#animation #deterministic #fixed-point #api-bindings #synchronization #ozz-animation #skeletal-animation

nightly ozz-animation-rs

A rust runtime library for ozz-animation with cross-platform deterministic

11 unstable releases (4 breaking)

0.10.1 Nov 14, 2024
0.9.2 Aug 15, 2024
0.9.1 Jun 15, 2024
0.8.1 Mar 26, 2024

#329 in Game dev

Download history 118/week @ 2024-09-13 58/week @ 2024-09-20 10/week @ 2024-09-27 2/week @ 2024-10-04 69/week @ 2024-11-08 51/week @ 2024-11-15 4/week @ 2024-11-22 1/week @ 2024-11-29 6/week @ 2024-12-06

131 downloads per month

MPL-2.0 license

480KB
10K SLoC

Release Doc Crate github actions CircleCI

Ozz-animation-rs

Ozz-animation-rs is a rust version skeletal animation library with cross-platform deterministic.

Ozz-animation-rs is based on ozz-animation library, an open source c++ 3d skeletal animation library and toolset. Ozz-animation-rs only implement ozz-animation's runtime part. You should use this library with ozz-animation's toolset.

In order to introduce cross-platform deterministic, ozz-animation-rs does not simply wrap ozz-animation's runtime, but rewrite the full runtime library in rust. So it can be used in network game scenarios, such as lock-step networking synchronize.

Features

The library supports almost all runtime features supported by C++ version ozz, including:

  • Animation playback
  • Joint attachment
  • Animation blending (partial/additive blending)
  • Two bone IK
  • Aim (Look-at) IK
  • User channels
  • Skinning
  • Multi-threading
  • SIMD (SSE2 + NEON)
  • WASM
  • Serialization (rkyv & serde)

The following functions are not supported yet:

  • Baked physic simulation (no plan)
  • All offline features (no plan, use C++ library instead)

Ozz-animation offline features are not supported, and no plans to support. Please use the original C++ library, which has a many tools and plug-ins.

Examples

A simple demo is in ./demo folder. Enter the folder and execute cargo run.

demo

The test cases under ./tests can be viewed as examples.

Ozz-animation-rs keeps the same API styles with original ozz-animation library. Therefore, you can also refer to the ozz-animation examples.

Here is a very sample example:

use glam::Mat4;
use ozz_animation_rs::*;
use std::cell::RefCell;
use std::rc::Rc;

// Load resources
let skeleton = Rc::new(Skeleton::from_path("./resource/playback/skeleton.ozz").unwrap());
let animation = Rc::new(Animation::from_path("./resource/playback/animation.ozz").unwrap());

// Init sample job (Rc style)
let mut sample_job: SamplingJobRc = SamplingJob::default();
sample_job.set_animation(animation.clone());
sample_job.set_context(SamplingContext::new(animation.num_tracks()));
let sample_out = Rc::new(RefCell::new(vec![SoaTransform::default(); skeleton.num_soa_joints()]));
sample_job.set_output(sample_out.clone());

// Init local to model job (Ref style)
let mut l2m_job: LocalToModelJobRef = LocalToModelJob::default();
l2m_job.set_skeleton(&skeleton);
let sample_out_ref = sample_out.borrow();
l2m_job.set_input(sample_out_ref.as_ref());
let mut l2m_out = vec![Mat4::default(); skeleton.num_joints()];
l2m_job.set_output(&mut l2m_out);

// Run the jobs
let ratio = 0.5;

sample_job.set_ratio(ratio);
sample_job.run().unwrap();

l2m_job.run().unwrap();
l2m_out.buf().unwrap(); // Outputs here, are model-space matrices

Toolchain

Since rust simd features are not stable, you need a nightly version rust to compile this library.

Platforms

In theory, ozz-animation-rs supports all platforms supported by rust. But I only tested on the following platforms:

  • Windows/Ubuntu/Mac x64 (Github actions)
  • X64/Arm64 docker (CircleCI)

Maybe you can run cross-platform deterministic test cases under ./tests on your target platform.

Compatibility

With the release of ozz-animation versions, .ozz files and some APIs will also be upgraded. Therefore ozz-animation-rs remains compatible with the corresponding version of ozz-animation, as shown in the following table:

ozz-animation-rs ozz-animation(C++)
0.10.x 0.15.x
0.9.x 0.14.x

Why not fixed-point?

Initially, I tried to implement similar functionality using fixed point numbers. But fixed-point performance is worse, and it is difficult to be compatible with other libraries.

With further research, I found that x64/arm64 platforms now have good support for the IEEE floating point standard. So I reimplemented this library based on f32.

Dependencies

~6.5MB
~176K SLoC