Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site cmcl2.UUCP
Path: utzoo!linus!decvax!harpo!floyd!cmcl2!kenner
From: kenner@cmcl2.UUCP
Newsgroups: net.math
Subject: Re: REPOSTING: the Register Exchange pro - (nf)
Message-ID: <33778@cmcl2.UUCP>
Date: Sun, 9-Oct-83 12:57:30 EDT
Article-I.D.: cmcl2.33778
Posted: Sun Oct  9 12:57:30 1983
Date-Received: Mon, 10-Oct-83 08:42:33 EDT
Sender: kenner@cmcl2.UUCP
Organization: New York University
Lines: 24

#R:sunybcs:-49100:cmcl2:27800001:000:1123
cmcl2!kenner    Oct  9 12:57:00 1983

This is actually another piece of CDC folklore.  It was discovered by Jack
Schwartz at NYU very early on in the life of the machine (it was likely
also independantly discovered by others).  In hindsight, it's rather simple,
though the code is indeed long.

Clearly, once you have saved the value of A6 (or A7), you are done.  The
trick is how to save that register.  If you think carefully, you will see
that there is only one other instruction than the SA6 instructions which
modifies memory -- the return jump instruction!

So the code goes as follows:  Pick a B register, say B1.  If the high bit of
B1 is set, do an RJ to a subroutine which simply returns.  Otherwise, skip
the RJ.  Then rotate B1 left one bit (by adding it to itself), test, and, if 
the sign bit is on, RJ to another similar subroutine.  In this manner, all 18
bits of B1 can be tested. 

Next, simply copy A6 to B1.  Then save all the other registers.  Finally,
reconstruct the contents of B1 by seeing which "subroutines" had their entry
word changed (and re-zero that entry word if it was called so the register
save routine can be called again).