STL/stl/inc/atomic

// atomic standard header

// Copyright (c) Microsoft Corporation.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

#pragma once
#ifndef _ATOMIC_
#define _ATOMIC_
#include <yvals.h>
#if _STL_COMPILER_PREPROCESSOR

#ifdef _M_CEE_PURE
#error <atomic> is not supported when compiling with /clr:pure.
#endif // _M_CEE_PURE

#include <stddef.h> // for size_t
#include <stdint.h>
#include <string.h>
#include <xatomic.h>

#pragma pack(push, _CRT_PACKING)
#pragma warning(push, _STL_WARNING_LEVEL)
#pragma warning(disable : _STL_DISABLED_WARNINGS)
_STL_DISABLE_CLANG_WARNINGS
#pragma push_macro("new")
#undef new

#define _Compiler_barrier() _STL_DISABLE_DEPRECATED_WARNING _ReadWriteBarrier() _STL_RESTORE_DEPRECATED_WARNING

#if defined(_M_ARM) || defined(_M_ARM64)
#define _Memory_barrier()             __dmb(0xB) // inner shared data memory barrier
#define _Compiler_or_memory_barrier() _Memory_barrier()

#define _ISO_VOLATILE_STORE8(_Storage, _Value)  __iso_volatile_store8(_Atomic_address_as<char>(_Storage), _Value)
#define _ISO_VOLATILE_STORE16(_Storage, _Value) __iso_volatile_store16(_Atomic_address_as<short>(_Storage), _Value)
#define _ISO_VOLATILE_STORE32(_Storage, _Value) __iso_volatile_store32(_Atomic_address_as<int>(_Storage), _Value)
#define _ISO_VOLATILE_STORE64(_Storage, _Value) __iso_volatile_store64(_Atomic_address_as<long long>(_Storage), _Value)
#define _ISO_VOLATILE_LOAD8(_Storage)           __iso_volatile_load8(_Atomic_address_as<const char>(_Storage))
#define _ISO_VOLATILE_LOAD16(_Storage)          __iso_volatile_load16(_Atomic_address_as<const short>(_Storage))

#elif defined(_M_IX86) || defined(_M_X64)
// x86/x64 hardware only emits memory barriers inside _Interlocked intrinsics
#define _Compiler_or_memory_barrier() _Compiler_barrier()

#define _ISO_VOLATILE_STORE8(_Storage, _Value)  (*_Atomic_address_as<char>(_Storage) = _Value)
#define _ISO_VOLATILE_STORE16(_Storage, _Value) (*_Atomic_address_as<short>(_Storage) = _Value)
#define _ISO_VOLATILE_STORE32(_Storage, _Value) (*_Atomic_address_as<long>(_Storage) = _Value)
#define _ISO_VOLATILE_STORE64(_Storage, _Value) (*_Atomic_address_as<long long>(_Storage) = _Value)
#define _ISO_VOLATILE_LOAD8(_Storage)           (*_Atomic_address_as<const char>(_Storage))
#define _ISO_VOLATILE_LOAD16(_Storage)          (*_Atomic_address_as<const short>(_Storage))

#else // ^^^ x86/x64 / unsupported hardware vvv
#error Unsupported hardware
#endif // hardware

#ifndef _INVALID_MEMORY_ORDER
#ifdef _DEBUG
#define _INVALID_MEMORY_ORDER _STL_REPORT_ERROR("Invalid memory order")
#else // ^^^ _DEBUG / !_DEBUG vvv
#define _INVALID_MEMORY_ORDER
#endif // _DEBUG
#endif // _INVALID_MEMORY_ORDER

#if 0 // TRANSITION, ABI
// MACRO _STD_COMPARE_EXCHANGE_128
#if _STD_ATOMIC_ALWAYS_USE_CMPXCHG16B || defined(_M_ARM64)
#define _STD_COMPARE_EXCHANGE_128 _InterlockedCompareExchange128
#endif // _STD_ATOMIC_ALWAYS_USE_CMPXCHG16B || defined(_M_ARM64)
#if defined(_M_X64) && !_STD_ATOMIC_ALWAYS_USE_CMPXCHG16B
// 16-byte atomics are separately compiled for x64, as not all x64 hardware has the cmpxchg16b
// instruction; in the event this instruction is not available, the fallback is a global
// CRITICAL_SECTION shared by all 16-byte atomics.
// (Note: machines without this instruction typically have 2 cores or fewer, so this isn't too bad)
// All pointer parameters must be 16-byte aligned.
_NODISCARD extern "C" unsigned char __cdecl __std_atomic_compare_exchange_128(
    _Inout_bytecount_(16) long long* _Destination, _In_ long long _ExchangeHigh, _In_ long long _ExchangeLow,
    _Inout_bytecount_(16) long long* _ComparandResult) noexcept;
_NODISCARD extern "C" bool __cdecl __std_atomic_has_cmpxchg16b() noexcept;
#define _STD_COMPARE_EXCHANGE_128 __std_atomic_compare_exchange_128
#endif // defined(_M_X64) && !_STD_ATOMIC_ALWAYS_USE_CMPXCHG16B

// MACRO _ATOMIC_HAS_DCAS
// Controls whether atomic::is_always_lock_free triggers for sizeof(void *) or 2 * sizeof(void *)
#if _STD_ATOMIC_ALWAYS_USE_CMPXCHG16B || !defined(_M_X64)
#define _ATOMIC_HAS_DCAS 1
#else // ^^ We always have DCAS / We only sometimes have DCAS vvv
#define _ATOMIC_HAS_DCAS 0
#endif // _STD_ATOMIC_ALWAYS_USE_CMPXCHG16B || !defined(_M_X64)
#endif // TRANSITION, ABI

// MACRO _ATOMIC_CHOOSE_INTRINSIC
#if defined(_M_IX86) || defined(_M_X64)
#define _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _Intrinsic, ...) \
    _Check_memory_order(_Order);                                   \
    _Result = _Intrinsic(__VA_ARGS__)
#elif defined(_M_ARM) || defined(_M_ARM64)
#define _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _Intrinsic, ...) \
    switch (_Order) {                                              \
    case memory_order_relaxed:                                     \
        _Result = _INTRIN_RELAXED(_Intrinsic)(__VA_ARGS__);        \
        break;                                                     \
    case memory_order_consume:                                     \
    case memory_order_acquire:                                     \
        _Result = _INTRIN_ACQUIRE(_Intrinsic)(__VA_ARGS__);        \
        break;                                                     \
    case memory_order_release:                                     \
        _Result = _INTRIN_RELEASE(_Intrinsic)(__VA_ARGS__);        \
        break;                                                     \
    default:                                                       \
        _INVALID_MEMORY_ORDER;                                     \
        /* [[fallthrough]]; */                                     \
    case memory_order_acq_rel:                                     \
    case memory_order_seq_cst:                                     \
        _Result = _Intrinsic(__VA_ARGS__);                         \
        break;                                                     \
    }
#endif // hardware

// LOCK-FREE PROPERTY
#define ATOMIC_BOOL_LOCK_FREE 2
#define ATOMIC_CHAR_LOCK_FREE 2
#ifdef __cpp_lib_char8_t
#define ATOMIC_CHAR8_T_LOCK_FREE 2
#endif // __cpp_lib_char8_t
#define ATOMIC_CHAR16_T_LOCK_FREE 2
#define ATOMIC_CHAR32_T_LOCK_FREE 2
#define ATOMIC_WCHAR_T_LOCK_FREE  2
#define ATOMIC_SHORT_LOCK_FREE    2
#define ATOMIC_INT_LOCK_FREE      2
#define ATOMIC_LONG_LOCK_FREE     2
#define ATOMIC_LLONG_LOCK_FREE    2
#define ATOMIC_POINTER_LOCK_FREE  2

_STD_BEGIN

// FUNCTION TEMPLATE kill_dependency
template <class _Ty>
_Ty kill_dependency(_Ty _Arg) noexcept { // "magic" template that kills dependency ordering when called
    return _Arg;
}

// FUNCTION _Check_memory_order
inline void _Check_memory_order(const memory_order _Order) noexcept {
    // check that _Order is a valid memory_order
    if (static_cast<unsigned int>(_Order) > static_cast<unsigned int>(memory_order_seq_cst)) {
        _INVALID_MEMORY_ORDER;
    }
}

// FUNCTION _Check_store_memory_order
inline void _Check_store_memory_order(const memory_order _Order) noexcept {
    switch (_Order) {
    case memory_order_relaxed:
    case memory_order_release:
    case memory_order_seq_cst:
        // nothing to do
        break;
    case memory_order_consume:
    case memory_order_acquire:
    case memory_order_acq_rel:
    default:
        _INVALID_MEMORY_ORDER;
        break;
    }
}

// FUNCTION _Check_load_memory_order
inline void _Check_load_memory_order(const memory_order _Order) noexcept {
    switch (_Order) {
    case memory_order_relaxed:
    case memory_order_consume:
    case memory_order_acquire:
    case memory_order_seq_cst:
        // nothing to do
        break;
    case memory_order_release:
    case memory_order_acq_rel:
    default:
        _INVALID_MEMORY_ORDER;
        break;
    }
}

// FUNCTION _Combine_cas_memory_orders
_NODISCARD inline memory_order _Combine_cas_memory_orders(
    const memory_order _Success, const memory_order _Failure) noexcept {
    // Finds upper bound of a compare/exchange memory order
    // pair, according to the following partial order:
    //     seq_cst
    //        |
    //     acq_rel
    //     /     \
    // acquire  release
    //    |       |
    // consume    |
    //     \     /
    //     relaxed
    static constexpr memory_order _Combined_memory_orders[6][6] = {// combined upper bounds
        {memory_order_relaxed, memory_order_consume, memory_order_acquire, memory_order_release, memory_order_acq_rel,
            memory_order_seq_cst},
        {memory_order_consume, memory_order_consume, memory_order_acquire, memory_order_acq_rel, memory_order_acq_rel,
            memory_order_seq_cst},
        {memory_order_acquire, memory_order_acquire, memory_order_acquire, memory_order_acq_rel, memory_order_acq_rel,
            memory_order_seq_cst},
        {memory_order_release, memory_order_acq_rel, memory_order_acq_rel, memory_order_release, memory_order_acq_rel,
            memory_order_seq_cst},
        {memory_order_acq_rel, memory_order_acq_rel, memory_order_acq_rel, memory_order_acq_rel, memory_order_acq_rel,
            memory_order_seq_cst},
        {memory_order_seq_cst, memory_order_seq_cst, memory_order_seq_cst, memory_order_seq_cst, memory_order_seq_cst,
            memory_order_seq_cst}};

    _Check_memory_order(_Success);
    _Check_load_memory_order(_Failure);
    return _Combined_memory_orders[static_cast<int>(_Success)][static_cast<int>(_Failure)];
}

// FUNCTION TEMPLATE _Atomic_reinterpret_as
template <class _Integral, class _Ty>
_NODISCARD _Integral _Atomic_reinterpret_as(const _Ty& _Source) noexcept {
    // interprets _Source as the supplied integral type
    static_assert(is_integral_v<_Integral>, "Tried to reinterpret memory as non-integral");
#if _HAS_IF_CONSTEXPR
    if constexpr (is_integral_v<_Ty> && sizeof(_Integral) == sizeof(_Ty)) {
        return static_cast<_Integral>(_Source);
    } else if constexpr (is_pointer_v<_Ty> && sizeof(_Integral) == sizeof(_Ty)) {
        return reinterpret_cast<_Integral>(_Source);
    } else
#endif // _HAS_IF_CONSTEXPR
    {
        _Integral _Result{}; // zero padding bits
        _CSTD memcpy(&_Result, _STD addressof(_Source), sizeof(_Source));
        return _Result;
    }
}

// FUNCTION _Load_barrier
inline void _Load_barrier(const memory_order _Order) noexcept { // implement memory barrier for atomic load functions
    switch (_Order) {
    case memory_order_relaxed:
        // no barrier
        break;
    default:
    case memory_order_release:
    case memory_order_acq_rel:
        _INVALID_MEMORY_ORDER;
        // [[fallthrough]];
    case memory_order_consume:
    case memory_order_acquire:
    case memory_order_seq_cst:
        _Compiler_or_memory_barrier();
        break;
    }
}

#if 1 // TRANSITION, ABI
template <class _Ty>
struct _Atomic_padded {
    alignas(sizeof(_Ty)) mutable _Ty _Value; // align to sizeof(T); x86 stack aligns 8-byte objects on 4-byte boundaries
};

#else // ^^^ don't break ABI / break ABI vvv
// STRUCT TEMPLATE _Atomic_storage_traits
template <class _Ty>
struct _Atomic_storage_traits { // properties for how _Ty is stored in an atomic
    static constexpr size_t _Storage_size =
        sizeof(_Ty) == 1 ? 1
                         : sizeof(_Ty) == 2 ? 2
                                            : sizeof(_Ty) <= 4 ? 4
                                                               : sizeof(_Ty) <= 8 ? 8
#if defined(_M_X64) || defined(_M_ARM64)
                                                                                  : sizeof(_Ty) <= 16 ? 16
#endif // 64 bits
                                                                                                      : sizeof(_Ty);

    static constexpr size_t _Padding_size = _Storage_size - sizeof(_Ty);
    static constexpr bool _Uses_padding   = _Padding_size != 0;
};

// STRUCT TEMPLATE _Atomic_padded
template <class _Ty, bool = _Atomic_storage_traits<_Ty>::_Uses_padding>
struct _Atomic_padded { // aggregate to allow explicit constexpr zeroing of padding
    alignas(_Atomic_storage_traits<_Ty>::_Storage_size) mutable _Ty _Value;
    mutable unsigned char _Padding[_Atomic_storage_traits<_Ty>::_Padding_size];
};

template <class _Ty>
struct _Atomic_padded<_Ty, false> {
    alignas(sizeof(_Ty)) mutable _Ty _Value; // align to sizeof(T); x86 stack aligns 8-byte objects on 4-byte boundaries
};
#endif // TRANSITION, ABI

// STRUCT TEMPLATE _Atomic_storage
#if 1 // TRANSITION, ABI
template <class _Ty, size_t = sizeof(_Ty)>
#else // ^^^ don't break ABI / break ABI vvv
template <class _Ty, size_t = _Atomic_storage_traits<_Ty>::_Storage_size>
#endif // TRANSITION, ABI
struct _Atomic_storage {
    // Provides operations common to all specializations of std::atomic, load, store, exchange, and CAS.
    // Locking version used when hardware has no atomic operations for sizeof(_Ty).

    _Atomic_storage() = default;

    /* implicit */ constexpr _Atomic_storage(const _Ty _Value) noexcept : _Storage(_Value) {
        // non-atomically initialize this atomic
    }

    void store(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // store with sequential consistency
        _Check_store_memory_order(_Order);
        _Lock();
        _Storage = _Value;
        _Unlock();
    }

    _NODISCARD _Ty load(const memory_order _Order = memory_order_seq_cst) const noexcept {
        // load with sequential consistency
        _Check_load_memory_order(_Order);
        _Lock();
        _Ty _Local(_Storage);
        _Unlock();
        return _Local;
    }

    _Ty exchange(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // exchange _Value with _Storage with sequential consistency
        _Check_memory_order(_Order);
        _Lock();
        _Ty _Result(_Storage);
        _Storage = _Value;
        _Unlock();
        return _Result;
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired,
        const memory_order _Order = memory_order_seq_cst) noexcept { // CAS with sequential consistency, plain
        _Check_memory_order(_Order);
        const auto _Storage_ptr  = _STD addressof(_Storage);
        const auto _Expected_ptr = _STD addressof(_Expected);
        bool _Result;
        _Lock();
        if (_CSTD memcmp(_Storage_ptr, _Expected_ptr, sizeof(_Ty)) == 0) {
            _CSTD memcpy(_Storage_ptr, _STD addressof(_Desired), sizeof(_Ty));
            _Result = true;
        } else {
            _CSTD memcpy(_Expected_ptr, _Storage_ptr, sizeof(_Ty));
            _Result = false;
        }

        _Unlock();
        return _Result;
    }

#if 1 // TRANSITION, ABI
    void _Lock() const noexcept { // lock the spinlock
        while (_InterlockedExchange(&_Spinlock, 1)) {
            _YIELD_PROCESSOR();
        }
    }

    void _Unlock() const noexcept { // unlock the spinlock
#if defined(_M_ARM) || defined(_M_ARM64)
        _Memory_barrier();
        __iso_volatile_store32(reinterpret_cast<int*>(&_Spinlock), 0);
        _Memory_barrier();
#else // ^^^ ARM32/ARM64 hardware / x86/x64 hardware vvv
        _InterlockedExchange(&_Spinlock, 0);
#endif // hardware
    }

private:
    mutable long _Spinlock = 0;

public:
    _Ty _Storage{};

#else // ^^^ don't break ABI / break ABI vvv
    void _Lock() const noexcept { // lock the spinlock
        while (_InterlockedExchange8(&_Spinlock, 1)) {
            _YIELD_PROCESSOR();
        }
    }

    void _Unlock() const noexcept { // unlock the spinlock
#if defined(_M_ARM) || defined(_M_ARM64)
        _Memory_barrier();
        __iso_volatile_store8(&_Spinlock, 0);
        _Memory_barrier();
#else // ^^^ ARM32/ARM64 hardware / x86/x64 hardware vvv
        _InterlockedExchange8(&_Spinlock, 0);
#endif // hardware
    }

    _Ty _Storage;
    mutable char _Spinlock = 0;
#endif // TRANSITION, ABI
};

template <class _Ty>
struct _Atomic_storage<_Ty, 1> { // lock-free using 1-byte intrinsics
    _Atomic_storage() = default;

    /* implicit */ constexpr _Atomic_storage(const _Ty _Value) noexcept : _Storage{_Value} {
        // non-atomically initialize this atomic
    }

    void store(const _Ty _Value) noexcept { // store with sequential consistency
        const auto _Mem      = _Atomic_address_as<char>(_Storage);
        const char _As_bytes = _Atomic_reinterpret_as<char>(_Value);
#if defined(_M_ARM) || defined(_M_ARM64)
        _Memory_barrier();
        __iso_volatile_store8(_Mem, _As_bytes);
        _Memory_barrier();
#else // ^^^ ARM32/ARM64 hardware / x86/x64 hardware vvv
        (void) _InterlockedExchange8(_Mem, _As_bytes);
#endif // hardware
    }

    void store(const _Ty _Value, const memory_order _Order) noexcept { // store with given memory order
        const char _As_bytes = _Atomic_reinterpret_as<char>(_Value);
        switch (_Order) {
        case memory_order_relaxed:
            _ISO_VOLATILE_STORE8(_Storage, _As_bytes);
            return;
        case memory_order_release:
            _Compiler_or_memory_barrier();
            _ISO_VOLATILE_STORE8(_Storage, _As_bytes);
            return;
        default:
        case memory_order_consume:
        case memory_order_acquire:
        case memory_order_acq_rel:
            _INVALID_MEMORY_ORDER;
            // [[fallthrough]];
        case memory_order_seq_cst:
            store(_Value);
            return;
        }
    }

    _NODISCARD _Ty load() const noexcept { // load with sequential consistency
        char _As_bytes = _ISO_VOLATILE_LOAD8(_Storage);
        _Compiler_or_memory_barrier();
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _NODISCARD _Ty load(const memory_order _Order) const noexcept { // load with given memory order
        char _As_bytes = _ISO_VOLATILE_LOAD8(_Storage);
        _Load_barrier(_Order);
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _Ty exchange(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // exchange with given memory order
        char _As_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _As_bytes, _InterlockedExchange8, _Atomic_address_as<char>(_Storage),
            _Atomic_reinterpret_as<char>(_Value));
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired,
        const memory_order _Order = memory_order_seq_cst) noexcept { // CAS with given memory order
        const char _Expected_bytes = _Atomic_reinterpret_as<char>(_Expected); // read before atomic operation
        char _Prev_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Prev_bytes, _InterlockedCompareExchange8, _Atomic_address_as<char>(_Storage),
            _Atomic_reinterpret_as<char>(_Desired), _Expected_bytes);
        if (_Prev_bytes == _Expected_bytes) {
            return true;
        }

        reinterpret_cast<char&>(_Expected) = _Prev_bytes;
        return false;
    }

    _Atomic_padded<_Ty> _Storage;
};

template <class _Ty>
struct _Atomic_storage<_Ty, 2> { // lock-free using 2-byte intrinsics
    _Atomic_storage() = default;

    /* implicit */ constexpr _Atomic_storage(const _Ty _Value) noexcept : _Storage{_Value} {
        // non-atomically initialize this atomic
    }

    void store(const _Ty _Value) noexcept { // store with sequential consistency
        const auto _Mem       = _Atomic_address_as<short>(_Storage);
        const short _As_bytes = _Atomic_reinterpret_as<short>(_Value);
#if defined(_M_ARM) || defined(_M_ARM64)
        _Memory_barrier();
        __iso_volatile_store16(_Mem, _As_bytes);
        _Memory_barrier();
#else // ^^^ ARM32/ARM64 hardware / x86/x64 hardware vvv
        (void) _InterlockedExchange16(_Mem, _As_bytes);
#endif // hardware
    }

    void store(const _Ty _Value, const memory_order _Order) noexcept { // store with given memory order
        const short _As_bytes = _Atomic_reinterpret_as<short>(_Value);
        switch (_Order) {
        case memory_order_relaxed:
            _ISO_VOLATILE_STORE16(_Storage, _As_bytes);
            return;
        case memory_order_release:
            _Compiler_or_memory_barrier();
            _ISO_VOLATILE_STORE16(_Storage, _As_bytes);
            return;
        default:
        case memory_order_consume:
        case memory_order_acquire:
        case memory_order_acq_rel:
            _INVALID_MEMORY_ORDER;
            // [[fallthrough]];
        case memory_order_seq_cst:
            store(_Value);
            return;
        }
    }

    _NODISCARD _Ty load() const noexcept { // load with sequential consistency
        short _As_bytes = _ISO_VOLATILE_LOAD16(_Storage);
        _Compiler_or_memory_barrier();
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _NODISCARD _Ty load(const memory_order _Order) const noexcept { // load with given memory order
        short _As_bytes = _ISO_VOLATILE_LOAD16(_Storage);
        _Load_barrier(_Order);
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _Ty exchange(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // exchange with given memory order
        short _As_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _As_bytes, _InterlockedExchange16, _Atomic_address_as<short>(_Storage),
            _Atomic_reinterpret_as<short>(_Value));
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired,
        const memory_order _Order = memory_order_seq_cst) noexcept { // CAS with given memory order
        const short _Expected_bytes = _Atomic_reinterpret_as<short>(_Expected); // read before atomic operation
        short _Prev_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Prev_bytes, _InterlockedCompareExchange16,
            _Atomic_address_as<short>(_Storage), _Atomic_reinterpret_as<short>(_Desired), _Expected_bytes);
        if (_Prev_bytes == _Expected_bytes) {
            return true;
        }

        _CSTD memcpy(_STD addressof(_Expected), &_Prev_bytes, sizeof(_Ty));
        return false;
    }

    _Atomic_padded<_Ty> _Storage;
};

template <class _Ty>
struct _Atomic_storage<_Ty, 4> { // lock-free using 4-byte intrinsics
    _Atomic_storage() = default;

    /* implicit */ constexpr _Atomic_storage(const _Ty _Value) noexcept : _Storage{_Value} {
        // non-atomically initialize this atomic
    }

    void store(const _Ty _Value) noexcept { // store with sequential consistency
#if defined(_M_ARM) || defined(_M_ARM64)
        _Memory_barrier();
        _ISO_VOLATILE_STORE32(_Storage, _Atomic_reinterpret_as<int>(_Value));
        _Memory_barrier();
#else // ^^^ ARM32/ARM64 hardware / x86/x64 hardware vvv
        (void) _InterlockedExchange(_Atomic_address_as<long>(_Storage), _Atomic_reinterpret_as<long>(_Value));
#endif // hardware
    }

    void store(const _Ty _Value, const memory_order _Order) noexcept { // store with given memory order
        const int _As_bytes = _Atomic_reinterpret_as<int>(_Value);
        switch (_Order) {
        case memory_order_relaxed:
            _ISO_VOLATILE_STORE32(_Storage, _As_bytes);
            return;
        case memory_order_release:
            _Compiler_or_memory_barrier();
            _ISO_VOLATILE_STORE32(_Storage, _As_bytes);
            return;
        default:
        case memory_order_consume:
        case memory_order_acquire:
        case memory_order_acq_rel:
            _INVALID_MEMORY_ORDER;
            // [[fallthrough]];
        case memory_order_seq_cst:
            store(_Value);
            return;
        }
    }

    _NODISCARD _Ty load() const noexcept { // load with sequential consistency
        auto _As_bytes = _ISO_VOLATILE_LOAD32(_Storage);
        _Compiler_or_memory_barrier();
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _NODISCARD _Ty load(const memory_order _Order) const noexcept { // load with given memory order
        auto _As_bytes = _ISO_VOLATILE_LOAD32(_Storage);
        _Load_barrier(_Order);
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _Ty exchange(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // exchange with given memory order
        long _As_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _As_bytes, _InterlockedExchange, _Atomic_address_as<long>(_Storage),
            _Atomic_reinterpret_as<long>(_Value));
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired,
        const memory_order _Order = memory_order_seq_cst) noexcept { // CAS with given memory order
        const long _Expected_bytes = _Atomic_reinterpret_as<long>(_Expected); // read before atomic operation
        long _Prev_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Prev_bytes, _InterlockedCompareExchange, _Atomic_address_as<long>(_Storage),
            _Atomic_reinterpret_as<long>(_Desired), _Expected_bytes);
        if (_Prev_bytes == _Expected_bytes) {
            return true;
        }

        _CSTD memcpy(_STD addressof(_Expected), &_Prev_bytes, sizeof(_Ty));
        return false;
    }

    _Atomic_padded<_Ty> _Storage;
};

template <class _Ty>
struct _Atomic_storage<_Ty, 8> { // lock-free using 8-byte intrinsics
    _Atomic_storage() = default;

    /* implicit */ constexpr _Atomic_storage(const _Ty _Value) noexcept : _Storage{_Value} {
        // non-atomically initialize this atomic
    }

#ifdef _M_IX86
    void store(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // store with (effectively) sequential consistency
        _Check_store_memory_order(_Order);
        (void) exchange(_Value, _Order);
    }
#else // ^^^ _M_IX86 / !_M_IX86 vvv

    void store(const _Ty _Value) noexcept { // store with sequential consistency
        const auto _Mem           = _Atomic_address_as<long long>(_Storage);
        const long long _As_bytes = _Atomic_reinterpret_as<long long>(_Value);
#ifdef _M_ARM64
        _Memory_barrier();
        __iso_volatile_store64(_Mem, _As_bytes);
        _Memory_barrier();
#else // ^^^ _M_ARM64 / ARM32, x64 vvv
        (void) _InterlockedExchange64(_Mem, _As_bytes);
#endif // _M_ARM64
    }

    void store(const _Ty _Value, const memory_order _Order) noexcept { // store with given memory order
        const long long _As_bytes = _Atomic_reinterpret_as<long long>(_Value);
        switch (_Order) {
        case memory_order_relaxed:
            _ISO_VOLATILE_STORE64(_Storage, _As_bytes);
            return;
        case memory_order_release:
            _Compiler_or_memory_barrier();
            _ISO_VOLATILE_STORE64(_Storage, _As_bytes);
            return;
        default:
        case memory_order_consume:
        case memory_order_acquire:
        case memory_order_acq_rel:
            _INVALID_MEMORY_ORDER;
            // [[fallthrough]];
        case memory_order_seq_cst:
            store(_Value);
            return;
        }
    }
#endif // _M_IX86

    _NODISCARD _Ty load() const noexcept { // load with sequential consistency
        const auto _Mem = _Atomic_address_as<const long long>(_Storage);
        long long _As_bytes;
#if defined(_M_ARM)
        _As_bytes = __ldrexd(_Mem);
        _Memory_barrier();
#elif defined(_M_IX86) || defined(_M_ARM64)
        _As_bytes = __iso_volatile_load64(_Mem);
        _Compiler_or_memory_barrier();
#else // _M_X64
        _As_bytes = *_Mem;
        _Compiler_barrier();
#endif // hardware
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

    _NODISCARD _Ty load(const memory_order _Order) const noexcept { // load with given memory order
        const auto _Mem = _Atomic_address_as<const long long>(_Storage);
#if defined(_M_ARM)
        long long _As_bytes = __ldrexd(_Mem);
#elif defined(_M_IX86) || defined(_M_ARM64)
        long long _As_bytes = __iso_volatile_load64(_Mem);
#else // _M_X64
        long long _As_bytes = *_Mem;
#endif // hardware
        _Load_barrier(_Order);
        return reinterpret_cast<_Ty&>(_As_bytes);
    }

#ifdef _M_IX86
    _Ty exchange(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // exchange with (effectively) sequential consistency
        _Ty _Temp{load()};
        while (!compare_exchange_strong(_Temp, _Value, _Order)) { // keep trying
        }

        return _Temp;
    }
#else // ^^^ _M_IX86 / !_M_IX86 vvv
    _Ty exchange(const _Ty _Value, const memory_order _Order = memory_order_seq_cst) noexcept {
        // exchange with given memory order
        long long _As_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _As_bytes, _InterlockedExchange64, _Atomic_address_as<long long>(_Storage),
            _Atomic_reinterpret_as<long long>(_Value));
        return reinterpret_cast<_Ty&>(_As_bytes);
    }
#endif // _M_IX86

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired,
        const memory_order _Order = memory_order_seq_cst) noexcept { // CAS with given memory order
        const long long _Expected_bytes = _Atomic_reinterpret_as<long long>(_Expected); // read before atomic operation
        long long _Prev_bytes;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Prev_bytes, _InterlockedCompareExchange64,
            _Atomic_address_as<long long>(_Storage), _Atomic_reinterpret_as<long long>(_Desired), _Expected_bytes);
        if (_Prev_bytes == _Expected_bytes) {
            return true;
        }

        _CSTD memcpy(_STD addressof(_Expected), &_Prev_bytes, sizeof(_Ty));
        return false;
    }

    _Atomic_padded<_Ty> _Storage;
};

#if 0 // TRANSITION, ABI
#if defined(_M_X64) || defined(_M_ARM64)
template <class _Ty>
struct _Atomic_storage<_Ty, 16> { // lock-free using 16-byte intrinsics
    _Atomic_storage() = default;

    /* implicit */ constexpr _Atomic_storage(const _Ty _Value) noexcept
        : _Storage{_Value} { // non-atomically initialize this atomic
    }

    void store(const _Ty _Value) noexcept { // store with sequential consistency
        (void) exchange(_Value);
    }

    void store(const _Ty _Value, const memory_order _Order) noexcept { // store with given memory order
        _Check_store_memory_order(_Order);
        (void) exchange(_Value, _Order);
    }

    _NODISCARD _Ty load() const noexcept { // load with sequential consistency
        long long* const _Storage_ptr = const_cast<long long*>(_Atomic_address_as<const long long>(_Storage));
        _Int128 _Result{}; // atomic CAS 0 with 0
        (void) _STD_COMPARE_EXCHANGE_128(_Storage_ptr, 0, 0, &_Result._Low);
        return reinterpret_cast<_Ty&>(_Result);
    }

    _NODISCARD _Ty load(const memory_order _Order) const noexcept { // load with given memory order
#ifdef _M_ARM64
        long long* const _Storage_ptr = const_cast<long long*>(_Atomic_address_as<const long long>(_Storage));
        _Int128 _Result{}; // atomic CAS 0 with 0
        switch (_Order) {
        case memory_order_relaxed:
            (void) _INTRIN_RELAXED(_InterlockedCompareExchange128)(_Storage_ptr, 0, 0, &_Result._Low);
            break;
        case memory_order_consume:
        case memory_order_acquire:
            (void) _INTRIN_ACQUIRE(_InterlockedCompareExchange128)(_Storage_ptr, 0, 0, &_Result._Low);
            break;
        default:
        case memory_order_release:
        case memory_order_acq_rel:
            _INVALID_MEMORY_ORDER;
            // [[fallthrough]];
        case memory_order_seq_cst:
            (void) _InterlockedCompareExchange128(_Storage_ptr, 0, 0, &_Result._Low);
            break;
        }

        return reinterpret_cast<_Ty&>(_Result);
#else // ^^^ _M_ARM64 / _M_X64 vvv
        _Check_load_memory_order(_Order);
        return load();
#endif // _M_ARM64
    }

    _Ty exchange(const _Ty _Value) noexcept { // exchange with sequential consistency
        _Ty _Result{_Value};
        while (!compare_exchange_strong(_Result, _Value)) { // keep trying
        }

        return _Result;
    }

    _Ty exchange(const _Ty _Value, const memory_order _Order) noexcept { // exchange with given memory order
        _Ty _Result{_Value};
        while (!compare_exchange_strong(_Result, _Value, _Order)) { // keep trying
        }

        return _Result;
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired,
        const memory_order _Order = memory_order_seq_cst) noexcept { // CAS with given memory order
        _Int128 _Desired_bytes{};
        _CSTD memcpy(&_Desired_bytes, _STD addressof(_Desired), sizeof(_Ty));
        _Int128 _Expected_temp{};
        _CSTD memcpy(&_Expected_temp, _STD addressof(_Expected), sizeof(_Ty));
        unsigned char _Result;
#ifdef _M_ARM64
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedCompareExchange128,
            _Atomic_address_as<long long>(_Storage), _Desired_bytes._High, _Desired_bytes._Low, &_Expected_temp._Low);
#else // ^^^ _M_ARM64 / _M_X64 vvv
        (void) _Order;
        _Result = _STD_COMPARE_EXCHANGE_128(
            &reinterpret_cast<long long&>(_Storage), _Desired_bytes._High, _Desired_bytes._Low, &_Expected_temp._Low);
#endif // _M_ARM64
        if (_Result == 0) {
            _CSTD memcpy(_STD addressof(_Expected), &_Expected_temp, sizeof(_Ty));
        }

        return _Result != 0;
    }

    struct _Int128 {
        alignas(16) long long _Low;
        long long _High;
    };

    _Atomic_padded<_Ty> _Storage;
};
#endif // defined(_M_X64) || defined(_M_ARM64)
#endif // TRANSITION, ABI

// STRUCT TEMPLATE _Atomic_integral
template <class _Ty, size_t = sizeof(_Ty)>
struct _Atomic_integral; // not defined

template <class _Ty>
struct _Atomic_integral<_Ty, 1> : _Atomic_storage<_Ty> { // atomic integral operations using 1-byte intrinsics
    using _Base = _Atomic_storage<_Ty>;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_integral() = default;
    /* implicit */ constexpr _Atomic_integral(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        char _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedExchangeAdd8, _Atomic_address_as<char>(this->_Storage),
            static_cast<char>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_and(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        char _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedAnd8, _Atomic_address_as<char>(this->_Storage), static_cast<char>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_or(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        char _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedOr8, _Atomic_address_as<char>(this->_Storage), static_cast<char>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_xor(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        char _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedXor8, _Atomic_address_as<char>(this->_Storage), static_cast<char>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty operator++(int) noexcept {
        return static_cast<_Ty>(_InterlockedExchangeAdd8(_Atomic_address_as<char>(this->_Storage), 1));
    }

    _Ty operator++() noexcept {
        unsigned char _Before =
            static_cast<unsigned char>(_InterlockedExchangeAdd8(_Atomic_address_as<char>(this->_Storage), 1));
        ++_Before;
        return static_cast<_Ty>(_Before);
    }

    _Ty operator--(int) noexcept {
        return static_cast<_Ty>(_InterlockedExchangeAdd8(_Atomic_address_as<char>(this->_Storage), -1));
    }

    _Ty operator--() noexcept {
        unsigned char _Before =
            static_cast<unsigned char>(_InterlockedExchangeAdd8(_Atomic_address_as<char>(this->_Storage), -1));
        --_Before;
        return static_cast<_Ty>(_Before);
    }
};


template <class _Ty>
struct _Atomic_integral<_Ty, 2> : _Atomic_storage<_Ty> { // atomic integral operations using 2-byte intrinsics
    using _Base = _Atomic_storage<_Ty>;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_integral() = default;
    /* implicit */ constexpr _Atomic_integral(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        short _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedExchangeAdd16, _Atomic_address_as<short>(this->_Storage),
            static_cast<short>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_and(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        short _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedAnd16, _Atomic_address_as<short>(this->_Storage),
            static_cast<short>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_or(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        short _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedOr16, _Atomic_address_as<short>(this->_Storage), static_cast<short>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_xor(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        short _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedXor16, _Atomic_address_as<short>(this->_Storage),
            static_cast<short>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty operator++(int) noexcept {
        unsigned short _After =
            static_cast<unsigned short>(_InterlockedIncrement16(_Atomic_address_as<short>(this->_Storage)));
        --_After;
        return static_cast<_Ty>(_After);
    }

    _Ty operator++() noexcept {
        return static_cast<_Ty>(_InterlockedIncrement16(_Atomic_address_as<short>(this->_Storage)));
    }

    _Ty operator--(int) noexcept {
        unsigned short _After =
            static_cast<unsigned short>(_InterlockedDecrement16(_Atomic_address_as<short>(this->_Storage)));
        ++_After;
        return static_cast<_Ty>(_After);
    }

    _Ty operator--() noexcept {
        return static_cast<_Ty>(_InterlockedDecrement16(_Atomic_address_as<short>(this->_Storage)));
    }
};

template <class _Ty>
struct _Atomic_integral<_Ty, 4> : _Atomic_storage<_Ty> { // atomic integral operations using 4-byte intrinsics
    using _Base = _Atomic_storage<_Ty>;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_integral() = default;
    /* implicit */ constexpr _Atomic_integral(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedExchangeAdd, _Atomic_address_as<long>(this->_Storage),
            static_cast<long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_and(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedAnd, _Atomic_address_as<long>(this->_Storage), static_cast<long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_or(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedOr, _Atomic_address_as<long>(this->_Storage), static_cast<long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_xor(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedXor, _Atomic_address_as<long>(this->_Storage), static_cast<long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty operator++(int) noexcept {
        unsigned long _After =
            static_cast<unsigned long>(_InterlockedIncrement(_Atomic_address_as<long>(this->_Storage)));
        --_After;
        return static_cast<_Ty>(_After);
    }

    _Ty operator++() noexcept {
        return static_cast<_Ty>(_InterlockedIncrement(_Atomic_address_as<long>(this->_Storage)));
    }

    _Ty operator--(int) noexcept {
        unsigned long _After =
            static_cast<unsigned long>(_InterlockedDecrement(_Atomic_address_as<long>(this->_Storage)));
        ++_After;
        return static_cast<_Ty>(_After);
    }

    _Ty operator--() noexcept {
        return static_cast<_Ty>(_InterlockedDecrement(_Atomic_address_as<long>(this->_Storage)));
    }
};

template <class _Ty>
struct _Atomic_integral<_Ty, 8> : _Atomic_storage<_Ty> { // atomic integral operations using 8-byte intrinsics
    using _Base = _Atomic_storage<_Ty>;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_integral() = default;
    /* implicit */ constexpr _Atomic_integral(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

#ifdef _M_IX86
    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        // effectively sequential consistency
        _Ty _Temp{this->load()};
        while (!this->compare_exchange_strong(_Temp, _Temp + _Operand, _Order)) { // keep trying
        }

        return _Temp;
    }

    _Ty fetch_and(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        // effectively sequential consistency
        _Ty _Temp{this->load()};
        while (!this->compare_exchange_strong(_Temp, _Temp & _Operand, _Order)) { // keep trying
        }

        return _Temp;
    }

    _Ty fetch_or(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        // effectively sequential consistency
        _Ty _Temp{this->load()};
        while (!this->compare_exchange_strong(_Temp, _Temp | _Operand, _Order)) { // keep trying
        }

        return _Temp;
    }

    _Ty fetch_xor(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        // effectively sequential consistency
        _Ty _Temp{this->load()};
        while (!this->compare_exchange_strong(_Temp, _Temp ^ _Operand, _Order)) { // keep trying
        }

        return _Temp;
    }

    _Ty operator++(int) noexcept {
        return fetch_add(static_cast<_Ty>(1));
    }

    _Ty operator++() noexcept {
        return fetch_add(static_cast<_Ty>(1)) + static_cast<_Ty>(1);
    }

    _Ty operator--(int) noexcept {
        return fetch_add(static_cast<_Ty>(-1));
    }

    _Ty operator--() noexcept {
        return fetch_add(static_cast<_Ty>(-1)) - static_cast<_Ty>(1);
    }

#else // ^^^ _M_IX86 / !_M_IX86 vvv
    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedExchangeAdd64,
            _Atomic_address_as<long long>(this->_Storage), static_cast<long long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_and(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedAnd64, _Atomic_address_as<long long>(this->_Storage),
            static_cast<long long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_or(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedOr64, _Atomic_address_as<long long>(this->_Storage),
            static_cast<long long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty fetch_xor(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        long long _Result;
        _ATOMIC_CHOOSE_INTRINSIC(_Order, _Result, _InterlockedXor64, _Atomic_address_as<long long>(this->_Storage),
            static_cast<long long>(_Operand));
        return static_cast<_Ty>(_Result);
    }

    _Ty operator++(int) noexcept {
        unsigned long long _After =
            static_cast<unsigned long long>(_InterlockedIncrement64(_Atomic_address_as<long long>(this->_Storage)));
        --_After;
        return static_cast<_Ty>(_After);
    }

    _Ty operator++() noexcept {
        return static_cast<_Ty>(_InterlockedIncrement64(_Atomic_address_as<long long>(this->_Storage)));
    }

    _Ty operator--(int) noexcept {
        unsigned long long _After =
            static_cast<unsigned long long>(_InterlockedDecrement64(_Atomic_address_as<long long>(this->_Storage)));
        ++_After;
        return static_cast<_Ty>(_After);
    }

    _Ty operator--() noexcept {
        return static_cast<_Ty>(_InterlockedDecrement64(_Atomic_address_as<long long>(this->_Storage)));
    }
#endif // _M_IX86
};

#if 1 // TRANSITION, ABI
template <size_t _TypeSize>
_INLINE_VAR constexpr bool _Is_always_lock_free = _TypeSize <= 8 && (_TypeSize & (_TypeSize - 1)) == 0;
#else // ^^^ don't break ABI / break ABI vvv
#if _ATOMIC_HAS_DCAS
template <size_t _TypeSize>
_INLINE_VAR constexpr bool _Is_always_lock_free = _TypeSize <= 2 * sizeof(void*);
#else // ^^^ _ATOMIC_HAS_DCAS / !_ATOMIC_HAS_DCAS vvv
template <size_t _TypeSize>
_INLINE_VAR constexpr bool _Is_always_lock_free = _TypeSize <= sizeof(void*);
#endif // _ATOMIC_HAS_DCAS
#endif // break ABI

template <class _Ty, bool _Is_lock_free = _Is_always_lock_free<sizeof(_Ty)>>
_INLINE_VAR constexpr bool _Deprecate_non_lock_free_volatile = true;

template <class _Ty>
_CXX20_DEPRECATE_VOLATILE _INLINE_VAR constexpr bool _Deprecate_non_lock_free_volatile<_Ty, false> = true;

// STRUCT TEMPLATE _Atomic_integral_facade
template <class _Ty>
struct _Atomic_integral_facade : _Atomic_integral<_Ty> {
    // provides operator overloads and other support for atomic integral specializations
    using _Base           = _Atomic_integral<_Ty>;
    using difference_type = _Ty;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_integral_facade() = default;
    /* implicit */ constexpr _Atomic_integral_facade(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    // _Deprecate_non_lock_free_volatile is unnecessary here.

    // note: const_cast-ing away volatile is safe because all our intrinsics add volatile back on.
    // We make the primary functions non-volatile for better debug codegen, as non-volatile atomics
    // are far more common than volatile ones.
    using _Base::fetch_add;
    _Ty fetch_add(const _Ty _Operand) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_add(_Operand);
    }

    _Ty fetch_add(const _Ty _Operand, const memory_order _Order) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_add(_Operand, _Order);
    }

    _NODISCARD static _Ty _Negate(const _Ty _Value) noexcept { // returns two's complement negated value of _Value
        return static_cast<_Ty>(0U - static_cast<make_unsigned_t<_Ty>>(_Value));
    }

    _Ty fetch_sub(const _Ty _Operand) noexcept {
        return fetch_add(_Negate(_Operand));
    }

    _Ty fetch_sub(const _Ty _Operand) volatile noexcept {
        return fetch_add(_Negate(_Operand));
    }

    _Ty fetch_sub(const _Ty _Operand, const memory_order _Order) noexcept {
        return fetch_add(_Negate(_Operand), _Order);
    }

    _Ty fetch_sub(const _Ty _Operand, const memory_order _Order) volatile noexcept {
        return fetch_add(_Negate(_Operand), _Order);
    }

    using _Base::fetch_and;
    _Ty fetch_and(const _Ty _Operand) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_and(_Operand);
    }

    _Ty fetch_and(const _Ty _Operand, const memory_order _Order) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_and(_Operand, _Order);
    }

    using _Base::fetch_or;
    _Ty fetch_or(const _Ty _Operand) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_or(_Operand);
    }

    _Ty fetch_or(const _Ty _Operand, const memory_order _Order) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_or(_Operand, _Order);
    }

    using _Base::fetch_xor;
    _Ty fetch_xor(const _Ty _Operand) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_xor(_Operand);
    }

    _Ty fetch_xor(const _Ty _Operand, const memory_order _Order) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_xor(_Operand, _Order);
    }

    using _Base::operator++;
    _Ty operator++(int) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::operator++(0);
    }

    _Ty operator++() volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::operator++();
    }

    using _Base::operator--;
    _Ty operator--(int) volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::operator--(0);
    }

    _Ty operator--() volatile noexcept {
        return const_cast<_Atomic_integral_facade*>(this)->_Base::operator--();
    }

    _Ty operator+=(const _Ty _Operand) noexcept {
        return static_cast<_Ty>(this->_Base::fetch_add(_Operand) + _Operand);
    }

    _Ty operator+=(const _Ty _Operand) volatile noexcept {
        return static_cast<_Ty>(const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_add(_Operand) + _Operand);
    }

    _Ty operator-=(const _Ty _Operand) noexcept {
        return static_cast<_Ty>(fetch_sub(_Operand) - _Operand);
    }

    _Ty operator-=(const _Ty _Operand) volatile noexcept {
        return static_cast<_Ty>(const_cast<_Atomic_integral_facade*>(this)->fetch_sub(_Operand) - _Operand);
    }

    _Ty operator&=(const _Ty _Operand) noexcept {
        return static_cast<_Ty>(this->_Base::fetch_and(_Operand) & _Operand);
    }

    _Ty operator&=(const _Ty _Operand) volatile noexcept {
        return static_cast<_Ty>(const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_and(_Operand) & _Operand);
    }

    _Ty operator|=(const _Ty _Operand) noexcept {
        return static_cast<_Ty>(this->_Base::fetch_or(_Operand) | _Operand);
    }

    _Ty operator|=(const _Ty _Operand) volatile noexcept {
        return static_cast<_Ty>(const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_or(_Operand) | _Operand);
    }

    _Ty operator^=(const _Ty _Operand) noexcept {
        return static_cast<_Ty>(this->_Base::fetch_xor(_Operand) ^ _Operand);
    }

    _Ty operator^=(const _Ty _Operand) volatile noexcept {
        return static_cast<_Ty>(const_cast<_Atomic_integral_facade*>(this)->_Base::fetch_xor(_Operand) ^ _Operand);
    }
};

#if _HAS_CXX20
template <class _Ty>
struct _Atomic_floating : _Atomic_storage<_Ty> {
    // provides atomic floating-point operations
    using _Base           = _Atomic_storage<_Ty>;
    using difference_type = _Ty;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_floating() = default;
    /* implicit */ constexpr _Atomic_floating(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        _Ty _Temp{this->load(memory_order_relaxed)};
        while (!this->compare_exchange_strong(_Temp, _Temp + _Operand, _Order)) { // keep trying
        }

        return _Temp;
    }

    // _Deprecate_non_lock_free_volatile is unnecessary here.

    // note: const_cast-ing away volatile is safe because all our intrinsics add volatile back on.
    // We make the primary functions non-volatile for better debug codegen, as non-volatile atomics
    // are far more common than volatile ones.
    _Ty fetch_add(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) volatile noexcept {
        return const_cast<_Atomic_floating*>(this)->fetch_add(_Operand, _Order);
    }

    _Ty fetch_sub(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) noexcept {
        _Ty _Temp{this->load(memory_order_relaxed)};
        while (!this->compare_exchange_strong(_Temp, _Temp - _Operand, _Order)) { // keep trying
        }

        return _Temp;
    }

    _Ty fetch_sub(const _Ty _Operand, const memory_order _Order = memory_order_seq_cst) volatile noexcept {
        return const_cast<_Atomic_floating*>(this)->fetch_sub(_Operand, _Order);
    }

    _Ty operator+=(const _Ty _Operand) noexcept {
        return fetch_add(_Operand) + _Operand;
    }

    _Ty operator+=(const _Ty _Operand) volatile noexcept {
        return const_cast<_Atomic_floating*>(this)->fetch_add(_Operand) + _Operand;
    }

    _Ty operator-=(const _Ty _Operand) noexcept {
        return fetch_sub(_Operand) - _Operand;
    }

    _Ty operator-=(const _Ty _Operand) volatile noexcept {
        return const_cast<_Atomic_floating*>(this)->fetch_sub(_Operand) - _Operand;
    }
};
#endif // _HAS_CXX20

// STRUCT TEMPLATE _Atomic_pointer
template <class _Ty>
struct _Atomic_pointer : _Atomic_storage<_Ty> {
    using _Base           = _Atomic_storage<_Ty>;
    using difference_type = ptrdiff_t;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    _Atomic_pointer() = default;
    /* implicit */ constexpr _Atomic_pointer(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    _Ty fetch_add(const ptrdiff_t _Diff, const memory_order _Order = memory_order_seq_cst) noexcept {
        const ptrdiff_t _Shift_bytes =
            static_cast<ptrdiff_t>(static_cast<size_t>(_Diff) * sizeof(remove_pointer_t<_Ty>));
        ptrdiff_t _Result;
#if defined(_M_IX86) || defined(_M_ARM)
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedExchangeAdd, _Atomic_address_as<long>(this->_Storage), _Shift_bytes);
#else // ^^^ 32 bits / 64 bits vvv
        _ATOMIC_CHOOSE_INTRINSIC(
            _Order, _Result, _InterlockedExchangeAdd64, _Atomic_address_as<long long>(this->_Storage), _Shift_bytes);
#endif // hardware
        return reinterpret_cast<_Ty>(_Result);
    }

    // _Deprecate_non_lock_free_volatile is unnecessary here.

    _Ty fetch_add(const ptrdiff_t _Diff) volatile noexcept {
        return const_cast<_Atomic_pointer*>(this)->fetch_add(_Diff);
    }

    _Ty fetch_add(const ptrdiff_t _Diff, const memory_order _Order) volatile noexcept {
        return const_cast<_Atomic_pointer*>(this)->fetch_add(_Diff, _Order);
    }

    _Ty fetch_sub(const ptrdiff_t _Diff) volatile noexcept {
        return fetch_add(static_cast<ptrdiff_t>(0 - static_cast<size_t>(_Diff)));
    }

    _Ty fetch_sub(const ptrdiff_t _Diff) noexcept {
        return fetch_add(static_cast<ptrdiff_t>(0 - static_cast<size_t>(_Diff)));
    }

    _Ty fetch_sub(const ptrdiff_t _Diff, const memory_order _Order) volatile noexcept {
        return fetch_add(static_cast<ptrdiff_t>(0 - static_cast<size_t>(_Diff)), _Order);
    }

    _Ty fetch_sub(const ptrdiff_t _Diff, const memory_order _Order) noexcept {
        return fetch_add(static_cast<ptrdiff_t>(0 - static_cast<size_t>(_Diff)), _Order);
    }

    _Ty operator++(int) volatile noexcept {
        return fetch_add(1);
    }

    _Ty operator++(int) noexcept {
        return fetch_add(1);
    }

    _Ty operator++() volatile noexcept {
        return fetch_add(1) + 1;
    }

    _Ty operator++() noexcept {
        return fetch_add(1) + 1;
    }

    _Ty operator--(int) volatile noexcept {
        return fetch_add(-1);
    }

    _Ty operator--(int) noexcept {
        return fetch_add(-1);
    }

    _Ty operator--() volatile noexcept {
        return fetch_add(-1) - 1;
    }

    _Ty operator--() noexcept {
        return fetch_add(-1) - 1;
    }

    _Ty operator+=(const ptrdiff_t _Diff) volatile noexcept {
        return fetch_add(_Diff) + _Diff;
    }

    _Ty operator+=(const ptrdiff_t _Diff) noexcept {
        return fetch_add(_Diff) + _Diff;
    }

    _Ty operator-=(const ptrdiff_t _Diff) volatile noexcept {
        return fetch_add(static_cast<ptrdiff_t>(0 - static_cast<size_t>(_Diff))) - _Diff;
    }

    _Ty operator-=(const ptrdiff_t _Diff) noexcept {
        return fetch_add(static_cast<ptrdiff_t>(0 - static_cast<size_t>(_Diff))) - _Diff;
    }
};

// STRUCT TEMPLATE atomic
#define ATOMIC_VAR_INIT(_Value) \
    { _Value }

template <class _Ty>
using _Choose_atomic_base2_t =
    typename _Select<is_integral_v<_Ty> && !is_same_v<bool, _Ty>>::template _Apply<_Atomic_integral_facade<_Ty>,
        typename _Select<is_pointer_v<_Ty> && is_object_v<remove_pointer_t<_Ty>>>::template _Apply<_Atomic_pointer<_Ty>,
            _Atomic_storage<_Ty>>>;

#if _HAS_CXX20
template <class _Ty>
using _Choose_atomic_base_t =
    typename _Select<is_floating_point_v<_Ty>>::template _Apply<_Atomic_floating<_Ty>, _Choose_atomic_base2_t<_Ty>>;
#else // ^^^ _HAS_CXX20 // !_HAS_CXX20 vvv
template <class _Ty>
using _Choose_atomic_base_t = _Choose_atomic_base2_t<_Ty>;
#endif //_HAS_CXX20

template <class _Ty>
struct atomic : _Choose_atomic_base_t<_Ty> { // atomic value
private:
    using _Base = _Choose_atomic_base_t<_Ty>;

public:
    // clang-format off
    static_assert(is_trivially_copyable_v<_Ty> && is_copy_constructible_v<_Ty> && is_move_constructible_v<_Ty>
        && is_copy_assignable_v<_Ty> && is_move_assignable_v<_Ty>,
        "atomic<T> requires T to be trivially copyable, copy constructible, move constructible, copy assignable, "
        "and move assignable.");
    // clang-format on

    using value_type = _Ty;

#ifdef __cplusplus_winrt // TRANSITION, VSO-1083296
    /* implicit */ constexpr atomic(const _Ty _Value) noexcept : _Base(_Value) {}
#else // ^^^ workaround / no workaround vvv
    using _Base::_Base;
#endif // ^^^ no workaround ^^^

    constexpr atomic() noexcept(is_nothrow_default_constructible_v<_Ty>) : _Base() {}

    atomic(const atomic&) = delete;
    atomic& operator=(const atomic&) = delete;

#if _HAS_CXX17
    static constexpr bool is_always_lock_free = _Is_always_lock_free<sizeof(_Ty)>;
#endif // _HAS_CXX17

#if 1 // TRANSITION, ABI
    _NODISCARD bool is_lock_free() const volatile noexcept {
        constexpr bool _Result = sizeof(_Ty) <= 8 && (sizeof(_Ty) & sizeof(_Ty) - 1) == 0;
        return _Result;
    }

#else // ^^^ don't break ABI / break ABI vvv

    _NODISCARD bool is_lock_free() const volatile noexcept {
#if _ATOMIC_HAS_DCAS
        return sizeof(_Ty) <= 2 * sizeof(void*);
#else // ^^^ _ATOMIC_HAS_DCAS / !_ATOMIC_HAS_DCAS vvv
        return sizeof(_Ty) <= sizeof(void*) || (sizeof(_Ty) <= 2 * sizeof(void*) && __std_atomic_has_cmpxchg16b());
#endif // _ATOMIC_HAS_DCAS
    }
#endif // TRANSITION, ABI

    _NODISCARD bool is_lock_free() const noexcept {
        return static_cast<const volatile atomic*>(this)->is_lock_free();
    }

    _Ty operator=(const _Ty _Value) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        this->store(_Value);
        return _Value;
    }

    _Ty operator=(const _Ty _Value) noexcept {
        this->store(_Value);
        return _Value;
    }

    // For the following, we do the real implementation in the non-volatile function, and const_cast
    // to call the non-volatile function in the volatile one. This is safe because all of the
    // non-volatile functions reapply volatile, as all our intrinsics accept only volatile T *.
    // We expect most atomic<T>s to be non-volatile, so making the real implementations
    // non-volatile should result in better debug codegen.
    using _Base::store;
    void store(const _Ty _Value) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        const_cast<atomic*>(this)->_Base::store(_Value);
    }

    void store(const _Ty _Value, const memory_order _Order) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        const_cast<atomic*>(this)->_Base::store(_Value, _Order);
    }

    using _Base::load;
    _NODISCARD _Ty load() const volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return const_cast<const atomic*>(this)->_Base::load();
    }

    _NODISCARD _Ty load(const memory_order _Order) const volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return const_cast<const atomic*>(this)->_Base::load(_Order);
    }

    using _Base::exchange;
    _Ty exchange(const _Ty _Value) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return const_cast<atomic*>(this)->_Base::exchange(_Value);
    }

    _Ty exchange(const _Ty _Value, const memory_order _Order) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return const_cast<atomic*>(this)->_Base::exchange(_Value, _Order);
    }

    using _Base::compare_exchange_strong;
    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return const_cast<atomic*>(this)->_Base::compare_exchange_strong(_Expected, _Desired);
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired, const memory_order _Order) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return const_cast<atomic*>(this)->_Base::compare_exchange_strong(_Expected, _Desired, _Order);
    }

    bool compare_exchange_strong(_Ty& _Expected, const _Ty _Desired, const memory_order _Success,
        const memory_order _Failure) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return this->compare_exchange_strong(_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
    }

    bool compare_exchange_strong(
        _Ty& _Expected, const _Ty _Desired, const memory_order _Success, const memory_order _Failure) noexcept {
        return this->compare_exchange_strong(_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
    }

    bool compare_exchange_weak(_Ty& _Expected, const _Ty _Desired) volatile noexcept {
        // we have no weak CAS intrinsics, even on ARM32/ARM64, so fall back to strong
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return this->compare_exchange_strong(_Expected, _Desired);
    }

    bool compare_exchange_weak(_Ty& _Expected, const _Ty _Desired) noexcept {
        return this->compare_exchange_strong(_Expected, _Desired);
    }

    bool compare_exchange_weak(_Ty& _Expected, const _Ty _Desired, const memory_order _Order) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return this->compare_exchange_strong(_Expected, _Desired, _Order);
    }

    bool compare_exchange_weak(_Ty& _Expected, const _Ty _Desired, const memory_order _Order) noexcept {
        return this->compare_exchange_strong(_Expected, _Desired, _Order);
    }

    bool compare_exchange_weak(_Ty& _Expected, const _Ty _Desired, const memory_order _Success,
        const memory_order _Failure) volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return this->compare_exchange_strong(_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
    }

    bool compare_exchange_weak(
        _Ty& _Expected, const _Ty _Desired, const memory_order _Success, const memory_order _Failure) noexcept {
        return this->compare_exchange_strong(_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
    }

    operator _Ty() const volatile noexcept {
        static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
        return this->load();
    }

    operator _Ty() const noexcept {
        return this->load();
    }
};

#if _HAS_CXX17
template <class _Ty>
atomic(_Ty) -> atomic<_Ty>;
#endif // _HAS_CXX17

// NONMEMBER OPERATIONS ON ATOMIC TYPES
template <class _Ty>
_NODISCARD bool atomic_is_lock_free(const volatile atomic<_Ty>* _Mem) noexcept {
    return _Mem->is_lock_free();
}

template <class _Ty>
_NODISCARD bool atomic_is_lock_free(const atomic<_Ty>* _Mem) noexcept {
    return _Mem->is_lock_free();
}

template <class _Ty>
_CXX20_DEPRECATE_ATOMIC_INIT void atomic_init(
    atomic<_Ty>* const _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
#if 1 // TRANSITION, ABI
    _CSTD memcpy(_STD addressof(_Mem->_Storage), _STD addressof(_Value), sizeof(_Ty));
#else // ^^^ don't break ABI / break ABI vvv
    _CSTD memcpy(_Mem, _STD addressof(_Value), sizeof(_Ty));
#endif // TRANSITION, ABI
    _CSTD memset(reinterpret_cast<unsigned char*>(_Mem) + sizeof(_Ty), 0, sizeof(atomic<_Ty>) - sizeof(_Ty));
}

template <class _Ty>
_CXX20_DEPRECATE_ATOMIC_INIT void atomic_init(
    volatile atomic<_Ty>* const _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    // NB: respecting volatility here appears unimplementable
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    _STD atomic_init(const_cast<atomic<_Ty>*>(_Mem), _Value);
}

template <class _Ty>
void atomic_store(volatile atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    _Mem->store(_Value);
}

template <class _Ty>
void atomic_store(atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value) noexcept {
    _Mem->store(_Value);
}

template <class _Ty>
void atomic_store_explicit(
    volatile atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value, const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    _Mem->store(_Value, _Order);
}

template <class _Ty>
void atomic_store_explicit(atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value, const memory_order _Order) noexcept {
    _Mem->store(_Value, _Order);
}

template <class _Ty>
_NODISCARD _Ty atomic_load(const volatile atomic<_Ty>* const _Mem) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->load();
}

template <class _Ty>
_NODISCARD _Ty atomic_load(const atomic<_Ty>* const _Mem) noexcept {
    return _Mem->load();
}

template <class _Ty>
_NODISCARD _Ty atomic_load_explicit(const volatile atomic<_Ty>* const _Mem, const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->load(_Order);
}

template <class _Ty>
_NODISCARD _Ty atomic_load_explicit(const atomic<_Ty>* const _Mem, const memory_order _Order) noexcept {
    return _Mem->load(_Order);
}

template <class _Ty>
_Ty atomic_exchange(volatile atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->exchange(_Value);
}

template <class _Ty>
_Ty atomic_exchange(atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value) noexcept {
    return _Mem->exchange(_Value);
}

template <class _Ty>
_Ty atomic_exchange_explicit(
    volatile atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value, const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->exchange(_Value, _Order);
}

template <class _Ty>
_Ty atomic_exchange_explicit(
    atomic<_Ty>* const _Mem, const _Identity_t<_Ty> _Value, const memory_order _Order) noexcept {
    return _Mem->exchange(_Value, _Order);
}

template <class _Ty>
bool atomic_compare_exchange_strong(
    volatile atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected, const _Identity_t<_Ty> _Desired) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->compare_exchange_strong(*_Expected, _Desired);
}

template <class _Ty>
bool atomic_compare_exchange_strong(
    atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected, const _Identity_t<_Ty> _Desired) noexcept {
    return _Mem->compare_exchange_strong(*_Expected, _Desired);
}

template <class _Ty>
bool atomic_compare_exchange_strong_explicit(volatile atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected,
    const _Identity_t<_Ty> _Desired, const memory_order _Success, const memory_order _Failure) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->compare_exchange_strong(*_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
}

template <class _Ty>
bool atomic_compare_exchange_strong_explicit(atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected,
    const _Identity_t<_Ty> _Desired, const memory_order _Success, const memory_order _Failure) noexcept {
    return _Mem->compare_exchange_strong(*_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
}

template <class _Ty>
bool atomic_compare_exchange_weak(
    volatile atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected, const _Identity_t<_Ty> _Desired) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->compare_exchange_strong(*_Expected, _Desired);
}

template <class _Ty>
bool atomic_compare_exchange_weak(
    atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected, const _Identity_t<_Ty> _Desired) noexcept {
    return _Mem->compare_exchange_strong(*_Expected, _Desired);
}

template <class _Ty>
bool atomic_compare_exchange_weak_explicit(volatile atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected,
    const _Identity_t<_Ty> _Desired, const memory_order _Success, const memory_order _Failure) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->compare_exchange_strong(*_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
}

template <class _Ty>
bool atomic_compare_exchange_weak_explicit(atomic<_Ty>* const _Mem, _Identity_t<_Ty>* const _Expected,
    const _Identity_t<_Ty> _Desired, const memory_order _Success, const memory_order _Failure) noexcept {
    return _Mem->compare_exchange_strong(*_Expected, _Desired, _Combine_cas_memory_orders(_Success, _Failure));
}

template <class _Ty>
_Ty atomic_fetch_add(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_add(_Value);
}

template <class _Ty>
_Ty atomic_fetch_add(atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value) noexcept {
    return _Mem->fetch_add(_Value);
}

template <class _Ty>
_Ty atomic_fetch_add_explicit(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value,
    const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_add(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_add_explicit(
    atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value, const memory_order _Order) noexcept {
    return _Mem->fetch_add(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_sub(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_sub(_Value);
}

template <class _Ty>
_Ty atomic_fetch_sub(atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value) noexcept {
    return _Mem->fetch_sub(_Value);
}

template <class _Ty>
_Ty atomic_fetch_sub_explicit(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value,
    const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_sub(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_sub_explicit(
    atomic<_Ty>* _Mem, const typename atomic<_Ty>::difference_type _Value, const memory_order _Order) noexcept {
    return _Mem->fetch_sub(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_and(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_and(_Value);
}

template <class _Ty>
_Ty atomic_fetch_and(atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    return _Mem->fetch_and(_Value);
}

template <class _Ty>
_Ty atomic_fetch_and_explicit(
    volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value, const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_and(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_and_explicit(
    atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value, const memory_order _Order) noexcept {
    return _Mem->fetch_and(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_or(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_or(_Value);
}

template <class _Ty>
_Ty atomic_fetch_or(atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    return _Mem->fetch_or(_Value);
}

template <class _Ty>
_Ty atomic_fetch_or_explicit(
    volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value, const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_or(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_or_explicit(
    atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value, const memory_order _Order) noexcept {
    return _Mem->fetch_or(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_xor(volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_xor(_Value);
}

template <class _Ty>
_Ty atomic_fetch_xor(atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value) noexcept {
    return _Mem->fetch_xor(_Value);
}

template <class _Ty>
_Ty atomic_fetch_xor_explicit(
    volatile atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value, const memory_order _Order) noexcept {
    static_assert(_Deprecate_non_lock_free_volatile<_Ty>, "Never fails");
    return _Mem->fetch_xor(_Value, _Order);
}

template <class _Ty>
_Ty atomic_fetch_xor_explicit(
    atomic<_Ty>* _Mem, const typename atomic<_Ty>::value_type _Value, const memory_order _Order) noexcept {
    return _Mem->fetch_xor(_Value, _Order);
}

// FENCES
extern "C" inline void atomic_thread_fence(const memory_order _Order) noexcept {
    if (_Order == memory_order_relaxed) {
        return;
    }

#if defined(_M_ARM) || defined(_M_ARM64)
    _Memory_barrier();
#else // ^^^ ARM32/ARM64 hardware / x86/x64 hardware vvv
    _Compiler_barrier();
    if (_Order == memory_order_seq_cst) {
        volatile long _Guard; // Not initialized to avoid an unnecessary operation; the value does not matter

        // _mm_mfence could have been used, but it is not supported on older x86 CPUs and is slower on some recent CPUs.
        // The memory fence provided by interlocked operations has some exceptions, but this is fine:
        // std::atomic_thread_fence works with respect to other atomics only; it may not be a full fence for all ops.
#pragma warning(suppress : 6001) // "Using uninitialized memory '_Guard'"
#pragma warning(suppress : 28113) // "Accessing a local variable _Guard via an Interlocked function: This is an unusual
                                  // usage which could be reconsidered."
        (void) _InterlockedIncrement(&_Guard);
        _Compiler_barrier();
    }
#endif // hardware
}

extern "C" inline void atomic_signal_fence(const memory_order _Order) noexcept {
    if (_Order != memory_order_relaxed) {
        _Compiler_barrier();
    }
}

// ATOMIC TYPEDEFS
using atomic_bool = atomic<bool>;

using atomic_char   = atomic<char>;
using atomic_schar  = atomic<signed char>;
using atomic_uchar  = atomic<unsigned char>;
using atomic_short  = atomic<short>;
using atomic_ushort = atomic<unsigned short>;
using atomic_int    = atomic<int>;
using atomic_uint   = atomic<unsigned int>;
using atomic_long   = atomic<long>;
using atomic_ulong  = atomic<unsigned long>;
using atomic_llong  = atomic<long long>;
using atomic_ullong = atomic<unsigned long long>;

#ifdef __cpp_lib_char8_t
using atomic_char8_t = atomic<char8_t>;
#endif // __cpp_lib_char8_t
using atomic_char16_t = atomic<char16_t>;
using atomic_char32_t = atomic<char32_t>;
using atomic_wchar_t  = atomic<wchar_t>;

using atomic_int8_t   = atomic<int8_t>;
using atomic_uint8_t  = atomic<uint8_t>;
using atomic_int16_t  = atomic<int16_t>;
using atomic_uint16_t = atomic<uint16_t>;
using atomic_int32_t  = atomic<int32_t>;
using atomic_uint32_t = atomic<uint32_t>;
using atomic_int64_t  = atomic<int64_t>;
using atomic_uint64_t = atomic<uint64_t>;

using atomic_int_least8_t   = atomic<int_least8_t>;
using atomic_uint_least8_t  = atomic<uint_least8_t>;
using atomic_int_least16_t  = atomic<int_least16_t>;
using atomic_uint_least16_t = atomic<uint_least16_t>;
using atomic_int_least32_t  = atomic<int_least32_t>;
using atomic_uint_least32_t = atomic<uint_least32_t>;
using atomic_int_least64_t  = atomic<int_least64_t>;
using atomic_uint_least64_t = atomic<uint_least64_t>;

using atomic_int_fast8_t   = atomic<int_fast8_t>;
using atomic_uint_fast8_t  = atomic<uint_fast8_t>;
using atomic_int_fast16_t  = atomic<int_fast16_t>;
using atomic_uint_fast16_t = atomic<uint_fast16_t>;
using atomic_int_fast32_t  = atomic<int_fast32_t>;
using atomic_uint_fast32_t = atomic<uint_fast32_t>;
using atomic_int_fast64_t  = atomic<int_fast64_t>;
using atomic_uint_fast64_t = atomic<uint_fast64_t>;

using atomic_intptr_t  = atomic<intptr_t>;
using atomic_uintptr_t = atomic<uintptr_t>;
using atomic_size_t    = atomic<size_t>;
using atomic_ptrdiff_t = atomic<ptrdiff_t>;
using atomic_intmax_t  = atomic<intmax_t>;
using atomic_uintmax_t = atomic<uintmax_t>;

#if _HAS_CXX20
// Though there are CMPXCHG8B and CMPXCHG16B,
// the largest atomics with a full set of efficient operations are pointer-sized.
using atomic_signed_lock_free   = atomic_intptr_t;
using atomic_unsigned_lock_free = atomic_uintptr_t;
#endif // _HAS_CXX20

// STRUCT atomic_flag
#define ATOMIC_FLAG_INIT \
    {}
struct atomic_flag { // flag with test-and-set semantics
#if _HAS_CXX20
    _NODISCARD bool test(const memory_order _Order = memory_order_seq_cst) const noexcept {
        return _Storage.load(_Order) != 0;
    }

    _NODISCARD bool test(const memory_order _Order = memory_order_seq_cst) const volatile noexcept {
        return _Storage.load(_Order) != 0;
    }
#endif // _HAS_CXX20

    bool test_and_set(const memory_order _Order = memory_order_seq_cst) noexcept {
        return _Storage.exchange(true, _Order) != 0;
    }

    bool test_and_set(const memory_order _Order = memory_order_seq_cst) volatile noexcept {
        return _Storage.exchange(true, _Order) != 0;
    }

    void clear(const memory_order _Order = memory_order_seq_cst) noexcept {
        _Storage.store(false, _Order);
    }

    void clear(const memory_order _Order = memory_order_seq_cst) volatile noexcept {
        _Storage.store(false, _Order);
    }

    constexpr atomic_flag() noexcept = default;

#if 1 // TRANSITION, ABI
    atomic<long> _Storage;
#else // ^^^ don't break ABI / break ABI vvv
    atomic<bool> _Storage;
#endif // TRANSITION, ABI
};


// atomic_flag NONMEMBERS
#if _HAS_CXX20
_NODISCARD inline bool atomic_flag_test(const volatile atomic_flag* const _Flag) noexcept {
    return _Flag->test();
}

_NODISCARD inline bool atomic_flag_test(const atomic_flag* const _Flag) noexcept {
    return _Flag->test();
}

_NODISCARD inline bool atomic_flag_test_explicit(
    const volatile atomic_flag* const _Flag, const memory_order _Order) noexcept {
    return _Flag->test(_Order);
}

_NODISCARD inline bool atomic_flag_test_explicit(const atomic_flag* const _Flag, const memory_order _Order) noexcept {
    return _Flag->test(_Order);
}
#endif // _HAS_CXX20

inline bool atomic_flag_test_and_set(atomic_flag* _Flag) noexcept {
    return _Flag->test_and_set();
}

inline bool atomic_flag_test_and_set(volatile atomic_flag* _Flag) noexcept {
    return _Flag->test_and_set();
}

inline bool atomic_flag_test_and_set_explicit(atomic_flag* _Flag, memory_order _Order) noexcept {
    return _Flag->test_and_set(_Order);
}

inline bool atomic_flag_test_and_set_explicit(volatile atomic_flag* _Flag, memory_order _Order) noexcept {
    return _Flag->test_and_set(_Order);
}

inline void atomic_flag_clear(atomic_flag* _Flag) noexcept {
    _Flag->clear();
}

inline void atomic_flag_clear(volatile atomic_flag* _Flag) noexcept {
    _Flag->clear();
}

inline void atomic_flag_clear_explicit(atomic_flag* _Flag, memory_order _Order) noexcept {
    _Flag->clear(_Order);
}

inline void atomic_flag_clear_explicit(volatile atomic_flag* _Flag, memory_order _Order) noexcept {
    _Flag->clear(_Order);
}

_STD_END

#undef _ATOMIC_CHOOSE_INTRINSIC
#undef _ATOMIC_HAS_DCAS
#undef _ISO_VOLATILE_LOAD8
#undef _ISO_VOLATILE_LOAD16
// #undef _ISO_VOLATILE_LOAD32 // Used in <memory>
#undef _ISO_VOLATILE_STORE8
#undef _ISO_VOLATILE_STORE16
#undef _ISO_VOLATILE_STORE32
#undef _ISO_VOLATILE_STORE64
#undef _STD_COMPARE_EXCHANGE_128
#undef _INVALID_MEMORY_ORDER
#undef _Compiler_or_memory_barrier
#undef _Memory_barrier
#undef _Compiler_barrier

#pragma pop_macro("new")
_STL_RESTORE_CLANG_WARNINGS
#pragma warning(pop)
#pragma pack(pop)
#endif // _STL_COMPILER_PREPROCESSOR
#endif // _ATOMIC_