https://github.com/skailasa/pyinvsqrt

Quake 3 Fast Inverse Square Root in Python
https://github.com/skailasa/pyinvsqrt

bit-manipulation fast-inverse-square-root hpc python3 quake-3

Last synced: 5 months ago
JSON representation

Quake 3 Fast Inverse Square Root in Python

• Host: GitHub
• URL: https://github.com/skailasa/pyinvsqrt
• Owner: skailasa
• License: mit
• Created: 2021-03-14T11:26:19.000Z (over 4 years ago)
• Default Branch: main
• Last Pushed: 2021-03-15T09:12:44.000Z (over 4 years ago)
• Last Synced: 2023-03-07T06:46:31.708Z (over 2 years ago)
• Topics: bit-manipulation, fast-inverse-square-root, hpc, python3, quake-3
• Language: C
• Homepage:
• Size: 29.3 KB
• Stars: 3
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
 Fast Inverse Square Root in Python 


[![Anaconda-Server Badge](https://img.shields.io/conda/v/skailasa/pyinvsqrt.svg)](https://anaconda.org/skailasa/pyinvsqrt) [![Anaconda-Server Badge](https://anaconda.org/skailasa/pyinvsqrt/badges/platforms.svg)](https://anaconda.org/skailasa/pyinvsqrt) [![Anaconda-Server Badge](https://anaconda.org/skailasa/pyinvsqrt/badges/latest_release_date.svg)](https://anaconda.org/skailasa/pyinvsqrt)

Made (in)famous by the original implementation in the Quake 3 Arena source code, where fast inverse square roots are vital for perfoming vector operations for game physics, the Fast Inverse Square Root is an extremely clever bit/memory hack, that avoids using slow division instructions, and replaces everything with rapid bit-shifts to calculate an approximation to, f(x) =  \frac{1}/{\sqrt{x}}.

## Requirements

- gcc

- conda/miniconda

## Install

```bash

conda install -c skailasa pyinvsqrt

```

## Usage

```python

import numpy as np

from pyinvsqrt import invsqrt

# Only 32 Bit compatible

x = np.float32(10)

# Compute inverse square root

y = invsqrt(x)

```

## Benchmark

Compared with a pure Numpy (BLAS backend) inverse square root, pyinvqrt is almost 400% faster, and almost 25% compared to a Numba-fied implementation.

![](static/benchmark.png)