std::ref 在传递对线程的引用时真的有必要吗？

Question

我正在通读 C++ 并发实战，在第 2 章中我被引导相信即使是函数原型，例如：

void MagicFunc(Data& myData);

旨在像下面这样使用：

Data dataExample;
thread t(MagicFunc,dataExample);

我真的应该这样做

Data dataExample
thread t(MagicFunc,std::ref(dataExample));

否则，我预计发生在 "dataExample" 上的变化将不会发生。具体来说，它是这样的：

Although MagicFunc expects the second parameter to be passed by reference, the std::thread constructor t doesn’t know that; it’s oblivious to the types of the arguments expected by the function and blindly copies the supplied values. When it calls Magicfunc, it will end up passing a reference to the internal copy of data and not a reference to data itself. Consequently, when the thread finishes, these updates will be discarded as the internal copies of the supplied arguments are destroyed, and process_widget_data will be passed an unchanged Data myData rather than a correctly updated version.

但是，使用以下程序对此进行测试

#include <iostream>
#include <thread>
#include <vector>
#include <chrono>
#include <assert.h>
using namespace std;
using namespace std::chrono;

const int NUM_VALS = 50000000;

#define _MULTICORE 

void AddValuesToSlots(vector<int>& vecVals,vector<int>::iterator& begin,
                      int num,int startNum){
    int i = startNum;
    auto end = begin + num;
    for (auto itr = begin; itr < end; ++itr){
        *itr = i++;
    }
}

int main()
{
    vector<int> vecVals;
    vecVals.resize(NUM_VALS);

    //get number of cores and divide up the workload
    unsigned int numCores = thread::hardware_concurrency();
    unsigned int slotsPerThread = NUM_VALS / numCores;

    //for timing
    high_resolution_clock::time_point t1 = high_resolution_clock::now();


    thread* t = new thread[numCores];

    //get the iterator to the beginning
    auto begin = vecVals.begin();

#ifdef _MULTICORE
    for (int core = 0; core < numCores; ++core){
        t[core] = thread(AddValuesToSlots, vecVals, begin + core*slotsPerThread,
            slotsPerThread, core*slotsPerThread);
    }

    for (int core = 0; core < numCores; ++core){
        t[core].join();
    }
#else
    AddValuesToSlots(vecVals, begin, NUM_VALS, 0);
#endif


    delete[] t;

    //how long did it take?
    high_resolution_clock::time_point t2 = high_resolution_clock::now();
    cout << duration_cast<milliseconds>(t2-t1).count() << endl;

#ifdef _DEBUG
    //test that the values are correct
    for (int slot = 0; slot < NUM_VALS; ++slot)
        assert(vecVals[slot] == slot);
#endif

    return 0;
}

我试过将 vecVals 封装在 std::ref 中，并且没有，两次都没有问题地执行。那么std::ref真的有必要，提供的信息有误吗？

谢谢

Answer 1

您没有直接更改 vecVals。迭代器正在工作，因为复制迭代器是可以的，它仍然指向相同的内存地址

Answer 2

您发布的代码在标准下实际上是非法的。 std::thread 应该使用参数的右值副本调用 AddValuesToSlots。

一些 C++ 编译器会出错，而是使用参数的左值副本调用它。

live example

测试编译器是否违反规则的简单方法是：

void func1(int&&) { std::cout << "func1\n"; }
void func2(int&) { std::cout << "func1\n"; }

int main() {
  int x;
  std::thread t1(func1, x);
  t1.join();
  std::thread t2(func2, x);
  t2.join();
}

如果 t1 ctor 被接受并且 t2 被拒绝，则您的编译器是兼容的。

如果t2 ctor被接受而t1被拒绝，说明你的编译器违反了标准。

See here for more about this MSVC compiler bug

std::ref 在传递对线程的引用时真的有必要吗？

Is std::ref really necessary when passing references to threads?

c++

multithreading

c++11

stdthread