C++ Armadillo and OpenMp: Parallelization of summation of outer products - define reduction for Armadillo matrix

问题

I am trying to parallelize a for loop using OpenMP which sums over Armadillo matrices. I have the following code:

#include <armadillo>
#include <omp.h>

int main()
{

        arma::mat A = arma::randu<arma::mat>(1000,700);
        arma::mat X = arma::zeros(700,700);
        arma::rowvec point = A.row(0);

        # pragma omp parallel for shared(A) reduction(+:X)
        for(unsigned int i = 0; i < A.n_rows; i++){
                arma::rowvec diff = point - A.row(i);
                X += diff.t() * diff; // Adding the matrices to X here
        }

}

I am getting this error:

[Legendre@localhost ~]$ g++ test2.cpp -o test2 -O2 -larmadillo -fopenmp
test2.cpp: In function ‘int main()’:
test2.cpp:11:52: error: user defined reduction not found for ‘X’

I read up on defining reductions, but I haven't found examples for working with Armadillo matrices. What is the best way to define a reduction for Armadillo matrices in my case?

回答1:

Those reductions are only available for built-in types (double, int, etc.). Thus you have to do the reduction yourself, which is simple. Just accumulate the results for each thread in a thread-local variable and add this to the global result within a critical section.

#include <armadillo>
#include <omp.h>

int main()
{

  arma::mat A = arma::randu<arma::mat>(1000,700);
  arma::mat X = arma::zeros(700,700);
  arma::rowvec point = A.row(0);

  #pragma omp parallel shared(A)
  {
    arma::mat X_local = arma::zeros(700,700);

    #pragma omp for
    for(unsigned int i = 0; i < A.n_rows; i++)
    {
      arma::rowvec diff = point - A.row(i);
      X_local += diff.t() * diff; // Adding the matrices to X here
    }

    #pragma omp critical
    X += X_local;
  }
}

With more recent OpenMP (4.5 I think?) you can also declare a user-defined reduction for your type.

#include <armadillo>
#include <omp.h>

#pragma omp declare reduction( + : arma::mat : omp_out += omp_in ) \
  initializer( omp_priv = omp_orig )

int main()
{

  arma::mat A = arma::randu<arma::mat>(1000,700);
  arma::mat X = arma::zeros(700,700);
  arma::rowvec point = A.row(0);

  #pragma omp parallel shared(A) reduction(+:X)
  for(unsigned int i = 0; i < A.n_rows; i++)
  {
    arma::rowvec diff = point - A.row(i);
    X += diff.t() * diff; // Adding the matrices to X here
  }
}

来源：https://stackoverflow.com/questions/44993310/c-armadillo-and-openmp-parallelization-of-summation-of-outer-products-defin

标签

c++

openmp

armadillo