Vectorization of Multibyte Floating Point Data Formats

Andrew Anderson, David Gregg

We propose a scheme for reduced-precision representation of floating point data on a continuum between IEEE-754 floating point types. Our scheme enables the use of lower precision…