[MathItYT]

Summary

Implements the browser-native js_webgpu backend for Pyodide and integrates its presentation context with rendercanvas.

Included

• navigator.gpu adapter/device acquisition
• JavaScript Promise → GPUPromise bridge
• Python snake_case → WebGPU camelCase descriptor conversion
• GPU resource wrappers
• Buffers, textures, views, samplers, bind groups and layouts
• Shader modules and compilation information
• Render/compute pipelines and async pipeline creation
• GPUCommandEncoder copy/clear/query operations
• Render/compute pass commands, indirect dispatch/draw, occlusion queries and bundles
• Browser GPUCanvasContext
• rendercanvas-compatible presentation context
  • configure()
  • unconfigure()
  • get_current_texture()
  • get_preferred_format()
  • physical-size propagation
  • present()
  • explicit resource release
• Initial API-shape tests

rendercanvas flow

rendercanvas.PyodideRenderCanvas supplies:

{
    "method": "screen",
    "platform": "browser",
    "window": HTMLCanvasElement,
}

The rendercanvas WgpuContextToScreen then calls wgpu.gpu.get_canvas_context(). The Pyodide backend wraps the native browser canvas.getContext("webgpu"), while keeping the public wgpu-py context API expected by rendercanvas.

Browser WebGPU presents automatically after queue submission; there is intentionally no native swapBuffers() equivalent.

Next

The next step is a real browser integration test with Pyodide + rendercanvas, including an actual render pass to an HTML canvas. After that, API parity and browser CI can be expanded systematically.

[Vipitis]

hey, I stumbled across this due to an issue search I have setup.

I really like the approach with call and getattr looks really clean. Since it took me several months (on and off) to figure out the quirks around pyodide ffi, I can offer you my solution (which makes use of wgpu.structs for descriptors) here: pygfx#753 (mainly _helpers.py things), feel free to have a look!