run-wasm helper (cargo-run-wasm) for testing the wgpu runner on the web.

[eddyb]

Originally submitted as:

• run-wasm helper (cargo-run-wasm) for testing the wgpu runner on the web. EmbarkStudios/rust-gpu#1016
  (kept below the main part of that description that is still relevant, and updated it)

---

Had to work around some sRGB complications (and push constants) but it finally works!

• rustup target add wasm32-unknown-unknown
    _{(should we add this to rust-toolchain.toml?)}
• cargo run-wasm -p example-runner-wgpu
    _{(defaults to serving the result of the build on http://localhost:8000)}
• google-chrome-stable --enable-features=Vulkan --enable-unsafe-webgpu http://localhost:8000

[eddyb]

I just wanted to rebase and share this branch, but I spent a few minutes on it and I think I fixed all remaining issues (the screenshot in the description is from the current state of the branch).

Ironically(?), the sky shader doesn't work because of it demoing pipeline-overridable constants (Vulkan specialization constants), but I am happy to default to the mouse shader because of the interactivity.

[LegNeato]

This feels like it should be somewhere reusable?

[eddyb]

It's been a while since I whipped up that transformation, because WebGPU launched w/o push constants:

• Investigation : Push Constants gpuweb/gpuweb#75

But given that I'm doing this on the SPIR-V module, before Naga sees it, it should maybe be:

• part of wgpu/Naga, which would let other wgpu users reuse this emulation
  • or at least implemented here on Naga IR, then passing WGSL/naga::Module to wgpu
• done as part of the build script, and potentially implemented in rustc_codegen_spirv/SPIR-T
  • @Firestar99's ideas around a WGSL-outputting target would also be relevant here
  • for wasm, it'd likely be best to disable runtime Naga, and embed the WGSL directly

[LegNeato]

Loved this demo when I saw it.

[eddyb]

• rustup target add wasm32-unknown-unknown
    _{(should we add this to rust-toolchain.toml?)}

Oops, forgot about this.
I'm also thinking that if we build this in CI, and use a crate like we_clap (which wraps clap to let ?foo=bar work for web targets, like --foo=bar does for CLI), we could have this on the website, w/ all the examples.
_{(might even be more impactful for shadertoys, tho I'm not sure the current approach there is a good idea)}

EDIT: oh and wgpu has a WebGL2 fallback we could enable, allowing simpler web demos to "always work".