Question

I’m normally used to using OpenMP for parallelizing tasks. However, I thought I might give Parallel Patterns Library a shot when I started working on a new project recently.

Problem is that I have to store data in the list structure, which would have varying sizes throughout the process. I could parallelize lists with OpenMP with creating tasks with each iteration, and I tried to accomplish the same thing with the following code

#include <list>
#include <ppl.h>
#include <iostream>
#include <Windows.h>
using namespace std;
using namespace Concurrency;

void expo(double x){
    double r = 1;

    for(int i = 0 ; i<1000000 ; ++i){
        r *= x;
    }
    cout << r << " ";
}

int wmain()
{
    int begin,end;
    list<int> numbers;
    for(int i=1; i<11 ; ++i){
        numbers.push_back(i);
    }

    list<int>::iterator lit;

    structured_task_group tasks;
    begin = GetTickCount();
    for(lit=numbers.begin() ; lit!=numbers.end() ; lit++){
        int k = *lit;
        auto task1 = make_task([&k](){ expo(k); });
        tasks.run(task1);
        tasks.wait();
    }
    end = GetTickCount();
    printf("elapsed: %d ms",end - begin);
    cin.get();
    return 0;
}

But, as you can see, I have to wait to the completion of the task inside the for-loop, which implies a serial execution. If I place “tasks.wait()” after the for-loop, I am getting an error, and the compiler says that there is a wait command missing after the run command.

How can I get around this problem?

By the way, do you know any good tutorial on PPL? I have found this page (http://msdn.microsoft.com/en-us/library/dd492418.aspx) which didn’t seem to me as the perfect tutorial for a beginner.

Thanks in advance!

Answer 1

You can replace your structured_task_group and for() loop with the following:

parallel_for_each(numbers.begin(), numbers.end(), [](int k){ expo(k); });