[julia-users] Weird performance difference; different ways to instantiate an immutable object

[Carlos Baptista]

I have a module consisting out of several files. In one file I define two 
immutables and a union:

immutable GlobalRBF
    eval::Function
end


immutable CompactRBF
    eval::Function
end

const global RBF = Union(GlobalRBF, CompactRBF)

and I define several immutable objects of which one is:

const global rbf_cp2 = CompactRBF(x -> (1.0 - x)^4*(4.0*x + 1.0))

In another file I have a function in which the above univariate scalar 
function is evaluated like this:

function myFunc(rbf::RBF)
    # Some stuff

    nested_for_loop
        rbf.eval(r[i, j])
    end

    # Some stuff

end

If I then run my code and use @time to measure the performance I get the 
following result:

elapsed time: 0.002800011 seconds (934720 bytes allocated)

However if I define the immutable object slightly differently like this:

f(x) = (1.0 - x)^4*(4.0*x + 1.0)
const global rbf_cp2 = CompactRBF(f)

then with @time I measure:

elapsed time: 0.001887599 seconds (606912 bytes allocated)

So, I would like to know what the big difference is between the two 
implementations of the object that would cause such a difference in 
performance?