MATLAB experiments with narrow-range floating-point formats

This repository contains the data plotted in [Sec. 5, 1] and the MATLAB code used to generate it.

The scripts rely on the CPFloat library for implementing custom-precision arithmetic.

The errors in data were produced with MATLAB version R2024a Update 2 and the CPFloat version under the SHA 0ddba87dd19adf5da154e228b68f86e60051f102. Run the script experiments.m to regenerate the data files in data or run the shell script plot.sh to regenerate the data files and plot the figures in diagrams.

References

[1] T. Mary and M. Mikaitis. Error Analysis of Matrix Multiplication with Narrow Range Floating-Point Arithmetic. hal-04671474. Aug. 2024.

License

This software is distributed under the terms of the 2-clause BSD software license (see LICENSE).

The CPFloat C library is distributed under the GNU Lesser General Public License, Version 2.1 or later.