Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/akhildevelops/cudaz/llms.txt

Use this file to discover all available pages before exploring further.

Before adding cudaz to your project, make sure you have two prerequisites in place: the NVIDIA CUDA Toolkit installed on your system (which provides the cuda.h header, and the libcuda, libnvrtc shared libraries), and a supported Zig toolchain (0.13.0 through 0.16.0). cudaz ships separate tagged releases for each Zig version, so you must fetch the correct archive for the compiler you are using.
cudaz does not support Windows. Building on Windows will immediately fail with error.WINDOWS_NOT_SUPPORTED. Use Linux or macOS.
1

Fetch cudaz and save it to build.zig.zon

Run the zig fetch --save command for your Zig version. This downloads the cudaz source archive and writes the URL and content hash into your project’s build.zig.zon file automatically.
zig fetch --save https://github.com/akhildevelops/cudaz/archive/0.4.0.tar.gz
2

Configure build.zig

Add the cudaz module to your executable and link the required system libraries. The key steps are: retrieve the cudaz dependency, import its cudaz module, then link libc, libcuda, and libnvrtc.
build.zig
const std = @import("std");

pub fn build(b: *std.Build) !void {
    // exe points to main.zig that uses cudaz
    const exe = b.addExecutable(.{
        .name = "main",
        .root_module = b.createModule(.{
            .root_source_file = b.path("src/main.zig"),
            .target = b.standardTargetOptions(.{}),
        }),
    });

    // Point to cudaz dependency
    const cudaz_dep = b.dependency(
        "cudaz",
        .{}, // replace with `.{ .CUDA_PATH = @as([]const u8, "<your cuda path>") }`
             // to specify a custom CUDA installation path
    );

    // Fetch and add the module from cudaz dependency
    const cudaz_module = cudaz_dep.module("cudaz");
    exe.root_module.addImport("cudaz", cudaz_module);

    // Dynamically link to libc, cuda, nvrtc
    exe.root_module.link_libc = true;
    exe.root_module.linkSystemLibrary("cuda", .{});
    exe.root_module.linkSystemLibrary("nvrtc", .{});

    // Run binary
    const run = b.step("run", "Run the binary");
    const run_step = b.addRunArtifact(exe);
    run.dependOn(&run_step.step);
}
3

Verify the build

With a CUDA-capable GPU present and the CUDA toolkit installed, run:
zig build
A successful build with no errors confirms that cudaz was fetched, the CUDA headers were located, and the linker found libcuda and libnvrtc. If the build fails, see the custom CUDA path section below.

​
Specifying a custom CUDA path

By default, cudaz searches for the CUDA toolkit in these standard locations (in order):
  • /usr
  • /usr/local/cuda
  • /opt/cuda
  • /usr/lib/cuda
It confirms a valid installation by checking for the presence of include/cuda.h under each candidate directory. If your CUDA toolkit is installed somewhere else — for example, a versioned directory like /usr/local/cuda-12.4 — you can override the search by passing the -DCUDA_PATH build option: 
zig build -DCUDA_PATH=/usr/local/cuda-12.4
You can also bake the custom path into your dependency declaration inside build.zig so every developer on the project uses the same path:
build.zig
const cudaz_dep = b.dependency("cudaz", .{
    .CUDA_PATH = @as([]const u8, "/usr/local/cuda-12.4"),
});

Build docs developers (and LLMs) love

Get started for freeTalk to us