"Need voxels? Reach for Voxelis." Powered by VoxTree — a deliciously crafted SVO DAG with batching. Drop it into your Rust, C++, Godot, or Bevy project and start carving worlds down to centimetre-level detail — while your memory bill stays shockingly low.
- Tiny voxels (4 cm resolution) without melting your RAM.
- Shared memory — DAG compression at 99.999% ratio.
- Batch edits — mutate hundreds of thousands of voxels in a blink.
- Zero garbage collection — just deterministic reference counting with generations.
- Built for gamedev — chunk grids, paging hooks, multithread-ready.
- Fearless Rust — no UB, no data races, only pure, undiluted speed.
This isn't just another voxel crate. It's a foundation for colossal, high-fidelity worlds.
| Operation | 32³ Voxels | Single Op | Batch Op | Notes |
|---|---|---|---|---|
| fill() | 32K | 9 ns | 10.6 ns | ⚡ Single leaf collapse |
| set_uniform() | 32K | 5.17 ms | 23.1 μs | 🚀 ~224× faster |
| set_checkerboard() | 32K | 2.52 ms | 116.6 μs | 🚀 ~22× faster |
| set_sum() (high-entropy) | 32K | 5.92 ms | 194.1 μs | 🌪️ Complex pattern, ~30× faster |
| perlin dunes (high-entropy) | 32K | - | ~12 μs | 🌎 ~1380 chunks/frame (60 FPS) |
Full raw results? Check benches-raw.md.
Summary tables? See benches-tables.md.
Full commentary and insights? Dive into benches.md.
use glam::IVec3;
use voxelis::{
MaxDepth, VoxInterner,
spatial::{VoxOpsBatch, VoxOpsRead, VoxOpsWrite, VoxTree},
};
fn main() {
let mut interner = VoxInterner::<u8>::with_memory_budget(256 * 1024 * 1024);
let mut tree = VoxTree::new(MaxDepth::new(5)); // 32³ voxels (chunk)
let mut batch = tree.create_batch();
batch.set(&mut interner, IVec3::new(3, 0, 4), 1); // stone
tree.apply_batch(&mut interner, &batch);
assert_eq!(tree.get(&interner, IVec3::new(3, 0, 4)), Some(1));
}Add via Cargo:
cargo add voxelis glam@0.29.3 # Requires Rust 1.86+, optionally use `wide` for SIMD meshing| Concept | Purpose |
|---|---|
| VoxTree | The SVO-DAG — compressed octree core. |
| VoxInterner | Shared memory for leaves/branches. Hash-consed. |
| Batch | Bottom-up batched editing — mutate at light speed. |
| VoxOps | Trait for per-voxel manipulation — set, get, fill, clear. |
| BlockId | 64-bit magic to encode voxel state compactly. |
| Mesher | SIMD greedy meshing (WIP) — turn voxels into worlds. |
More? Crack open The Voxelis Bible (docs/The Voxelis Bible_ From Pixels to Worlds - An In-Depth Guide v2.3.pdf) — 38 pages of dangerously concentrated nerdery.
- Multithreaded interner with Rayon.
- GreedyMesh v2 (distance-field LOD magic).
- GPU frustum-culling traversals.
- OBJ / glTF import/export.
- True out-of-core paging (MMAP + LRU).
PRs welcome. Bonus points if your patch makes the CI bot 🐈🔄 purr.
cargo bench -p voxelis_benchHardware: Apple M3 Max, Rust 1.86 stable, -C target-cpu=native, final profile.
Want real numbers? We've got them — benches-tables.md and benches-raw.md await.
- Fork and branch (
feat/my-magnificent-contribution). cargo test && cargo bench- Open PR, include new benchmark delta.
- Bask in voxel-induced glory.
Dual licensed under MIT / Apache-2.0. Pick your poison, build something massive.
Voxelis may cause extreme enthusiasm, uncontrollable world-building, and compulsive Rust evangelism. Consult your GPU before operating heavy voxel engines. 😜