i tried play std library ,in particular thread one. idea fill matrix , without multithreading. have 8 cores.
void fillmatrix(int id, std::vector<std::vector<double>> & lmatrix,int inthread){ (int = 0; < 10000; ++i){ if (i % inthread == id){ (int j = 0; j < 10000; ++j){ lmatrix[i][j] = 123456.0; } } } } void testmt(int inthread){ std::vector<std::thread> lpool; std::vector<std::vector<double>> lmatrix(10000, std::vector<double>(10000)); (int = 0; < inthread; ++i){ lpool.push_back(std::thread(std::bind(&fillmatrix,i, lmatrix,inthread))); } (std::thread & t : lpool){ t.join(); } } the main code :
int main(){ const clock_t begin_time1 = clock(); testmt(1); std::cout << float(clock() - begin_time1) / clocks_per_sec; } testmt(1) takes 2.1 seconds run, while testmt(8) takes 7.032 seconds.
any idea ? thanks.
your code not run want to. after threads run, you'll notice none of elements of lmatrix set 123456.0. because std::bind copies vector that's sent thread. need use std::ref(lmatrix) in bind call work , use same matrix.
change
std::bind(&fillmatrix,i, lmatrix,inthread) to
std::bind(&fillmatrix, i, std::ref(lmatrix), inthread) a significant portion of increased execution time therefore caused copy overhead; being 10000 x 10000 allocations , copies being made per thread launch (and deallocations afterwards).
