C ++ 17：如何在并行STL中获取工作项的索引

Question

Is there a way to know an ID/index of a work item in parallel algorithms?

That would be useful for example in generating iota:

std::vector<int> vec(max_size);
std::for_each(std::execution::par, vec.begin(), vec.end(), 
   [](int& elem) {
    elem = work_item_index;
});

but how to compute that work_item_index value efficiently?

Answer 1

EDIT: This approach is not guaranteed to work, as the element identity (ie the address) cannot be relied upon in parallel execution.

A correct solution is to tie the iterator range with a range of indices, as shown in Caleth's answer.

You can use pointer arithmetic:

 
 
 
  
  int* beg = vec.data(); std::for_each(std::execution::par, vec.begin(), vec.end(), [beg](int& elem) { elem = &elem - beg; });
 
  

Obviously, this only works for containers which store the elements contiguously.

Answer 2

You operate on a range that includes an index.

std::vector<int> vec(max_size);
auto indexed_vec = ranges::view::zip(vec, ranges::view::indices(vec));
std::for_each(std::execution::par, indexed_vec.begin(), indexed_vec.end(), 
   [](std::tuple<int&, int> elem) {
    std::get<0>(elem) = std::get<1>(elem);
});

Answer 3

Actually it's not standardized. It depends on parallel engine that is used for algorithms' parallelization. For example, omp_get_thread_num() for OpenMP.

However, if you just want to get index of current processing element of vector you can use counting_iterator . As soon as it's also non-standardized feature, you can use it from libraries: Boost , TBB , Intel's Parallel STL

std::vector<int> vec(max_size);
auto first = vec.begin();
std::for_each(std::execution::par, counting_iterator<int>(0), counting_iterator<int>(max_size), 
   [first](int& work_item_index) {
    first[work_item_index] = work_item_index;
});