + void swap( StringRef& other ) noexcept; + + public: // operators + auto operator == ( StringRef const& other ) const noexcept -> bool; + auto operator != ( StringRef const& other ) const noexcept -> bool; + + auto operator[] ( size_type index ) const noexcept -> char; + + public: // named queries + auto empty() const noexcept -> bool { + return m_size == 0; + } + auto size() const noexcept -> size_type { + return m_size; + } + + auto numberOfCharacters() const noexcept -> size_type; + auto c_str() const -> char const*; + + public: // substrings and searches + auto substr( size_type start, size_type size ) const noexcept -> StringRef; + + // Returns the current start pointer. + // Note that the pointer can change when if the StringRef is a substring + auto currentData() const noexcept -> char const*; + + private: // ownership queries - may not be consistent between calls + auto isOwned() const noexcept -> bool; + auto isSubstring() const noexcept -> bool; + }; + + auto operator + ( StringRef const& lhs, StringRef const& rhs ) -> std::string; + auto operator + ( StringRef const& lhs, char const* rhs ) -> std::string; + auto operator + ( char const* lhs, StringRef const& rhs ) -> std::string; + + auto operator += ( std::string& lhs, StringRef const& sr ) -> std::string&; + auto operator << ( std::ostream& os, StringRef const& sr ) -> std::ostream&; + + inline auto operator "" _sr( char const* rawChars, std::size_t size ) noexcept -> StringRef { + return StringRef( rawChars, size ); + } + +} // namespace Catch + +inline auto operator "" _catch_sr( char const* rawChars, std::size_t size ) noexcept -> Catch::StringRef { + return Catch::StringRef( rawChars, size ); +} + +// end catch_stringref.h +// start catch_type_traits.hpp + + +#include + +namespace Catch{ + +#ifdef CATCH_CPP17_OR_GREATER + template + inline constexpr auto is_unique = std::true_type{}; + + template + inline constexpr auto is_unique = std::bool_constant< + (!std::is_same_v && ...) && is_unique + >{}; +#else + +template +struct is_unique : std::true_type{}; + +template +struct is_unique : std::integral_constant +::value + && is_unique::value + && is_unique::value +>{}; + +#endif +} + +// end catch_type_traits.hpp +// start catch_preprocessor.hpp + + +#define CATCH_RECURSION_LEVEL0(...) __VA_ARGS__ +#define CATCH_RECURSION_LEVEL1(...) CATCH_RECURSION_LEVEL0(CATCH_RECURSION_LEVEL0(CATCH_RECURSION_LEVEL0(__VA_ARGS__))) +#define CATCH_RECURSION_LEVEL2(...) CATCH_RECURSION_LEVEL1(CATCH_RECURSION_LEVEL1(CATCH_RECURSION_LEVEL1(__VA_ARGS__))) +#define CATCH_RECURSION_LEVEL3(...) CATCH_RECURSION_LEVEL2(CATCH_RECURSION_LEVEL2(CATCH_RECURSION_LEVEL2(__VA_ARGS__))) +#define CATCH_RECURSION_LEVEL4(...) CATCH_RECURSION_LEVEL3(CATCH_RECURSION_LEVEL3(CATCH_RECURSION_LEVEL3(__VA_ARGS__))) +#define CATCH_RECURSION_LEVEL5(...) CATCH_RECURSION_LEVEL4(CATCH_RECURSION_LEVEL4(CATCH_RECURSION_LEVEL4(__VA_ARGS__))) + +#ifdef CATCH_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR +#define INTERNAL_CATCH_EXPAND_VARGS(...) __VA_ARGS__ +// MSVC needs more evaluations +#define CATCH_RECURSION_LEVEL6(...) CATCH_RECURSION_LEVEL5(CATCH_RECURSION_LEVEL5(CATCH_RECURSION_LEVEL5(__VA_ARGS__))) +#define CATCH_RECURSE(...) CATCH_RECURSION_LEVEL6(CATCH_RECURSION_LEVEL6(__VA_ARGS__)) +#else +#define CATCH_RECURSE(...) CATCH_RECURSION_LEVEL5(__VA_ARGS__) +#endif + +#define CATCH_REC_END(...) +#define CATCH_REC_OUT + +#define CATCH_EMPTY() +#define CATCH_DEFER(id) id CATCH_EMPTY() + +#define CATCH_REC_GET_END2() 0, CATCH_REC_END +#define CATCH_REC_GET_END1(...) CATCH_REC_GET_END2 +#define CATCH_REC_GET_END(...) CATCH_REC_GET_END1 +#define CATCH_REC_NEXT0(test, next, ...) next CATCH_REC_OUT +#define CATCH_REC_NEXT1(test, next) CATCH_DEFER ( CATCH_REC_NEXT0 ) ( test, next, 0) +#define CATCH_REC_NEXT(test, next) CATCH_REC_NEXT1(CATCH_REC_GET_END test, next) + +#define CATCH_REC_LIST0(f, x, peek, ...) , f(x) CATCH_DEFER ( CATCH_REC_NEXT(peek, CATCH_REC_LIST1) ) ( f, peek, __VA_ARGS__ ) +#define CATCH_REC_LIST1(f, x, peek, ...) , f(x) CATCH_DEFER ( CATCH_REC_NEXT(peek, CATCH_REC_LIST0) ) ( f, peek, __VA_ARGS__ ) +#define CATCH_REC_LIST2(f, x, peek, ...) f(x) CATCH_DEFER ( CATCH_REC_NEXT(peek, CATCH_REC_LIST1) ) ( f, peek, __VA_ARGS__ ) + +#define CATCH_REC_LIST0_UD(f, userdata, x, peek, ...) , f(userdata, x) CATCH_DEFER ( CATCH_REC_NEXT(peek, CATCH_REC_LIST1_UD) ) ( f, userdata, peek, __VA_ARGS__ ) +#define CATCH_REC_LIST1_UD(f, userdata, x, peek, ...) , f(userdata, x) CATCH_DEFER ( CATCH_REC_NEXT(peek, CATCH_REC_LIST0_UD) ) ( f, userdata, peek, __VA_ARGS__ ) +#define CATCH_REC_LIST2_UD(f, userdata, x, peek, ...) f(userdata, x) CATCH_DEFER ( CATCH_REC_NEXT(peek, CATCH_REC_LIST1_UD) ) ( f, userdata, peek, __VA_ARGS__ ) + +// Applies the function macro `f` to each of the remaining parameters, inserts commas between the results, +// and passes userdata as the first parameter to each invocation, +// e.g. CATCH_REC_LIST_UD(f, x, a, b, c) evaluates to f(x, a), f(x, b), f(x, c) +#define CATCH_REC_LIST_UD(f, userdata, ...) CATCH_RECURSE(CATCH_REC_LIST2_UD(f, userdata, __VA_ARGS__, ()()(), ()()(), ()()(), 0)) + +#define CATCH_REC_LIST(f, ...) CATCH_RECURSE(CATCH_REC_LIST2(f, __VA_ARGS__, ()()(), ()()(), ()()(), 0)) + +#define INTERNAL_CATCH_EXPAND1(param) INTERNAL_CATCH_EXPAND2(param) +#define INTERNAL_CATCH_EXPAND2(...) INTERNAL_CATCH_NO## __VA_ARGS__ +#define INTERNAL_CATCH_DEF(...) INTERNAL_CATCH_DEF __VA_ARGS__ +#define INTERNAL_CATCH_NOINTERNAL_CATCH_DEF +#define INTERNAL_CATCH_STRINGIZE(...) INTERNAL_CATCH_STRINGIZE2(__VA_ARGS__) +#ifndef CATCH_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR +#define INTERNAL_CATCH_STRINGIZE2(...) #__VA_ARGS__ +#define INTERNAL_CATCH_STRINGIZE_WITHOUT_PARENS(param) INTERNAL_CATCH_STRINGIZE(INTERNAL_CATCH_REMOVE_PARENS(param)) +#else +// MSVC is adding extra space and needs another indirection to expand INTERNAL_CATCH_NOINTERNAL_CATCH_DEF +#define INTERNAL_CATCH_STRINGIZE2(...) INTERNAL_CATCH_STRINGIZE3(__VA_ARGS__) +#define INTERNAL_CATCH_STRINGIZE3(...) #__VA_ARGS__ +#define INTERNAL_CATCH_STRINGIZE_WITHOUT_PARENS(param) (INTERNAL_CATCH_STRINGIZE(INTERNAL_CATCH_REMOVE_PARENS(param)) + 1) +#endif + +#define INTERNAL_CATCH_MAKE_NAMESPACE2(...) ns_##__VA_ARGS__ +#define INTERNAL_CATCH_MAKE_NAMESPACE(name) INTERNAL_CATCH_MAKE_NAMESPACE2(name) + +#define INTERNAL_CATCH_REMOVE_PARENS(...) INTERNAL_CATCH_EXPAND1(INTERNAL_CATCH_DEF __VA_ARGS__) + +#ifndef CATCH_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR +#define INTERNAL_CATCH_MAKE_TYPE_LIST2(...) decltype(get_wrapper()) +#define INTERNAL_CATCH_MAKE_TYPE_LIST(...) INTERNAL_CATCH_MAKE_TYPE_LIST2(INTERNAL_CATCH_REMOVE_PARENS(__VA_ARGS__)) +#else +#define INTERNAL_CATCH_MAKE_TYPE_LIST2(...) INTERNAL_CATCH_EXPAND_VARGS(decltype(get_wrapper())) +#define INTERNAL_CATCH_MAKE_TYPE_LIST(...) INTERNAL_CATCH_EXPAND_VARGS(INTERNAL_CATCH_MAKE_TYPE_LIST2(INTERNAL_CATCH_REMOVE_PARENS(__VA_ARGS__))) +#endif + +#define INTERNAL_CATCH_MAKE_TYPE_LISTS_FROM_TYPES(...)\ + CATCH_REC_LIST(INTERNAL_CATCH_MAKE_TYPE_LIST,__VA_ARGS__) + +#define INTERNAL_CATCH_REMOVE_PARENS_1_ARG(_0) INTERNAL_CATCH_REMOVE_PARENS(_0) +#define INTERNAL_CATCH_REMOVE_PARENS_2_ARG(_0, _1) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_1_ARG(_1) +#define INTERNAL_CATCH_REMOVE_PARENS_3_ARG(_0, _1, _2) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_2_ARG(_1, _2) +#define INTERNAL_CATCH_REMOVE_PARENS_4_ARG(_0, _1, _2, _3) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_3_ARG(_1, _2, _3) +#define INTERNAL_CATCH_REMOVE_PARENS_5_ARG(_0, _1, _2, _3, _4) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_4_ARG(_1, _2, _3, _4) +#define INTERNAL_CATCH_REMOVE_PARENS_6_ARG(_0, _1, _2, _3, _4, _5) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_5_ARG(_1, _2, _3, _4, _5) +#define INTERNAL_CATCH_REMOVE_PARENS_7_ARG(_0, _1, _2, _3, _4, _5, _6) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_6_ARG(_1, _2, _4, _5, _6) +#define INTERNAL_CATCH_REMOVE_PARENS_8_ARG(_0, _1, _2, _3, _4, _5, _6, _7) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_7_ARG(_1, _2, _3, _4, _5, _6, _7) +#define INTERNAL_CATCH_REMOVE_PARENS_9_ARG(_0, _1, _2, _3, _4, _5, _6, _7, _8) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_8_ARG(_1, _2, _3, _4, _5, _6, _7, _8) +#define INTERNAL_CATCH_REMOVE_PARENS_10_ARG(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_9_ARG(_1, _2, _3, _4, _5, _6, _7, _8, _9) +#define INTERNAL_CATCH_REMOVE_PARENS_11_ARG(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10) INTERNAL_CATCH_REMOVE_PARENS(_0), INTERNAL_CATCH_REMOVE_PARENS_10_ARG(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10) + +#define INTERNAL_CATCH_VA_NARGS_IMPL(_0, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, N, ...) N + +#define INTERNAL_CATCH_TYPE_GEN\ + template struct TypeList {};\ + template\ + constexpr auto get_wrapper() noexcept -> TypeList { return {}; }\ + \ + template class L1, typename...E1, template class L2, typename...E2> \ + constexpr auto append(L1, L2) noexcept -> L1 { return {}; }\ + template< template class L1, typename...E1, template class L2, typename...E2, typename...Rest>\ + constexpr auto append(L1, L2, Rest...) noexcept -> decltype(append(L1{}, Rest{}...)) { return {}; }\ + template< template class L1, typename...E1, typename...Rest>\ + constexpr auto append(L1, TypeList, Rest...) noexcept -> L1 { return {}; }\ + \ + template< template class Container, template class List, typename...elems>\ + constexpr auto rewrap(List) noexcept -> TypeList> { return {}; }\ + template< template class Container, template class List, class...Elems, typename...Elements>\ + constexpr auto rewrap(List,Elements...) noexcept -> decltype(append(TypeList>{}, rewrap(Elements{}...))) { return {}; }\ + \ + template