On Sun, Oct 17, 1999 at 10:23:33PM -0400, Rick Byers wrote:
> I just bought an old Pentium 90, and it turns out it has the FDIV bug (I
> forgot all about that it's been so long).  What is the amount of
> compensation for the FDIV bug in NetBSD-1.4.1?  I searched the mailing
> lists and found some reference to gcc using a work around, is this still
> the case with EGCS (i.e. any code I compile shouldn't be a problem?).  I
> assume there is no binary level workaround other than disabling the FPU?
> 
> How serious is the bug anyway?  I don't expect to be doing much (if any)
> floating point stuff on the machine (it's basically just a NAT box).  Is
> it just a few rare circumstanances that result in slightly innacurate
> divisions, or is it a good portion of divisions are completely wrong?

NetBSD doesn't compensate for this one, as it is not possible to do so
for the operating system. Except maybe for using FP emulation and
checking every fdiv instruction, resulting in a significant slowdown..

I don't think that the workaround every became part of the gcc sources,
but I don't know that for a fact.

If the machine is just a NAT box, don't worry about it. It's very
unlikely that it'll show up in day-to-day operations. Just don't do
any serious FP calculations on it.

- Frank