Why does Julia allocate memory when I already pre-allocated?

General Usage Performance
1

Look at the code here.

julia> results = Vector{Int}(undef, 10^7);

julia> @time for i in 1:10^7
           results[i] = i
       end
  0.315568 seconds (20.00 M allocations: 305.160 MiB, 5.86% gc time)

Why is there memory allocation and GC run, when the vector was allocated beforehand?

2

results is global variable.

3

Do you mean that when an array is declared in global scope, I actually get an array of pointers to heap-allocated objects?

4

More likely there’s a bit of boxing and unboxing going on in each iteration of the loop while executing an unspecialized setindex!.

If you place the loop in a function with results as an argument or make results a const global you get more efficient code.

2 Likes
5

Thanks for the tips. I fixed the problem. I just had to add const before declaring the array results, and the memory allocation disappeared. I was aware that non-const globals are bad, but didn’t realize it would affect code as simple as this.

6

The simplicity of code matters not. If variable is global it allocates, so you should either use const or wrap in let block or wrap in a function. You can read more here: Performance Tips · The Julia Language

1 Like