Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site brl-tgr.ARPA
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!talcott!harvard!seismo!brl-tgr!tgr!FIRTH@TL-20B.ARPA
From: FIRTH@TL-20B.ARPA
Newsgroups: net.lang.c
Subject: Loop unrolling
Message-ID: <9617@brl-tgr.ARPA>
Date: Sun, 31-Mar-85 08:28:24 EST
Article-I.D.: brl-tgr.9617
Posted: Sun Mar 31 08:28:24 1985
Date-Received: Wed, 3-Apr-85 00:39:39 EST
Sender: news@brl-tgr.ARPA
Lines: 21

The loop

    L1:
	movb (r2)+,(r1)+
	bneq L1

is indeed improved by unrolling:

    L1:
	movb (r2)+,(r1)+
	beql L2
	movb (r2)+,(r1)+
	bneq L1
    L2:

Each unrolling replaces a conditional branch taken by a conditional
branch NOT taken, which is a gain in speed

Robert Firth
Tartan Labs Inc
-------