This is a C++11 implementation of an object pool allocator.
For more information on object pool allocators and their purpose see http://gameprogrammingpatterns.com/object-pool.html.
Both a fixed size pool (FixedObjectPool
) and a dynamically growing pool
(DynamicObjectPool
) implementation are included.
The main features of this implementation are:
new_object
method uses C++11 std::forward to pass construction arguments to the constructor of the new object being created in the poolfor_each
method will iterate over all live objects in the pool calling the given function on themdelete_all
method will free all pool objects at once, skipping the destructor call for trivial types- maintains a freelist of next available pool entry for fast allocation
These object pool classes are not designed with exceptions in mind as most game code avoids using exceptions.
// some type to be pooled
struct Enemy {
/* some data */
Enemy(const char* name);
void update(double delta_time);
};
// created fixed size pool with space for 64 enemies
FixedObjectPool<Enemy> enemy_pool(64);
// allocate an enemy - construction parameters are forwarded
Enemy* baddie = enemy_pool.new_object("The Mekon");
// update all live enemies by executing the lambda on all allocated objects
enemy_pool.for_each([delta_time](Enemy* enemy)
{
enemy->update(delta_time);
});
// delete a single enemy
enemy_pool.delete_object(baddie);
// delete all enemies at once (beware of dangling pointers)
enemy_pool.delete_all();
Both FixedObjectPool
and DynamicObjectPool
are implemented using the
ObjectPoolBlock
class.
ObjectPoolBlock
is a single allocation containing the ObjectPoolBlock
instance, indices of used pool entries and the pool memory itself.
Occupancy is tracked using indexes into available entries in the block for
constant time allocation. The ObjectPoolBlock
keeps the next free index head.
This index can be used to find the next available block entry when allocating
a new entry.
A separate list of indices is used to track occupancy versus reusing object pool memory for this purpose to avoid polluting CPU caches with objects which are deleted and thus no longer in use.
Unit tests are written using the Catch
unit testing framework. Unit tests are run through the runtest
executable.
This repository also includes bench.hpp
which is a single header file
micro-benchmarking framework inspired by Catch and Rust's
benchmarking tests.
It's my intention to make this standalone at some point but at the moment it's very much a work in progress.
Currently each micro-benchmark compares the performance of the following:
- Fixed pool
- Dynamic pool with 64, 128 and 256 entry blocks
- The default allocator
Benchmarks output nanoseconds per iteration (lower is better) and megabytes per second throughput (higher is better).
The test and benchmarking applications require CMake to generate build files.
To generate a build, compile and run follow these steps:
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make
./runtests
./runbench
This software is licensed under the zlib license, see the LICENSE file for details.