The main motivation for this project was to complete the assignment of Intro to Parallel Programming using as little C++ as possible.

The project started in 2021 with Zig 0.7.0 and was in limbo until 2025 where I upgraded it to Zig 0.15.2 leveraging all the improvement the compiler and toolchain got int the meantime.

Cuda is a superset of C++ with custom annotation to distinguish between device (GPU) functions and host (CPU) functions. They also have special variables for GPU thread IDs and special syntax to schedule a GPU function.

You're supposed to compile this Cuda code using nvcc NVidia proprietary compiler, to Nvidia PTX "assembly" format. NVCC creates complicated object containing the ptx and automotically load it when the executable/library is loaded.

But Zig has a LLVM backend that can also output ptx. With build.zig we can control the full build magic done by nvcc and expose it in a more modular way to the user.

This repo is divided in several parts:

• cuda.zig: a wrapper around cuda.h the CPU part of Cuda.

• nvptx.zig: PTX intrinsics that help write device code in Zig.

• build.zig: contains code to compile the kernel into ptx, and expose it to CPU code.

• CS344 contains code for all the lesson and homework. Not ported yet