AXIR — Universal IR for GPU/AI Code

Write in CUDA, HIP, SYCL, OpenCL. Run anywhere.

🌐 GitHub Repo ✉ Contact Me 📄 White Paper

Why AXIR?

Every GPU vendor has its own locked ecosystem: NVIDIA (CUDA), AMD (HIP), Intel (SYCL/oneAPI), and Khronos (OpenCL). Normally, code written in one ecosystem doesn’t run in another.

AXIR is a universal intermediate representation (IR) that makes GPU/AI code portable without rewrites.

Proof of Concept v0.5

Frontends: CUDA, HIP, OpenCL, SYCL
Backend: CPU (NumPy)
Kernels: Vector Add, Saxpy, Reduce Sum

All 12 combinations (4 languages × 3 kernels) run successfully:

vector_add → [0, 3, 6, 9, 12]
saxpy      → [0, 4, 8, 12, 16]
reduce_sum → 120.0

Get in Touch

Interested in collaboration, research, or investment?

📩 Contact Form