While a lot of this stuff is still in flux (naming wise, most code is in a terrible location, etc). I wanted to share the work so-far to gather some feedback.

There are several reasons for adding bindless support to rust-gpu;

• Much lower CPU cost for updating bindings (bindings are created only at resource creation time, instead of on-demand like in a more traditional renderer); see for example this post https://ourmachinery.com/post/moving-the-machinery-to-bindless/
• More flexibility when using resource on the GPU, meaning one can create nested data structures which was previously impossible to do.

Relevant MCP: #389

Example of how to use:

#[derive(Copy, Clone)]
pub struct Bindings {
    pub buffer: Buffer,
}

#[derive(Copy, Clone)]
#[spirv(block)]
pub struct ShaderConstants {
    pub bindings: SimpleBuffer<Bindings>,
    pub user_data0: u32,
    pub user_data1: u32,
    pub user_data2: u32,
}

struct Data {
    a: f32,
    b: u32,
    c: f32,
    d: u32,
}

#[spirv(gl_compute)]
pub fn main(constants: PushConstant<ShaderConstants>) {
    let data = Data {
        a: 1.0,
        b: 0,
        c: 6.0,
        d: 1782,
    };

    let bindings = constants.bindings.load();

    unsafe {
        bindings.buffer.store(0, data)
    };
}

Draft because; lots of work still left to do.

• store method on Buffer
• StorageTexture
• StructuredBuffer<T> equivalent that can function as an array instead of a "bag of bytes" like Buffer
• Remaining functions on Texture
• Update Texture to use the generics that recently got added.
• Other Texture1d and Texture3d
• Test for the implementations of each, instead of just passing spirv-val and eye-balling the SPIR-V
• Add NonUniformResourceIndex equivalent