libwebgpu

Utilities for WebGPU development

Currently heavily based on: https://eliemichel.github.io/LearnWebGPU/index.html

Pre-reqs

I'm using WSL2 with Debian along with CLion. https://www.jetbrains.com/toolbox-app/

If using Toolbox and you want to login to your Jetbrains account, make sure to install a browser and the xdg-utils package.

Useful packages

sudo apt install curl wget zip unzip gpg vim xdg-utils xz-utils libwayland-dev libxkbcommon-dev

Native tools

Since the Emscripten toolchain uses Clang, we use it also for native builds. GCC also works but can give weird results with Clang-Tidy.

sudo apt install clang cmake python3 pipx git ninja-build pkgconf cppcheck clang-tidy
sudo update-alternatives --config c++ # set to `clang++`

Emscripten

See https://emscripten.org/docs/getting_started/downloads.html. It is not necessary to add the Emscripten environment to your PATH.

Conan

See https://docs.conan.io/2/installation.html. I used the pipx install.

Init conan: conan profile detect --force

You may wish to modify the auto-detected profile a bit ($HOME/.conan2/profiles/default). Mine is:

[settings]
arch=x86_64
build_type=Release
compiler=clang
compiler.cppstd=20
compiler.version=19
os=Linux

[conf]
tools.build:compiler_executables={'c':'/usr/bin/clang', 'cpp':'/usr/bin/c++'}
tools.cmake.cmaketoolchain:generator=Ninja
tools.system.package_manager:mode=install
tools.system.package_manager:sudo=true
tools.system.package_manager:sudo_askpass=true

Also add a profile for Emscripten. Modify the paths under [conf] -> tools.build:compiler_executables to match your Emscripten installation. $HOME/.conan2/profiles/emscripten:

[settings]
arch=wasm
build_type=Release
compiler=emcc
compiler.cppstd=20
compiler.libcxx=libc++
# Optional settings to enable multithreading (see note below)
# compiler.threads=posix
compiler.version=4.0.10
os=Emscripten

[conf]
# Optional settings to enable memory allocation
tools.build:exelinkflags=['-sALLOW_MEMORY_GROWTH=1', '-sMAXIMUM_MEMORY=4GB', '-sINITIAL_MEMORY=64MB']
tools.build:sharedlinkflags=['-sALLOW_MEMORY_GROWTH=1', '-sMAXIMUM_MEMORY=4GB', '-sINITIAL_MEMORY=64MB']
tools.build:compiler_executables={'c':'/home/meg/dev/emsdk/upstream/emscripten/emcc', 'cpp':'/home/meg/dev/emsdk/upstream/emscripten/em++'}
tools.cmake.cmaketoolchain:generator=Ninja

[buildenv]
CC=emcc
CXX=em++
AR=emar
NM=emnm
RANLIB=emranlib
STRIP=emstrip

Clone the libwebgpu project.

Build Dawn Native

https://github.com/google/dawn/blob/main/docs/building.md. Use the CMake and Ninja steps, but add -DDAWN_USE_WAYLAND=ON -DDAWN_ENABLE_INSTALL=ON (along with default -GNinja and -DCMAKE_BUILD_TYPE=[Release|Debug]) to the CMake command.

Install Dawn Native

https://github.com/google/dawn/blob/main/docs/quickstart-cmake.md#build-and-install-dawn-with-cmake. You can install system-wide, but I install to the external subdirectory of libwebgpu

cd <root Dawn directory> # ../.. if coming from the build steps
cmake --install out/Debug --prefix <libwebgpu root>/external/Linux/Debug

Download and install emdawnwebgpu port

This is the Emscripten port needed for wasm builds. Download the 'pkg' .zip (not remoteport .py) from here: https://github.com/google/dawn/releases Unzip and copy to external subdirectory of libwebgpu

unzip <zip file>
mkdir -p <libwebgpu root>/external/Emscripten/Debug
cp -r emdawnwebgpu_pkg <libwebgpu root>/external/Emscripten/Debug/

Build

conan install . --build=missing --settings=build_type=Debug --profile=default
conan install . --build=missing --settings=build_type=Debug --profile=emscripten

cmake . --preset=conan-linux-debug
cmake . --preset=conan-emscripten-debug

cmake --build --preset=conan-linux-debug
cmake --build --preset=conan-emscripten-debug

Run

# Native
build/Linux/Debug/testapp

# Web: Browse to http://localhost:8000/testapp.html
python3 -m http.server -d build/Emscripten/Debug