Taking Python's New JIT for a Spin.md

The speedup provided by Python's new JIT is pretty amazing. A 20% speedup over 3.12! And: no code changes necessary.

Benchmark 1: python3.12 main.py
  Time (mean ± σ):      1.213 s ±  0.028 s    [User: 1.162 s, System: 0.008 s]
  Range (min … max):    1.181 s …  1.281 s    10 runs

Benchmark 2: PYTHON_JIT=0 python3.13 main.py
  Time (mean ± σ):      1.124 s ±  0.013 s    [User: 1.114 s, System: 0.004 s]
  Range (min … max):    1.097 s …  1.137 s    10 runs

Benchmark 3: PYTHON_JIT=1 python3.13 main.py
  Time (mean ± σ):      1.007 s ±  0.004 s    [User: 0.998 s, System: 0.004 s]
  Range (min … max):    1.001 s …  1.013 s    10 runs

Summary
  PYTHON_JIT=1 python3.13 main.py ran
    1.12 ± 0.01 times faster than PYTHON_JIT=0 python3.13 main.py
    1.20 ± 0.03 times faster than python3.12 main.py

How to install Python 3.13 and run the benchmark

brew install llvm
CONFIGURE_OPTS="--enable-experimental-jit=yes-off" pyenv install 3.13.0b1
export PATH="$(pyenv root)/versions/3.13.0b1/bin:$PATH"
hyperfine "python3.12 main.py" "PYTHON_JIT=0 python3.13 main.py" "PYTHON_JIT=1 python3.13 main.py"

The code

import time
import random


def matrix_multiply(a, b):
    result = [[0] * len(b[0]) for _ in range(len(a))]

    for i in range(len(a)):
        for j in range(len(b[0])):
            for k in range(len(b)):
                result[i][j] += a[i][k] * b[k][j]

    return result


def generate_matrix(size):
    return [[random.random() for _ in range(size)] for _ in range(size)]


if __name__ == "__main__":
    SIZE = 300

    random.seed(42)

    a = generate_matrix(SIZE)
    b = generate_matrix(SIZE)

    start_time = time.time()
    result = matrix_multiply(a, b)
    end_time = time.time()

    print(f"Time taken for matrix multiplication: {end_time - start_time} seconds")

Python and system info

$ python3.12 -VV
Python 3.12.2 (main, Apr 22 2024, 17:20:27) [Clang 15.0.0 (clang-1500.3.9.4)]
$ python3.13 -VV
Python 3.13.0b1 (main, May 30 2024, 15:39:04) [Clang 15.0.0 (clang-1500.3.9.4)]

$ neofetch --off --stdout | rg "^(OS|Host|Kernel|[CG]PU|Memory)"
OS: macOS 14.5 23F79 arm64
Host: Mac14,15
Kernel: 23.5.0
CPU: Apple M2
GPU: Apple M2
Memory: 3408MiB / 24576MiB