acl

Animation Compression Library

Animation compression is a fundamental aspect of modern video game engines. Not only is it important to keep the memory footprint down but it is also critical to keep the animation clip sampling performance fast.

The more memory an animation clip consumes, the slower it will be to sample it and extract a character pose at runtime. For these reasons, any game that attempts to push the boundaries of what the hardware can achieve will at some point need to implement some form of animation compression.

While some degree of compression can easily be achieved with simple tricks, achieving high compression ratios, fast decompression, while simultaneously not compromising the accuracy of the resulting compressed animation requires a great deal of care.

Goals

This library has four primary goals:

• Implement state of the art and production ready animation compression algorithms
• Be easy to integrate into modern video game engines
• Serve as a benchmark to compare various techniques against one another
• Document what works and doesn't work

Algorithms are optimized with a focus on (in this particular order):

• Minimizing the compression artifacts in order to reach high cinematographic quality
• Fast decompression on all our supported hardware
• A small memory footprint to lower memory pressure at runtime as well as reducing disk and network usage

Decompression speed will not be sacrificed for a smaller memory footprint nor will accuracy be compromised under any circumstances.

Philosophy

Much thought was put into designing the library for it to be as flexible and powerful as possible. To this end, the following decisions were made:

• The library consists of 100% C++11 header files and is thus easy to integrate in any game engine
• An intermediary clip format is supported in order to facilitate debugging and bug reporting
• All allocations use a game provided allocator
• All asserts use a game provided macro

Supported platforms

Continuous integration tests a variety of platforms and configurations but it generally runs as-is anywhere where C++11 (or later) is supported. CI currently tests:

• Windows VS2022: x86, x64, ARM64, ARM64EC
• Windows VS2026: x86, x64, ARM64, ARM64EC
• Linux GCC 14+: x86, x64, ARMv7, ARM64
• Linux Clang 22+: x86, x64, ARMv7, ARM64
• OS X XCode 16+: ARM64
• Emscripten 6.0.11: WASM (with and without SIMD)

Each releases is also manually tested on iOS and Android.

On RISC-V (riscv64), ACL uses the scalar code path of RTM because RTM has no SIMD code for RISC-V. CI does not test riscv64. make.py -cpu riscv64 supports native builds on a Linux riscv64 host. To cross-compile, use CMake with a RISC-V toolchain file.

The Unreal Engine is supported through a plugin originally found here although it is now distributed with a more recent version as part of each engine release since UE 5.13.

Getting started

This library is 100% headers as such you just need to include them in your own project to start using it. However, if you wish to run the unit tests, regression tests, to contribute to ACL or use it for research, head on over to the getting started section in order to setup your environment and make sure to check out the contributing guidelines.

If you would like to integrate ACL into your own game engine, follow the integration instructions here.

You can install nfrechette-acl with Conan.

Performance metrics

• Carnegie-Mellon University database performance
• Paragon database performance
• Matinee fight scene performance
• Decompression performance

External dependencies

You don't need anything else to get started: everything is self contained. See here for details.

License, copyright, and code of conduct

This project uses the MIT license.

Copyright (c) 2017 Nicholas Frechette & Animation Compression Library contributors

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.

Contributors ✨

Thanks goes to these wonderful people (emoji key):

This project follows the all-contributors specification. Contributions of any kind welcome!