"M. R. Brown" <mrbrown@0xd6.org> writes:

> >   Double-precision math must be emulated, if I understand correctly.
> > 
> 
> It's available in hardware.


It's much slower than single-precision though, so it might be
"emulated" in hardware using multiple single-precision operations.
Even a double precision add takes 6 cycles.  (All single precision
instructions except division, square root and vector multiplication
execute in a single cycle.)


  // Marcus