Zaneham/Booth

Open-source CUDA, Triton and HIP compiler targeting multiple GPU and CPU architectures.

What it solves

Booth is a compiler designed to break the lock-in of GPU-specific languages. It allows developers to write kernels in CUDA C, HIP, or Triton and compile them for a wide variety of hardware targets, including AMD GPUs (RDNA 2/3/4), NVIDIA GPUs (via PTX or native SASS/cubin), Tenstorrent Metalium, and even standard x86-64 CPUs for those without a GPU.

How it works

Booth acts as a cross-compiler that takes source code from CUDA, HIP, or Triton and lowers it into its own internal representation (IR). From there, it can emit machine code or low-level instructions for the target architecture. It supports multiple frontends, including LFortran for Fortran do concurrent kernels and OCaml functions, and it implements advanced compiler techniques like SSA register allocation and divergence analysis to optimize GPU execution.

Who it’s for

It is built for GPU developers, researchers, and hobbyists who want to target multiple GPU architectures from a single codebase or run GPU-style kernels on CPU hardware for testing and development.

Highlights

  • Multi-GPU Target: Supports NVIDIA, AMD, and Tenstorrent hardware.
  • Triton-to-CPU: Ability to compile Triton kernels directly to native x86-64 code without needing LLVM.
  • Native NVIDIA Backend: Can write cubin files that the GPU loads directly, bypassing the driver's JIT.
  • Broad Language Support: Accepts CUDA C, HIP, Triton, Fortran (via LFortran), and OCaml.
  • Mainframe-style Debugging: Includes structured crash dumps and parameter snapshots for kernel faults.

Related

  • Project
    ROCm/hip

    HIP is a C++ runtime API and kernel language for AMD GPUs that enables developers to port NVIDIA CUDA code to AMD hardware for portable GPU acceleration.

  • Project
    ByteDance-Seed/Triton-distributed

    A distributed compiler based on OpenAI Triton that enables the development of high-performance kernels with overlapping computation and communication for Nvidia and AMD GPUs.

  • Project
    triton-lang/triton

    A language and compiler for writing highly efficient custom deep-learning primitives with higher productivity than CUDA and greater flexibility than other DSLs.

  • Dispatch
    OpenAI Triton 1.0 Release

    OpenAI has released Triton 1.0, an open-source Python-like language and compiler that allows researchers to write highly efficient GPU kernels without deep CUDA expertise.

  • Project
    m4rs-mt/ILGPU

    A JIT compiler for high-performance GPU programming written in C#, enabling .NET developers to write GPU kernels without native dependencies.