TeXmacs/Qt on intel graphics (Re: [Texmacs-dev] Two fixes for delayed execution)

[Norbert Nemec]

Alex D wrote:
> If I understand correctly, the -graphicssystem raster switch tells the 
> Qt library to use a different graphics backend that performs rendering 
> and text rasterization in software (on the client) instead of relying 
> on the server (X11).
>
> I heard that these performance issues were due to performance 
> regressions recently introduced into the Intel graphics drivers. 
> However, the upcoming version of Ubuntu, Karmic Koala, introduces 
> newer drivers which are supposed to fix all these issues. I'll do some 
> testing when it comes out later this month.

I just checked: my system is already running with the new UXA setting. 
It does not solve the problem.

Running TeXmacs/Qt without switches, even simple cursor movements 
through a plain text bring the CPU load of the X server up to 98% and 
TeXmacs becomes painfully sluggish.

Setting the "-graphicssystem raster" command line option fixes the problem.

Greetings,
Norbert



> Setting the switch is the only way to force graphicssystem raster for 
> a particular application short of recompiling the libraries. This can 
> be done with a wrapper script.
>
> --Alex Dobkin
>
> On Wed, Oct 14, 2009 at 11:19 AM, Norbert Nemec 
> <address@hidden <mailto:address@hidden>> wrote:
>
>     Wow, indeed! X server load goes down to 30% when I move through a
>     long text. Sluggishness is completely gone...
>
>     What does that switch do? Could it be set permanently? Does it
>     affect only Intel graphics hardware?
>
>     Greetings,
>     Norbert
>
>
>
>     Alex D wrote:
>
>         Hi,
>
>         In case anyone is runing the Qt port on Intel graphics
>         hardware on Ubuntu or similar systems and experiencing
>         sluggish performance, a suggestion is to run the application
>         with the '-graphicssystem raster' switch enabled. This yields
>         greatly improved performance in my case.
>
>         --Alex Dobkin
>
>         On Wed, Oct 14, 2009 at 5:04 AM, Gubinelli Massimiliano
>         <address@hidden <mailto:address@hidden>
>         <mailto:address@hidden <mailto:address@hidden>>>
>         wrote:
>
>            Hi,
>
>
>            On 13 oct. 09, at 22:32, Norbert Nemec wrote:
>
>                In fact, I found a clear measurement that indicates a
>         problem:
>
>                Running TeXmacs/Qt on my fairly modern notebook (fully
>                optimized compile).
>
>                I fill one page with plain text. Now simply moving
>         through the
>                text with the right arrow (no editing!) brings up the
>         CPU load
>                to 100%. Of this, 95% are spent in the X server. TeXmacs
>                itself only takes 5% of the CPU power.
>
>
>            I can confirm this observation. In my ubuntu virtual machine
>            (VirtualBox/MacOSX) the above protocol gives \sim 75% CPU
>         load to
>            Xorg and \sim 23% to texmacs. You should check that you do
>         not use
>            cairo since at the moment the cairo rendering is slightly
>            unoptimal in QT (explicit and naive double-buffer).
>            On my mac the load remains reasonable (30% which should include
>            the GUI load also since the total CPU load is not much higher).
>
>            Proprer implementation of check_event require knowing if
>         there are
>            pending key events in the queue and this is not possible in
>         Qt (or
>            even in Cocoa). Moderns GUI API explicitly discurage looking at
>            the event queue so we should design a different mechanism to
>            reimplement what it is currently done in the X11 port.
>
>
>            best
>            max
>
>
>
>                I tried activating the code in check_event. The display is
>                incomplete now, but the CPU load is down to 30% and TeXmacs
>                feels fully reactive.
>
>
>                Seems like this optimization is crucial not only for slow
>                machines after all...
>
>
>
>
>
>                Greetings,
>                Norbert
>
>
>
>
>                Joris van der Hoeven wrote:
>
>                    On Tue, Oct 13, 2009 at 05:14:02PM +0200, Álvaro Tejero
>                    Cantero wrote:
>
>                        I didn't notice any problems, but TeXmacs-QT feels
>                        slower already when
>                        typing at the end of 3-line paragraphs.
>
>
>                    That is shortcoming of the current Qt version:
>                    interrupts during the rendering phase have not yet been
>                    implemented.
>                    Now that delayed_event has become cleaner, maybe
>         Max can
>                    fix that one too.
>
>
>                        Is there any more quantitative way to test
>         performance?
>
>
>                    This is rather a matter of reactivity. Hard to measure.
>
>                    Best wishes, --Joris
>
>
>                    _______________________________________________
>                    Texmacs-dev mailing list
>                    address@hidden <mailto:address@hidden>
>         <mailto:address@hidden <mailto:address@hidden>>
>
>                    http://lists.gnu.org/mailman/listinfo/texmacs-dev
>
>
>
>
>
>                _______________________________________________
>                Texmacs-dev mailing list
>                address@hidden <mailto:address@hidden>
>         <mailto:address@hidden <mailto:address@hidden>>
>
>                http://lists.gnu.org/mailman/listinfo/texmacs-dev
>
>
>
>
>            _______________________________________________
>            Texmacs-dev mailing list
>            address@hidden <mailto:address@hidden>
>         <mailto:address@hidden <mailto:address@hidden>>
>
>            http://lists.gnu.org/mailman/listinfo/texmacs-dev
>
>
>         ------------------------------------------------------------------------
>
>         _______________________________________________
>         Texmacs-dev mailing list
>         address@hidden <mailto:address@hidden>
>         http://lists.gnu.org/mailman/listinfo/texmacs-dev
>          
>
>
>
>
>     _______________________________________________
>     Texmacs-dev mailing list
>     address@hidden <mailto:address@hidden>
>     http://lists.gnu.org/mailman/listinfo/texmacs-dev
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Texmacs-dev mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/texmacs-dev
>