Julia's type inference is where you get performance

type_inference.jl

# Always ignore one run

x = 0

@elapsed for i in 1:100_000_000
	x += 1
end # => 3.330939854

# Second time is more accurate

x = 0

@elapsed for i in 1:100_000_000
	x += 1
end # => 3.287221881

function foo()
	x = 0

	t = @elapsed for i in 1:100_000_000
		x += 1
	end

	println(t)
end

foo() # => 0.028839841

foo() # => 0.028527174

Running identical code in a function is 100x faster. Why? Because the compiler can determine that the type of x is invariant, which isn't guaranteed in the global namespace.

Note, for example, that this code gets compiled to code that's nearly maximally efficient:

julia> function bar()
       x = 0
       for i in 1:100_000_000
       x += 1
       end
       return x
       end
# methods for generic function bar
bar() at none:2

julia> code_native(bar, ())
    .section    __TEXT,__text,regular,pure_instructions
Filename: none
Source line: 3
    push    RBP
    mov RBP, RSP
    mov EAX, 100000000
Source line: 3
    dec RAX
    jne -9
    mov EAX, 100000000
Source line: 6
    pop RBP
    ret