[FIXED] Cairo performance regression?

Hi, as stated here:
http://www.nvnews.net/vbulletin/showthread.php?p=2418346
The cairo-1.10.0-buggy_gradients.patch shouldnt be needed anymore,
but i tried unpatched cairo 1.12.8 with nvidia drivers 310.14(beta),
304.60 and 304.51.
With those drivers gtk applications are very slow, even firefox is
slow when scrolling pages using native widgets.
Applying the cairo patch OR using nouveau make things smooth again.

Users complaining:
https://bbs.archlinux.org/viewtopic.php?id=152212

Bug report rejected by Arch developers:
https://bugs.archlinux.org/task/32491

Nvidia bug report
http://wpage.unina.it/aorefice/sharevari/nvidia-bug-report.log.gz

Please, i’d like to have a word on this issue by an nvidia developer.
No answer here, no answer to the email i sent to the nvidia linux-bugs@nvidia.com.

Shameless bump ;) Any information about this issue would be great.

I’ve made a simple presentation showing the difference for Phoronix forum user, and I was requested to post it here. http://www.youtube.com/watch?v=D1B2hqEOGJ8

Best regards.

The driver accelerates gradients whenever possible, but certain gradient configurations can’t be easily accelerated and fall back to software. I filed a bug to look into exactly why that’s happening and whether there’s something we can do to improve it, but Cairo could help the situation dramatically by sticking to gradients that the GPU can render.

@Aaron
Thanks for replying.

Since patched cairo (that relies entirely on the cpu) is much faster,maybe the “fallback to software” doesn’t work that good.

I’ve much better performance when using (underclocked/no reclocked) the same card via nouveau driver.
Even an old Sapphire Radeon 9500SE with ‘radeon’ driver performs better.

Yes, there are improvements we could make in the speed of the software fallbacks, but it will never be as fast as the fully hardware-accelerated path.

I may be missed it, but has this thing been investigated/fixed?
Thanks.

This should have been improved in the 319.* timeframe.

Ops, i missed your reply, sorry.
I just come here to say thanks, because after installing 325.15 and testing with a benchmark with 100 buttons and gradients in a scrollview,
plain cairo scrolled them in 14.7 seconds; while the patched cairo that bypasses any acceleration performed in 20 seconds, while it was faster than plain cairo in the past.
Firefox scrolling is ok too.

Hello Nvidia developers,
I am new to this forum, because I encountered a problem with a new Linux-PC.
As other users noted, sometimes the scrolling (eg. special HTML-pages in firefox)
is done VERY slow and the XServer completely frozen for up to 1 Minute!!!
I see with ‘top’ that Xorg consumes up to 100% cpu then.

This was never the case before on other PCs.

I use the newest NVidia driver version 2.1.20-3.1.1 (build time 27.1.2013)
and x11-video-nvidiaG03 version 319.32-15.1 (build time 28.6.2013)

Could you please fix it!!!

An easy test for whether Cairo’s “buggy gradients” patch is faster, is to open a new blank tab in firefox, and measure with our eyes how long it takes the “preview windows” (or whatever they’re called) to be drawn (I get 9 of them, arranged as 3x3 rectangles). This is still noticeably slower with Nvidia and unpatched Cairo.