Reduce the allocations in the for loop

General Usage
,
1

With the increase of Num, the allocations become larger and larger, can someone help me to optimize this program, here is an easy example
ps: in my program, Num is quite larger, ~100000

function test(x,y)
    c=rand(2).*[pi, 2pi]
    x+=c[1]
    y+=c[2]
    return x,y
end

Num=100
a=zeros(Num)
b=ones(Num)

@time begin

@inbounds for ii in 1:Num-1
    global a[ii+1],b[ii+1]=test(a[ii],b[ii])
end

end
2

To optimize:
First, wrap the thing in a function. Accessing a global variable is slowwwwwwww.
Second, set the an and b to undef and a[1] and b[1] to 0 and 1 respectively.
Finally, avoid using an array, you can use a pair of variables instead.

3

Read: Performance Tips · The Julia Language

And probably: Common allocation mistakes

1 Like
4

rand(2) will allocate a length=2 array, as will [pi,2pi], and then the product will allocate a third.

Consider using StaticArrays when you need small, fixed size arrays and you want them to be fast. With SVector, this could be non-allocating.

But in this specific case, it’s simple enough to avoid any array at all (although a SVector should be equally fast). For example, something like with test(x,y) = (x + rand()*pi, y + rand()*2pi).

Timing the loop in global scope will lead to many allocations. See the link to the performance tips from a previous post.

4 Likes