Path: utzoo!attcan!uunet!snorkelwacker!usc!samsung!munnari.oz.au!metro!ipso!runxtsa!applix
From: applix@runxtsa.runx.oz.au (Andrew Morton)
Newsgroups: comp.sys.att
Subject: Re: An MGR modification & bug list
Summary: New MGR bitblit code
Keywords: MGR
Message-ID: <901@runxtsa.runx.oz.au>
Date: 11 Apr 90 02:17:52 GMT
References: <879@runxtsa.runx.oz.au> <35720@brunix.UUCP>
Organization: RUNX Unix Timeshare.  Sydney, Australia.
Lines: 33

In article <35720@brunix.UUCP>, cs132074@cs.brown.edu (Wildebeest) writes:
> In article <879@runxtsa.runx.oz.au> applix@runxtsa.runx.oz.au (Andrew Morton) writes:
> >My version of MGR runs on an Applix 1616 68000 system
> .
> .
> .
> >I ended up rewriting the bitblit code from scratch in assembler.
> 
> Uh-huh.  Do you think you could get around to posting this little
> tidbit?  The Unix-PC is a 68010 machine...

	I was afraid this would happen :-)

	My new bitblit code gives about a 2:1 performance increase
over the original (oblit/*) code on simple character output and
a similar increase in scrolling performance. The inner loop of
the scroll code consists of:

	move.l	(a0)+,(a1)+
	move.l	(a0)+,(a1)+
		.
		.

	etc, which I reckon is about as quick as it will get.

	I blushingly admit to having used self-modifying code to
implement the 16 different pixel processing modes, but this can
be simply avoided.

	Converting it to run on the Unix PC will take some effort, so
I think the best approach will be for someone who is familiar with
the Unix PC version of MGR to mail me and we can privately get the new
stuff up and running, then distribute it.