Int's vs. BigInt's

costly_bigints.jl

function foo(n::Integer)
    x::Int = 0
    t1::Float64 = @elapsed for i in 1:n
        x = x + 1
    end
    @printf "Int: %f\n" t1

    y::BigInt = BigInt(0)
    t2::Float64 = @elapsed for i in 1:n
        y = y + BigInt(1)
    end
    @printf "BigInt: %f\n" t2
end

foo(100_000_000)

@erasmospunk, Mark Roddy showed that xrange and sum instead of a loop gives you a big speedup, but still not the full Julia speed. (In fairness, beating Julia here is close to impossible: the machine code for the loop is about as tight as you can get.)

@shabbychef, You're right that the BigInt constructor is slowing this loop down a lot. I wrote things that way for simplicity, but this example shows that the Julia compiler is still not smart enough to fold many constants out during optimization. At the same time, I don't understand how the R and Python interpreters actually process these lines very well, so I wanted things to be as close to parallel as I knew how to make them.