Return to BSD News archive
Path: sserve!manuel.anu.edu.au!munnari.oz.au!news.hawaii.edu!ames!olivea!uunet!psinntp!dg-rtp!ponds.uucp!rivers
From: rivers@ponds.uucp (Thomas David Rivers)
Newsgroups: comp.unix.bsd
Subject: Re: [386bsd] installing fpu (ULSI) results in floating exceptions
Message-ID: <1992Dec1.124507.10959@ponds.uucp>
Date: 1 Dec 92 12:45:07 GMT
References: <1992Nov30.060552.7912@kithrup.COM> <ByJIpL.69w@unx.sas.com> <1992Nov30.190622.2701@kithrup.COM>
Lines: 49
In article <1992Nov30.190622.2701@kithrup.COM> sef@kithrup.COM (Sean Eric Fagan) writes:
>In article <ByJIpL.69w@unx.sas.com> sastdr@torpid.unx.sas.com (Thomas David Rivers) writes:
>> Yes, apparently the definition ULSI followed doesn't strictly imply
>> superset. That's the flaw.
>
>The '387-only instructions are the transcendental instructions.
>
>I have an IIT 387 in my machine, and I guarantee that it executes the entire
>suite of instructions.
>
>--
>Sean Eric Fagan | "we will probably just crash immediately; but at least
>sef@kithrup.COM | we will have written less code."
>-----------------+ -- Chris Torek (torek@ee.lbl.gov)
>Any opinions expressed are my own, and generally unpopular with others.
Sean -
I don't mean to debate this here; but I must reply that I had direct
contact with ULSI; who gave me this information. There *are* apparently
some instructions available on a INTEL (or Cyrix since it's basically
the same mask) 80387 with which the ULSI part doesn't function.
ULSI was quite sure of this fact, and informed me that -
1) It worked under DOS.
2) SCO UNIX/Xenix initially discovered the problem (i.e. the
difficulty exists there as well.)
3) A new part would not be available until sometime late '93.
ULSI even offered to refund my money!
So, although I'm quite sure you are correct in your assertion - there must
be something going on here. Perhaps it's not the math instructions at all,
but some other instruction/hardware implementation (synch for process switch,
interrupt handling or something else altogether.)
If you would like further details, I encourage you to contact ULSI
directly, and ask for Roland.
- Dave Rivers -
(rivers@ponds.uucp (home))
(sastdr@unx.sas.com (work))
p.s. I can readily demonstrate the problem in 386bsd with some compiles;
everything runs fine until I try to build something which contains
floating point constants (i.e. actually *use* the FPU.)