1001407 - Consider setting gfx.xrender.enabled to false on trunk for testing, possibly on a subset of hardware

Status: RESOLVED DUPLICATE of bug 1241832

(Core :: Graphics: Layers, defect)

Description

[A, T]

I've got two machines, one with Intel HD4000 one and the one I'm currently using to post this and mostly test this:

NOTE: OMTC and layers.acceleration.force-enabled are turned off, they make the browser unusable on Fx28.
So simply xrender is turned off. 

Graphics:  Card: Advanced Micro Devices [AMD/ATI] RS690M [Radeon Xpress 1200/1250/1270]
           Display Server: Fedora X.org 1.14.4 drivers: ati,radeon (unloaded: fbdev,vesa)
           Resolution: 1280x800@60.0hz
           GLX Renderer: Gallium 0.4 on ATI RS690 GLX Version: 2.1 Mesa 10.1.0
Both running the same Fedora system. 

I have recently disabled xrender and noticed rather large improvements in rendering pages, with image heavy pages or large images making me believe I'm on a different set of hardware than on the anemic RS690M ATi GPU, but still noticable on the much stronger i5-3570k with HD4000 graphics. 
The first thought that would come to mind by flipping this is image corruption, weirldy enough I haven't noticed anything of the sort in the last 5 days.
Now, this might just be good luck with my GPUs and the free drivers they use, I honestly have no idea how this would impact different hardware, like nVidia with nouveau or highend AMD ones with fglrx. 

Would it be possible enable this on the trunk on a 'safe' set of hardware and gauge what to do further from feedback and telemetry ? I believe Mozilla already has a blocklist of linux drivers/hardware that don't play nicely. 

The benefit should at least be worth the effort, a speed increase across the board on the linux platform. 

If you need about:support or glxinfo , just tell and I'll post what I have.

Comment 1

[Nicolas Silva [:nical]]

Our xrender code has been more trouble than help over the past 2 (at least) years, and it's linux-specific. If removing it is an improvement I'm all for doing it now (we might even take that decision for the sake of having one less linux-specific thing to worry about anyway). Milan, what do you think? if anything breaks we can just flip back the pref.

Comment 2

[Milan Sreckovic [:milan] (needinfo for best results)]

We're going to give bug 738937 (and bug 720523) a try, they are actively being worked on.