Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!gem.mps.ohio-state.edu!ginosko!uunet!mcsun!mcvax!dik
From: dik@cwi.nl (Dik T. Winter)
Newsgroups: comp.arch
Subject: Re: (Im)precise exceptions
Message-ID: <8444@boring.cwi.nl>
Date: 5 Oct 89 01:24:41 GMT
References: <2353@oakhill.UUCP> <261500010@S34.Prime.COM> <34701@apple.Apple.COM> <2451@wyse.wyse.com> <477@ctycal.UUCP> <265@ssp1.idca.tds.philips.nl> <4796@orca.WV.TEK.COM> <27659@shemp.CS.UCLA.EDU> <269@ssp1.idca.tds.philips.nl> <28764@winchester.mips.COM> <33004@
Organization: CWI, Amsterdam
Lines: 33

In article <33004@ames.arc.nasa.gov> lamaster@ames.arc.nasa.gov (Hugh LaMaster) writes:
 > Of course, the CDC 7600 had not
 > only imprecise exceptions, but also "delayed exceptions" (is there a
 > name) when certain floating point conditions produced an indefinite,
 > which was an *error* at the point you *used* it in another operation,
 > millions of instructions later...

Yes that could be an headache.
 > 
 > I don't remember imprecise exceptions being that big a deal, although
 > it did create a small cottage industry for debugging certain difficult
 > programs :-)
 > 
Depends on how imprecise your exceptions are.  If you are able to determine
by means of a core dump and other information whereabouts the exception did
occur, there is no problem.  It is difficult when the exception occurs in
such a way that there is no means to determine even approximately the place
where it did occur.  On the Cybers one such exception was jump to address 0
(ARITHMETIC ERROR MODE=0 for those who speak Cybernese).  In most cases one
of the two standard places where the address of the offending instruction
ought to be stored was correct.  But there were cases when this was 0 too,
in both places.  I have never found a reasonable way to debug such a program.
A similar thing can occur (I am told) on IBM 360's.  It is called a register
collapse, which may mean that all registers and the PC are invalid.

In any case, for debugging purposes it is not a big deal if the instruction
counter can be a bit wrong (or a few bits), as long as it points roughly to
the offending place.  It gets a bit more difficult if a few jumps are taken
in between, but it is still do-able.
-- 
dik t. winter, cwi, amsterdam, nederland
INTERNET   : dik@cwi.nl
BITNET/EARN: dik@mcvax