y u no work, RcppParallel?

yunoparallel.cpp

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::depends(RcppParallel)]]
#include <RcppParallel.h>
using namespace RcppParallel;

// accumulate # of elements seen on the input iterator.
void accum_nel_int(RVector<double>::const_iterator vecit, 
                   RVector<double>::const_iterator end, 
                   RVector<int>::iterator xret) {
    for (;vecit != end;++vecit) { *xret += 1; }
}
void join_em_int(RVector<int> lhs, const RVector<int> rhs) {
    lhs[0] += rhs[0];
}

struct NEL_int : public Worker
{   
   // source vector
   const RVector<double> input;
   // accumulated value
   RVector<int> output;
   
   // constructors
   NEL_int(const NumericVector input, IntegerVector output) : input(input), output(output) {}
   NEL_int(const NEL_int& othr, Split) : input(othr.input), output(othr.output) {}
   
   // accumulate just the element of the range I've been asked to
   void operator()(std::size_t begin, std::size_t end) {
       accum_nel_int(input.begin() + begin, input.begin() + end, output.begin());
   }
     
   // join my value with that of another NEL_int
   void join(const NEL_int& rhs) { 
       join_em_int(output,rhs.output);
   }
};

// dumb example counting # of elements on input:
// [[Rcpp::export]]
IntegerVector dumbCount_int(NumericVector x) {
   // allocate the output
   IntegerVector output(1);
   // declare the instance
   NEL_int wrkr(x,output);
   // call parallel_reduce to start the work
   parallelReduce(0, x.length(), wrkr, 5000);
   // return the computed number of non-na elements
   return output;
}

// in R do:
// library(Rcpp)
// source('yunoparallel.cpp')
// x <- rnorm(1e5)
// dumbCount_int(x)
// dumbCount_int(x)
// # fine, right? no.
// replicate(10,dumbCount_int(x))
// sd(replicate(100,dumbCount_int(x)))

//for vim modeline: (do not edit)
// vim:ts=4:sw=4:tw=79:fdm=marker:fmr=FOLDUP,UNFOLD:cms=//%s:syn=cpp:ft=cpp:ai:si:et:cin:nu:fo=croql:cino=p0t0c5(0:

yunoparallel.R

library(Rcpp)
source('yunoparallel.cpp')
x <- rnorm(1e5)
dumbCount_int(x)
dumbCount_int(x)
# fine, right? no.
replicate(10,dumbCount_int(x))
sd(replicate(100,dumbCount_int(x)))