GPU.js on Expo — example

A minimal Expo app that runs GPU.js kernels on the device GPU through @gpujs/expo-gl.

Running it

This needs a development build, not Expo Go. Expo Go on the App Store tracks an older SDK and cannot open an SDK 57 project.

npm install
npx expo run:ios      # or: npx expo run:android

That prebuilds the native project, compiles it, and installs it on a connected device or simulator. iOS needs Xcode and a signing identity; the first build takes several minutes, later ones are quick.

On a simulator the GPU is emulated in software, so kernels are correct but extremely slow — a 512x512 multiply took ~28 seconds on an iPhone 17 Pro simulator versus 71 ms on the physical device. Benchmark on hardware.

The app is deliberately a diagnostic rather than a minimal demo. It reports each stage separately:

1. Create GL context — GLView.createContextAsync()
2. Create GPU — new GPU({ context }), which selects the Expo kernel
3. Detect features — compiles and runs a probe kernel, so this is the first real GL work
4. Compile kernel — a 512×512 matrix multiply
5. Run kernel — press the button; reports elapsed milliseconds

If something breaks, the failing stage and its error appear on screen, which is far more useful in a bug report than "it didn't work".

Kernels are strings here

React Native runs on Hermes, which discards function source, so a kernel written as a function cannot be compiled — see @gpujs/expo-gl. This app passes its kernel as a template string for that reason.

Web is not supported

@gpujs/expo-gl targets Expo's native GL. On web, use GPU.js directly — it already runs on WebGL there.

What CI can and cannot check

CI typechecks and runs expo export, which makes Metro resolve and compile every import. That catches a broken shim, a GPU.js that will not bundle for React Native, and version drift against the Expo SDK.

It cannot run a kernel: there is no Node implementation of Expo's GL. Anything past "it bundles" has to be verified on a device — which is the reason this app exists.