Multi-Threading in Eigen (OpenMP is not used)

Question

I have a problem in using multi-threading in Eigen library. This is my code:

#include <QCoreApplication>
#include <iostream>
#include "Eigen/Core"
#include <QDebug>

using namespace Eigen;

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

    Eigen::setNbThreads(6);
    qDebug()  << Eigen::nbThreads( );

    int n = 1000;
    MatrixXd A = MatrixXd::Ones(n,n);
    MatrixXd B = MatrixXd::Ones(n,n);
    MatrixXd C = MatrixXd::Ones(n,n);
    C.noalias() += A*B;
    std::cout << C.sum() << "\n";


    return a.exec();
}

Whatever I did Eigen::nbThreads( ) always returns 1 no matter what number I used in Eigen::setNbThreads(6)!

I read here but it actually doesn't say clearly about how can we actually run Eigen in parallel mode when OpenMP is not present!

I also did a lot of searches but all of them are used with OpenMP!

What just happened? Is Eigen only supports OpenMP for multi-threading or it has a built-in multi-threading too?

Thanks in advance!

Answer 1

Eigen's built-in multithreading only works with activated OpenMP. If you did not compile with OpenMP, then Eigen::setNbThreads(6); does essentially nothing, otherwise it is essentially equivalent to calling omp_set_num_threads (from Eigen's point of view).

You can use Eigen in an application which itself is multi-threaded (the main caveat to be careful with are calls to setRandom() and related, as described on the page you linked to).

Also, if your own multi-threading is based on OpenMP, but you don't want Eigen to exploit multithreading, you can disable it, at compile time by defining EIGEN_DONT_PARALLELIZE, or locally by setting Eigen::setNbThreads(1);.