WasmGPU.python¶

Summary¶

WasmGPU.python provides ndarray-oriented data transfer utilities for Python-JS interop workflows. It supports sending arrays into Wasm memory, receiving typed views/copies back, in-place updates, and freeing heap-owned transfers. This API is designed to pair with Pyodide or any bridge that exposes Python buffers. It is separate from WasmGPU.webassembly, which reads directly from foreign WebAssembly modules.

Syntax¶

WasmGPU.python: PythonInterop
const py = wgpu.python;

Parameters¶

This accessor does not take parameters.

Returns¶

PythonInterop - Object with send/view/bytes/copy/receive/free methods.

Type Details¶

type WasmNdarrayHandle = {
    kind: "heap" | "frame" | "arena";
    dtype: "i8" | "u8" | "i16" | "u16" | "i32" | "u32" | "f32" | "f64";
    shape: number[];
    ptr: number;
    length: number;
    byteLength: number;
    epoch?: number;
};

type PythonInterop = {
    sendNdarray(src: PythonArraySource, options?: SendNdarrayOptions): WasmNdarrayHandle;
    view(handle: WasmNdarrayHandle): NumberTypedArray;
    bytes(handle: WasmNdarrayHandle): Uint8Array;
    copyInto(handle: WasmNdarrayHandle, src: PythonArraySource, options?: { dtype?: DType }): void;
    receiveNdarray(handle: WasmNdarrayHandle, options?: { copy?: boolean }): { dtype: DType; shape: number[]; data: NumberTypedArray };
    free(handle: WasmNdarrayHandle): void;
};

Example¶

const canvas = document.querySelector("canvas");
const wgpu = await WasmGPU.create(canvas);

const source = new Float32Array([1, 2, 3, 4, 5, 6]);
const handle = wgpu.python.sendNdarray(source, { dtype: "f32", shape: [2, 3], allocator: "heap" });
const view = wgpu.python.view(handle);
view[0] = 99;

const transfer = wgpu.python.receiveNdarray(handle, { copy: true });
console.log(transfer.dtype, transfer.shape, transfer.data[0]);
wgpu.python.free(handle);