OpenMP sample with therad_local

test.cpp

/*
# Copyright H. Watanabe 2017
# Distributed under the Boost Software License, Version 1.0.
# (See accompanying file LICENSE_1_0.txt or copy at
# http://www.boost.org/LICENSE_1_0.txt)
*/
#include <iostream>
#include <stdio.h>
#include <vector>
#include <chrono>
#include <random>
#include <algorithm>
#include <numeric>
#include <omp.h>
#include <memory>
class Worker {
private:
  std::vector<double> mytask;
public:
  int id;
  void work(int task) {
    mytask.push_back(task);
    printf("%d: %d\n", id, task);
    auto start = std::chrono::system_clock::now();
    while (true) {
      auto end = std::chrono::system_clock::now();
      auto diff = std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count();
      if (diff > task)break;
    }
  }
  void show_summary(void) {
    int sum = std::accumulate(mytask.begin(), mytask.end(), 0);
    printf("%d: total num = %d, total tasks = %d\n", id, mytask.size(), sum);
  }
};

int
main(void) {
  int nthreads = 0;
  #pragma omp parallel
  {
    nthreads = omp_get_num_threads();
  }
  std::vector <std::unique_ptr<Worker> > workers(nthreads);
  thread_local Worker worker;
  #pragma omp parallel for
  for (int i = 0; i < nthreads; i++) {
    worker.id = i;
  }
  std::mt19937 mt;
  std::uniform_int_distribution<int> ud(0, 1000);
  std::vector<int> tasks;
  const int N = 20;
  for (int i = 0; i < N; i++) {
    int task = ud(mt);
    if (task > 900) {
      task *= 3;
    }
    tasks.push_back(task);
    std::cout << task << std::endl;
  }
  auto start = std::chrono::system_clock::now();
  #pragma omp parallel for schedule(dynamic)
  for (int i = 0; i < N; i++) {
    worker.work(tasks[i]);
  }
  printf("------------\n");
  #pragma omp parallel for 
  for (int i = 0; i < nthreads; i++) {
    worker.show_summary();
  }
  printf("------------\n");
  auto end = std::chrono::system_clock::now();
  int sum = std::accumulate(tasks.begin(), tasks.end(), 0);
  double ideal_task = static_cast<double>(sum) / static_cast<double>(nthreads);
  auto diff = std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count();
  printf("Total Tasks: %d\n", sum);
  printf("Ideal Tasks per thread: %d\n", static_cast<int>(ideal_task));
  printf("Elapsed time [ms]: %d\n", diff);
  printf("------------\n");
}

$ g++ -O2 -std=c++11 -fopenmp test.cpp -o a.out
$ ./a.out
(snip)
------------
1: total num = 3, total tasks = 6452
2: total num = 4, total tasks = 6043
3: total num = 4, total tasks = 6444
0: total num = 9, total tasks = 8801
------------
Total Tasks: 27740
Ideal Tasks per thread: 6935
Elapsed time [ms]: 8821
------------