The test indicates that there's no sense in parallelizing simple loops over large datasets

speedup_test.cpp

#include <iostream>
#include <iomanip>
#include <random>
#include <chrono>

using namespace std;
using namespace chrono;

vector<pair<size_t, double>> TestParallel(const vector<size_t>& sizes) {
  random_device rd;
  mt19937 gen(rd());
  uniform_real_distribution<> dis(10, 100);
  vector<pair<size_t, double>> result;

  for (size_t size : sizes) {
    std::vector<double> lhs;
    std::vector<double> rhs;

    lhs.reserve(size);
    rhs.reserve(size);

    for (size_t i = 0; i < size; i++) {
      lhs.push_back(dis(gen));
      rhs.push_back(dis(gen));
    }

    size_t count = 0;
    auto start = steady_clock::now();

    #pragma opm parallel for shared(count, lhs, rhs) reduce(+: count)
    for (size_t i = 0; i < size; i++)
      count += lhs[i] < rhs[i];
    auto finish = steady_clock::now();

    double duration = duration_cast<nanoseconds>(finish - start).count() * 1e-9;

    result.push_back(make_pair(count, duration));
  }

  return result;
}

vector<pair<size_t, double>> TestSequential(const vector<size_t>& sizes) {
  random_device rd;
  mt19937 gen(rd());
  uniform_real_distribution<> dis(10, 100);
  vector<pair<size_t, double>> result;

  for (size_t size : sizes) {
    std::vector<double> lhs;
    std::vector<double> rhs;

    lhs.reserve(size);
    rhs.reserve(size);

    for (size_t i = 0; i < size; i++) {
      lhs.push_back(dis(gen));
      rhs.push_back(dis(gen));
    }

    size_t count = 0;
    auto start = steady_clock::now();
    for (size_t i = 0; i < size; i++)
      count += lhs[i] < rhs[i];
    auto finish = steady_clock::now();

    double duration = duration_cast<nanoseconds>(finish - start).count() * 1e-9;

    result.push_back(make_pair(count, duration));
  }

  return result;
}

int main() {
  vector<size_t> sizes = { 
    10, 100, 1'000, 10'000, 100'000, 1'000'000,
    10'000'000, 100'000'000
  };
  
  vector<pair<size_t, double>> parallel = TestParallel(sizes);
  vector<pair<size_t, double>> sequential = TestSequential(sizes);

  cout << scientific;

  cout << setw(12) << "Size" << " "
       << setw(12) << "Count Par." << " " << setw(12) <<  "Parallel (s)" << " "
       << setw(12) << "Count Seq." << " " << setw(12) <<  "Sequential (s)"
       << "\n";

  for (size_t i = 0; i < sizes.size(); i++) {
    cout << setw(12) << sizes[i] << " "
         << setw(12) << parallel[i].first << " " << parallel[i].second << " "
         << setw(12) << sequential[i].first << " " << sequential[i].second
         << "\n";
  }
  return 0;
}