sorenmulli / pyspeed

Comparing speeds of python implementations

Python implementations speed test

A project concerned with comparing speeds of different python low-level language integrations.

Tested implementations

• Pure python

Using core python language features and nothing else.

• Numpy

Using built in functions from the numpy library

• CPU PyTorch

Using built in function from pytorch library on cpu

• Cython pure

Compiling the pure python script as a cython library and importing it

• Cython C

Adding cython (C) language features to the pure python implementation.

• Ctypes C

Using the ctypes module to load in a dll (so) written in C and compiled

• Ctypes Rust

Using the ctypes module to load in a dll (so) written in Rust and compiled

• Numba

Using the pure python code and adding @jit from numba

Test cases

• rootloop

Compute the sum of the square root of every even number up to N.

Setup

Run build.sh.

Usage

Run run.py.