Why is this Rcpp code not faster than pure R?

Question

Setting up

I tried some R and Rcpp benchmarks with Fibonacci numbers.

Pure R

We use the fibR function for the code that was written in pure R.

fibR <- function(n) {
    if (n < 2)
       n
    else
       fib(n - 1) + fib(n - 2)
}

Rcpp

We use another fib function for the code written in Rcpp.

library(Rcpp)

Rcpp::cppFunction('int fib(int x) {
    if((x < 2)) 
       return(x);
    else 
       return(fib(x-1) + fib(x-2));
   }')

---

The results

Here are the results according to several benchmarks tests.

library(rbenchmark)

After that, I did several tests.

benchmark(fib(30), fibR(30))[,1:3]
      test replications elapsed
1  fib(30)          100    0.46
2 fibR(30)          100    0.45

benchmark(fib(31), fibR(31))[,1:3]
      test replications elapsed
1  fib(31)          100    0.77
2 fibR(31)          100    0.80

benchmark(fib(32), fibR(32))[,1:3]
      test replications elapsed
1  fib(32)          100    1.37
2 fibR(32)          100    1.33

benchmark(fib(33), fibR(33))[,1:3]
      test replications elapsed
1  fib(33)          100    1.88
2 fibR(33)          100    2.02

benchmark(fib(34), fibR(34))[,1:3]
      test replications elapsed
1  fib(34)          100    3.09
2 fibR(34)          100    3.26

benchmark(fib(35), fibR(35))[,1:3]
      test replications elapsed
1  fib(35)          100    5.57
2 fibR(35)          100    5.60

benchmark(fib(36), fibR(36))[,1:3]
      test replications elapsed
1  fib(36)          100    8.95
2 fibR(36)          100    8.51

benchmark(fib(37), fibR(37))[,1:3]
      test replications elapsed
1  fib(37)          100   16.94
2 fibR(37)          100   14.47

benchmark(fib(38), fibR(38))[,1:3]
      test replications elapsed
1  fib(38)          100   22.92
2 fibR(38)          100   23.67

benchmark(fib(39), fibR(39))[,1:3]
      test replications elapsed
1  fib(39)          100   35.80
2 fibR(39)          100   38.83

I would expect that the Rcpp code is steadily faster than the equivalent R code, but no – it is sometimes slower!

Why does it work like this, and what am I missing?

Answer 1

When something can't be true you should check closely what is going on :)

Here your R function called your C++ function as you were not careful about where fib() was coming from. A repaired, more explicit version follow.

Code

fibR <- function(n) {
    if (n < 2)
        n
    else
        fibR(n - 1) + fibR(n - 2)
}

library(Rcpp)
Rcpp::cppFunction('int fibRcpp(int x) {
   if ((x < 2))
       return(x);
   else
       return(fibRcpp(x-1) + fibRcpp(x-2));
}')

library(rbenchmark)
benchmark(fibRcpp(30), fibR(30))[,1:3]

Output

         test replications elapsed
2    fibR(30)          100  80.846
1 fibRcpp(30)          100   0.153

This is a rather well-studied example. Look at the directory examples/Misc/ in your Rcpp installation for examples last updated years ago. Also look at prior questions here on SO.