Comment by Taniwha

Yup - I've just finished bringing up some FP hardware - you can't do all the 64-bit double space but you can do all the 32-bit single op instructions (converts, sqrt, etc) - but add/sub/multiple (32x32) are not doable

But this is even more appropriate for 16-bit floats - you can realistically do all the adds/multiplies/etc, but maybe not the 3op multiply-accumulates