Allocating a atomic on shared memory

Question

I am trying to allocate a atomic on a shared memory block (on linux). The atomic will be accessed and modified my multiple threads simultaneously. The rational behind allocating it on shared memory is because I want to persist the values so if my process is restarted the previous state can be recovered. I know for a fact if i use a mutex in shared memory i have to initialize it as a SHARED. Is there any such requirement for atomics? Is this feasible?

Answer 1

Yes, you can do that. Here is an example I ripped from Quora (ripped code from Quora), not my code and includes boost so I have not tested it:

#include <atomic>
#include <string>
#include <iostream>
#include <cstdlib>
#include <boost/interprocess/managed_shared_memory.hpp>

using namespace std::string_literals;
namespace bip = boost::interprocess;
static_assert(ATOMIC_INT_LOCK_FREE == 2,
              "atomic_int must be lock-free");
int main(int argc, char *argv[])
{
  if(argc == 1) //Parent process
  {
    struct shm_remove {
      shm_remove() { bip::shared_memory_object::remove("szMem");}
      ~shm_remove(){ bip::shared_memory_object::remove("szMem");}
    } remover;
    bip::managed_shared_memory segment(bip::create_only,
                                       "szMem", 65536);
    auto ap = segment.construct<std::atomic_int>("the counter")(0);
    //Launch 5 child processes
    std::string s = argv[0] +" child"s;
    std::system((s + '&' + s + '&' + s + '&' + s + '&' + s).c_str());
    std::cout << "parent existing: counter = " << *ap << '\n';
    segment.destroy<std::atomic_int>("the counter");
 } else { // child
    bip::managed_shared_memory segment(bip::open_only, "szMem");
    auto res = segment.find<std::atomic_int>("the counter");
    for(int n = 0; n < 100000; ++n)
        ++*res.first; // C++17 will remove the dumb "first"
    std::cout << "child exiting, counter = " << *res.first << '\n';
  }
}

Here is documentation: link to Boost docs