Xinerama infrastructure

[Steve Salazar]

Ok, sorry.  Like I said I don't know X except as a user so I'll summarize 
from that point of view and not even try to use any X jargon:

1. When I use XFree on one head, it is fast
2. When I use XFree on two heads, it is really slow
3. When I use Xig on two heads, it is fast

Done.

Oh, by "fast" I don't just mean glxgears.  I also mean glxgears but mainly I 
am talking about normal desktop performance of things like scrolling web 
pages, dragging windows around, minimizing and maximizing, etc.  I'm not a 
gamer so the desktop is the part I am really asking about.  I also use 
several cad visualization tools that are really really slow on dual head 
XFree and fast on dual head Xig.

Here is where I got the notion that "acceleration" was disabled for Xinerama 
in XFree:

    http://dri.sourceforge.net/cgi-bin/moin.cgi/Xinerama

I guess I didn't realize that "accelaration" was not the same thing as 
having DRI enabled which is what it actually looks like the issue is.

Anyway, now that that is cleared up (I know no one initially was really 
interested in my personal desktop issues), my original question still 
stands:  Is this performance deficiency something that is likely to also be 
a problem with Xserver?  Also, again using the "I know nothing" disclaimer, 
a couple replies indicated that the dual head performance was bad because of 
the doubling of traffic on the vga bus but just to clear that up, the Xig 
server has very good performance in dual head mode.

Thanks again.

Also, it has never been clear to me if Xserver is really being developed as 
an ultimate replacement for XFree (or alternative) or if it is just an 
experimental server to develop new extensions?  A lot of people are assuming 
(and are exited) that in a year or so the fd.o xserver will replace XFree.  
Is that true?


>From: "Mike A. Harris" <mharris@www.linux.org.uk>
>To: Benjamin Herrenschmidt <benh@kernel.crashing.org>
>CC: Jaymz Julian <jaymz@artificial-stupidity.net>,   Steve Salazar 
><eagsalazar@hotmail.com>, xserver@freedesktop.org
>Subject: Re: Xinerama infrastructure
>Date: Mon, 16 Feb 2004 08:43:44 -0500 (EST)
>
>On Mon, 16 Feb 2004, Benjamin Herrenschmidt wrote:
>
> >> (to be fair, on certain cards, in my specific case the matrix g400,
> >> xinerama is indeed as slow as molasses, for as yet unknown reasons -
> >> although it's nice to know that it's tied only to certain cards, and
> >> not the entire universe :)
> >
> >I suspect the biggest reason is that by having 2 outputs, you
> >double the load on the card's bus, thus slowing down everything
>
>Yes, but this isn't any weakness in X, or in the video drivers
>per se.  The person's email implies that Xinerama itself needs to
>be somehow accelerated, which makes no sense.  2D graphics are
>what is accelerated or not.  If Xinerama is used, then 2 or more
>screens are being used.  Yes the bandwidth over the bus may
>double, but it's _still_ being 2D accelerated.
>
>I don't understand the concept of "accelerated Xinerama", unless
>perhaps merged framebuffer pseudo-xinerama is faster somehow for
>2D.
>
>What exactly is "accelerated Xinerama" supposed to mean?  Unless
>we have a clear technical explanation of what _specifically_ is
>meant by that in technical terms, it is just guesswork as to what
>the original poster means really.
>
>For example, it is not like Windows drivers are going to do
>accelerated graphics to two displays any faster than X can
>becaues of some magic equivalent of whatever "Xinerama
>acceleration" is supposed to mean.  Speaking purely in terms of
>2D acceleration - Xinerama is already accelerated unless a given
>driver disables it, or is broken in Xinerama mode.
>
>Perhaps I just don't grok the non-technical aspect of the
>discussion however.  ;o)
>
>--
>Mike A. Harris
>

_________________________________________________________________
Find great local high-speed Internet access value at the MSN High-Speed 
Marketplace. http://click.atdmt.com/AVE/go/onm00200360ave/direct/01/