Skip to content

xman/sgpositpy

Repository files navigation

sgpositpy

Join the chat at https://gitter.im/sgpositpy/Lobby sgpositpy regression test status on Travis CI sgpositpy regression test status on Appveyor sgpositpy under MIT license

Posit is a new binary format for representing decimal numbers, an alternative to IEEE 754 floating-point number. Different configurations on the number bit size and the scaling are possible. Details can be found at the posithub and the notebook on posit.

We are interested to experiment with different posit configurations the nbits and the es sizes, and attempt to use smaller bit sizes in our applications while achieving the required accuracy to operate. However, an implementation conforming to the posit design is still lacking for popular programming languages.

This project, sgpositpy, is our attempt to develop a prototype implementation that is correct by the posit design. This can then serve as a reference for subsequent development of performance optimized posit library. Hence, the computation efficiency is currently not the focus of the project.

Current status

The current implementation is research oriented and experimental at pre-alpha stage. This is NOT currently suitable for production use.

Install

Install from PyPI

$ pip install sgposit

Install from code repository

$ git clone https://github.com/xman/sgpositpy
$ cd sgpositpy
$ pip install -r requirements.txt
$ python setup.py install

# Run regression tests.
$ python setup.py test

# Enable long running tests.
$ SGPOSIT_LONG_TESTS=1 python setup.py test

Getting started

The PCPosit class is our reference implementation with extensive regression tests. We shall have performance optimized version in Posit class for general use in the future.

The following code snippet creates posit objects from the given bit patterns, and the posit configuration, nbits and es.

# file: samples/pcposit_sample.py

from sgposit.pcposit import PCPosit

a = PCPosit(0x0C, mode='bits', nbits=6, es=2)
b = PCPosit(0x0F, mode='bits', nbits=6, es=2)

c = a + b
print("{} + {} => {}".format(a, b, c))

d = a - b
print("{} - {} => {}".format(a, b, d))

e = a * b
print("{} * {} => {}".format(a, b, e))

f = a / b
print("{} + {} => {}".format(a, b, f))

g = -a
print("uminus {} => {}".format(a, g))
$ python samples/pcposit_sample.py
1/4 + 3/4 => 1
1/4 - 3/4 => -1/2
1/4 * 3/4 => 3/16
1/4 / 3/4 => 3/8
uminus 1/4 => -1/4

License

sgpositpy is licensed under MIT License.

About

Probably correct posit prototype for python

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages