RcppParallel with RcppArmadillo

## installation
library(devtools)
install_github("RcppCore/RcppParallel")

## code
library(Rcpp)
library(RcppArmadillo)
library(RcppParallel)
sourceCpp(code = '
// [[Rcpp::depends(RcppArmadillo)]]
// [[Rcpp::depends(RcppParallel)]]
#define ARMA_DONT_USE_CXX11
#include <RcppArmadillo.h>
#include <RcppParallel.h>
using namespace Rcpp;
using namespace arma;
using namespace RcppParallel;

// [[Rcpp::export]]
vec times_two_f1(NumericVector xr) {
  vec x(xr.begin(), xr.size()), y(xr.size());
  for(uword i = 0; i < x.n_elem; i++) y(i) = x(i)*2;
  return y;
}

struct TimesTwo: public Worker {
  vec& input;
  vec& output;
  TimesTwo(vec& input, vec& output) : input(input), output(output) {}
  void operator()(std::size_t begin, std::size_t end) {
    for(std::size_t i = begin; i < end; i++) output(i) = input(i)*2;
  }
};

// [[Rcpp::export]]
vec times_two_f2(NumericVector xr) {
  vec x(xr.begin(), xr.size()), y(xr.size());
  TimesTwo timesTwo(x, y);
  parallelFor(0, x.n_elem, timesTwo);
  return y;
}

// [[Rcpp::export]]
vec times_two_f3(NumericVector xr) {
  vec x(xr.begin(), xr.size());
  return x*2;
}')

library(rbenchmark)
x = rnorm(3e7)
benchmark(cpp = times_two_f1(x),
    cppParallel = times_two_f2(x),
    Armadillo = times_two_f3(x),
    columns=c("test", "replications","elapsed", "relative"),
    replications=20, order="relative")
#          test replications elapsed relative
# 2 cppParallel           20    4.06    1.000
# 3   Armadillo           20    4.39    1.081
# 1         cpp           20    4.70    1.158