Path: utzoo!attcan!uunet!aplcen!uakari.primate.wisc.edu!zaphod.mps.ohio-state.edu!swrinde!ucsd!ucbvax!decwrl!shelby!portia!dhinds
From: dhinds@portia.Stanford.EDU (David Hinds)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Replacing an 80386 with an 80486
Message-ID: <9729@portia.Stanford.EDU>
Date: 2 Mar 90 18:59:43 GMT
References: <29108@amdcad.AMD.COM> <1640057@hpspcoi.HP.COM> <9830@sequoia.UUCP> <1304@watserv1.waterloo.edu>
Sender: David Hinds <dhinds@portia.stanford.edu>
Organization: Stanford University
Lines: 22

In article <1304@watserv1.waterloo.edu>, alexew@watserv1.waterloo.edu (Alex E. Wielhouwer) writes:
> 
> Please correct me if I'm wrong, but in a multi-tasking environment, the 
> floating point registers and stack need to be handled seperately on a
> '386/'387 system during task swaps, or, in some cases the '387 is ignored.
> (OS/2?). With a '486 environment, this should be much more efficient, i.e.,
> faster.

    During task switches, the 386 saves its own context, but does not
automatically save the floating point processor's context.  It generates
an interrupt at the first floating point operation following a task switch,
so the system software can handle switching 387 contexts.  As this feature
was not present in the 80286, OS/2 probably doesn't know about it.  OS/2
may do it manually whenever a task switch happens, or never do it - I don't
know.  The 386 method has the advantage of not unnecessarily saving and
restoring the 387 context for tasks that don't use floating point.  I don't
know how the 486 does this, but one of the "new features to improve
multitasking" may be to quickly save/restore the floating point context
automatically.

 -David Hinds
  dhinds@popserver.stanford.edu