<atomic>

The latest version of this topic can be found at <atomic>.

Defines classes and template classes to use to create types that support atomic operations.

Syntax

#include <atomic>  

Remarks

An atomic operation has two key properties that help you use multiple threads to correctly manipulate an object without using mutex locks.

• Because an atomic operation is indivisible, a second atomic operation on the same object from a different thread can obtain the state of the object only before or after the first atomic operation.

• Based on its memory_order argument, an atomic operation establishes ordering requirements for the visibility of the effects of other atomic operations in the same thread. Consequently, it inhibits compiler optimizations that violate the ordering requirements.

On some platforms, it might not be possible to efficiently implement atomic operations for some types without using mutex locks. An atomic type is lock-free if no atomic operations on that type use locks.

C++11: In signal-handlers you can perform atomic operations on an object obj if obj.is_lock_free() or atomic_is_lock_free(x) are true.

The class atomic_flag provides a minimal atomic type that holds a bool flag. Its operations are always lock-free.

The template class atomic<T> stores an object of its argument type T and provides atomic access to that stored value. You can instantiate it by using any type that can be copied by using memcpy and tested for equality by using memcmp. In particular, you can use it with user-defined types that meet these requirements and, in many cases, with floating-point types.

The template also has a set of specializations for integral types and a partial specialization for pointers. These specializations provide additional operations that are not available through the primary template.

Pointer Specializations

The atomic<T *> partial specializations apply to all pointer types. They provide methods for pointer arithmetic.

Integral Specializations

The atomic<integral> specializations apply to all integral types. They provide additional operations that are not available through the primary template.

Each atomic<integral> type has a corresponding macro that you can use in an if directive to determine at compile time whether operations on that type are lock-free. If the value of the macro is zero, operations on the type are not lock-free. If the value is 1, operations might be lock-free, and a runtime check is required. If the value is 2, operations are lock-free. You can use the function atomic_is_lock_free to determine at runtime whether operations on the type are lock-free.

For each of the integral types, there is a corresponding named atomic type that manages an object of that integral type. Each atomic_integral type has the same set of member functions as the corresponding instantiation of atomic<T> and can be passed to any of the non-member atomic functions.

Typedef names exist for specializations of the atomic template for some of the types that are defined in the header <inttypes.h>.

Structs

Enums

Functions

In the following list, the functions that do not end in _explicit have the semantics of the corresponding _explicit, except that they have the implicit memory_order arguments of memory_order_seq_cst.

See Also

Header Files Reference
Standard Template Library