phash2 isn't very resistant to collisions?

[Roger Lipscombe]

I'm sure this has come up previously, but all I could find talked about atoms.

I know it's not intended to be cryptographically strong, but
erlang:phash2 is woefully collidey with 2-tuples:

Test = fun(N) ->
    L = [{X, Y} || X <- lists:seq(0, N - 1), Y <- lists:seq(0, N - 1)],
    Expected = N * N,
    Expected = length(L),
    Expected = length(lists:usort([erlang:phash2(Term) || Term <- L]))
end.

1> Test(10).
100

2> Test(100).
** exception error: no match of right hand side value 9999

3> Test(99).
** exception error: no match of right hand side value 9800

4> Test(500).
** exception error: no match of right hand side value 249763

What alternatives are there to phash2, if I still care about performance?