使用ISO C ++ 0x线程会降低性能

Question

I'm writing a program in C++ using threads to improve the efficiency.

Basically I'm just creating a huge vector of integers (1 GB) and filling it with random numbers. I executed the program without threads and calculated the time needed.

Now I want to use 2 threads and see the time improvement, but the program is taking much more time with 2 threads than without. Don't know what I'm doing wrong :S

#includes...

using namespace std;

//This function just make a for from first to final. Each iteration write a random
//number in the position i of the vector
void generateRandomVector(vector<int> &vec,int first, int final);

//Inside this function i take timestamp2 and calculate the executing time
void calculateTime(clock_t start);


int main(int argc, char *argv[]){

clock_t start;
    double logaritmo;
    int n = 256*1024*1024;


//Taking timestamp 1
start = clock();

    vector<int> vec(n);

    thread t1(generateRandomVector, ref(vec), 0, n/2);
thread t2(generateRandomVector, ref(vec), n/2, n);

t1.join();
t2.join();

calculateTime(start);

I'm passing by reference the vector to both threads because I'm giving them different ranges so they will never be accessing the same position.

If needed, I can also post the generateRandomVector function.

Hope someone can help :D

EDIT - generateRandomVector function:

void generarRandomVector(vector<int> &vec,int first, int final){
    srand((unsigned)time(0));
    //PID of each thread
    cout << "PID: " << this_thread::get_id() << "\n";

    for(int i = first; i < final; i++){
        vec[i] = static_cast<int> ((double)rand()*(double)(vec.size()) / ((double)RAND_MAX+1.0));
        vec[i] = vec[i] % 10;
    }
}

Answer 1

Here's the complete C++0x solution, using <random> and <chrono> :

#include <random>
#include <chrono>
#include <thread>
#include <iostream>
#include <vector>
#include <functional>

void generarRandomVector(std::vector<int> &vec, int first, int final)
{
    std::mt19937_64 e(time(0));
    std::uniform_int_distribution<> d(0, 9);
    //PID of each thread
    std::cout << "PID: " << std::this_thread::get_id() << "\n";

    for(int i = first; i < final; i++)
        vec[i] = d(e);
}

int main()
{
    typedef std::chrono::high_resolution_clock Clock;
    typedef std::chrono::duration<double> sec;
    int n = 256*1024*1024;
    Clock::time_point start = Clock::now();
    std::vector<int> vec(n);
    std::thread t1(generarRandomVector, std::ref(vec), 0, n/2);
    std::thread t2(generarRandomVector, std::ref(vec), n/2, n);
    t1.join();
    t2.join();
    Clock::time_point end = Clock::now();
    std::cout << sec(end-start).count() << " seconds\n";
}

Answer 2

Don't use rand().

rand() uses a global state to generate the next number. Also some sources on the web [1] claim that it is 'thread-safe', that means that it may use a lock, thus serializing all the calls to rand() and eliminating all the concurrency.

Answer 3

上面给出的代码没有数据竞争的风险，但是在向量的两半部分中获得相同数据的风险很大。