From: Holger Muenx (muenx@heike.informatik.uni-dortmund.de)
Date: 03/13/93


From: muenx@heike.informatik.uni-dortmund.de (Holger Muenx)
Subject: Re: X and math co-processor; A Speed up?
Date: 14 Mar 1993 00:10:58 GMT

Hello!

Posting on Saturday night -- seems as if I should get a life!

In article <1nq49tINNp0@srvr1.engin.umich.edu>, lcd@umcc.ais.org (Leon Dent) writes:
|> I run XFree86 1.2 under Linux 99p6 on a 386sx-25 with 8 megs. On the
|> whole I'm satisfied with performance (since I can't afford faster!).
|> But after watching two xterms try to keep up with scrolling (one at 2400bps,
|> the other with a sort output) I began to wonder if the addition of
|> a math chip would be a noticeable addition (i.e. in regards to scrolling).

I've upgraded my VGA graphics adapter from ET3000 to ET4000 since X was not
usuable with ET3000 (slow scrolling in xterm etc). There was a reasonable
speed up. I measured the xstones on my system (386/33/128k 8MB) yesterday:

        TOTAL 51530 lineStones
        TOTAL 9430 fillStones
        TOTAL 5119 blitStones
        TOTAL 479224 arcStones
        TOTAL 32312 textStones
        TOTAL 18300 complexStones
        TOTAL 13468 xStones

With this configuration X is definately usuable. It is, judged by "feeling"
and xstones, comparable to a _monochrome_ NCD16 X terminal (68k based).

If you have a fast graphics adapter and still there is slow scrolling you
should examine the global resources (ie. your files in the app-defaults dir
and the .Xdefaults). It is necessary to switch the jump scroll mode
of xterm on (see the xterm man page).

I think a math coprocessor would not speed up the performance of X. In
particular, scrolling involves movement of big memory areas which cannot be
accelerated by a coprocessor.

Holger Muenx (muenx@heike.informatik.uni-dortmund.de)