Add (Js)ArrayBuffer version for create_buffer_init

[zeroexcuses]

Problem: I want to upload to a Vertex Buffer Object directly from a https://rustwasm.github.io/wasm-bindgen/api/js_sys/struct.ArrayBuffer.html

Current Best Solution: We have to do ArrayBuffer -> &[u8], then internally, wgpu probably does &[u8] -> ArrayBuffer. This is via this function / struct pair:

impl DeviceExt for crate::Device {
    fn create_buffer_init(&self, descriptor: &BufferInitDescriptor<'_>) -> crate::Buffer {

pub struct BufferInitDescriptor<'a> {
    /// Debug label of a buffer. This will show up in graphics debuggers for easy identification.
    pub label: crate::Label<'a>,
    /// Contents of a buffer on creation.
    pub contents: &'a [u8],
    /// Usages of a buffer. If the buffer is used in any way that isn't specified here, the operation
    /// will panic.
    pub usage: crate::BufferUsages,
}

On the wasm32 platform, I'd like a way to bypass these two extra copies and directly do ArrayBuffer -> VBO.

Thanks!

[Wumpf]

As of #4042 you can map buffers directly into ArrayBuffer which solves the issue, but create_buffer_init is missing an ArrayBuffer version

[zeroexcuses]

I apologize for missing something obvious.

All the functions I see in #4042 return a js::ArrayBuffer , rather than take a js::ArrayBuffer as an argument.

How am I supposed to use the functions in #4042 when my goal is to upload the contents of a js::ArrayBuffer into a vertex-buffer-object ?

Thanks!

[Wumpf]

Hmm not sure if I'm a bit confused about this myself, the PR I linked does after all advertise it for readback as prime motivator.
However, what it allows doing is to get the mapping of a buffer directly as ArrayBuffer, so you can now directly copy to that and then unmap the buffer, so that's already a little bit better. Before that change, a mapping would again go through a &[u8] step on the wasm heap.
So, having the array buffer on a variant of the BufferInitDescriptor would ofc be even better because then that would be just what's piped through to the buffer creation, meaning you wouldn't need to do a copy from your ArrayBuffer to the mapped one

[Wumpf]

brought this up on the wgpu maintainer call and confirmed the above.
We'll need ArrayBuffer versions both for BufferInitDescriptor and Queue::write_buffer. But using the method from #4042 should work albeit it still requires a copy (one less!)