Bevy App Compute

Dispatch and run compute shaders on bevy from App World .

Getting Started

Add the following line to your Cargo.toml

[dependencies]
bevy_app_compute = "0.13.0"

Usage

Setup

Declare your shaders in structs implementing ComputeShader. The shader() fn should point to your shader source code. You need to derive TypePath as well:

#[derive(TypePath)]
struct SimpleShader;

impl ComputeShader for SimpleShader {
    fn shader() -> ShaderRef {
        "shaders/simple.wgsl".into()
    }
}

Next, declare a struct implementing ComputeWorker to declare the bindings and the logic of your worker:

#[derive(Resource)]
struct SimpleComputeWorker;

impl ComputeWorker for SimpleComputeWorker {
    fn build(world: &mut World) -> AppComputeWorker<Self> {
        let worker = AppComputeWorkerBuilder::new(world)
            // Add a uniform variable
            .add_uniform("uni", &5.)

            // Add a staging buffer, it will be available from
            // both CPU and GPU land.
            .add_staging("values", &[1., 2., 3., 4.])

            // Create a compute pass from your compute shader
            // and define used variables
            .add_pass::<SimpleShader>([4, 1, 1], &["uni", "values"])
            .build();

        worker
    }
}

Don't forget to add a shader file to your assets/ folder:

@group(0) @binding(0)
var<uniform> uni: f32;

@group(0) @binding(1)
var<storage, read_write> my_storage: array<f32>;

@compute @workgroup_size(1)
fn main(@builtin(global_invocation_id) invocation_id: vec3<u32>) {
    my_storage[invocation_id.x] = my_storage[invocation_id.x] + uni;
}

Add the AppComputePlugin plugin to your app, as well as one AppComputeWorkerPlugin per struct implementing ComputeWorker:

use bevy::prelude::*;
use bevy_app_compute::AppComputePlugin;

fn main() {
    App::new()
        .add_plugins(AppComputePlugin)
        .add_plugins(AppComputeWorkerPlugin::<SimpleComputeWorker>::default());
}

Your compute worker will now run every frame, during the PostUpdate stage. To read/write from it, use the AppComputeWorker<T> resource!

fn my_system(
    mut compute_worker: ResMut<AppComputeWorker<SimpleComputeWorker>>
) {
    if !compute_worker.available() {
        return;
    };

    let result: Vec<f32> = compute_worker.read_vec("values");

    compute_worker.write_slice("values", [2., 3., 4., 5.]);

    println!("got {:?}", result)
}

(see simple.rs)

Multiple passes

You can have multiple passes without having to copy data back to the CPU in between:

let worker = AppComputeWorkerBuilder::new(world)
    .add_uniform("value", &3.)
    .add_storage("input", &[1., 2., 3., 4.])
    .add_staging("output", &[0f32; 4])
    // add each item + `value` from `input` to `output`
    .add_pass::<FirstPassShader>([4, 1, 1], &["value", "input", "output"]) 
    // multiply each element of `output` by itself
    .add_pass::<SecondPassShader>([4, 1, 1], &["output"]) 
    .build();

    // the `output` buffer will contain [16.0, 25.0, 36.0, 49.0]

(see multi_pass.rs)

One shot computes

You can configure your worker to execute only when requested:

let worker = AppComputeWorkerBuilder::new(world)
    .add_uniform("uni", &5.)
    .add_staging("values", &[1., 2., 3., 4.])
    .add_pass::<SimpleShader>([4, 1, 1], &["uni", "values"])

    // This `one_shot()` function will configure your worker accordingly
    .one_shot()
    .build();

Then, you can call execute() on your worker when you are ready to execute it:

// Execute it only when the left mouse button is pressed.
fn on_click_compute(
    buttons: Res<ButtonInput<MouseButton>>,
    mut compute_worker: ResMut<AppComputeWorker<SimpleComputeWorker>>
) {
    if !buttons.just_pressed(MouseButton::Left) { return; }

    compute_worker.execute();
} 

It will run at the end of the current frame, and you'll be able to read the data in the next frame.

(see one_shot.rs)

Examples

See examples

Features being worked upon

• Ability to read/write between compute passes.
• add more options to the api, like deciding BufferUsages or size of buffers.
• Optimization. Right now the code is a complete mess.
• Tests. This badly needs tests.

Bevy version mapping

Bevy	bevy_app_compute
main	main
0.10	0.10.3
0.12	0.10.5
0.13	0.13.0