A function is said to have a cycle of length n if there exists an x in its domain such that fn(x) = x and fm(x) ≠ x for 0 < m < n, where the superscript n denotes n-fold application of f. Note that a cycle of length 1 is a fixed point f(x) = x.
Your task is to implement a bijective function from the integers to themselves, which has exactly one cycle of every positive length n. A bijective function is a one-to-one correspondence, i.e. every integer mapped to exactly once. Having exactly one cycle of length n means that there are exactly n distinct numbers x for which fn(x) = x and fm(x) ≠ x for 0 < m < n.
Here is an example of what such a function might look like around x = 0:
x ... -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 ...
f(x) ... 2 4 6 -3 -1 1 -4 0 -2 5 7 -7 -6 3 -5 ...
This excerpt contains cycles of length 1 to 5:
n cycle
1 0
2 -2 1
3 -4 -3 -1
4 -5 6 3 7
5 -7 2 5 -6 4
...
Note that above I'm using "function" only in the mathematical sense. You may write either a function or a full program in your language of choice, as long as it takes a single (signed) integer as input and returns a single (signed) integer. As usual you may take input via STDIN, command-line argument, function argument etc. and output via STDOUT, function return value or function (out) argument etc.
Of course, many languages don't (easily) support arbitrary-precision integers. It's fine if your implementation only works on the range of your language's native integer type, as long as that covers at least the range [-127, 127] and that it would work for arbitrary integers if the language's integer type was replaced with arbitrary-precision integers.
Standard code-golf rules apply.