Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!ucla-cs!marc
From: marc@ucla-cs.UUCP
Newsgroups: comp.arch
Subject: Re: Branch prediction in the 532
Message-ID: <5662@shemp.UCLA.EDU>
Date: Thu, 23-Apr-87 12:19:08 EST
Article-I.D.: shemp.5662
Posted: Thu Apr 23 12:19:08 1987
Date-Received: Sat, 25-Apr-87 05:16:24 EST
References: <324@dumbo.UUCP> <165100008@uiucdcsb>
Sender: root@CS.UCLA.EDU
Reply-To: marc@CS.UCLA.EDU (Marc Tremblay)
Organization: UCLA Computer Science Department
Lines: 30

In article <165100008@uiucdcsb> robison@uiucdcsb.cs.uiuc.edu writes:
>
>Are there any good papers on branch prediction?  I've seen two schemes:
>
>	1) Put an extra bit in the instruction indicating which way the branch
>	   will probably go.  This is a compile-time prediction.
>
>	2) Keep a record of past branches, i.e. run-time prediction.   
>
>Though I've heard of examples of machines with each scheme, I've never
>seen any studies of their efficiency.
>
>Arch D. Robison
>University of Illinois at Urbana-Champaign
>	

The best paper I have read about branch prediction is:

"Branch Prediction Strategies and Branch Target Buffer Design"
Johnny K. F. Lee, Hewlett-Packard
Alan Jay Smith, UCB.
Computer (IEEE), January 1984, pp 6-22.

Also you might want to look at the architecture issues concerning the
Branch Target Cache of the Am29000, they claim a 60% hit ratio.

			Enjoy,
					Marc Tremblay
					UCLA Computer Science Dept.