Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!amdcad!ames!hao!boulder!sunybcs!rutgers!iuvax!pur-ee!uiucdcs!uxc.cso.uiuc.edu!ccvaxa!aglew
From: aglew@ccvaxa.UUCP
Newsgroups: comp.arch
Subject: Re: register windows
Message-ID: <28200057@ccvaxa>
Date: Fri, 23-Oct-87 17:54:00 EST
Article-I.D.: ccvaxa.28200057
Posted: Fri Oct 23 17:54:00 1987
Date-Received: Wed, 28-Oct-87 20:36:05 EST
References: <201@PT.CS.CMU.EDU>
Lines: 29
Nf-ID: #R:PT.CS.CMU.EDU:201:ccvaxa:28200057:000:1361
Nf-From: ccvaxa.UUCP!aglew    Oct 23 15:54:00 1987


>>Henry Spencer:
>>Register windows may end up being a net win despite context-switch overhead.
>
>There isn't that much of a context-switch overhead, either.  Saving
>the register windows does not dominate the cost of a
>context-switch.  Context switches require hundreds of instructions,
>while a pipelined system should be able to save the register file
>in about 100-150 instructions.
>
>Greg Frazier

Why do context switches require hundreds of instructions? On Real-Time OSes
a context switch can require much less than a hundred instructions - and
that's without a CISCy dispatch instruction. Even on UNIX a context switch
can be made to occur in not much more time than a syscall (which is another
story).

The 'hidden' loss due to cache flushing, etc., can cost a lot, but on real
memory systems that shouldn't bother you.

Andy "Krazy" Glew. Gould CSD-Urbana.    USEnet:  ihnp4!uiucdcs!ccvaxa!aglew
1101 E. University, Urbana, IL 61801    ARPAnet: aglew@gswd-vms.arpa

I always felt that disclaimers were silly and affected, but there are people
who let themselves be affected by silly things, so: my opinions are my own,
and not the opinions of my employer, or any other organisation with which I am
affiliated. I indicate my employer only so that other people may account for
any possible bias I may have towards my employer's products or systems.