Custom hash in std::unordered map in C++

Question

I want to meet some performance requirements in my C++ project. I have found that using push_back to an std::vector has a performance penalty. Also when I use find for the elements of an std::vector it also takes more time than using an unordered_map .

Is it OK if I use emplace on an unordered_map ? Can the save/get cycle be done faster?

g++ driver.cpp -std=c++11

The header file is

#ifndef CUSTOM_UNOR_MAP_HPP
#define CUSTOM_UNOR_MAP_HPP


#include <boost/algorithm/string/predicate.hpp>
#include <boost/functional/hash.hpp>
#include <unordered_map>


namespace pe
{

    struct pe_hash
    {

        size_t operator()(const std::string& key) const
        {

            std::size_t seed = 0;
            std::locale locale;

            for(auto c : key)
            {
                boost::hash_combine(seed, std::toupper(c, locale));
            }

            return seed;

        }
    };



    struct pe_key_eq
    {

        bool operator()(const std::string& l, const std::string& r) const
        {
            return boost::iequals(l, r);
        }
    };

    using pe_map = std::unordered_map<std::string, std::string, pe_hash, pe_key_eq>;

}

#endif

The driver file (main.cpp) is

#include "pemap.hpp"
#include <iostream>


template <typename T>

    inline const std::string& get_map_value(const T& container, const std::string& key)

    {

        if (container.count(key))
        {

            return container.find(key)->second;

        }

        static std::string empty;
        return empty;

    }

int main(int argc, char** argv) {


    std::string key = "key";
    std::string val = "value1";
    std::string val2 = "value2";

    pe::pe_map container;
    container.emplace(std::move(key), std::move(val));
    container.emplace(std::move(key), std::move(val2));

    std::cout << get_map_value( container, "key") << std::endl;

}

Answer 1

I want to meet some performance requirements in my C++ project.

Here are the steps you should do:

• run profiler
• identify which code and data manipulation takes the most time
• think about better algorithm and/or data organization for your case (container type is one of them)
• if performance now is enough, you are done, if not try to optimize code, that takes most time

You would not get a good answer just to look at a container in vanilla example. You need to optimize your program, not example.