线程共享的全局变量的原子性

Question

Okay, trivial question. Problem is, every answer I find on here has 3/4 conflicting answers.

I have a very simple problem. I have a global variable called ABORT_SIGNAL. As of now, this is declared volatile int ABORT_SIGNAL . I understand that this does not do what I wish... which is...

I have two threads which may write to ABORT_SIGNAL. It will start at 0, and go to 1 after a number of seconds. Every other thread will be reading this variable on a regular basis, checking to see if the value has been set to 1.

Would the following achieve this ....

#include <stdint.h>
atomic_int ABORT_SIGNAL;
...

// When updating the value ...
atomic_store(&ABORT_SIGNAL, someValue);

// When reading the value ...
if (atomic_load_explicit(&ABORT_SIGNAL, memory_order_relaxed))
    doSomething()

Others have also suggested that I would need to do something like the following. After each write issue atomic_thread_fence(memory_order_acq_rel); and before each read issue atomic_thread_fence(memory_order_acq_rel);

Answer 1

Using relaxed consistency makes not much sense if you want acquire semantics.

But, frankly, all of this looks much too complicated for not much gain.

Just use

_Bool _Atomic flag;

and then the usual ops, you don't need to use all these atomic_... pseudo functions. You get sequential consistency with that.

Then, you could investigate if this is really a bottleneck, and replace with another consistency at a specific place.