Zaneham

Zaneham / BarraCUDA

Public

Open-source CUDA compiler targeting multiple GPU architectures. Compiles .cu to AMD and Tenstorrent GPU's

c99 compiler cuda gpu ml
1,461
59
100% credibility
Found Feb 17, 2026 at 109 stars 13x -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C
AI Summary

BarraCUDA compiles NVIDIA CUDA C code directly to runnable binaries for AMD RDNA 3 GPUs without needing translation layers or large dependencies.

How It Works

1
🔍 Discover a game-changer

You hear about a clever tool that lets your NVIDIA CUDA programs run on AMD graphics cards without rewriting a single line of code.

2
📥 Grab the tool

Download the simple package from the project page and get ready to build it on your computer.

3
🔨 Set it up quickly

Follow the one-command instruction to build your personal CUDA-to-AMD converter.

4
💻 Prepare your code

Take your existing CUDA program file and place it where you can work with it.

5
⚡ Convert with one command

Run the easy command to transform your NVIDIA code into an AMD-ready program file.

6
▶️ Load and run

Use your favorite AMD GPU runner to load the new file and watch your kernels execute.

🎉 Success on AMD hardware

Your CUDA programs now run smoothly on AMD GPUs, unlocking new hardware options without hassle.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 109 to 1,461 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is BarraCUDA?

BarraCUDA is an open source CUDA compiler that targets AMD GPUs, compiling standard .cu files directly to RDNA3 GFX11 machine code and ELF .hsaco binaries. It bypasses NVIDIA's ecosystem, letting you run CUDA kernels on AMD hardware without HIP translation layers or LLVM dependencies. Built in plain C99 with a one-line make build, it handles core CUDA features like shared memory, atomics, warp shuffles, and barriers via simple CLI: `./barracuda --amdgpu-bin kernel.cu -o kernel.hsaco`.

Why is it gaining traction?

As an open source CUDA alternative for AMD, it stands out by ditching heavy tools—no nvcc, no ROCm HIP porting headaches. Developers get working kernels fast, with validated encodings against LLVM objdump and a test suite covering vector adds to cooperative groups. It's a lightweight open source CUDA AMD project that just works for basic GPU compute.

Who should use this?

AMD GPU programmers porting NVIDIA kernels, researchers benchmarking CUDA code on RDNA3 cards, or teams seeking an open source CUDA compiler amid vendor lock-in. Ideal for github open source tools enthusiasts tackling "cannot open source file cuda_runtime.h" pains without full HIP rewrites. Skip if you need textures, constants, or production optimizations.

Verdict

Promising open source CUDA project for AMD at 16 stars and 1.0% credibility—early alpha with solid tests but missing compounds, consts, and perf tuning. Try for proofs-of-concept; watch the roadmap for Intel/others.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.