C++ std async vs regular loops performance

fast.cpp

// Tutorial by The Cherno on Youtube
// https://youtu.be/5HWCsmE9DrE

// Compile using
// g++ -std=c++17 fast.cpp -o fast

#include <vector>
#include <chrono>
#include <thread>
#include <future>
#include <iostream>
#include <cmath>

std::vector<double> repetitive_math() {
    std::vector<double> subvector;
    double result = 0;
    for (double i = pow(5,7); i >= 0; i--) {
        result += atan(i) * tan(i);
        subvector.push_back(result);
    }
    return subvector;
}

int main() {
    using std::chrono::high_resolution_clock;
    using std::chrono::duration_cast;
    using std::chrono::duration;
    using std::chrono::milliseconds;

    std::vector<std::future<std::vector<double>>> futures;

    auto start = high_resolution_clock::now();
    
    // The main task
    for (int i=0; i<500; i++) {
        futures.push_back(std::async(std::launch::async, repetitive_math));
    }

    auto end = high_resolution_clock::now();
    auto ms_int = duration_cast<milliseconds>(end - start);
    std::cout << "Duration: " << ms_int.count() << " ms" << std::endl;

    std::cout << "The size of the outer vector: " << futures.size() << std::endl;
    std::cout << "The size of the inner vector: " << futures[1].get().size() << std::endl;

    return 0;
}

// Result
// Duration: 391 ms
// The size of the outer vector: 500
// The size of the inner vector: 78126

slow.cpp

// Compile using
// g++ -std=c++17 slow.cpp -o slow

#include <vector>
#include <chrono>
#include <thread>
#include <future>
#include <iostream>
#include <cmath>

std::vector<double> repetitive_math() {
    std::vector<double> subvector;
    double result = 0;
    for (double i = pow(5,7); i >= 0; i--) {
        result += atan(i) * tan(i);
        subvector.push_back(result);
    }
    return subvector;
}

int main() {
    using std::chrono::high_resolution_clock;
    using std::chrono::duration_cast;
    using std::chrono::duration;
    using std::chrono::milliseconds;

    // Instead of using futures, we are using a regular 2d vector
    std::vector<std::vector<double>> multivector;

    auto start = high_resolution_clock::now();
    
    // The main task
    for (int i=0; i<500; i++) {
        multivector.push_back(repetitive_math());
    }

    auto end = high_resolution_clock::now();
    auto ms_int = duration_cast<milliseconds>(end - start);
    std::cout << "Duration: " << ms_int.count() << " ms" << std::endl;

    std::cout << "The size of the outer vector: " << multivector.size() << std::endl;
    std::cout << "The size of the inner vector: " << multivector[1].size() << std::endl;

    return 0;
}

// Result
// Duration: 3273 ms
// The size of the outer vector: 500
// The size of the inner vector: 78126