How to profile and optimise already fast code in R

Question

I have a function in R which I want to optimize. It is part of the innermost loop so it runs millions of times and I know from profiling that this function takes up about 80% of the total computation time.

I have been using the profvis package to understand where my code is slow and by making incremental improvements a single call of the function now takes less than 10 ms to run.

But at this point, profvis stops working and does not give a useful breakdown of which lines of code are using the most time. For example:

func <- function(x){
  x1 <- x ** 2
  x2 <- x * 3
  x3 <- sum(1:x)
  x4 <- x1 + x2 + x3
}
profvis::profvis(func(10))
Error in parse_rprof(prof_output, expr_source) : 
  No parsing data available. Maybe your function was too fast?

Are their alternative packages or methods that work well to profile functions that take less than 10ms to run?

Answer 1

profvis uses Rprof , so you may want to read about its limits at ?Rprof .

If all else fails, you can run small-scale experiments: change something in your function (eg reduce the number of assignments), and then measure the total running time of a loop in which the function is called. This is somewhat complicated by your compiler settings (see ?compiler::compile ). But you have the ultimate way to see if some change helped: run your code (many times), and see if it has become faster.

func <- function(x){
  x1 <- x ** 2
  x2 <- x * 3
  x3 <- sum(1:x)
  x4 <- x1 + x2 + x3
}
func2 <- function(x)
  x*x + x*3 + sum(seq_len(x)) 


library("compiler")
ii <- 1:1000000
enableJIT(0) 
system.time(for (i in ii) func (100))
system.time(for (i in ii) func2(100))

enableJIT(3) 
system.time(for (i in ii) func (100))
system.time(for (i in ii) func2(100))