exo-lang/exo

Exocompilation for productive programming of hardware accelerators

What it solves

Exo is a language and compiler designed to make programming hardware accelerators more productive. It bridges the gap between high-level productivity and the low-level control required to optimize performance on specialized hardware.

How it works

Exo allows developers to write code that can be executed directly via Python or compiled into C and header files using the exocc command. It leverages a build system that utilizes CMake and Ninja to compile and link the generated C code. For advanced hardware feature testing (such as AVX-512 or AMX), it can integrate with the Intel Software Development Emulator (SDE).

Who it’s for

It is intended for developers and researchers working with hardware accelerators who need a way to schedule and optimize computations efficiently without sacrificing productivity.

Highlights

  • Exocompilation: A specialized compilation process for hardware accelerators.
  • C Code Generation: Ability to generate C and header files from Exo files.
  • Hardware Support: Integration with tools like the Intel SDE for testing x86-specific features like AMX.
  • Academic Foundation: Based on research published at PLDI 2022 and ASPLOS 2025.

Related

  • Project
    cornell-zhang/allo

    A Python-embedded, MLIR-based language and compiler for the modular design and programming of high-performance machine learning accelerators.

  • Project
    exoharness/exo

    Exo is an AI agent harness designed for recursive self-improvement, allowing agents to autonomously modify their own source code, prompts, and policies to evolve their capabilities.

  • Project
    openxla/xla

    XLA is an open-source machine learning compiler that optimizes models from PyTorch, TensorFlow, and JAX for high-performance execution on GPUs, CPUs, and ML accelerators.

  • Project
    goplus/xgo

    XGo is a natural-language-style programming language that simplifies Go's syntax and integrates C/C++, Python, and JavaScript/TypeScript libraries into a unified ecosystem.

  • 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.