WebGPU on iOS

The iOS platform supports WebGPU as a rendering backend alongside Metal. There are two WebGPU implementations: Dawn (Google) and wgpu-native (Rust). Both use Metal as the GPU backend on iOS.

Building with Dawn

Dawn is fetched and built from source via CMake FetchContent, so no extra setup is needed. The first build takes a while because Dawn pulls in abseil, SPIRV-Tools, Tint, etc. Subsequent builds are incremental.

cmake --preset ios-webgpu-dawn
cmake --build build-ios-webgpu-dawn -- -sdk iphonesimulator -arch arm64

Building with wgpu-native

wgpu-native requires a pre-built static library for the target architecture. The CMake build will attempt to build it from source if cargo is available, but cross-compilation for iOS needs some setup.

1. Install the Rust iOS targets

# For simulator (Apple Silicon Mac)
rustup target add aarch64-apple-ios-sim

# For device
rustup target add aarch64-apple-ios

2. Cross-compile wgpu-native

For the iOS Simulator:

cd vendor/wgpu-native
BINDGEN_EXTRA_CLANG_ARGS="--target=arm64-apple-ios -isysroot $(xcrun --sdk iphoneos --show-sdk-path)" \
  cargo build --release --target aarch64-apple-ios-sim \
  --no-default-features --features metal,wgsl,spirv,glsl

For a physical device:

cd vendor/wgpu-native
BINDGEN_EXTRA_CLANG_ARGS="--target=arm64-apple-ios -isysroot $(xcrun --sdk iphoneos --show-sdk-path)" \
  cargo build --release --target aarch64-apple-ios \
  --no-default-features --features metal,wgsl,spirv,glsl

The BINDGEN_EXTRA_CLANG_ARGS environment variable is required so that bindgen’s macro fallback can find iOS system headers when generating FFI bindings.

This produces vendor/wgpu-native/target/<target>/release/libwgpu_native.a, which the CMake build picks up automatically.

3. Configure and build

cmake --preset ios-webgpu-wgpu
cmake --build build-ios-webgpu-wgpu -- -sdk iphonesimulator -arch arm64

If the pre-built library is not found and cargo is available, CMake will attempt to build wgpu-native automatically during configuration.

Implementation notes

The native backend lives in platform/ios/src/MLNMapView+WebGPU.mm. It uses #if MLN_WEBGPU_IMPL_DAWN / #elif MLN_WEBGPU_IMPL_WGPU conditionals for the API differences between the two implementations, following the same pattern as the Android and desktop backends.

Both backends use a plain UIView with a CAMetalLayer (not MTKView) for the WebGPU surface. The rendering view has userInteractionEnabled = NO so touch events pass through to the parent MLNMapView.

Keyboard shortcuts