@matrix
Sorry - never used 16 bit floating point.
For a few years I had to deal with fixed point arithmetic on 16 bit machines. You had so many bits of fraction (no exponent). You could do +/- 1 (roughly) with one sign bit and 15 bits of fraction.
Let's say you had a maximum speed of 30 mph. You could represent that with 5 bits of integer & 10 bits of fraction.
Do all that in assembly language and track the binary point by hand.