Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp Path: utzoo!watmath!clyde!bonnie!akgua!whuxlm!harpo!decvax!decwrl!sun!gnu From: gnu@sun.uucp (John Gilmore) Newsgroups: net.arch Subject: Re: A feature, not a bug? Message-ID: <2362@sun.uucp> Date: Fri, 28-Jun-85 05:41:27 EDT Article-I.D.: sun.2362 Posted: Fri Jun 28 05:41:27 1985 Date-Received: Mon, 1-Jul-85 06:34:25 EDT References: <1680@amdcad.UUCP> <36900007@ima.UUCP> Organization: Sun Microsystems, Inc. Lines: 16 John Levine, ima!johnl, said: > Sounds pretty smart to me. Why waste chip real estate with locking circuits > that'll be used .0001% of the time? I expect that the WE 32100 chip > special-cases the interlocking between branches and the tests they depend > on. For that matter, the 360/91 did that 15 years ago. Of course, the /91 had to do it while running object code that ran on all the other models. It therefore needed the wasteful locking circuits. I agree that on a new machine, the frequency of "read condition codes" is so small that effectively making it a double-size opcode (NOP,READCONDITIONCODES) is a win. Of course, when they build a chip that runs two instructions simultaneously, they'll need those ubiquitous locking circuits again for object code compatability with today's chips.