Chunk a map with Intel TBB parallel_for

Question

I want to traverse my std::map in parallel with parallel_for.
As the tutorial says, we cannot randomly access it, so I chose parallel_for_each.

std::map<std::string, std::string> map_; // student name, student age
tbb::parallel_for_each(map_.begin(), map_.end(), [&](const auto& item) { 
 processOne worker;
 worker(item.first, map_.size());
});

But it seems that parallel_for_each will not chunk the map into subtasks, so it is quite slow.

Is it possible to do something like this?

tbb::parallel_for(tbb::blocked_range<int>(0, map_.size(), 10000), [&](tbb::blocked_range<int> r) {
 ...
});

• 463035818_is_not_an_ai: consider to use a std::vector<std::pair<Key,Value>> instead of the map
• Weimin Chan: I tried but it is twice as slow as parellel_for_each, weird.
• Öö Tiib: You can get better results with tbb::concurrent_hash_map if it suits you ... or with array where you actually keep boost::intrusive::set elements. That std::map just isn't designed to be very efficient with lot of data, even when single threaded.