Back to browse
GitHub Repository

An LLVM based multiplatform Python Jit with GA optimizations

2 starsPython

Hana JIT – LLVM-backed Python JIT with genetic-algorithm superoptimizer

by iqbal1980·Jul 10, 2026·3 points·0 comments

AI Analysis

●●●BangerWizardryBig BrainBold Bet

Genetic-algorithm superoptimizer pushes LLVM beyond what Numba achieves on numeric loops.

Strengths
  • Genetic algorithm searches instruction schedules and vectorization patterns Numba's fixed passes miss.
  • Zero type annotations required; infers types from first call and caches native specializations automatically.
  • Safe fallback to CPython interpreter for unsupported code paths instead of crashing or miscompiling.
Weaknesses
  • AMD GPU and FPGA targets explicitly marked as needing help; current support likely NVIDIA-centric.
  • Alpha status with zero stars suggests limited real-world testing beyond author's benchmarks.
Target Audience

Python developers needing numerical performance without rewriting in C or CUDA

Similar To

Numba · PyPy · Cython

Post Description

I wanted a replacement for Numba that supports other GPU types and FPGA, I need help in devlopment and testing especially in AMD targets and FPGA.

Thanks

Similar Projects