Export (0) Print
Expand All

How to: Use combinable to Combine Sets

This topic shows how to use the concurrency::combinable class to compute the set of prime numbers.

The following example computes the set of prime numbers two times. Each computation stores the result in a std::bitset object. The example first computes the set serially and then computes the set in parallel. The example also prints to the console the time that is required to perform both computations.

This example uses the concurrency::parallel_for algorithm and a combinable object to generate thread-local sets. It then uses the concurrency::combinable::combine_each method to combine the thread-local sets into the final set.

// parallel-combine-primes.cpp 
// compile with: /EHsc
#include <windows.h>
#include <ppl.h>
#include <bitset>
#include <iostream>

using namespace concurrency;
using namespace std;

// Calls the provided work function and returns the number of milliseconds  
// that it takes to call that function. 
template <class Function>
__int64 time_call(Function&& f)
{
   __int64 begin = GetTickCount();
   f();
   return GetTickCount() - begin;
}

// Determines whether the input value is prime. 
bool is_prime(int n)
{
   if (n < 2)
      return false;
   for (int i = 2; i < n; ++i)
   {
      if ((n % i) == 0)
         return false;
   }
   return true;
}

const int limit = 40000;

int wmain()
{
   // A set of prime numbers that is computed serially.
   bitset<limit> primes1;

   // A set of prime numbers that is computed in parallel.
   bitset<limit> primes2;

   __int64 elapsed;

   // Compute the set of prime numbers in a serial loop.
   elapsed = time_call([&] 
   {
      for(int i = 0; i < limit; ++i) {
         if (is_prime(i))
            primes1.set(i);
      }
   });
   wcout << L"serial time: " << elapsed << L" ms" << endl << endl;

   // Compute the same set of numbers in parallel.
   elapsed = time_call([&] 
   {
      // Use a parallel_for loop and a combinable object to compute  
      // the set in parallel.  
      // You do not need to synchronize access to the set because the  
      // combinable object provides a separate bitset object to each thread.
      combinable<bitset<limit>> working;
      parallel_for(0, limit, [&](int i) {
         if (is_prime(i))
            working.local().set(i);
      });

      // Merge each thread-local computation into the final result.
      working.combine_each([&](bitset<limit>& local) {
         primes2 |= local;
      });
   });
   wcout << L"parallel time: " << elapsed << L" ms" << endl << endl;
}

The following sample output is for a computer that has four processors.

serial time: 312 ms

parallel time: 78 ms

Copy the example code and paste it in a Visual Studio project, or paste it in a file that is named parallel-combine-primes.cpp and then run the following command in a Visual Studio Command Prompt window.

cl.exe /EHsc parallel-combine-primes.cpp

Show:
© 2014 Microsoft