@vgpu/core

0.0.8 — early preview

@vgpu/core is the runtime foundation for vgpu. It gives you a small wrapper layer around WebGPU devices, queues, buffers, textures, and shader modules, plus an App.create() entry point that asks an adapter for a device and returns a compact app instance. In 0.0.8 the goal is portability and a minimal public surface, not a fully opinionated engine.

Install

pnpm add @vgpu/core

Exports

Classes

App
Buffer
Device
Queue
Shader
Texture
VGPUError
ValidationError

Functions

bind
createBindGroupLayout
createPipelineLayout
createBindGroup
createSampler
createMockGPUDevice

Types

AppCreateOptions
AppInstance
VGPUAdapter
BufferOptions
BufferUsageName
BufferWriteData
TextureOptions
TextureUsageName
CreateDeviceOptions
BindVisibility
CreateBindGroupLayoutOptions
CreatePipelineLayoutOptions
CreateBindGroupOptions
DeviceLike
ShaderInput

Usage

import { App } from "@vgpu/core";
import { createMockAdapter } from "@vgpu/adapter-mock";

const { device } = await App.create({ adapter: createMockAdapter() });
const data = new Float32Array([1, 2, 3, 4]);
const buffer = device.createBuffer({
  size: data.byteLength,
  usage: ["copy_dst", "copy_src", "storage"],
});

buffer.write(data);
const copy = new Float32Array(await buffer.read(data.byteLength));
device.destroy();

Explicit binding helpers keep layout control in user code without WGSL reflection or layout: "auto":

import { bind, createBindGroupLayout } from "@vgpu/core";

const sceneLayout = createBindGroupLayout(device, {
  entries: [bind.uniform(0, "vertex|fragment")],
});

The .gpu property is an unmanaged raw WebGPU escape hatch for APIs that VGPU does not wrap yet. Prefer wrapper lifecycle methods (buffer.destroy(), texture.destroy(), device.destroy()) instead of destroying .gpu directly.

License

MIT.