Redcraft/Redcraft.Utility/Source/Public/Templates/Atomic.h

357 lines
28 KiB
C++

#pragma once
#include "CoreTypes.h"
#include "Templates/Invoke.h"
#include "Memory/Alignment.h"
#include "Templates/Function.h"
#include "TypeTraits/TypeTraits.h"
#include "Templates/Noncopyable.h"
#include <atomic>
NAMESPACE_REDCRAFT_BEGIN
NAMESPACE_MODULE_BEGIN(Redcraft)
NAMESPACE_MODULE_BEGIN(Utility)
/**
* EMemoryOrder specifies how memory accesses, including regular, non-atomic memory accesses,
* are to be ordered around an atomic operation. Absent any constraints on a multi-core system,
* when multiple threads simultaneously read and write to several variables, one thread can observe
* the values change in an order different from the order another thread wrote them. Indeed,
* the apparent order of changes can even differ among multiple reader threads. Some similar effects
* can occur even on uniprocessor systems due to compiler transformations allowed by the memory model.
*
* @see https://en.cppreference.com/w/cpp/atomic/memory_order
*/
enum class EMemoryOrder : uint8
{
Relaxed = static_cast<TUnderlyingType<NAMESPACE_STD::memory_order>>(NAMESPACE_STD::memory_order_relaxed),
Consume = static_cast<TUnderlyingType<NAMESPACE_STD::memory_order>>(NAMESPACE_STD::memory_order_consume),
Acquire = static_cast<TUnderlyingType<NAMESPACE_STD::memory_order>>(NAMESPACE_STD::memory_order_acquire),
Release = static_cast<TUnderlyingType<NAMESPACE_STD::memory_order>>(NAMESPACE_STD::memory_order_release),
AcquireRelease = static_cast<TUnderlyingType<NAMESPACE_STD::memory_order>>(NAMESPACE_STD::memory_order_acq_rel),
SequentiallyConsistent = static_cast<TUnderlyingType<NAMESPACE_STD::memory_order>>(NAMESPACE_STD::memory_order_seq_cst),
};
#if BUILD_DEBUG
NAMESPACE_PRIVATE_BEGIN
FORCEINLINE void MemoryOrderCheck(EMemoryOrder Order, uint8 Require)
{
switch (Order)
{
case EMemoryOrder::Relaxed: checkf((Require) & 0x01, TEXT("Invalid memory order.")); break;
case EMemoryOrder::Consume: checkf((Require) & 0x02, TEXT("Invalid memory order.")); break;
case EMemoryOrder::Acquire: checkf((Require) & 0x04, TEXT("Invalid memory order.")); break;
case EMemoryOrder::Release: checkf((Require) & 0x08, TEXT("Invalid memory order.")); break;
case EMemoryOrder::AcquireRelease: checkf((Require) & 0x10, TEXT("Invalid memory order.")); break;
case EMemoryOrder::SequentiallyConsistent: checkf((Require) & 0x20, TEXT("Invalid memory order.")); break;
default: check_no_entry();
}
}
NAMESPACE_PRIVATE_END
#define MEMORY_ORDER_CHECK(Order, Require) NAMESPACE_PRIVATE::MemoryOrderCheck(Order, Require)
#else
#define MEMORY_ORDER_CHECK(Order, Require)
#endif
NAMESPACE_PRIVATE_BEGIN
template <typename T, bool bIsRef>
struct TAtomicImpl : FSingleton
{
protected:
using NativeAtomicType = TConditional<bIsRef, NAMESPACE_STD::atomic_ref<T>, NAMESPACE_STD::atomic<T>>;
public:
using ValueType = T;
/** Indicates that the type is always lock-free */
static constexpr bool bIsAlwaysLockFree = NativeAtomicType::is_always_lock_free;
/** Indicates the required alignment of an object to be referenced by TAtomicRef. */
static constexpr size_t RequiredAlignment = NAMESPACE_STD::atomic_ref<T>::required_alignment;
/** Constructs an atomic object. */
FORCEINLINE constexpr TAtomicImpl() requires (!bIsRef) : NativeAtomic() { };
FORCEINLINE constexpr TAtomicImpl(ValueType Desired) requires (!bIsRef) : NativeAtomic(Desired) { };
/** Constructs an atomic reference. */
FORCEINLINE explicit TAtomicImpl(ValueType& Desired) requires (bIsRef) : NativeAtomic(Desired) { check(Memory::IsAligned(&Desired, RequiredAlignment)); };
FORCEINLINE TAtomicImpl(TAtomicImpl& InValue) requires (bIsRef) : NativeAtomic(InValue) { };
/** Stores a value into an atomic object. */
FORCEINLINE ValueType operator=(ValueType Desired) { return NativeAtomic = Desired; }
FORCEINLINE ValueType operator=(ValueType Desired) volatile requires (bIsAlwaysLockFree) { return NativeAtomic = Desired; }
/** Atomically replaces the value of the atomic object with a non-atomic argument. */
FORCEINLINE void Store(ValueType Desired, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { MEMORY_ORDER_CHECK(Order, 0x01 | 0x08 | 0x20); NativeAtomic.store(Desired, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE void Store(ValueType Desired, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (bIsAlwaysLockFree) { MEMORY_ORDER_CHECK(Order, 0x01 | 0x08 | 0x20); NativeAtomic.store(Desired, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically obtains the value of the atomic object. */
NODISCARD FORCEINLINE ValueType Load(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) const { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); return NativeAtomic.load(static_cast<NAMESPACE_STD::memory_order>(Order)); }
NODISCARD FORCEINLINE ValueType Load(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) const volatile requires (bIsAlwaysLockFree) { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); return NativeAtomic.load(static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Loads a value from an atomic object. */
NODISCARD FORCEINLINE operator ValueType() const { return static_cast<ValueType>(NativeAtomic); }
NODISCARD FORCEINLINE operator ValueType() const volatile requires (bIsAlwaysLockFree) { return static_cast<ValueType>(NativeAtomic); }
/** Atomically replaces the value of the atomic object and obtains the value held previously. */
NODISCARD FORCEINLINE ValueType Exchange(ValueType Desired, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { return NativeAtomic.exchange(Desired, static_cast<NAMESPACE_STD::memory_order>(Order)); }
NODISCARD FORCEINLINE ValueType Exchange(ValueType Desired, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (bIsAlwaysLockFree) { return NativeAtomic.exchange(Desired, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically compares the value of the atomic object with non-atomic argument and performs atomic exchange if equal or atomic load if not. */
NODISCARD FORCEINLINE bool CompareExchange(ValueType& Expected, ValueType Desired, EMemoryOrder Success, EMemoryOrder Failure, bool bIsWeak = false)
{
MEMORY_ORDER_CHECK(Failure, 0x01 | 0x02 | 0x04 | 0x20);
if (bIsWeak) return NativeAtomic.compare_exchange_weak(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Success), static_cast<NAMESPACE_STD::memory_order>(Failure));
else return NativeAtomic.compare_exchange_strong(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Success), static_cast<NAMESPACE_STD::memory_order>(Failure));
}
/** Atomically compares the value of the atomic object with non-atomic argument and performs atomic exchange if equal or atomic load if not. */
NODISCARD FORCEINLINE bool CompareExchange(ValueType& Expected, ValueType Desired, EMemoryOrder Success, EMemoryOrder Failure, bool bIsWeak = false) volatile requires (bIsAlwaysLockFree)
{
MEMORY_ORDER_CHECK(Failure, 0x01 | 0x02 | 0x04 | 0x20);
if (bIsWeak) return NativeAtomic.compare_exchange_weak(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Success), static_cast<NAMESPACE_STD::memory_order>(Failure));
else return NativeAtomic.compare_exchange_strong(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Success), static_cast<NAMESPACE_STD::memory_order>(Failure));
}
/** Atomically compares the value of the atomic object with non-atomic argument and performs atomic exchange if equal or atomic load if not. */
NODISCARD FORCEINLINE bool CompareExchange(ValueType& Expected, ValueType Desired, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent, bool bIsWeak = false)
{
if (bIsWeak) return NativeAtomic.compare_exchange_weak(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Order));
else return NativeAtomic.compare_exchange_strong(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Order));
}
/** Atomically compares the value of the atomic object with non-atomic argument and performs atomic exchange if equal or atomic load if not. */
NODISCARD FORCEINLINE bool CompareExchange(ValueType& Expected, ValueType Desired, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent, bool bIsWeak = false) volatile requires (bIsAlwaysLockFree)
{
if (bIsWeak) return NativeAtomic.compare_exchange_weak(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Order));
else return NativeAtomic.compare_exchange_strong(Expected, Desired, static_cast<NAMESPACE_STD::memory_order>(Order));
}
/** Blocks the thread until notified and the atomic value changes. */
FORCEINLINE void Wait(ValueType Old, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); NativeAtomic.wait(Old, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE void Wait(ValueType Old, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); NativeAtomic.wait(Old, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Notifies at least one or all threads blocked waiting on the atomic object. */
FORCEINLINE void Notify(bool bIsAll = false) { if (bIsAll) NativeAtomic.notify_all(); else NativeAtomic.notify_one(); }
FORCEINLINE void Notify(bool bIsAll = false) volatile { if (bIsAll) NativeAtomic.notify_all(); else NativeAtomic.notify_one(); }
/** Atomically executes the 'Func' on the value stored in the atomic object and obtains the value held previously. */
template <typename F> requires (CInvocableResult<ValueType, F, ValueType>)
FORCEINLINE ValueType FetchFn(F&& Func, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent)
{
ValueType Temp(Load(EMemoryOrder::Relaxed));
// We do a weak read here because we require a loop.
while (!CompareExchange(Temp, InvokeResult<ValueType>(Forward<F>(Func), Temp), Order, true));
return Temp;
}
/** Atomically executes the 'Func' on the value stored in the atomic object and obtains the value held previously. */
template <typename F> requires (CInvocableResult<ValueType, F, ValueType> && bIsAlwaysLockFree)
FORCEINLINE ValueType FetchFn(F&& Func, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile
{
ValueType Temp(Load(EMemoryOrder::Relaxed));
// We do a weak read here because we require a loop.
while (!CompareExchange(Temp, InvokeResult<ValueType>(Forward<F>(Func), Temp), Order, true));
return Temp;
}
/** Atomically adds the argument to the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchAdd(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> || CFloatingPoint<T>) { return NativeAtomic.fetch_add(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchAdd(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree { return NativeAtomic.fetch_add(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically adds the argument to the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchAdd(ptrdiff InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CPointer<T> ) { return NativeAtomic.fetch_add(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchAdd(ptrdiff InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CPointer<T> && bIsAlwaysLockFree) { return NativeAtomic.fetch_add(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically subtracts the argument from the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchSub(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> || CFloatingPoint<T>) { return NativeAtomic.fetch_sub(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchSub(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree { return NativeAtomic.fetch_sub(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically subtracts the argument from the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchSub(ptrdiff InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CPointer<T> ) { return NativeAtomic.fetch_sub(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchSub(ptrdiff InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CPointer<T> && bIsAlwaysLockFree) { return NativeAtomic.fetch_sub(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically multiples the argument from the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchMul(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> || CFloatingPoint<T>) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old * InValue; }); }
FORCEINLINE ValueType FetchMul(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree { return FetchFn([InValue](ValueType Old) -> ValueType { return Old * InValue; }); }
/** Atomically divides the argument from the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchDiv(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> || CFloatingPoint<T>) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old / InValue; }); }
FORCEINLINE ValueType FetchDiv(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree { return FetchFn([InValue](ValueType Old) -> ValueType { return Old / InValue; }); }
/** Atomically models the argument from the value stored in the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchMod(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> ) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old % InValue; }); }
FORCEINLINE ValueType FetchMod(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old % InValue; }); }
/** Atomically performs bitwise AND between the argument and the value of the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchAnd(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> ) { return NativeAtomic.fetch_and(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchAnd(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return NativeAtomic.fetch_and(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically performs bitwise OR between the argument and the value of the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchOr(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> ) { return NativeAtomic.fetch_or(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchOr(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return NativeAtomic.fetch_or(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically performs bitwise XOR between the argument and the value of the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchXor(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> ) { return NativeAtomic.fetch_xor(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE ValueType FetchXor(ValueType InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return NativeAtomic.fetch_xor(InValue, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically performs bitwise LSH between the argument and the value of the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchLsh(size_t InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> ) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old << InValue; }); }
FORCEINLINE ValueType FetchLsh(size_t InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old << InValue; }); }
/** Atomically performs bitwise RSH between the argument and the value of the atomic object and obtains the value held previously. */
FORCEINLINE ValueType FetchRsh(size_t InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) requires (CIntegral<T> ) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old >> InValue; }); }
FORCEINLINE ValueType FetchRsh(size_t InValue, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return FetchFn([InValue](ValueType Old) -> ValueType { return Old >> InValue; }); }
/** Increments the atomic value by one. */
FORCEINLINE ValueType operator++() requires ((CIntegral<T> || CPointer<T>) ) { return ++NativeAtomic; }
FORCEINLINE ValueType operator++() volatile requires ((CIntegral<T> || CPointer<T>) && bIsAlwaysLockFree) { return ++NativeAtomic; }
/** Increments the atomic value by one. */
FORCEINLINE ValueType operator++(int) requires ((CIntegral<T> || CPointer<T>) ) { return NativeAtomic++; }
FORCEINLINE ValueType operator++(int) volatile requires ((CIntegral<T> || CPointer<T>) && bIsAlwaysLockFree) { return NativeAtomic++; }
/** Decrements the atomic value by one. */
FORCEINLINE ValueType operator--() requires ((CIntegral<T> || CPointer<T>) ) { return --NativeAtomic; }
FORCEINLINE ValueType operator--() volatile requires ((CIntegral<T> || CPointer<T>) && bIsAlwaysLockFree) { return --NativeAtomic; }
/** Decrements the atomic value by one. */
FORCEINLINE ValueType operator--(int) requires ((CIntegral<T> || CPointer<T>) ) { return NativeAtomic--; }
FORCEINLINE ValueType operator--(int) volatile requires ((CIntegral<T> || CPointer<T>) && bIsAlwaysLockFree) { return NativeAtomic--; }
/** Adds with the atomic value. */
FORCEINLINE ValueType operator+=(ValueType InValue) requires ((CIntegral<T> || CFloatingPoint<T>) ) { return NativeAtomic += InValue; }
FORCEINLINE ValueType operator+=(ValueType InValue) volatile requires ((CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree) { return NativeAtomic += InValue; }
/** Adds with the atomic value. */
FORCEINLINE ValueType operator+=(ptrdiff InValue) requires (CPointer<T> ) { return NativeAtomic += InValue; }
FORCEINLINE ValueType operator+=(ptrdiff InValue) volatile requires (CPointer<T> && bIsAlwaysLockFree) { return NativeAtomic += InValue; }
/** Subtracts with the atomic value. */
FORCEINLINE ValueType operator-=(ValueType InValue) requires ((CIntegral<T> || CFloatingPoint<T>) ) { return NativeAtomic -= InValue; }
FORCEINLINE ValueType operator-=(ValueType InValue) volatile requires ((CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree) { return NativeAtomic -= InValue; }
/** Subtracts with the atomic value. */
FORCEINLINE ValueType operator-=(ptrdiff InValue) requires (CPointer<T> ) { return NativeAtomic -= InValue; }
FORCEINLINE ValueType operator-=(ptrdiff InValue) volatile requires (CPointer<T> && bIsAlwaysLockFree) { return NativeAtomic -= InValue; }
/** Multiples with the atomic value. */
FORCEINLINE ValueType operator*=(ValueType InValue) requires ((CIntegral<T> || CFloatingPoint<T>) ) { return FetchMul(InValue) * InValue; }
FORCEINLINE ValueType operator*=(ValueType InValue) volatile requires ((CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree) { return FetchMul(InValue) * InValue; }
/** Divides with the atomic value. */
FORCEINLINE ValueType operator/=(ValueType InValue) requires ((CIntegral<T> || CFloatingPoint<T>) ) { return FetchDiv(InValue) / InValue; }
FORCEINLINE ValueType operator/=(ValueType InValue) volatile requires ((CIntegral<T> || CFloatingPoint<T>) && bIsAlwaysLockFree) { return FetchDiv(InValue) / InValue; }
/** Models with the atomic value. */
FORCEINLINE ValueType operator%=(ValueType InValue) requires (CIntegral<T> ) { return FetchMod(InValue) % InValue; }
FORCEINLINE ValueType operator%=(ValueType InValue) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return FetchMod(InValue) % InValue; }
/** Performs bitwise AND with the atomic value. */
FORCEINLINE ValueType operator&=(ValueType InValue) requires (CIntegral<T> ) { return NativeAtomic &= InValue; }
FORCEINLINE ValueType operator&=(ValueType InValue) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return NativeAtomic &= InValue; }
/** Performs bitwise OR with the atomic value. */
FORCEINLINE ValueType operator|=(ValueType InValue) requires (CIntegral<T> ) { return NativeAtomic |= InValue; }
FORCEINLINE ValueType operator|=(ValueType InValue) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return NativeAtomic |= InValue; }
/** Performs bitwise XOR with the atomic value. */
FORCEINLINE ValueType operator^=(ValueType InValue) requires (CIntegral<T> ) { return NativeAtomic ^= InValue; }
FORCEINLINE ValueType operator^=(ValueType InValue) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return NativeAtomic ^= InValue; }
/** Performs bitwise LSH with the atomic value. */
FORCEINLINE ValueType operator<<=(size_t InValue) requires (CIntegral<T> ) { return FetchLsh(InValue) << InValue; }
FORCEINLINE ValueType operator<<=(size_t InValue) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return FetchLsh(InValue) << InValue; }
/** Performs bitwise RSH with the atomic value. */
FORCEINLINE ValueType operator>>=(size_t InValue) requires (CIntegral<T> ) { return FetchRsh(InValue) >> InValue; }
FORCEINLINE ValueType operator>>=(size_t InValue) volatile requires (CIntegral<T> && bIsAlwaysLockFree) { return FetchRsh(InValue) >> InValue; }
protected:
NativeAtomicType NativeAtomic;
};
NAMESPACE_PRIVATE_END
template <typename T> requires (CTriviallyCopyable<T>
&& CCopyConstructible<T> && CMoveConstructible<T>
&& CCopyAssignable<T> && CMoveAssignable<T>)
struct TAtomic final : STRONG_INHERIT(NAMESPACE_PRIVATE::TAtomicImpl<T, false>);
template <typename T> requires (CTriviallyCopyable<T>)
struct TAtomicRef final : STRONG_INHERIT(NAMESPACE_PRIVATE::TAtomicImpl<T, true>);
template <typename T>
TAtomic(T) -> TAtomic<T>;
template <typename T>
TAtomicRef(T&) -> TAtomicRef<T>;
static_assert(sizeof(TAtomic<int32>) == sizeof(int32), "The byte size of TAtomic is unexpected");
/**
* FAtomicFlag is an atomic boolean type. Unlike all specializations of TAtomic, it is guaranteed to be lock-free.
* Unlike TAtomic<bool>, FAtomicFlag does not provide load or store operations.
*/
struct FAtomicFlag final : FSingleton
{
public:
/** Constructs an atomic flag. */
FORCEINLINE constexpr FAtomicFlag() : NativeAtomic() { };
/** Atomically sets flag to false. */
FORCEINLINE void Clear(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { MEMORY_ORDER_CHECK(Order, 0x01 | 0x08 | 0x20); NativeAtomic.clear(static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE void Clear(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile { MEMORY_ORDER_CHECK(Order, 0x01 | 0x08 | 0x20); NativeAtomic.clear(static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically sets the flag to true and obtains its previous value. */
FORCEINLINE bool TestAndSet(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { return NativeAtomic.test_and_set(static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE bool TestAndSet(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) volatile { return NativeAtomic.test_and_set(static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Atomically returns the value of the flag. */
NODISCARD FORCEINLINE bool Test(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) const { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); return NativeAtomic.test(static_cast<NAMESPACE_STD::memory_order>(Order)); }
NODISCARD FORCEINLINE bool Test(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) const volatile { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); return NativeAtomic.test(static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Blocks the thread until notified and the atomic value changes. */
FORCEINLINE void Wait(bool Old, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) const { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); const_cast<const NAMESPACE_STD::atomic_flag&>(NativeAtomic).wait(Old, static_cast<NAMESPACE_STD::memory_order>(Order)); }
FORCEINLINE void Wait(bool Old, EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) const volatile { MEMORY_ORDER_CHECK(Order, 0x01 | 0x02 | 0x04 | 0x20); const_cast<const NAMESPACE_STD::atomic_flag&>(NativeAtomic).wait(Old, static_cast<NAMESPACE_STD::memory_order>(Order)); }
/** Notifies at least one or all threads blocked waiting on the atomic object. */
FORCEINLINE void Notify(bool bIsAll = false) { if (bIsAll) const_cast<NAMESPACE_STD::atomic_flag&>(NativeAtomic).notify_all(); else const_cast<NAMESPACE_STD::atomic_flag&>(NativeAtomic).notify_one(); }
FORCEINLINE void Notify(bool bIsAll = false) volatile { if (bIsAll) const_cast<NAMESPACE_STD::atomic_flag&>(NativeAtomic).notify_all(); else const_cast<NAMESPACE_STD::atomic_flag&>(NativeAtomic).notify_one(); }
private:
NAMESPACE_STD::atomic_flag NativeAtomic;
};
template <typename T>
NODISCARD inline T KillDependency(T InValue)
{
T Temp(InValue);
return Temp;
}
extern "C" FORCEINLINE void AtomicThreadFence(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { NAMESPACE_STD::atomic_thread_fence(static_cast<NAMESPACE_STD::memory_order>(Order)); }
extern "C" FORCEINLINE void AtomicSignalFence(EMemoryOrder Order = EMemoryOrder::SequentiallyConsistent) { NAMESPACE_STD::atomic_signal_fence(static_cast<NAMESPACE_STD::memory_order>(Order)); }
#undef MEMORY_ORDER_CHECK
NAMESPACE_MODULE_END(Utility)
NAMESPACE_MODULE_END(Redcraft)
NAMESPACE_REDCRAFT_END