Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!bloom-beacon!athena.mit.edu!tada
From: tada@athena.mit.edu (Michael J Zehr)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Compaq find problem with chip
Summary: description of problems
Keywords: defect in all '486 chips thus far!
Message-ID: <15533@bloom-beacon.MIT.EDU>
Date: 31 Oct 89 20:02:20 GMT
References: <443@ssp1.idca.tds.philips.nl> <2719@ganymede.inmos.co.uk> <2725@ganymede.inmos.co.uk> <2591@munnari.oz.au> <788@awdprime.UUCP>
Sender: daemon@bloom-beacon.MIT.EDU
Reply-To: tada@athena.mit.edu (Michael J Zehr)
Organization: Massachusetts Institute of Technology
Lines: 31

In article <443@ssp1.idca.tds.philips.nl>, dolf@idca.tds.PHILIPS.
Grunbauer) writes:
> What is the actual problem ? What chips 80486 and/or 80487 ?
> Could someone give more details or did they only say there is a problem
> and not what the precise problem is ?

(we're probably going to get hundreds of postings on this, but i haven't
seen any here yet so i'll post this anyway...)

according to Info World (vol 11, no 44, oct. 30):

there are two problems the first occurs only when doing a tangent and
"1) a particular rounding mode is used; [it doesn't say which one] 2)
mantissa data overflows; 3) the result is an exact integer power of two;
and 4) one of a particular set of floating instructions follows within
67 clock cycles.  if all of these conditions are met then the floating
point stack can become corrupted."

"the second problem concerns the special exception handling operation,
sometimes referred to as special error detection. ... [most exceptions
are handled by the application, but if] a programmer has chosen to
unmask the operation, allowing it instead to pass through to the math
processor, and a wait cycle occurs, the exception will be missed and an
incorrect value will be returned."

does anyone have any information on how the error was discovered?  it
sounds like a pain for the first problem because most attempts to debug
a problem section of code would affect what is executed in the 67 cycle
window and make the problem disappear....

-michael j zehr