Approximating Trigonometric Functions for Posits Using the CORDIC Method

This archive is the artifact for the CF-2020 paper "Approximating Trigonometric Functions for Posits Using the CORDIC Method"

There are two methods to install the artifact:

(1) Installation using Dockerfile:

Pull the docker image from the docker hub:

$ docker pull jpl169/cordicwithposit
$ docker run -it jpl169/cordicwithposit

(2) Manual installation with Ubuntu 18.04"

Install the required packages:

$ sudo apt−get update
$ sudo apt−get install −yq −−no−install−recommends apt−utils
$ sudo apt−get install −yq build−essential python3 python3−pip libgmp3−dev libmpfr−dev git
$ python3 −m pip install numpy matplotlib

Install SoftPosit library

$ git clone https://gitlab.com/cerlane/SoftPosit.git
$ cd SoftPosit/build/Linux−x86_64−GCC/
$ make
$ cd ../../..

Finally, untar the artifact and build the code:

$ export SOFTPOSITPATH=<path to SoftPosit directory>
$ tar −xvf CordicWithPosit.tar.gz
$ cd CordicWithPosit && make

(2) Experiments

There are three experiments included in the artifact.

(a) Graph Generation

This experiment creates three graphs describing the accuracy of sin, cos, and atan of our CORDIC compared to the naive floating point implementation of CORDIC. This script generates three graphs, sin.pdf, cos.pdf, and atan.pdf graph directory.

$ python3 runGraphGeneration.py

(b) Accuracy Experiment (Fast). This experiment is the short version of the accuracy evaluation script.

$ python3 runSimplifiedAccAnalysis.py
$ python3 GenerateTableForSimplifiedAcc.py

(c) Accuracy Experiment (Full). This experiment is the full accuracy evaluation script.

$ python3 runAccAnalysis.py
$ python3 GenerateTableForAcc.py