uuids::to_string and hash containers performances

[rickyviking]

Hi,
using uuids as keys in stl unordered containers I have noticed a certain performance impact, both in filling and searching. I've come up with the conclusion that this is due to the hash function, which converts the uuids to their string representation and then computes the hash.
Here are some benchmarks I've run for 1'000'000 entries (on i7-7820X 3.6GHz):

operation	millisecs
create uuids	85
to_string()	5183
fill std::set	771
fills std::unordered_set	11003
search std::set	942
search std::unordered_set	5455

As the numbers show, it's much faster both to fill and to search standard containers rather than hash ones, which is counter-intuitive (especially for the search part).

1. is this an expected/known behavior?
2. would it be possible for find a more efficient hashing algorithm?

[UPDATE] I've done a quick test replacing the hash algorithm with the simple one used here: updated results below are more in line with what you would expect from a hash container. So is there a specific design implementation behind the string version? Thanks!

operation	millisecs
create uuids	84
to_string()	5134
fill std::set	783
fills std::unordered_set	336
search std::set	924
search std::unordered_set	261

[mariusbancila]

A faster implementation of std::hash<uuid> is now available.

Although probably now needed, the previous one is still available if you define the UUID_HASH_STRING_BASED macro.

[rickyviking]

Great, benchmarks with your new implementation are aligned to the ones I reported above.