ladybird/AK/Atomic.h
Andreas Kling ae3ffdd521 AK: Make it possible to not using AK classes into the global namespace
This patch adds the `USING_AK_GLOBALLY` macro which is enabled by
default, but can be overridden by build flags.

This is a step towards integrating Jakt and AK types.
2022-11-26 15:51:34 +01:00

449 lines
14 KiB
C++

/*
* Copyright (c) 2018-2020, Andreas Kling <kling@serenityos.org>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#pragma once
#include <AK/Concepts.h>
#include <AK/Platform.h>
#include <AK/Types.h>
namespace AK {
static inline void atomic_signal_fence(MemoryOrder order) noexcept
{
return __atomic_signal_fence(order);
}
static inline void atomic_thread_fence(MemoryOrder order) noexcept
{
return __atomic_thread_fence(order);
}
static inline void full_memory_barrier() noexcept
{
atomic_signal_fence(AK::MemoryOrder::memory_order_acq_rel);
atomic_thread_fence(AK::MemoryOrder::memory_order_acq_rel);
}
template<typename T>
static inline T atomic_exchange(volatile T* var, T desired, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_exchange_n(var, desired, order);
}
template<typename T, typename V = RemoveVolatile<T>>
static inline V* atomic_exchange(volatile T** var, V* desired, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_exchange_n(var, desired, order);
}
template<typename T, typename V = RemoveVolatile<T>>
static inline V* atomic_exchange(volatile T** var, std::nullptr_t, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_exchange_n(const_cast<V**>(var), nullptr, order);
}
template<typename T>
[[nodiscard]] static inline bool atomic_compare_exchange_strong(volatile T* var, T& expected, T desired, MemoryOrder order = memory_order_seq_cst) noexcept
{
if (order == memory_order_acq_rel || order == memory_order_release)
return __atomic_compare_exchange_n(var, &expected, desired, false, memory_order_release, memory_order_acquire);
return __atomic_compare_exchange_n(var, &expected, desired, false, order, order);
}
template<typename T, typename V = RemoveVolatile<T>>
[[nodiscard]] static inline bool atomic_compare_exchange_strong(volatile T** var, V*& expected, V* desired, MemoryOrder order = memory_order_seq_cst) noexcept
{
if (order == memory_order_acq_rel || order == memory_order_release)
return __atomic_compare_exchange_n(var, &expected, desired, false, memory_order_release, memory_order_acquire);
return __atomic_compare_exchange_n(var, &expected, desired, false, order, order);
}
template<typename T, typename V = RemoveVolatile<T>>
[[nodiscard]] static inline bool atomic_compare_exchange_strong(volatile T** var, V*& expected, std::nullptr_t, MemoryOrder order = memory_order_seq_cst) noexcept
{
if (order == memory_order_acq_rel || order == memory_order_release)
return __atomic_compare_exchange_n(const_cast<V**>(var), &expected, nullptr, false, memory_order_release, memory_order_acquire);
return __atomic_compare_exchange_n(const_cast<V**>(var), &expected, nullptr, false, order, order);
}
template<typename T>
static inline T atomic_fetch_add(volatile T* var, T val, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_fetch_add(var, val, order);
}
template<typename T>
static inline T atomic_fetch_sub(volatile T* var, T val, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_fetch_sub(var, val, order);
}
template<typename T>
static inline T atomic_fetch_and(volatile T* var, T val, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_fetch_and(var, val, order);
}
template<typename T>
static inline T atomic_fetch_or(volatile T* var, T val, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_fetch_or(var, val, order);
}
template<typename T>
static inline T atomic_fetch_xor(volatile T* var, T val, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_fetch_xor(var, val, order);
}
template<typename T>
static inline T atomic_load(volatile T* var, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_load_n(var, order);
}
template<typename T, typename V = RemoveVolatile<T>>
static inline V* atomic_load(volatile T** var, MemoryOrder order = memory_order_seq_cst) noexcept
{
return __atomic_load_n(const_cast<V**>(var), order);
}
template<typename T>
static inline void atomic_store(volatile T* var, T desired, MemoryOrder order = memory_order_seq_cst) noexcept
{
__atomic_store_n(var, desired, order);
}
template<typename T, typename V = RemoveVolatile<T>>
static inline void atomic_store(volatile T** var, V* desired, MemoryOrder order = memory_order_seq_cst) noexcept
{
__atomic_store_n(var, desired, order);
}
template<typename T, typename V = RemoveVolatile<T>>
static inline void atomic_store(volatile T** var, std::nullptr_t, MemoryOrder order = memory_order_seq_cst) noexcept
{
__atomic_store_n(const_cast<V**>(var), nullptr, order);
}
template<typename T>
static inline bool atomic_is_lock_free(volatile T* ptr = nullptr) noexcept
{
return __atomic_is_lock_free(sizeof(T), ptr);
}
template<typename T, MemoryOrder DefaultMemoryOrder = AK::MemoryOrder::memory_order_seq_cst>
class Atomic {
// FIXME: This should work through concepts/requires clauses, but according to the compiler,
// "IsIntegral is not more specialized than IsFundamental".
// Additionally, Enums are not fundamental types except that they behave like them in every observable way.
static_assert(IsFundamental<T> | IsEnum<T>, "Atomic doesn't support non-primitive types, because it relies on compiler intrinsics. If you put non-primitives into it, you'll get linker errors like \"undefined reference to __atomic_store\".");
T m_value { 0 };
public:
Atomic() noexcept = default;
Atomic& operator=(Atomic const&) volatile = delete;
Atomic& operator=(Atomic&&) volatile = delete;
Atomic(Atomic const&) = delete;
Atomic(Atomic&&) = delete;
constexpr Atomic(T val) noexcept
: m_value(val)
{
}
volatile T* ptr() noexcept
{
return &m_value;
}
T exchange(T desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
// We use this hack to prevent unnecessary initialization, even if T has a default constructor.
// NOTE: Will need to investigate if it pessimizes the generated assembly.
alignas(T) u8 buffer[sizeof(T)];
T* ret = reinterpret_cast<T*>(buffer);
__atomic_exchange(&m_value, &desired, ret, order);
return *ret;
}
[[nodiscard]] bool compare_exchange_strong(T& expected, T desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
if (order == memory_order_acq_rel || order == memory_order_release)
return __atomic_compare_exchange(&m_value, &expected, &desired, false, memory_order_release, memory_order_acquire);
return __atomic_compare_exchange(&m_value, &expected, &desired, false, order, order);
}
ALWAYS_INLINE operator T() const volatile noexcept
{
return load();
}
ALWAYS_INLINE T load(MemoryOrder order = DefaultMemoryOrder) const volatile noexcept
{
alignas(T) u8 buffer[sizeof(T)];
T* ret = reinterpret_cast<T*>(buffer);
__atomic_load(&m_value, ret, order);
return *ret;
}
// NOLINTNEXTLINE(misc-unconventional-assign-operator) We want operator= to exchange the value, so returning an object of type Atomic& here does not make sense
ALWAYS_INLINE T operator=(T desired) volatile noexcept
{
store(desired);
return desired;
}
ALWAYS_INLINE void store(T desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
__atomic_store(&m_value, &desired, order);
}
ALWAYS_INLINE bool is_lock_free() const volatile noexcept
{
return __atomic_is_lock_free(sizeof(m_value), &m_value);
}
};
template<Integral T, MemoryOrder DefaultMemoryOrder>
class Atomic<T, DefaultMemoryOrder> {
T m_value { 0 };
public:
Atomic() noexcept = default;
Atomic& operator=(Atomic const&) volatile = delete;
Atomic& operator=(Atomic&&) volatile = delete;
Atomic(Atomic const&) = delete;
Atomic(Atomic&&) = delete;
constexpr Atomic(T val) noexcept
: m_value(val)
{
}
volatile T* ptr() noexcept
{
return &m_value;
}
T exchange(T desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_exchange_n(&m_value, desired, order);
}
[[nodiscard]] bool compare_exchange_strong(T& expected, T desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
if (order == memory_order_acq_rel || order == memory_order_release)
return __atomic_compare_exchange_n(&m_value, &expected, desired, false, memory_order_release, memory_order_acquire);
return __atomic_compare_exchange_n(&m_value, &expected, desired, false, order, order);
}
ALWAYS_INLINE T operator++() volatile noexcept
{
return fetch_add(1) + 1;
}
ALWAYS_INLINE T operator++(int) volatile noexcept
{
return fetch_add(1);
}
ALWAYS_INLINE T operator+=(T val) volatile noexcept
{
return fetch_add(val) + val;
}
ALWAYS_INLINE T fetch_add(T val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_add(&m_value, val, order);
}
ALWAYS_INLINE T operator--() volatile noexcept
{
return fetch_sub(1) - 1;
}
ALWAYS_INLINE T operator--(int) volatile noexcept
{
return fetch_sub(1);
}
ALWAYS_INLINE T operator-=(T val) volatile noexcept
{
return fetch_sub(val) - val;
}
ALWAYS_INLINE T fetch_sub(T val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_sub(&m_value, val, order);
}
ALWAYS_INLINE T operator&=(T val) volatile noexcept
{
return fetch_and(val) & val;
}
ALWAYS_INLINE T fetch_and(T val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_and(&m_value, val, order);
}
ALWAYS_INLINE T operator|=(T val) volatile noexcept
{
return fetch_or(val) | val;
}
ALWAYS_INLINE T fetch_or(T val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_or(&m_value, val, order);
}
ALWAYS_INLINE T operator^=(T val) volatile noexcept
{
return fetch_xor(val) ^ val;
}
ALWAYS_INLINE T fetch_xor(T val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_xor(&m_value, val, order);
}
ALWAYS_INLINE operator T() const volatile noexcept
{
return load();
}
ALWAYS_INLINE T load(MemoryOrder order = DefaultMemoryOrder) const volatile noexcept
{
return __atomic_load_n(&m_value, order);
}
// NOLINTNEXTLINE(misc-unconventional-assign-operator) We want operator= to exchange the value, so returning an object of type Atomic& here does not make sense
ALWAYS_INLINE T operator=(T desired) volatile noexcept
{
store(desired);
return desired;
}
ALWAYS_INLINE void store(T desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
__atomic_store_n(&m_value, desired, order);
}
ALWAYS_INLINE bool is_lock_free() const volatile noexcept
{
return __atomic_is_lock_free(sizeof(m_value), &m_value);
}
};
template<typename T, MemoryOrder DefaultMemoryOrder>
class Atomic<T*, DefaultMemoryOrder> {
T* m_value { nullptr };
public:
Atomic() noexcept = default;
Atomic& operator=(Atomic const&) volatile = delete;
Atomic& operator=(Atomic&&) volatile = delete;
Atomic(Atomic const&) = delete;
Atomic(Atomic&&) = delete;
constexpr Atomic(T* val) noexcept
: m_value(val)
{
}
volatile T** ptr() noexcept
{
return &m_value;
}
T* exchange(T* desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_exchange_n(&m_value, desired, order);
}
[[nodiscard]] bool compare_exchange_strong(T*& expected, T* desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
if (order == memory_order_acq_rel || order == memory_order_release)
return __atomic_compare_exchange_n(&m_value, &expected, desired, false, memory_order_release, memory_order_acquire);
return __atomic_compare_exchange_n(&m_value, &expected, desired, false, order, order);
}
T* operator++() volatile noexcept
{
return fetch_add(1) + 1;
}
T* operator++(int) volatile noexcept
{
return fetch_add(1);
}
T* operator+=(ptrdiff_t val) volatile noexcept
{
return fetch_add(val) + val;
}
T* fetch_add(ptrdiff_t val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_add(&m_value, val * sizeof(*m_value), order);
}
T* operator--() volatile noexcept
{
return fetch_sub(1) - 1;
}
T* operator--(int) volatile noexcept
{
return fetch_sub(1);
}
T* operator-=(ptrdiff_t val) volatile noexcept
{
return fetch_sub(val) - val;
}
T* fetch_sub(ptrdiff_t val, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
return __atomic_fetch_sub(&m_value, val * sizeof(*m_value), order);
}
operator T*() const volatile noexcept
{
return load();
}
T* load(MemoryOrder order = DefaultMemoryOrder) const volatile noexcept
{
return __atomic_load_n(&m_value, order);
}
// NOLINTNEXTLINE(misc-unconventional-assign-operator) We want operator= to exchange the value, so returning an object of type Atomic& here does not make sense
T* operator=(T* desired) volatile noexcept
{
store(desired);
return desired;
}
void store(T* desired, MemoryOrder order = DefaultMemoryOrder) volatile noexcept
{
__atomic_store_n(&m_value, desired, order);
}
bool is_lock_free() const volatile noexcept
{
return __atomic_is_lock_free(sizeof(m_value), &m_value);
}
};
}
#if USING_AK_GLOBALLY
using AK::Atomic;
using AK::full_memory_barrier;
#endif