In some email I received from Christos Zoulas, sie wrote:
> On Jun 9,  8:33am, lennart@augustsson.net (Lennart Augustsson) wrote:
> -- Subject: Re: kern/1515 - Changing netstat variables to 64 bit
> 
> | Indeed, going from 32 to 64 bit counters should add one or two instructions
> | every time you increase the counter (add to the accumulator).  I don't see
> | how that can suck performance so badly.
> | 
> 
> It is not just two instructions, look in adddi3.c. Yes, on a pentium class
> machine this might not matter, but it surely does on an ss2 or any m68k
> boxes, when those instructions end up in the critical loops. I think that
> at this point the most productive thing to do is test the changes on a slow
> machine

Maybe they need adddi3.s then, which uses carry-bit, etc, to run faster if
that adddi3.c doesn't compile down fast enough ?

Darren