Intel Threading Building Blocks

Name

atomic<T> Template Class — Template class for atomic operations.

Synopsis

#include "tbb/atomic.h"

template<typename T> atomic;

Description

An atomic<T> supports atomic read, write, fetch-and-add, fetch-and-store, and compare-and-swap. Type T may be an integral type or a pointer type. When T is a pointer type, arithmetic operations are interpreted as pointer arithmetic. For example, if x has type atomic<float*> and a float occupies four bytes, ++x advances x by four bytes. The specialization atomic<void*> does not allow pointer arithmetic.

Some of the methods offer two method variants, one of which is a templated form that permits more selective memory fencing using memory_semantics. For instance, the first of the following fetch_and_add calls is templated to specify a release fence, whereas the second call is not templated:

	refcount.fetch_and_add<release>(-1);
	refcount.fetch_and_add(1);

On IA-32 and Intel 64 processors, the templated calls have the same effect as the nontemplated variants because the processors do not support weaker memory consistency. On processors with weak memory consistency, the templated calls may improve performance by allowing the memory subsystem more latitude on the orders of reads and writes; therefore, using them may improve performance. Table 7-5 shows the fencing defaults (nontemplated form).

Tip

The template class atomic<T> does not have any nontrivial constructors because such constructors could lead to accidental introduction of compiler temporaries ...