Subject: Floating Point Woes, part 2
To: None <port-vax@netbsd.org>
From: Henry R. Bent <hbent@cs.oberlin.edu>
List: port-vax
Date: 12/09/2002 03:52:28
Hi all,

I've been investigating the problem I mailed to the list earlier about floating
point issues with the vax port.  I've got some more hard info, but things are a
little beyond my abilities to solve alone at this point.

Basically, here's what I can determine.  I'll use xmountains as my example
program: I compiled it from pkgsrc, so this should be easily duplicatable by
pretty much anyone.  On my 4000/90 the program dies almost instantly.  Running
gdb tells me that it's dying in /usr/lib/libm.so.0 trying to run a standard
atan().  The part that has me stumped, however, is that my 4000/200 has no such
problem.  I can run xmountains the first time, every time.  Both systems are
running release copies of 1.6 with customized kernels, but libm is the same on
both.

I suspect that there's something wrong with NetBSD's floating-point support on
the VAX that's chip dependent, but I have no way to prove this.  I would be very
interested to hear from other people running NVAX, SOC, or other systems to hear
if they do or do not experience this problem.  I think if we can pinpoint what
systems this is/isn't happening on then it will make debugging a lot easier.

-- 
Henry Bent
hbent@cs.oberlin.edu