Anoymous Functions Slow in sort

Is it expected that an anonymous function is 300x slower than a named function in sort?

For example:

test = rand(1000)
my_id(x) = x
sort(test, by = x -> x)   # 0.074194 sec
sort(test, by = identity) # 0.000070 sec
sort(test, by = my_id)   # 0.000092 sec

In some code I am using have had moved to using named functions for this because of the crazy drop in speed. I thought anonymous functions where equivalent to named functions for speed these days. Am I doing something silly?

Thanks so much.

1 Like

Looks like you are counting compile time. Consider using BenchmarkTools.

julia> @btime sort($test, by=$(x->x));
  39.904 μs (2 allocations: 7.95 KiB)

julia> @btime sort($test, by=$my_id);
  41.423 μs (2 allocations: 7.95 KiB)

julia> @btime sort($test, by=$identity);
  15.551 μs (2 allocations: 7.95 KiB)

Not sure why identity is so much faster, that seems unfortunate.


Sweet! That makes a tonne of sense. Now I need to find out why my full function is slow as clearly it is not this!

1 Like

Identity sorting chooses the algorithm based on element type, and can use quick sort for numbers. I think that non-identity sorting uses merge sort instead to guarantee stability. You can choose a different sorting algorithm explicitly though.