Rcpp omp_set_num_threads within a package - c++

Question

I wrote the following simple example with Rcpp and OpenMP that works fine when I source the cpp file from RStudio:
#include <Rcpp.h>
#include <omp.h>
// [[Rcpp::plugins(openmp)]]
using namespace Rcpp;
// [[Rcpp::export]]
NumericMatrix my_matrix(int I, int J, int nthreads) {
NumericMatrix A(I,J);
int i,j,tid;
omp_set_num_threads(nthreads);
#pragma omp parallel for private(i, j, tid)
for(int i = 0; i < I; i++) {
for(int j = 0; j < J; j++) {
tid = omp_get_thread_num();
A(i,j) = tid ;
}
}
return A;
}
/*** R
set.seed(42)
my_matrix(10,10,5)
*/
[,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
[1,] 0 0 0 0 0 0 0 0 0 0
[2,] 0 0 0 0 0 0 0 0 0 0
[3,] 1 1 1 1 1 1 1 1 1 1
[4,] 1 1 1 1 1 1 1 1 1 1
[5,] 2 2 2 2 2 2 2 2 2 2
[6,] 2 2 2 2 2 2 2 2 2 2
[7,] 3 3 3 3 3 3 3 3 3 3
[8,] 3 3 3 3 3 3 3 3 3 3
[9,] 4 4 4 4 4 4 4 4 4 4
[10,] 4 4 4 4 4 4 4 4 4 4
However, the same code does not work as expected if I create a package:
> rcpphello::my_matrix(10,10,5)
[,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
[1,] 0 0 0 0 0 0 0 0 0 0
[2,] 0 0 0 0 0 0 0 0 0 0
[3,] 0 0 0 0 0 0 0 0 0 0
[4,] 0 0 0 0 0 0 0 0 0 0
[5,] 0 0 0 0 0 0 0 0 0 0
[6,] 0 0 0 0 0 0 0 0 0 0
[7,] 0 0 0 0 0 0 0 0 0 0
[8,] 0 0 0 0 0 0 0 0 0 0
[9,] 0 0 0 0 0 0 0 0 0 0
[10,] 0 0 0 0 0 0 0 0 0 0
Why is the same code only using one thread if I call it from within my package? In case it helps, I pushed all the code to this github repo

Answer 1

Add to src/Makevars and src/Makevars.win:
PKG_CXXFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS)
This enables the -fopenmp flag. Otherwise, you will not end up having OpenMP enabled in your package.
Note: When using:
// [[Rcpp::plugins(openmp)]]
This sets the -fopenmp parameter only when run with sourceCpp(). This option does not transfer into a package. Hence, we must establish the setting in Makevars and Makevars.win.
A short example can be found here:
https://github.com/r-pkg-examples/rcpp-and-openmp
Though, I'll need to clean it up a bit.

Answer 2

#coatless answered the question already. I would like to add one caveat: Don't use data structures from R or Rcpp within parallel code. You can use RcppParallel, though:
#include <Rcpp.h>
// [[Rcpp::plugins(openmp)]]
#include <omp.h>
// [[Rcpp::depends(RcppParallel)]]
#include <RcppParallel.h>
using namespace Rcpp;
// [[Rcpp::export]]
NumericMatrix my_matrix(int I, int J, int nthreads) {
NumericMatrix A(I,J);
// create a thread safe accessor for A
RcppParallel::RMatrix<double> a(A);
int tid;
omp_set_num_threads(nthreads);
#pragma omp parallel for private(tid)
for(int j = 0; j < J; j++) {
for(int i = 0; i < I; i++) {
tid = omp_get_thread_num();
a(i, j) = tid ;
}
}
return A;
}
/*** R
set.seed(42)
my_matrix(12,10,5)
*/
Note that I have also changed the access to column major and removed the additional declaration of i and j. Note that variables declared inside of a parallel section are automatically private.
And in case you want to use R's RNG (since you are setting the seed), there is another "don't do that". Have a look at packages like sitmo or dqrng for RNGs that can be used with parallel code.