Is there any reason we cannot implement a parallel_reduce for a rvalue range?

[sukeya]

Hello,

I'm writing a program merging many std::sets. I would like to parallelize it with joining two sets by std::set::merge, but oneTBB doesn't seem to provide any move join function like T join(T&&, T&&) as far as I can see Reduction, parallel_reduce.h and enumerable_thread_specific.h.
Is there any reason we cannot implement it?

[dnmokhov]

You should be able to use the lvalue-reference version of std::set::merge, e.g.,

struct Body {
  std::vector<std::set<int>> &sets;
  std::set<int> result;

  Body(std::vector<std::set<int>> &data) : sets(data) {}
  Body(Body &left, tbb::split) : sets(left.sets) {}

  void operator()(tbb::blocked_range<int> range) {
    for (int i = range.begin(); i < range.end(); ++i) {
      result.merge(sets[i]);
    }
  }

  void join(Body &right) { result.merge(right.result); }
};

In the example above, right is an lvalue reference to the subtask andright.result is an lvalue reference to its result.

Specifying a parallel_reduce body is further documented at https://oneapi-src.github.io/oneTBB/main/tbb_userguide/parallel_reduce.html

[sukeya]

I didn't know that an Body can have the join member function which argument is a lvalue-reference without const.
Thank you!