mach gpu-dawn

Google’s Dawn WebGPU implementation, cross-compiled with Zig into a single static library, requiring nothing more than zig and git to build and cross-compile a static Dawn library for every OS:

No cmake
No ninja
No gn
No system dependencies (xcode, etc.)
Automagic cross compilation out of the box with nothing more than zig and git!
Builds a single static libdawn.a

Building from source

This will take ~10 minutes to finish, so by default when using this package a prebuilt binary version of Dawn (see the ‘binary releases’ section below) but you can always build Dawn from source using:

DAWN_FROM_SOURCE=true zig build

Binary releases

Dawn (specifically all the shader compilers, and the DirectXShaderCompiler) is a large C++ codebase and takes 5-10 minutes to build on a modern laptop. Since waiting is no fun, we also have binary releases produced by our GitHub actions:

View binary releases

Here’s how to read the downloads provided:

_debug.tar.gz and _release-fast.tar.gz are tarballs of the static library + headers for each OS and debug/release mode, respectively.
headers.json.gz is a JSON archive of all the Dawn/WebGPU headers.
Files ending in .a.gz and .lib.gz are the individual static libdawn.a and dawn.lib (Windows) gzippped and distributed. These are provided as individual downloads so there is no need to extract a tarball.

Important: Building WebGPU API symbols

Dawn and other WebGPU implementations (like the Rust one) do not agree on a standard webgpu.h API. Aspirationally, they aim to target the same https://github.com/webgpu-native/webgpu-headers header, but in practice they expose different APIs which are not ABI compatible.

When you call a Dawn webgpu.h function, Dawn internally diverts this call through a vtable which must be initialized using a call to dawnProcSetProcs.

mach/gpu-dawn builds since Oct 17th 2022 no longer include the webgpu.h symbols by default. If you intend to actually call the WebGPU API, you should build these two source files as part of your application:

And call dawnProcSetProcs to set up the proc table.

A warning about API stability

You should be aware:

WebGPU’s API is not formalized yet.
Dawn’s API is still changing.
The webgpu.h API is still changing
Dawn and gfx-rs/wgpu, although both try to implement webgpu.h, do not exactly implement the same interface. There are subtle differences in device discovery & creation for example.

Generated code

Dawn itself relies on a fairly large amount of dependencies, generated code, etc. To avoid having any dependency on Google build tools, code generation, etc. we maintain a minor fork of Dawn which has generated code and third-party dependencies comitted in “generated” branches. We are usually up-to-date with the upstream within a few weeks on average.

It also provides a few small patches to enable building Dawn with the Zig compiler which we plan to upstream soon, as well as some patches to build the DirectXShaderCompiler with Zig.

Getting started

Create a build.zig.zon in your project (replace LATEST_COMMIT with the latest commit hash):

.{
    .name = "mypkg",
    .version = "0.1.0",
    .dependencies = .{
        .mach_gpu_dawn = .{
            .url = "https://pkg.machengine.org/mach-gpu-dawn/LATEST_COMMIT.tar.gz",
        },
    },
}

Run zig build in your project, and the compiler will instruct you to add a .hash = "..." field next to .url:

note: expected .hash = "12209838fcfb7a77d2d6931efdc7448c033a1b7dad11d082c94bbeeba9d1038cd311",

Then use the dependency in your build.zig:

pub fn build(b: *std.Build) void {
    ...
    exe.addModule("mach-gpu-dawn", b.dependency("mach_gpu_dawn", .{
        .target = target,
        .optimize = optimize,
    }).module("mach-gpu-dawn"));
}

You can now use it in your src/main.zig file:

const gpu_dawn = @import("mach-gpu-dawn");

Ran into trouble?

Triple-check you followed the build.zig.zon instructions correctly, it’s easy to mess that part up.

Feel free to join the Mach Discord community for help.