Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!purdue!decwrl!jumbo!dillon
From: dillon@jumbo.dec.com (John Dillon)
Newsgroups: comp.arch
Subject: Re: Register Windows
Message-ID: <13381@jumbo.dec.com>
Date: 13 Oct 88 15:39:29 GMT
References: <46500028@uxe.cso.uiuc.edu> <2768@ima.ima.isc.com>
Distribution: na
Organization: DEC Systems Research Center, Palo Alto
Lines: 57

In article <2768@ima.ima.isc.com>, johnl@ima.ima.isc.com (John R. Levine) writes:
> In article <46500028@uxe.cso.uiuc.edu> mcdonald@uxe.cso.uiuc.edu writes:
> >What is a register window?

> It's a little piece of Lexan in the chip package of a microprocessor that lets
> you see into the machine's registers. Writers of optimizing compilers find
> these windows invaluable for debugging and tuning object code. For example, in
> case of register overflow you can see bits floating near the register that
> overflowed. In case of register collisions, which all compiler writers will
> agree are a big pain, if you examine the registers very closely with a
> magnifying glass, you can usually see little scratches and dents where the
> collision occurred.
> 
> Register windows are such a useful tool that it's a shame that the SPARC is
> the only commercial chip that provides them, although even on the SPARC the
> Lexan gets scratched after a few months, making them hard to see through. I
> wish that quartz windows weren't so expensive, they last a lot longer.
> -- 
> John R. Levine, IECC, PO Box 349, Cambridge MA 02238-0349, +1 617 492 3869
> { bbn | think | decvax | harvard | yale }!ima!johnl, Levine@YALE.something
> Rome fell, Babylon fell, Scarsdale will have its turn.  -G. B. Shaw


I must take Mr. Levine to task for forgetting the AMD 29000.

AMD goes to the trouble of providing the user with pictures of the
metallization, since metal shows the little scratches and dents much
better than silicon.

The 29000 does not provide the same register windows as SPARC.  Instead,
AMD adopted the neat strategy of trapping to supervisor code when
the register window overflows and underflows.

This strategy dovetails nicely with AMD's commitment to shadow registers
and serial scan.  The trap handler, using the shadow register (actually
a miniature on-chip ccd camera obscura), takes a picture of the
metallization and scans it out to the companion AMD 291984 image
processing chip, which reconstructs the picture for the user.

Consider the advantages and disadvantages (as I see them :-)
  + no need for either lexan or quartz window
  + the packaged part is insensitive to light
  + no need for a magnifying glass: you put the image on your display
  + only compiler writers need purchase the 291984
  - the 291984 is unlikely to be in your machine when you need it
  - it takes time to reconstruct and scan the image: no real-time window

No, I have no connection with either AMD or Sun.

   -- John

disclaimer
n.
1. the act of disclaiming; the renouncing, repudiating, or denying of a
   claim; disavowal.
2. a person who disclaims.
3. a statement, document, or the like, that disclaims.