> 
> On Tue, 18 Mar 1997 23:37:26 -0600 (CST),
> Scott Reynolds <scottr@og.org> wrote:
> > On Wed, 12 Mar 1997, Allen Briggs wrote:
> > > 
> > > This is in the kernel compilation to make sure that there are no floating
> > > point instructions in the kernel.
> > 
> > Underline this double.  As a matter of principle, the kernel should avoid
> > all floating point instructions, always.
> 
> Not that I have any objection to this principle, but what's the
> rationale?  Or have I not tried hard enough to read through the Daemon
> book... ;-)

If there are no kernel fp instructions, among other things we don't have
to save fp state on entering kernel code, either for an interrupt or
for a trap. Since 6888X state can sometimes be HUGE, we win.

For interest: the problem is saving state in the middle of a long
instruction.  There are some points in FPU instructions where the time
needed to finish an instruction and just save FPU registers is shorter
than the time to save mid-instruction state, so the FPU finishes
the instruction before saving state.

Take care,

Bill