Hash an arbitrary precision value (boost::multiprecision::cpp_int)

I need to get the hash of a value with arbitrary precision (from Boost.Multiprecision); I use the cpp_int backend. I came up with the following code:

boost::multiprecision::cpp_int x0 = 1;
const auto seed = std::hash<std::string>{}(x0.str());

I don't need the code to be as fast as possible, but I find it very clumsy to hash the string representation.

So my question is twofold:

• Keeping the arbitrary precision, can I hash the value more efficiently?
• Maybe I should not insisting on keeping the arbitrary precision and I should convert to a double which I could hash easily (I would still however make the comparison needed for the hash table using the arbitrary precision value)?

Solution 1:

You can (ab)use the serialization support:

Support for serialization comes in two forms: Classes number, debug_adaptor, logged_adaptor and rational_adaptor have "pass through" serialization support which requires the underlying backend to be serializable.

Backends cpp_int, cpp_bin_float, cpp_dec_float and float128 have full support for Boost.Serialization.

So, let me cobble something together that works with boost and std unordered containers:

template <typename Map>
void test(Map const& map) {
    std::cout << "\n" << __PRETTY_FUNCTION__ << "\n";
    for(auto& p : map)
        std::cout << p.second << "\t" << p.first << "\n";
}

int main() {
    using boost::multiprecision::cpp_int;

    test(std::unordered_map<cpp_int, std::string> {
        { cpp_int(1) << 111, "one"   },
        { cpp_int(2) << 222, "two"   },
        { cpp_int(3) << 333, "three" },
    });

    test(boost::unordered_map<cpp_int, std::string> {
        { cpp_int(1) << 111, "one"   },
        { cpp_int(2) << 222, "two"   },
        { cpp_int(3) << 333, "three" },
    });
}

Let's forward the relevant hash<> implementations to our own hash_impl specialization that uses Multiprecision and Serialization:

namespace std {
    template <typename backend> 
    struct hash<boost::multiprecision::number<backend> > 
        : mp_hashing::hash_impl<boost::multiprecision::number<backend> > 
    {};
}

namespace boost {
    template <typename backend> 
    struct hash<multiprecision::number<backend> > 
        : mp_hashing::hash_impl<multiprecision::number<backend> > 
    {};
}

Now, of course, this begs the question, how is hash_impl implemented?

template <typename T> struct hash_impl {
    size_t operator()(T const& v) const {
        using namespace boost;
        size_t seed = 0;
        {
            iostreams::stream<hash_sink> os(seed);
            archive::binary_oarchive oa(os, archive::no_header | archive::no_codecvt);
            oa << v;
        }
        return seed;
    }
};

This looks pretty simple. That's because Boost is awesome, and writing a hash_sink device for use with Boost Iostreams is just the following straightforward exercise:

namespace io = boost::iostreams;

struct hash_sink {
    hash_sink(size_t& seed_ref) : _ptr(&seed_ref) {}

    typedef char         char_type;
    typedef io::sink_tag category;

    std::streamsize write(const char* s, std::streamsize n) {
        boost::hash_combine(*_ptr, boost::hash_range(s, s+n));
        return n;
    }
  private:
    size_t* _ptr;
};

Full Demo:

Live On Coliru

#include <iostream>
#include <iomanip>

#include <boost/archive/binary_oarchive.hpp>
#include <boost/multiprecision/cpp_int.hpp>
#include <boost/multiprecision/cpp_int/serialize.hpp>
#include <boost/iostreams/device/back_inserter.hpp>
#include <boost/iostreams/stream_buffer.hpp>
#include <boost/iostreams/stream.hpp>

#include <boost/functional/hash.hpp>

namespace mp_hashing {
    namespace io = boost::iostreams;

    struct hash_sink {
        hash_sink(size_t& seed_ref) : _ptr(&seed_ref) {}

        typedef char         char_type;
        typedef io::sink_tag category;

        std::streamsize write(const char* s, std::streamsize n) {
            boost::hash_combine(*_ptr, boost::hash_range(s, s+n));
            return n;
        }
      private:
        size_t* _ptr;
    };

    template <typename T> struct hash_impl {
        size_t operator()(T const& v) const {
            using namespace boost;
            size_t seed = 0;
            {
                iostreams::stream<hash_sink> os(seed);
                archive::binary_oarchive oa(os, archive::no_header | archive::no_codecvt);
                oa << v;
            }
            return seed;
        }
    };
}

#include <unordered_map>
#include <boost/unordered_map.hpp>

namespace std {
    template <typename backend> 
    struct hash<boost::multiprecision::number<backend> > 
        : mp_hashing::hash_impl<boost::multiprecision::number<backend> > 
    {};
}

namespace boost {
    template <typename backend> 
    struct hash<multiprecision::number<backend> > 
        : mp_hashing::hash_impl<multiprecision::number<backend> > 
    {};
}

template <typename Map>
void test(Map const& map) {
    std::cout << "\n" << __PRETTY_FUNCTION__ << "\n";
    for(auto& p : map)
        std::cout << p.second << "\t" << p.first << "\n";
}

int main() {
    using boost::multiprecision::cpp_int;

    test(std::unordered_map<cpp_int, std::string> {
        { cpp_int(1) << 111, "one"   },
        { cpp_int(2) << 222, "two"   },
        { cpp_int(3) << 333, "three" },
    });

    test(boost::unordered_map<cpp_int, std::string> {
        { cpp_int(1) << 111, "one"   },
        { cpp_int(2) << 222, "two"   },
        { cpp_int(3) << 333, "three" },
    });
}

Prints

void test(const Map&) [with Map = std::unordered_map<boost::multiprecision::number<boost::multiprecision::backends::cpp_int_backend<> >, std::basic_string<char> >]
one 2596148429267413814265248164610048
three   52494017394792286184940053450822912768476066341437098474218494553838871980785022157364316248553291776
two 13479973333575319897333507543509815336818572211270286240551805124608

void test(const Map&) [with Map = boost::unordered::unordered_map<boost::multiprecision::number<boost::multiprecision::backends::cpp_int_backend<> >, std::basic_string<char> >]
three   52494017394792286184940053450822912768476066341437098474218494553838871980785022157364316248553291776
two 13479973333575319897333507543509815336818572211270286240551805124608
one 2596148429267413814265248164610048

As you can see, the difference in implementation between Boost's and the standard library's unordered_map show up in the different orderings for identical hashes.

Solution 2:

Just to say that I've just added native hashing support (for Boost.Hash and std::hash) to git develop. It works for all the number types including those from GMP etc. Unfortunately that code won't be released until Boost-1.62 now.

The answer above that (ab)uses serialization support, is actually extremely cool and really rather clever ;) However, it wouldn't work if you wanted to use a vector-based hasher like CityHash, I added an example of using that by accessing the limbs directly to the docs: https://htmlpreview.github.io/?https://github.com/boostorg/multiprecision/blob/develop/doc/html/boost_multiprecision/tut/hash.html Either direct limb-access or the serialization tip will work with all previous releases of course.