Boost pool for unordered set of custom objects

Question

I'm having a hard time finding an example for this. My code looks like this:

typedef boost::unordered_set<CustomObject, boost::hash<CustomObject>, 
  CustomObjectEqual, allocator<CustomObject> > CustomObjectSet;

I've tried to use fast_pool_allocator directly but this leads to a compiler error (using std::allocator works). My questions are:

1. Do I need to create a custom allocator for CustomObject?
2. Does this increase the speed of my program?

Answer 1

Question 1: Do I need to create a custom allocator for CustomObject?

No. It will compile without it. The allocator uses default parameters. In an example below are these the same:

using fast_allocator = boost::fast_pool_allocator<
    CustomObject,
    boost::default_user_allocator_new_delete,
    boost::mutex,
    32,
    0>;

using fast_allocator = boost::fast_pool_allocator<CustomObject>;

Example

#include <boost/unordered_set.hpp>
#include <boost/pool/pool.hpp>
#include <boost/pool/pool_alloc.hpp>

struct CustomObject {
    CustomObject(std::size_t value)
        : value(value)
    {
    }

    std::size_t value;
};

struct CustomObjectKeyEq {
    bool operator()(CustomObject const& l, CustomObject const& r) const 
    {
        return l.value == r.value;
    }
};

std::size_t hash_value(CustomObject const& value)
{
    return value.value;
}

int main()
{
    typedef boost::unordered_set<CustomObject,
                                 boost::hash<CustomObject>,
                                 CustomObjectKeyEq> StandardObjectSet;

    StandardObjectSet set1;
    set1.insert(10);
    set1.insert(20);
    set1.insert(30);

    using fast_allocator = boost::fast_pool_allocator<CustomObject>;
    typedef boost::unordered_set<CustomObject,
                                 boost::hash<CustomObject>,
                                 CustomObjectKeyEq,
                                 fast_allocator> CustomObjectSet;

    CustomObjectSet set2;
    set2.insert(10);
    set2.insert(20);
    set2.insert(30);

    return 0;
}

Question 2: Does this increase the speed of my program?

Generally, you must measure it. It has no significant impact on example above. Inserting one million object into set1 took:

0.588423s wall, 0.570000s user + 0.020000s system = 0.590000s CPU (100.3%)

and into set2:

0.584661s wall, 0.560000s user + 0.010000s system = 0.570000s CPU (97.5%) 1000000