Tbb based row sum of an armadillo matrix

Following the tbb example for parallel_reduce here with a basic datatypes, I wanted to try and implement a tbb version of a row sum of an Armadillo matrix. (I realize the use of arma::sum for this purpose, but this was an attempt at getting to know tbb, Armadillo and how they work together a bit better. )

The following code was aimed to loop over columns of a matrix X, and add the column values to a running total vector. It compiles fine but crashes at run time. I am wondering where I am going wrong. Is the X.col(j) use is incorrect? Thank you!

arma::vec parallelRowsumTBB_withArma(arma::mat X){

   arma::vec y=tbb::parallel_reduce(
      tbb::blocked_range<size_t>(0,X.n_cols),
      arma::vec(X.n_rows).fill(0),
      [&](const tbb::blocked_range<size_t>& r, arma::vec runningTotal) { 
         for(int j=r.begin(); j!=r.end(); ++j){
            runningTotal = runningTotal + X.col(j);
         }
         return runningTotal;
      },
      [](arma::vec a, arma::vec b){
         return a + b;
      }
   );
   return y;
}

That is a 'frequently asked question': you cannot have any parallel code have any callback or interaction with the main R process which is not safe for such callback. And Armadillo data structures provided by RcppArmadillo reuse (by default) the R memory so you can potentially have interactions with R's processs via the gc().

See the RcppParallel package (with relies on TBB) and its documentation for an alternative: RMatrix and RVector. Or ensure your RcppArmadillo objects are in fact distinct copies.