openmp_select.cc

#include <omp.h>
#include <random>
#include <vector>
#include <iostream>
#include <chrono>

using namespace std;
using namespace std::chrono;

std::vector<int> transform(const std::vector<int> &v) {
  std::vector<int> r;
  for (auto i : v) {
    if (i == 0) {
      r.push_back(i);
    }
  }
  return r;
}

std::vector<int> transform_mt(const std::vector<int> &v) {

  std::vector<int> sizes(100);
#pragma omp parallel for schedule(static) shared(v, sizes)
  for (unsigned i = 0; i < v.size(); i++) {
    auto e = v[i];
    if (e == 0) {
      sizes[omp_get_thread_num()] += 1;
    }
  }

  int rsize = 0;
  for (auto size : sizes) {
    rsize += size;
  }
  std::vector<int> r(rsize);

  unsigned j = 0;
#pragma omp parallel for schedule(static) shared(r, v, sizes) firstprivate(j)
  for (unsigned i = 0; i < v.size(); i++) {
    if (j == 0) {
      j = omp_get_thread_num() > 0 ? sizes[omp_get_thread_num() - 1] : 0;
    }
    auto e = v[i];
    if (e == 0) {
      r[j] = e;
      j++;
    }
  }
  return r;
}

int main() {

  std::random_device rd;
  std::mt19937_64 gen(rd());

  std::uniform_int_distribution<unsigned long long> dis;
  const int vec_size = 1000000;

  std::vector<int> v(vec_size);

  for (int i = 0; i < vec_size; i++) {
    v[i] = dis(gen) % 2;
  }

  high_resolution_clock::time_point t1 = high_resolution_clock::now();
  /* transform(v); */
  transform_mt(v);
  high_resolution_clock::time_point t2 = high_resolution_clock::now();

  auto duration = duration_cast<microseconds>(t2 - t1).count();

  cout << "TIME: " << duration << std::endl;
  return 0;
}