Example Code
Download

Memory Management

Eric's Memory Management API (EMMA) is a C++ memory management framework developed from the ground-up. The goal of this library is to "make memory management manageable" in C++, removing the mystery beyond handling memory in C++.

Motivation
I was doing some research one weekend, looking for a way to easily handle pooled objects in memory. I needed the ability to handle creation pool alignment seamlessly, with the option to track free and used memory in a variety of ways via different in-memory data structures (lists, trees, etc.). I quickly realized that the C++ standard libraries didn't provide much in terms of an extensible and lightweight memory management framework that could scale to meet the needs of any sized project.

So... like any good C++ developer, I took it as a challenge and decided to write my own.

When I started implementing EMMA, all I really wanted to do was to learn more about memory. As I began to accomplish this goal, I realized that I could keep going, turning what was a small handful of code into a professional memory framework. This gave me the ability to add cool new features, such as support for the popular TLSF algorithm and Doug Lea's dlmalloc allocator.

With the C++0x vote around the corner, I thought it would be fun to add some new features to the library. You'll see some use of auto, nullptr and static_cast throughout the code.

Overview
The key to effective C++ memory management (like many challenges in game development) is to reduce the problem's complexity into something manageable, if not simple. To do this I needed to abstract memory management into the following components:

  • Allocators — acquire and release memory
  • Algorithms — manipulate a subset of memory, responding to malloc and free requests
  • Trackers — track used and free memory
  • Containers — wrap algorithms to facilitate use and enable additional functionality (e.g. thread safety)
  • Managers — keep track of all containers/algorithms and any calls to acquire/release memory


Completed Algorithms
The library comes with the following algorithms already implemented.

  • Two Level Segregate Fit (TLSF) — A general-purpose algorithm that adapts well to applications with high flexibility requirements.
  • Buddy — A general-purpose algorithm that minimizes fragmentation by allocating memory in powers of 2.
  • Fixed-sized Pools — Pools with memory divided up into pre-determined allotments.
  • Dynamically-sized Pools — Pools that adapt to the runtime allocation requirements.
  • Object-based Pools — Pools that store objects and manage their creation and destruction events.
  • Fixed-sized Stacks — Stacks that prohibit individual free requests. Perfect for venue loading/unloading.
  • Dynamically-sized Stacks — Basic LIFO stacks with the ability to free large portions of memory at any time.
  • Permanent Stacks — Immutable stacks designed for permanent use. Perfect for allocations that must persist for the lifetime of the game.
  • Circular Queues — Array-based queues that provide for a pseudo-garbage collection capability.
  • Fixed-sized Ranges — A basic multi-pool algorithm.
  • Delta Ranges — A multi-pool algorithm based on Pascal's Triangle that favors smaller memory requests.
  • Nabla Ranges — A multi-pool algorithm based on Pascal's Triangle that favors larger memory requests.


In-Progress Algorithms
I'm currently adding the following memory algorithms to EMMA.

  • Slab — A general-purpose algorithm that minimizes external fragmentation and page cache misses by pre-allocating contiguous blocks of memory that service multiple requests for the same size.
  • Arena — A general-purpose algorithm that pre-allocates larger contiguous blocks of memory.
  • Metamorph — A custom-designed general-purpose algorithm that minimizes creation overhead.


Source Code
The following example project only contains a subset of EMMA. I'm holding off on releasing the entire library until I've had a chance to finalize the licensing. Please note that EMMA can only be used only for non-commercial purposes at this time.

You will need the following in order to build and test the example project.

  • Visual Studio 2010
  • I've included all the referenced Boost files in the download.
  • EMMA is still a work-in-progress. As a result, I haven't tested a wide range of operating systems and compilers. Therefore, I can only attest to the code working properly under a 32bit version of Windows XP. Theoretically the library should work fine in other environments, including 64 bit and big-endian.


Download

Shipped Titles
Guitar Hero® Warriors of Rock
"Join with friends on an epic rock journey to save rock n’ roll."

Band Hero®
"Featuring the hottest chart-topping hits from everyone's favorite acts."

Guitar Hero® 5
"Living room legends will rock any way they want."

Copyright 2011 - All Rights Reserved