Hi,
I am using wine to play World of warcraft (this game is using opengl, and is threaded - 2/3 threads IIRC). I did recently find a reproductable performance regression on versions 0.9.24+.
Under the same conditions (always difficult to get in an MMORPG game, but this is reproductible), version 0.9.23 gives me 20/25% more FPS than versions 0.9.24->0.9.26 (have not tried .27 yet).
- Is this performance regression known/expected on opengl/threaded applications (new codebase) ? - Is someone using WOW here (and has found a similar preformance regression) ? - It is worth for me to start looking at it (more measures, throw oprofile at it, bissect the change...)
Thanks,
Norbert
Hi,
I am using wine to play World of warcraft (this game is using opengl, and is threaded - 2/3 threads IIRC). I did recently find a reproductable performance regression on versions 0.9.24+.
Under the same conditions (always difficult to get in an MMORPG game, but this is reproductible), version 0.9.23 gives me 20/25% more FPS than versions 0.9.24->0.9.26 (have not tried .27 yet).
- Is this performance regression known/expected on opengl/threaded
applications (new codebase) ?
- Is someone using WOW here (and has found a similar preformance
regression) ?
- It is worth for me to start looking at it (more measures, throw oprofile
at it, bissect the change...)
Thanks,
Norbert
I have done lots of OpenGL work lately. I have mainly moved some code from opengl32.dll over to gdi32.dll and winex11.drv. I don't think I have done anything heavy between 0.9.23 and 0.9.24 which might have killed performance. A regression test could be usefull. Are you sure that the regression didn't happen between lets say 0.9.24 and 0.9.25? In there I changed the location of the DisabledExtensions registry key from hkey_local_machine to hkey_current_user. When you disabled VBOs using this extension you could gain more performance on some videocards+drivers. I guess that you are seeing this.
Regards, Roderick
Hi,
I have done lots of OpenGL work lately. I have mainly moved some code from
opengl32.dll over to gdi32.dll and winex11.drv. I don't think I have done anything heavy between 0.9.23 and 0.9.24 which might have killed
performance.
A regression test could be usefull. Are you sure that the regression
didn't happen
between lets say 0.9.24 and 0.9.25?
Well, I am investigating this now (see below).
I remember staying at .23 because when .24 was out the degradation was quite noticable.
I am now building the .23, .to .27 trees with the exact same compiler, options (-fno-stack-protector -O2), driver (nv 9xxx), wow configuration, and without any tree changes (I did hack opengl_ext.* since some time for performance). I keep you informed.
Thanks,
Norbert