jWebGPU

jWebGPU is a Java binding stack for WebGPU APIs across desktop, Android, and web targets.

• Browser example: Studio demo
• Android app: Google Play

What This Project Is For

• Build and ship WebGPU-capable Java apps across multiple runtimes.
• Reuse the same Java-facing API while swapping backend/runtime packaging per platform.
• Generate bindings from webgpu/base templates plus webgpu/builder IDL/native sources using the jParser Gradle plugin.

Targets:

• JNI desktop runtime
• FFM desktop runtime
• TeaVM C desktop runtime bootstrap (Windows x64 WGPU)
• TeaVM/WebAssembly web runtime
• Android JNI runtime

Project Structure

Canonical modules are defined in settings.gradle.kts.

Library modules:

• :webgpu:download - native dependency download tasks.
• :webgpu:builder - jParser plugin configuration and native build driver.
• :webgpu:base - hand-authored Java templates and directive blocks.
• :webgpu:core - generated Java API layer.
• :webgpu:shared:jni - generated JNI Java runtime.
• :webgpu:shared:c - generated TeaVM C implementations and portable native resources.
• :webgpu:desktop:c - desktop TeaVM C native-resource packaging.
• :webgpu:desktop:jni - desktop JNI native packaging.
• :webgpu:desktop:ffm - generated FFM Java runtime and desktop native packaging.
• :webgpu:web:wasm - TeaVM/WebAssembly runtime packaging.
• :webgpu:android:jni - WGPU/Dawn Android JNI runtime packaging.

Demo modules:

• :demos:backend:core
• :demos:backend:desktop
• :demos:backend:web
• :demos:backend:android
• :demos:app:core
• :demos:app:desktop-jni
• :demos:app:desktop-ffm
• :demos:app:web
• :demos:app:android

Build Commands

Download tasks are manual prerequisites for native builds. jParser_build_* tasks do not run them automatically.

Generate Java bindings:

.\gradlew.bat :webgpu:builder:jParser_generate

Windows JNI/FFM:

.\gradlew.bat :webgpu:download:webgpu_download_glfw_windows
.\gradlew.bat :webgpu:download:webgpu_download_windows_x86_64_wgpu
.\gradlew.bat :webgpu:download:webgpu_download_windows_x86_64_dawn
.\gradlew.bat :webgpu:builder:jParser_build_windows64_jni_wgpu :webgpu:builder:jParser_build_windows64_jni_dawn
.\gradlew.bat :webgpu:builder:jParser_build_windows64_ffm_wgpu :webgpu:builder:jParser_build_windows64_ffm_dawn

Desktop TeaVM C publishes one artifact per native backend and host platform:

Backend	Windows x64	Linux x64	macOS x64	macOS arm64
WGPU	webgpu-desktop-c-wgpu_windows_x64	webgpu-desktop-c-wgpu_linux_x64	webgpu-desktop-c-wgpu_mac_x64	webgpu-desktop-c-wgpu_mac_arm64
Dawn	webgpu-desktop-c-dawn_windows_x64	webgpu-desktop-c-dawn_linux_x64	webgpu-desktop-c-dawn_mac_x64	webgpu-desktop-c-dawn_mac_arm64

Add exactly one matching artifact to a TeaVM C application. Backend selection is a build-time choice. Linkage is a separate consumer choice: JPARSER_JWEBGPU_TEAVMC_LINKAGE accepts STATIC, SHARED_LINKED, or RUNTIME_LOADED, and jParser defaults it to STATIC when the application does not select a mode.

The WGPU/Dawn consumer requirements are declared in webgpu/builder/build.gradle.kts. jParser folds all platform/backend declarations into the existing generated jparser_jwebgpu_teavm_c.cmake inside webgpu-c; backend native artifacts contain only headers and native libraries, with no handwritten CMake hook. The packaged include/webgpu/wgpu.h or include/dawn/webgpu.h selector identifies the chosen backend. Supplying both backend artifacts, or neither matching artifact for the current platform, fails during CMake configuration instead of silently linking the wrong implementation.

The published Windows static payloads currently use the dynamic MSVC runtime and are stored as the md variant. A static consumer therefore selects /MD. The shared bridge remains usable by /MD and /MT applications because the CRT stays behind the DLL boundary. A static /MT consumer can rebuild the jWebGPU bridge and selected WebGPU dependency with /MT, package them under the matching mt payload directory, or provide them with JPARSER_JWEBGPU_TEAVMC_LIBRARY and JPARSER_JWEBGPU_{WGPU|DAWN}_NATIVE_LIBRARY. jWebGPU does not silently change the application's runtime or linkage choice.

Windows TeaVM C:

.\gradlew.bat :webgpu:download:webgpu_download_glfw_windows
.\gradlew.bat :webgpu:download:webgpu_download_windows_x86_64_wgpu
.\gradlew.bat :webgpu:download:webgpu_download_windows_x86_64_dawn
.\gradlew.bat :webgpu:builder:jParser_build_windows64_teavm_c_wgpu :webgpu:builder:jParser_build_windows64_teavm_c_dawn
.\gradlew.bat :webgpu:shared:c:jar
.\gradlew.bat :webgpu:desktop:c:nativeJar_wgpu_windows_x64 :webgpu:desktop:c:nativeJar_dawn_windows_x64

Linux JNI/FFM:

./gradlew :webgpu:download:webgpu_download_linux_x86_64_wgpu
./gradlew :webgpu:download:webgpu_download_linux_x86_64_dawn
./gradlew :webgpu:download:webgpu_download_glfw_windows
./gradlew :webgpu:builder:jParser_build_linux64_jni_wgpu :webgpu:builder:jParser_build_linux64_jni_dawn
./gradlew :webgpu:builder:jParser_build_linux64_ffm_wgpu :webgpu:builder:jParser_build_linux64_ffm_dawn

Linux TeaVM C:

./gradlew :webgpu:builder:jParser_build_linux64_teavm_c_wgpu :webgpu:builder:jParser_build_linux64_teavm_c_dawn
./gradlew :webgpu:desktop:c:nativeJar_wgpu_linux_x64 :webgpu:desktop:c:nativeJar_dawn_linux_x64

macOS JNI/FFM:

./gradlew :webgpu:download:webgpu_download_macos_x86_64_wgpu
./gradlew :webgpu:download:webgpu_download_macos_aarch64_wgpu
./gradlew :webgpu:download:webgpu_download_macos_x86_64_dawn
./gradlew :webgpu:download:webgpu_download_macos_aarch64_dawn
./gradlew :webgpu:download:webgpu_download_glfw_windows
./gradlew :webgpu:builder:jParser_build_mac64_jni_wgpu :webgpu:builder:jParser_build_mac64_jni_dawn :webgpu:builder:jParser_build_macArm_jni_wgpu :webgpu:builder:jParser_build_macArm_jni_dawn
./gradlew :webgpu:builder:jParser_build_mac64_ffm_wgpu :webgpu:builder:jParser_build_mac64_ffm_dawn :webgpu:builder:jParser_build_macArm_ffm_wgpu :webgpu:builder:jParser_build_macArm_ffm_dawn

macOS TeaVM C:

./gradlew :webgpu:builder:jParser_build_mac64_teavm_c_wgpu :webgpu:builder:jParser_build_mac64_teavm_c_dawn :webgpu:builder:jParser_build_macArm_teavm_c_wgpu :webgpu:builder:jParser_build_macArm_teavm_c_dawn
./gradlew :webgpu:desktop:c:nativeJar_wgpu_mac_x64 :webgpu:desktop:c:nativeJar_dawn_mac_x64 :webgpu:desktop:c:nativeJar_wgpu_mac_arm64 :webgpu:desktop:c:nativeJar_dawn_mac_arm64

TeaVM/WebAssembly:

.\gradlew.bat :webgpu:download:webgpu_download_emdawnwebgpu
.\gradlew.bat :webgpu:builder:jParser_build_web_wasm

Android JNI:

.\gradlew.bat :webgpu:download:webgpu_download_android_x86_64_wgpu
.\gradlew.bat :webgpu:download:webgpu_download_android_i686_wgpu
.\gradlew.bat :webgpu:download:webgpu_download_android_armv7_wgpu
.\gradlew.bat :webgpu:download:webgpu_download_android_aarch64_wgpu
.\gradlew.bat :webgpu:download:webgpu_download_android_x86_64_dawn
.\gradlew.bat :webgpu:download:webgpu_download_android_i686_dawn
.\gradlew.bat :webgpu:download:webgpu_download_android_armv7_dawn
.\gradlew.bat :webgpu:download:webgpu_download_android_aarch64_dawn
.\gradlew.bat :webgpu:builder:jParser_build_android_jni_wgpu :webgpu:builder:jParser_build_android_jni_dawn

Generated native outputs are placed under webgpu/builder/build/c++/libs/.

Native compiler policy is configured through jParser's generic target hooks. The current Windows bridge payloads, and the downloaded Dawn archive they can statically link, use the dynamic CRT, so the producer target declarations pass compileFlag("/MD") explicitly; there is no jWebGPU or jParser msvcRuntime type. Rebuilding the bridge and dependencies with another compatible ABI only requires changing the ordinary compiler flag and packaging the result in the matching runtime directory. Linux, macOS, Android, and iOS use their normal toolchain options; MSVC /MT and /MD flags are Windows-only.

Run Demos

.\gradlew.bat :demos:app:desktop-jni:webgpu_demo_app_desktop_jni_wgpu_run
.\gradlew.bat :demos:app:desktop-jni:webgpu_demo_app_desktop_jni_dawn_run
.\gradlew.bat :demos:app:desktop-ffm:webgpu_demo_app_desktop_ffm_wgpu_run
.\gradlew.bat :demos:app:desktop-ffm:webgpu_demo_app_desktop_ffm_dawn_run
.\gradlew.bat :demos:app:web:webgpu_demo_app_web_run
.\gradlew.bat :demos:app:android:installWgpuDebug
.\gradlew.bat :demos:app:android:installDawnDebug

Development Notes

• Edit binding templates in webgpu/base/src/main/java/**.
• Edit IDL and native glue in webgpu/builder/src/main/cpp/**.
• Do not hand-edit generated Java under webgpu/core, webgpu/shared/jni, webgpu/shared/c, webgpu/desktop/ffm, or webgpu/web/wasm.

Installation

Use group ID com.github.xpenatan.jWebGPU.

dependencies {
    implementation("com.github.xpenatan.jWebGPU:webgpu-core:<version>")

    // Desktop JNI: choose one backend native artifact for your platform.
    implementation("com.github.xpenatan.jWebGPU:webgpu-desktop-jni:<version>")
    runtimeOnly("com.github.xpenatan.jWebGPU:webgpu-desktop-jni-wgpu_windows_x64:<version>")
    // runtimeOnly("com.github.xpenatan.jWebGPU:webgpu-desktop-jni-dawn_windows_x64:<version>")

    // Desktop FFM: choose one backend native artifact for your platform.
    // implementation("com.github.xpenatan.jWebGPU:webgpu-desktop-ffm:<version>")
    // runtimeOnly("com.github.xpenatan.jWebGPU:webgpu-desktop-ffm-wgpu_windows_x64:<version>")
    // runtimeOnly("com.github.xpenatan.jWebGPU:webgpu-desktop-ffm-dawn_windows_x64:<version>")

    // TeaVM C, Windows x64 WGPU: includes webgpu-c and the jParser C runtime transitively.
    // implementation("com.github.xpenatan.jWebGPU:webgpu-desktop-c-wgpu_windows_x64:<version>")

    // TeaVM/WebAssembly requires both the Java runtime and WebAssembly payload.
    // implementation("com.github.xpenatan.jWebGPU:webgpu-web:<version>")
    // implementation("com.github.xpenatan.jWebGPU:webgpu-web_wasm:<version>")

    // Android: choose exactly one backend AAR.
    // implementation("com.github.xpenatan.jWebGPU:webgpu-android-wgpu:<version>")
    // implementation("com.github.xpenatan.jWebGPU:webgpu-android-dawn:<version>")
}

Android backend AARs are mutually exclusive because both package libjWebGPU.so. Applications using webgpu-android-dawn must initialize with JWebGPULoader.init(JWebGPUBackend.DAWN, listener); the overload without a backend keeps WGPU as its default.

Ecosystem

For LibGDX projects, gdx-webgpu is a higher-level backend extension built on top of jWebGPU.

Support

If you find this project valuable and want to fuel its continued growth, please consider sponsoring.

License

jWebGPU is licensed under the Apache License 2.0.