On Tue, 2009-04-21 at 23:57 +0800, Michael Clark wrote:
> Michel Dänzer wrote:
> > On Sat, 2009-04-18 at 13:46 +0800, Michael Clark wrote:
> >   
> >> I have made a port of gears that uses the OpenGL ES 1.1 Common API 
> >> subset (using glDrawElements instead of display lists and optionally 
> >> using vertex buffer objects). It should be visually equivalent to 
> >> gears.c, although the flat faces are rendered with GL_SMOOTH to avoid 
> >> additional draw calls.
> >>
> >> While testing it, I noticed a severe screen corruption issue when using 
> >> vertex buffer objects with the intel driver in ubuntu (both ubuntu 9.04 
> >> driver and ubuntu xorg-edgers 2.6.99.1+git20090416). I have verified the 
> >> code is working with the nvidia-180 proprietary driver on another system 
> >> so I believe it is an intel driver issue. I am wondering if the same 
> >> issue exists upstream...
> >>
> >> The attached gearsvbo.c source code can be compiled with:
> >>
> >>   gcc gearsvbo.c -o gearsvbo -lGL -lglut
> >>
> >> It can be run with or without vertex buffer objects (add -novbo command 
> >> line option to disable).
> >>
> >> With the Intel driver I see a 40% speedup with glDrawElements alone. 
> >> With vertex buffer objects enabled I see almost 200% speedup although 
> >> the front-buffer doesn't appear to being cleared or swapped correctly 
> >> (see attached png).
> >>     
> >
> > Fixed in Git.
> >   
> 
> Thanks. I just noticed the fix has arrived in ubuntu xorg-edgers as of  
> mesa_7.4.1~git20090420+mesa-7-4-branch.
> 
> Before trying this I built from mesa master earlier today and noticed a 
> performance regression with the 965GM on one of my GL apps - going from 
> 15fps to 7.5fps (both running on top of latest drm-intel 2.6.30-rc2). 
> 7.4.1 doesn't exhibit the slow down.
> 
> It seemed to be a timing issue as the app would no longer be using all 
> cpu of a single core (noticeably less cpu than on the 7.4-branch 
> version), and paradoxically, running another cpu intensive task would 
> result in the fps increasing. I'll do some more testing (and double 
> check the two builds were configured equivalently)...

bisecting would be the best thing to do if you've got a performance drop
and it's not a CPU-bound app.

-- 
Eric Anholt
e...@anholt.net                         eric.anh...@intel.com


Attachment: signature.asc
Description: This is a digitally signed message part

------------------------------------------------------------------------------
Stay on top of everything new and different, both inside and 
around Java (TM) technology - register by April 22, and save
$200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco.
300 plus technical and hands-on sessions. Register today. 
Use priority code J9JMT32. http://p.sf.net/sfu/p
_______________________________________________
Mesa3d-dev mailing list
Mesa3d-dev@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev

Reply via email to