C++11：atomic::compare_exchange_weak 是否支持 none-原始类型？

Question

我有以下代码：

#include<atomic>
#include<iostream>
using namespace std;

struct Big{
    int i;
    int j;
    int k[100];
};
int main(){
    atomic<int> i;
    cout<<i.load()<<endl;
    i.store(20);
    cout<<i.load()<<endl;
    i.exchange(30);
    cout<<i.load()<<endl;

    atomic<Big> ab,expect,value;
    ab.compare_exchange_weak(expect,value,memory_order_release,memory_order_relaxed);//error
    return 0;
}

好吧，atomic 工作得很好，但我想看看 compare_exchange_weak 的 none-lock 函数是否适用于复杂的数据结构。用 --std=c++11 编译它给了我：

error: no matching member function for call to 'compare_exchange_weak'
    ab.compare_exchange_weak(expect,value,memory_order_release,memory_order_relaxed);
    ~~~^~~~~~~~~~~~~~~~~~~~~
candidate function not viable: no known conversion from 'atomic<Big>' to 'Big &' for 1st argument
    bool compare_exchange_weak(_Tp& __e, _Tp __d,

所以我的问题是：

Does std::atomic::compare_exchange_weak work with complex structures?

If intel cpu hardware CMPEXG only works within 64 bit length cache line, does structures larger than 8 bytes work with CMPEXG? Is it still atomic operation?

How to fix my program?

谢谢。

Answer 1

Does std::atomic::compare_exchange_weak work with complex structures?

是的，但是还有conditions that include trivially copyable and trivially constructible。

If intel cpu hardware CMPEXG only works within 64 bit length cache line, does structures larger than 8 bytes work with CMPEXG?

没有。它不是那样工作的。如果你创建像你那里那样的疯狂大结构，你的代码将不会是 "lockfree"。您的编译器将发出总线锁以确保线程安全，这就是为什么您永远不应该使用大数据结构来做您正在做的事情。你会把你的程序减慢数百倍，如果不是更多的话。考虑原子地交换指针。

Is it still atomic operation?

不，它使用锁。您可以使用 std::atomic::is_lock_free()

进行测试

How to fix my program?

好了：

#include <atomic>
#include <iostream>

using namespace std;

struct Big {
  int i;
  int j;
  int k[100];
};

int main() {
  Big value, expect;
  atomic<Big> ab;
  ab.compare_exchange_weak(expect, value);
  return 0;
}

C++11：atomic::compare_exchange_weak 是否支持 none-原始类型？

C++11: does atomic::compare_exchange_weak support none-primitive types?

c++

multithreading

atomic

c++11