To test our compiler and make sure that Rcpp is installed, execute the following to run and compile a short Rcpp snippet

library("Rcpp")
evalCpp("2 + 2")

## [1] 4

The above code, although seemingly short, is in fact doing a whole lot in the background. This call to evalCpp is taking the passed in string, constructing the necessary boilerplate code for a minimal C++ program, then compiling and executing the resulting program.

It’s also possible to compile a string into a function that is then accessible to you in R

library("Rcpp")

cppFunction("
            bool is_odd_cpp(int num = 10) {
              bool result = (num % 2 == 1);
              return result;
            }
            ")

is_odd_cpp(42L)

## [1] FALSE

One of the primary reasons to use Rcpp is for gains in speed as we’re working at a lower abstraction level. Let’s see how this cpp function compares to the R version

library(microbenchmark)

is_odd_r <- function(num = 10L) {
  result <- (num %% 2L == 1L)
  return(result)
}

results <- microbenchmark(
  is_odd_r   = is_odd_r(12L),
  is_odd_cpp = is_odd_cpp(12L)
)

results

## Unit: nanoseconds
##        expr min   lq     mean median     uq     max neval cld
##    is_odd_r 310  358 22917.39    515  632.5 2230425   100   a
##  is_odd_cpp 964 1209 12875.92   1663 3204.0 1060537   100   a

The C++ version of is_odd is about twice as fast as the R version.

It’s easy to see, though, that this is inline usage of strings is not a scalable way of writing complex C++ code to be called from R. Luckily, Rcpp works very well using the original C++ file format .cpp