A fast BVH library written in Rust. This library contains implementations of the following BVH building algorithms: - SAH with Spatial splits (Highest quality, slow build times) - Binned SAH (High quality, decently fast) - Locally Ordered Clustering Builder (Very fast build times)
All building algorithms make efficient use of multithreading where possible. An MBVH implementation is included as well and all the BVH structures implement traversal algorithms for both single rays and quad rays. This library emits a C/C++ library which is used in my hobby renderer.
In a scene with a teapot of ~6300 triangles, my MacBook Pro with an 8-core i9 9980HK achieves the following performance:
Single-threaded rays: 1.000.000 rays in 192.859 ms, 5.1851354 million rays per second
16 threads rays: 1.000.000 rays in 22.546 ms, 44.353767 million rays per second
Single-threaded packets: 1.000.000 rays in 42.497 ms, 23.531073 million rays per second
16 threads packets: 1.000.000 rays in 4.992 ms, 200.32051 million rays per second
```rust use rtbvh::*;
struct Triangle { vertex0: [f32; 3], vertex1: [f32; 3], vertex2: [f32; 3], }
impl Primitive for Triangle { fn center(&self) -> [f32; 3] { let mut result = [0.0; 3]; for i in 0..3 { result[i] = (self.vertex0[i] + self.vertex1[i] + self.vertex2[i]) / 3.0; } result }
fn aabb(&self) -> crate::AABB {
let mut aabb = crate::AABB::new();
aabb.grow(self.vertex0);
aabb.grow(self.vertex1);
aabb.grow(self.vertex2);
aabb
}
}
impl SpatialTriangle for Triangle { fn vertex0(&self) -> [f32; 3] { self.vertex0 }
fn vertex1(&self) -> [f32; 3] {
self.vertex1
}
fn vertex2(&self) -> [f32; 3] {
self.vertex2
}
}
let vertices: Vec<[f32; 3]> = vec![ [-1.0, -1.0, 0.0], [1.0, -1.0, 0.0], [1.0, 1.0, 0.0], [-1.0, 1.0, 0.0], ];
let primitives: Vec
// Choose one of these algorithms: let bvh = builder.clone().constructlocallyorderedclustered(); let bvh = builder.clone().constructbinnedsah(); let bvh = builder.constructspatial_sah(); ```