straggler thread in new_parallel?

[crclark]

I gave the no key hash map new_parallel function a vec of about 41 billion u64s. I left it unattended for several hours on a machine with 100 cores and came back to find one thread still busy. perf record shows all time is being spent in boomphf::Mphf<T>::hash and boomphf::hashmap::NoKeyBoomHashMap<K,D1>::new_parallel.

I haven't looked at the code closely but I am guessing it may be a straggler thread in one of the two par_chunks loops here.

Any possible mitigations I can take to avoid this?