At 6:07 Uhr -0600 20.2.2003, Frederick Bruckman wrote:
>> Maybe I have misunderstood the issue, but ISTR that kernels are built with
>> --soft-float so that fpu state does not have to preserved, since the kernel
>> does not touch the fpu registers.
>
>That doesn't make sense to me. What would happen after returning
>control to a different process? From examing mac68k's
>locore.s:savectx(), I gather that we do, in fact, save the FP state
>with a context switch.

Well, if you've actually looked at the code, that gives you a clear
advantage.  ;) I was probably thinking of context switches inside the
kernel (interrupts). But, to borrow the .sig of a NetBSD developer, "I
probably don't know what I am talking about".

>For what it's worth, I've been running kernels built with no
>soft-float on my Q840AV and Q630 since November, with no unusual
>problems, albeit they're not doing much real work (occasionally
>compiling packages, and one's serving a printer which has no ink).

That's strong evidence, then. If the situation had been anything like I
suspected, a kernel without --soft-float would have fallen flat on the face.

	hauke

--
/~\  The ASCII Ribbon Campaign
\ /    No HTML/RTF in email
 X     No Word docs in email
/ \  Respect for open standards