In order to stress C2 components that rely on profile information gathered by the interpreter and C1 it would be useful to randomize such information. This can be particularly useful with CTW test and when using RepeatCompilation. It is also likely to help reproducing intermittent bugs that are highly dependent on profile data (e.g. JDK-8355354 or JDK-8342772).
Obvious randomization candidates are method invocation and loop backedge counters.