Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!ames!lll-winken!uunet!lotus!esegue!johnl
From: johnl@esegue.segue.boston.ma.us (John R. Levine)
Newsgroups: comp.arch
Subject: Re: Filling branch delay slot with test
Message-ID: <1989Sep5.145727.2311@esegue.segue.boston.ma.us>
Date: 5 Sep 89 14:57:27 GMT
References: <1432@atanasoff.cs.iastate.edu> <26859@winchester.mips.COM> <1437@atanasoff.cs.iastate.edu>
Reply-To: johnl@esegue.segue.boston.ma.us (John R. Levine)
Distribution: na
Organization: Segue Software, Cambridge MA
Lines: 19

In article <1437@atanasoff.cs.iastate.edu> hascall@atanasoff.cs.iastate.edu.UUCP (John Hascall) writes:
>     No.  What I was alluding to was "starting down both paths" of the
>     branch and then "dumping the loser".

Back in 1969, the 360/91 prefetched both paths.  Since it had a prefetch
queue but no cache, it had to prefetch both in order to be sure it had
something to do after the branch.  Like a lot of other stuff the /91 did
(most notoriously executing instructions out of order and punting if one of
them caused an exception) people seem to have found other ways to get high
performance with less complication.  Much of the /91 technology was
inherited from the STRETCH, so it may have prefetched both ways as well.

Of course, "high performance" is relative.  The /91 I used had a 60ns cycle
time and 2MB of RAM, while the desktop machine on which I am typing this has
a 40ns cycle time and 4MB, soon to be 8MB, of RAM.  It has a cache, too.
-- 
John R. Levine, Segue Software, POB 349, Cambridge MA 02238, +1 617 492 3869
{ima|lotus}!esegue!johnl, johnl@ima.isc.com, Levine@YALE.something
Massachusetts has 64 licensed drivers who are over 100 years old.  -The Globe