[Intel-gfx] Performance regression using Firefox between 2.14 and 2.15

[Clemens Eisserer]

I was able to bisect the performance regression, the commit which
introduced it was:

commit 686018f283f1d131073ef5917213e6a8ac013f26
Author: Chris Wilson <chris at chris-wilson.co.uk>
Date:   Tue Apr 12 08:23:04 2011 +0100

    Turn relaxed-fencing off by default for older (pre-G33) chipset

Are there plans to re-enable relaxed fencing again?

Thanks, Clemens


2011/9/29 Clemens Eisserer <linuxhippy at gmail.com>:
> Hi,
>
> For quite some time now I experience slow scrolling on some pages when
> using FireFox,
> now I took the time to find the culprit.
> The slow scrolling only manifests when maximizing FireFox' window
> (screen size is 1920x1280),
> and manifests in "after-scrolling" (scrolling although mouse already stopped).
> During scrolling, X doesn't respond to new input.
>
> It seems with 2.15 there is a performance regression, causing some
> pages to scroll slow.
> On my system the following page scrolls slow:
> http://news.orf.at/stories/2081416/2081417/,
> especially if the small images on the top need to be repainted.
> Also
>
> During slow scrolling, intel_gpu_top reports a completly busy ring (0%
> idle), ring space at ~10%
> and the following load:
> Color clalculator: 99%
> Bypass FIFO: 99%
> Map filter: 97%
> Intermediate Z: 62%
> Windowizer: 55%
> Pixel Shader: 1%
> Setup engine: 1%
>
> My System is:
> - Fedora Core 15 (updated)
> - intel 2.15 & 2.16 (both show the slowdown, 2.14 was ok)
> - Intel 945GM
> - Core2Duo T7200 (2x2ghz, 4mb shared L2 cache)
>
> I pasted my Xorg.log to: http://pastebin.com/74NXbrTp
>
> Is there anything further I should try to track this problem down?
>
> Thank you in advance, Clemens
>