This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
HELP ! Running LAPACK's selftests crashes my system !
- To: egcs at cygnus dot com
- Subject: HELP ! Running LAPACK's selftests crashes my system !
- From: Toon Moene <toon at moene dot indiv dot nluug dot nl>
- Date: Fri, 21 Aug 98 14:52:40 +0200
- Organization: Moene Computational Physics, Maartensdijk, The Netherlands
Heh, heh,
Now that I have your attention, let me explain how.
(this is on m68k-next-nextstep3, BTW).
Because everything on this 25 Mhz machine is an overnight job, I
compiled and ran LAPACK's libraries and self test programs during
the night. When I came back the next morning - system crashed.
Believing that it was just bad luck, I tried again last night:
crash.
So I restarted it this morning to watch what happened: crash.
Interestingly it only crashed during running the self tests, never
during compilation. Now that I watched it I saw that the system
actually panic'd, so I decided to take a look in /usr/adm/messages:
Aug 21 11:44:43 moene mach: panic: (Cpu 0) fpsp: frame format error
"fpsp frame format error" ?!?!?
Heck, I compiled everything with -fomit-frame-pointer !
Apparently, in the LAPACK libraries so many routines are actually
compiled without frame pointer, that this error (obviously a
interrupt or exception handler assuming that it always gets a valid
fp) happens often enough to be noticable.
Well, that's it, then - NEXTSTEP is a "hide-the-source" operating
system, so the bus stops here. I think I'll leave testing with
-fomit-frame-pointer on m68k to the NetBSD guys and Kate (who owns a
Sun3).
Cheers,
Toon.