Bug 42891

Created attachment 53492 [details]
screenshot

Could be damage related, could be flush related. Or it might indeed be a whole new bug, only time will tell!

Another strange artifact I only saw with skype was, that text sometimes appeared bold for a short periode of time, while selecting it - will try to get a video of it.

video of test going missing when resizing the chat window: http://www.youtube.com/watch?v=LDHQONHy850

(most visible at the bottom, but happens on a couple of other places too).

video of bold looking text rendered while changing selection: http://www.youtube.com/watch?v=Y90ST9o77V8

Happens quite at the top of the selection at 11:22:06 "es ist ein normaler funktionsaufruf", as well as in the middle of the selection at 11:23:22 "in 5ms beantwortet"

I suspect I might have to actually break the old 945gm out of storage and use it for a while. I'm having difficultly, so far, reproducing these text related bugs on PineView.

I am pretty sure this is the same bug as bug 44929, I just changed QT's background color to something != white, and indeed the text is rendered, although white (with subpixel AA gone crazy).

Created attachment 55916 [details]
text with different background-color

ok sorry, false alarm, sometimes the text is bold/white but sometimes its still completly missing.

Ta, yes that would also explain it quite nicely. And having realised I made the basic error of not actually testing the CA paths recently on my PNV box, time to see I can trigger a few errors.

Created attachment 57420 [details]
text weirdness and flickering in oxygen-gtk-demo

With SNA enabled at driver compilation I experience this in most if not all GTK programs. Curiously it's not prevalent in any Qt4 applications. The address bar in Chromium is driving me spare. :>

The machine is running Kubuntu 11.10 with extra xorg packages from the xorg-edgers ppa. Notably;
  xserver-xorg-core=1.11.2.902+git20111209+server-1.11-branch.0ca8869e-0ubuntu0sarvatt~oneiric
  libpixman-1-0=0.24.4-1~oneiric1

intel driver and libdrm are from git and up to date as of Feb 21st.


From #intel-gfx@irc.freenode.net;
> [21:12] <ickle_> half-rendered CA glyphs
> [21:13] <ickle_> your guess is as good as mine why the second half of the glyph rendering only sometimes disappears
> [21:13] <zorael> all right :> thanks
> [21:30] <zorael> ickle_: is this bug 42891 (https://bugs.freedesktop.org/show_bug.cgi?id=42891), or should I file a new one?
> [21:31] <ickle_> yes, that's now the same issue
> [21:31] <zorael> now or not?
> [21:33] <ickle_> now - originally it was a nasty fallback case with uninitialised data pulled back from the cpu over the temporary glyph mask, all that remains now is the mystery of the white glyphs
> [21:33] <ickle_> (i.e. only the DEST_OUT pass hits the scanout, the ADD pass is woefully absent)
> [21:49] <zorael> ickle_: I'll quote that in my bug report reply if you don't mind
> [21:59] <ickle_> depends whether you have a patch or if it is just me to talking to myself ;-)
> [22:02] <zorael> D:


See the attached short screencast. Selecting text like that is a very easy way to reproduce the issue.

Created attachment 57421 [details]
lspci -vnn of machine in comment 11

Created attachment 57422 [details]
Test case

This reproduces some of the issues with cairo.git.

Created attachment 57423 [details]
testcase result on clemens eisserer's i945GM

sorry, the screenshot was with cairo 1.10.2 - not with cairo.git

Ok, I think I've finally got a handle on this problem.

commit fe6602cbbc4eed1b88ac731a30b46cc970ea444f
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Feb 21 21:26:29 2012 +0000

    sna/gen3+: Only flush the vertices after checking for end-of-batch
    
    Or upon actually closing the vertex buffer.
    
    However, the underlying issue remains. That is we are failing to re-emit
    the first-pass for CA text after flushing the vertex buffer (and so
    emitting the second-pass for the flushed vertices).
    
    Reported-by: lemens Eisserer <linuxhippy@gmail.com>
    References: https://bugs.freedesktop.org/show_bug.cgi?id=42891
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

Now just need that eureka moment to find a clean solution, but hopefully this makes the problem mostly go away for the time being...

*** Bug 44929 has been marked as a duplicate of this bug. ***

(In reply to comment #16)
With that commit everything seems to be working great; the Chromium address bar is sane once more. So far, at least.

Many thanks!

Thanks a lot - white text issues are completly gone :)

commit 6038cede83e7f360428b4625d288411794f9d052
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Feb 21 21:26:29 2012 +0000

    sna/gen3+: Re-emit composite state after flushing CA vertices
    
    Reported-by: Clemens Eisserer <linuxhippy@gmail.com>
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42891
    Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

And that will prevent the bug completely.