These MATALB codes provide efficient and accurate way to simulate low precision floating point arithmetic, namely fp16 and bfloat16. They also provide options to use subnormal numbers as well.

• N. J. Higham, S. Pranesh. Simulating Low Precision Floating-Point Arithmetic

• Test.m Generates all the tables in the manuscript into the folder results/.

• The codes have been developed and tested with MATLAB 2018b.
• This code requires fp16 class from Cleve's Labaratory. It can be downloaded from https://uk.mathworks.com/matlabcentral/fileexchange/59085-cleve_s-laboratory
• codes from https://github.com/higham/chop

• Nicholas. J. Higham
• Srikara Pranesh