Closed Bug 782311 Opened 12 years ago Closed 10 years ago

Scrolling causes artifacts on some NVIDIA GPUs due to removed D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag in bug #775215

Categories

(Core :: Graphics, defect)

Product:
Core
Component:
Graphics
Version:
17 Branch
Platform:
All
Windows 7
Type:
defect
Priority:
Not set
Severity:
major

Tracking

()

Status:
VERIFIED FIXED
Milestone:
mozilla19
Tracking Flags:
Tracking Status
firefox17 + fixed
firefox18 + fixed

People

(Reporter: Virtual, Assigned: nrc)

References

(
URL
)

Details

(Keywords: nightly-community, regression)

Attachments

(8 files, 4 obsolete files)

artefacts.png
26.59 KB, image/png
Details
screencast.mp4
3.58 MB, application/octet-stream
Details
backout patch :-(
1.82 KB, patch
bas.schouten
: review+
lsblakk
: approval-mozilla-aurora+
lsblakk
: approval-mozilla-beta+
Details | Diff | Splinter Review
AIDA64 - Report
173.11 KB, text/html
Details
chrome:gpu
78.12 KB, text/html
Details
opera:gpu
1.94 KB, text/html
Details
about:support
552 bytes, text/plain
Details
normal view and "blurred" view of webpage after scrolling
30.79 KB, image/png
Details
Attached image artefacts.png β€” 
1. Open https://forum.utorrent.com/viewtopic.php?id=121568
2. Scroll to the bottom of the site
3. Now scroll with mouse up and down or 2x up and 1x down to see the artefacts
I have same issue on Google Reader sometimes (rare), but not in this uTorrent board page.
Using the middle click to scroll the page can easy reproduce it, scroll to down completely, and scroll up, when reach.
Hmm, still a lot of stuff in that range, do you think you could narrow it down anymore using inbound builds from http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/ ?
Thanks.

Hmm the only thing in there that looks suspicious is bug 775215.
You think you could test win32 and non-pgo builds so we can pin it to a specific bug? I don't think we get win64 and pgo builds as often.
Bug 775215 is the only thing plausible to be causing this.
Blocks: 775215
Can we mark this bug as confirmed please, so it won't be forgotten ;)
tracking-firefox17: --- → ?
Nick - can you take a look at this? It'd also be good to understand what benefit bug 775215, since we may want to back it out.
Assignee: nobody → ncameron
Status: UNCONFIRMED → ASSIGNED
tracking-firefox17: ? → +
Ever confirmed: true
Keywords: regression
It also happens when I scroll with scroll bar. It's especially visible on the end of this page https://forum.utorrent.com/viewtopic.php?id=123284
Summary: Scrolling with mouse causes artefacts → Scrolling causes artefacts
Here are try builds with bug 775215 backed out so we can confirm it is the cause: https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/tnikkel@gmail.com-41c65519b3da/
Virtual_ManPL could you copy the graphics section of about:support here please? I can't recreate the problem and am interested to know your graphics setup. Thanks.
It seems unlikely that removing D3D10_RESOURCE_MISC_GDI_COMPATIBLE would cause this, but Bas thinks it is possible due to buggy drivers, which would explain why I can't reproduce the artefacts.

tn: can you reproduce this also? If so, could you copy your HW/driver details too please?
No, I couldn't reproduce it.
Sure.

Graphics
Adapter Description - NVIDIA GeForce GTX 460 v2
Vendor ID - 0x10de
Device ID - 0x1205
Adapter RAM - 1023
Adapter Drivers - nvd3dumx,nvwgf2umx,nvwgf2umx nvd3dum,nvwgf2um,nvwgf2um
Driver Version - 9.18.13.602
Driver Date - 8-22-2012
Direct2D Enabled - true
DirectWrite Enabled - true (6.1.7601.17789)
ClearType Parameters - ClearType parameters not found
WebGL Renderer - Google Inc. -- ANGLE (NVIDIA GeForce GTX 460 v2) -- OpenGL ES 2.0 (ANGLE 1.0.0.1242)
GPU Accelerated Windows - 1/1 Direct3D 10
AzureCanvasBackend - direct2d
AzureFallbackCanvasBackend - cairo
AzureContentBackend - direct2d

tl;dr - NVIDIA GeForce GTX 460 v2 with 306.02 drivers on Windows 7 64bit
FYI - Installed yesterday latest stable 306.23 drivers and it still happens.
Thanks for the info Virtual_ManPL, we're looking into it...
It sounds like that is a bug in the driver, we really shouldn't need D3D10_RESOURCE_MISC_GDI_COMPATIBLE around, and so we should blacklist drivers where not having it causes these bugs. But, we need to find out which drivers are buggy.
Don't forget that even latest drivers from NVIDIA have this bug. So if you want to ban all bugged drivers now, you will block all of them. Making Hardware Acceleration on NVIDIA cards disabled due to this.
FYI - it also happens with latest beta 306.63
and it's very visible on most forums

Was NVIDIA informed about this bug by Mozilla dev team? maybe we should CC someone from NVIDIA team to this bug?

and how about backing out bug #775215 and wait for NVIDIA to fix it on stable driver?
I also recommending to push the fixed stable driver to Windows Update in drivers section. This will prevent users using a bugged drivers and seeing these artifacts or having hardware acceleration disabled when Mozilla dev team will be blocking bugged drivers.

What do you guys think about it? :)
ping...

Merging day is Monday. So this bug will be also in Bete release now. Will we have possibility to "fix" it in Beta when code is mostly frozen there?
(In reply to Virtual_ManPL [:Virtual] from comment #25)
> ping...
> 
> Merging day is Monday. So this bug will be also in Bete release now. Will we
> have possibility to "fix" it in Beta when code is mostly frozen there?

Hi, I think that if we have a low-risk fix, then we can push to Beta. Chances are we will just add the driver to our black list, which would (I hope) be fine to push to Beta.

Sorry I haven't had much time for this, we've had trouble identifying exactly where we get problems, and there is a lot of higher priority work around at the moment.
(In reply to Nick Cameron [:nrc] from comment #26)
> (In reply to Virtual_ManPL [:Virtual] from comment #25)
> > ping...
> > 
> > Merging day is Monday. So this bug will be also in Bete release now. Will we
> > have possibility to "fix" it in Beta when code is mostly frozen there?
> 
> Hi, I think that if we have a low-risk fix, then we can push to Beta.
> Chances are we will just add the driver to our black list, which would (I
> hope) be fine to push to Beta.
> 
> Sorry I haven't had much time for this, we've had trouble identifying
> exactly where we get problems, and there is a lot of higher priority work
> around at the moment.

Can't we just back out bug 775215 given that? What's the user benefit to that bug?
(In reply to Alex Keybl [:akeybl] from comment #27)
> (In reply to Nick Cameron [:nrc] from comment #26)
> > (In reply to Virtual_ManPL [:Virtual] from comment #25)
> > > ping...
> > > 
> > > Merging day is Monday. So this bug will be also in Bete release now. Will we
> > > have possibility to "fix" it in Beta when code is mostly frozen there?
> > 
> > Hi, I think that if we have a low-risk fix, then we can push to Beta.
> > Chances are we will just add the driver to our black list, which would (I
> > hope) be fine to push to Beta.
> > 
> > Sorry I haven't had much time for this, we've had trouble identifying
> > exactly where we get problems, and there is a lot of higher priority work
> > around at the moment.
> 
> Can't we just back out bug 775215 given that? What's the user benefit to
> that bug?

775215 prevents increased memory use and slower performance for most D2D drivers (I think, Bas can confirm).
Attached file AIDA64 GPU Report (obsolete) β€” 
Adding some info about my GPU from AIDA64, Chrome and Opera.
Yesterday I installed latest stable WHQL 306.97 drivers. It were even pushed to Windows Update as option and bug still occurs.



I would like to hear some answers and opinions about my previews comment :)
(In reply to Virtual_ManPL [:Virtual] from comment #24)
> FYI - it also happens with latest beta 306.63
> and it's very visible on most forums
> 
> Was NVIDIA informed about this bug by Mozilla dev team? maybe we should CC
> someone from NVIDIA team to this bug?
> 
> and how about backing out bug #775215 and wait for NVIDIA to fix it on
> stable driver?
> I also recommending to push the fixed stable driver to Windows Update in
> drivers section. This will prevent users using a bugged drivers and seeing
> these artifacts or having hardware acceleration disabled when Mozilla dev
> team will be blocking bugged drivers.
> 
> What do you guys think about it? :)



reply to Nick Cameron [:nrc] from comment #28)
> 775215 prevents increased memory use and slower performance for most D2D
> drivers (I think, Bas can confirm).
Will Firefox with disabled HW Acceleration due to bugged drivers (currently all NVIDIA drivers are affected) have greater performance than using Firefox with not removed D3D10_RESOURCE_MISC_GDI_COMPATIBLE "slow" and "memory hog" flg? I doubt :P



I also searched net about "D3D10_RESOURCE_MISC_GDI_COMPATIBLE" and find some info, maybe they will be helpful in some way. Just search on these sites for therm "D3D10_RESOURCE_MISC_GDI_COMPATIBLE":
1 - https://code.google.com/p/slimdx/source/browse/branches/v2/SlimDX2/Sources/SlimDX2.DXGI/Interfaces.cs?r=1704#388
2 - http://www.gamedev.net/topic/547920-how-to-use-d2d-with-d3d11/page__st__40__p__4534404#entry4534404(In
I don't think all Nvidia drivers are affected, in particular, mine are not :-) Also, no-one else I can get to try this has this problem.

One thought is that it might be the presence of DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE without D3D10_RESOURCE_MISC_GDI_COMPATIBLE which is causing the artifacts, since that is not a supported configuration. Removing DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE too can cause problems with menus and dropdowns on some Nvidia drivers, although I did not have that problem.

Virtual_ManPL - do you have Optimus on your system? If you do, could you turn it on and see if the bug goes away (we don't use DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE with Optimus), alternatively if I post a patch to test this, can you build Firefox to test if this solves the problem?

If that is the case, then we could try taking out DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE too, although that might re-break menus/dropdowns. Or we could put D3D10_RESOURCE_MISC_GDI_COMPATIBLE back in, if that is justified by the need for DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE.

See also bug 613790 and bug 623807.
Flags: needinfo?(BernesB)
(In reply to Virtual_ManPL [:Virtual] from comment #2)
> Using the middle click to scroll the page can easy reproduce it, scroll to
> down completely, and scroll up, when reach.

Confirm it in https://forum.utorrent.com for middle click.
(In reply to Raul Malea from comment #34)
> (In reply to Virtual_ManPL [:Virtual] from comment #2)
> > Using the middle click to scroll the page can easy reproduce it, scroll to
> > down completely, and scroll up, when reach.
> 
> Confirm it in https://forum.utorrent.com for middle click.

Video card: 
Grafică
Accelerare GPU Windows 1/1 
Direct3D 10
Dată driver 10-2-2012
Dată driver (GPU #2)3-19-201
2Descriere adaptor (GPU #2)Intel(R) HD Graphics 3000
Descrierea adaptorului NVIDIA GeForce GTX 560 Ti  
Direct2D activat true
DirectWrite activatt rue (6.1.7601.17789)
Drivere pentru adaptor nvd3dumx,nvwgf2umx,nvwgf2umx nvd3dum,nvwgf2um,nvwgf2um
GPU #2 activ false
ID dispozitiv 0x1200
ID dispozitiv (GPU #2)0x0112
ID furnizor 0x10de
ID vÒnzător (GPU #2)0x8086
Motor de afiΘ™are WebGL no information
RAM adaptor (GPU #2)Unknown
RAM pentru adaptor 2047
Versiune driver (GPU #2)8.15.10.2696
 AzureCanvasBackenddirect 2d
AzureContentBackenddirect 2d
AzureFallbackCanvasBackend cairo

Driver: WHQL 306.97 drivers
This bug is also present in Gmail interface (see bug 801415)
(In reply to Nick Cameron [:nrc] from comment #33)
> I don't think all Nvidia drivers are affected, in particular, mine are not
> :-) Also, no-one else I can get to try this has this problem.
This could be related more to GPU models I think. Like some models have this bug, others don't. Maybe NVIDIA didn't block something, which isn't supported by affected GPU models and we see this artifacts
or simply like my links are said and you too said it, it could be problem with DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE without D3D10_RESOURCE_MISC_GDI_COMPATIBLE. I hope these links where helpful in some way and contain a hint. ;)

I also created thread about this bug on mozillaZine. Maybe this will help in some way diagnosing and analyzing this bug (specific GPUs, drivers, system etc.)
http://forums.mozillazine.org/viewtopic.php?f=23&t=2573963



(In reply to Nick Cameron [:nrc] from comment #33)
> Virtual_ManPL - do you have Optimus on your system? If you do, could you
> turn it on and see if the bug goes away (we don't use
> DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE with Optimus),
I don't have Optimus, because I use standalone PC with CPU: C2D E6550 2,33GHz@3,5GHz without integrated GPU.



(In reply to Nick Cameron [:nrc] from comment #33)
> alternatively if I post a
> patch to test this, can you build Firefox to test if this solves the problem?
Unfortunately I never created Firefox build, so I prefer building try-builds by someone else, if this won't be too much burden. :)



(In reply to Raul Malea from comment #34)
Thank you for confirmation and Graphic section info from about:config
Flags: needinfo?(BernesB)
Virtual_ManPL: could you try the build from this try push please: https://tbpl.mozilla.org/?tree=Try&rev=57de4df594d8. Could you let me know if it has any affect on the proplem you reported (might make it better or worse) and whether there are any other problems, in particular menus or drop downs not appearing?

Thank you!
I still can confirm this bug on build you created. No worse, no better, no other problems, menus or drop downs are appearing. 


Mozilla/5.0 (Windows NT 6.1; WOW64; rv:19.0) Gecko/19.0 Firefox/19.0
buildID: 20121015211000

Source
Built from http://hg.mozilla.org/try/rev/57de4df594d8
(In reply to Virtual_ManPL [:Virtual] from comment #41)
> Created attachment 673809 [details] [diff] [review]
> last resort patch - readd D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag (completly
> undo patch from bug #775215)
> 
> Based on patch from #775215.

We really need to blacklist and -not- re-add this flag. It makes things slower and worse in all kinds of ways on all other devices. This driver is broken, we should find out where this happens and kill it.
(In reply to Bas Schouten (:bas) from comment #42)
> (In reply to Virtual_ManPL [:Virtual] from comment #41)
> > Created attachment 673809 [details] [diff] [review]
> > last resort patch - readd D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag (completly
> > undo patch from bug #775215)
> > 
> > Based on patch from #775215.
> 
> We really need to blacklist and -not- re-add this flag. It makes things
> slower and worse in all kinds of ways on all other devices. This driver is
> broken, we should find out where this happens and kill it.

I suppose we could also re-add the flag just for NVidia x60 devices. Which so far seems to be the only correlating factor. I wish we could figure out what makes these devices so broken though.
All drivers are affected in this case. Like you can also see in mozillaZine thread not only GTX460 GPU devices are affected. Quadro NVS 3100M are too.

So I will repeat my questions as I still don't get any answers on them...:
-did you guys contact NVIDIA, so they will know about this issue and fix it in their drivers? I see many guys with nvidia.com end in email in CC search
-what will be more slower:
     Firefox without D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag with blacklisted GPU drivers without Hardware Acceleration
     Firefox with D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag without blaclisted GPU drivers with Hardware Acceleration
I still think the second option will be much faster, so why not to use this flag only for these "bugged" devices like creating exception etc.?
How much this D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag cost performance and how much it uses memory compared to build without this flag?

Also this patch is last resort patch if anything fails. ;d
Comment on attachment 673809 [details] [diff] [review]
last resort patch - readd D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag (completly undo patch from bug #775215)

Review of attachment 673809 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks for the patch. It seems to me we probably should put this back for known broken drivers that benefit from it, but this definitely needs to be guarded on those drivers.
Attachment #673809 - Flags: review?(ncameron)
bjacob: would you be able to look into the driver black listing stuff as part of the new blacklist work? I don't really know what is going on there, maybe I can fill you in on the wider issues here on IRC if that looks feasible.
Comment on attachment 673809 [details] [diff] [review]
last resort patch - readd D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag (completly undo patch from bug #775215)

Review of attachment 673809 [details] [diff] [review]:
-----------------------------------------------------------------

We really need to guard this as Nick concluded.
Attachment #673809 - Flags: review?(bas)
Yep, this will be the best solution.
What's more, 8600GT are also affected as I see ( http://forums.mozillazine.org/viewtopic.php?p=12405641#p12405641 ). This bug getting more and more strange... ;/

Was NVIDIA informed about this bug? I would do that by myself, but unfortunately I don't know technical details.
FYI - latest beta drivers 310.33 released 2 days ago are also affected...
(In reply to Bas Schouten (:bas.schouten) from comment #47)
> Comment on attachment 673809 [details] [diff] [review]
> last resort patch - readd D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag (completly
> undo patch from bug #775215)
> 
> Review of attachment 673809 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> We really need to guard this as Nick concluded.

Will this be done in the next couple of days? If not, let's revert to known working code asap, since the regression appears to be worse than the benefit of bug 775215.
tracking-firefox19: ? → +
(In reply to Alex Keybl [:akeybl] from comment #51)
> (In reply to Bas Schouten (:bas.schouten) from comment #47)
> > Comment on attachment 673809 [details] [diff] [review]
> > last resort patch - readd D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag (completly
> > undo patch from bug #775215)
> > 
> > Review of attachment 673809 [details] [diff] [review]:
> > -----------------------------------------------------------------
> > 
> > We really need to guard this as Nick concluded.
> 
> Will this be done in the next couple of days? If not, let's revert to known
> working code asap, since the regression appears to be worse than the benefit
> of bug 775215.

Unlikely, we both have higher priority work atm (b2g, Win8), so we're not going to have time to look at this for the next few weeks, sorry. Backing out 775215 is neither here nor there, the problem is that to fix this we also need to back out bug 767337 and that is a horrible out of memory crash, we thought the crash was more important than scrolling artefacts, but we have no idea how widespread either is to make a call. I could reproduce the crash but not these artefacts, which is obviously a bad thing to go on, but is all I really have. I asked QA for input, but they were not keen to help out because it is looking for rendering artefacts, rather than a crash.
OK - thought about this some more. Bug 775215 landed in FF15, which in turn caused 767337, which was fixed in FF16. So presumably FF16 is also affected by this bug. If that's the case, we can maintain the status quo here for FF17 given so few have reported this issue.

Please let me know if I've got that right Nick.
(In reply to Alex Keybl [:akeybl] from comment #53)
> OK - thought about this some more. Bug 775215 landed in FF15, which in turn
> caused 767337, which was fixed in FF16. So presumably FF16 is also affected
> by this bug. If that's the case, we can maintain the status quo here for
> FF17 given so few have reported this issue.
> 
> Please let me know if I've got that right Nick.

Not quite. The fix to bug 767337 was to remove some occurrences of D3D10_RESOURCE_MISC_GDI_COMPATIBLE (and landed in FF16), the rest were removed in bug 775215 which I thin landed in FF17. I can't be sure whether the artefacts appear with only the 767337 change or both changes, my guess would be only 767337, but it is only a guess.

If we do get a load of problems, this is very easy to fix quickly, it's just fixing it properly that is a bit more time consuming.
(In reply to Nick Cameron [:nrc] from comment #54)
> I can't be sure whether the artefacts appear
> with only the 767337 change or both changes, 
> my guess would be only 767337, but it is only a guess.

See comment #39 and comment #40


(In reply to Nick Cameron [:nrc] from comment #52)
> I asked QA for input, but they were not keen to help
> out because it is looking for rendering artefacts, rather than a crash.

Seems like wrong people in wrong seats. ;)



Was NVIDIA informed about this bug? I asking this question for the fifth time... ;d
Summary: Scrolling causes artefacts → Scrolling causes artifacts on some NVIDIA GPUs due to removed D3D10_RESOURCE_MISC_GDI_COMPATIBLE flag in bug #775215
(In reply to Virtual_ManPL [:Virtual] from comment #55)
> (In reply to Nick Cameron [:nrc] from comment #54)
> > I can't be sure whether the artefacts appear
> > with only the 767337 change or both changes, 
> > my guess would be only 767337, but it is only a guess.
> 
> See comment #39 and comment #40
> 
Sorry, that patch was completely different, it didn't back out 767337 or 775215, but tried something different. If you could test it, I can prepare a build without 767337 to see if that helps (although I suspect it will need 775215 too). 
> 
> (In reply to Nick Cameron [:nrc] from comment #52)
> > I asked QA for input, but they were not keen to help
> > out because it is looking for rendering artefacts, rather than a crash.
> 
> Seems like wrong people in wrong seats. ;)
> 
> 
Heh, I think it is just that without a crash it is very hard to automate and they have a lot of high priority stuff at the moment.
> 
> Was NVIDIA informed about this bug? I asking this question for the fifth
> time... ;d

Bas? (Afraid I don't have any contacts at Nvidia).
(In reply to Nick Cameron [:nrc] from comment #56)
> Sorry, that patch was completely different, it didn't back out 767337 or
> 775215, but tried something different. If you could test it, I can prepare a
> build without 767337 to see if that helps (although I suspect it will need
> 775215 too).

Sure thing! I will be glad to help as always.


(In reply to Nick Cameron [:nrc] from comment #56)
>(Afraid I don't have any contacts at Nvidia).

I see many guys with "@nvidia.com" end in email in CC search
(In reply to Virtual_ManPL [:Virtual] from comment #57)
> (In reply to Nick Cameron [:nrc] from comment #56)
> > Sorry, that patch was completely different, it didn't back out 767337 or
> > 775215, but tried something different. If you could test it, I can prepare a
> > build without 767337 to see if that helps (although I suspect it will need
> > 775215 too).
> 
> Sure thing! I will be glad to help as always.
> 

https://tbpl.mozilla.org/?tree=Try&rev=040112e997c4

If you could let us know if this improves things, that would be useful, thanks!

The build undoes 775215 but leaves 767337, that means we shouldn't get the out of memory crash that we need to not return, but we still use the GDI compatible flags in other places, so this is a pretty obnoxious combination, but if it works is a possible temporary solution.
tracking-firefox19: + → ---
(In reply to Nick Cameron [:nrc] from comment #56)
> (In reply to Virtual_ManPL [:Virtual] from comment #55)
> > (In reply to Nick Cameron [:nrc] from comment #54)
> > > I can't be sure whether the artefacts appear
> > > with only the 767337 change or both changes, 
> > > my guess would be only 767337, but it is only a guess.
> > 
> > See comment #39 and comment #40
> > 
> Sorry, that patch was completely different,
> it didn't back out 767337 or 775215,
> but tried something different.

Awww, sorry, I quoted wrong comment.
I mean the comment #14 and the comment #15
In your first build after all we tested absence of DXGI_SWAP_CHAIN_FLAG_GDI_COMPATIBLE flag
(In reply to Timothy Nikkel (:tn) from comment #14)
> Here are try builds with bug 775215 backed out so we can confirm it is the
> cause:
> https://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/tnikkel@gmail.com-
> 41c65519b3da/


(In reply to Nick Cameron [:nrc] from comment #58)
> https://tbpl.mozilla.org/?tree=Try&rev=040112e997c4
> 
> If you could let us know if this improves things, that would be useful,
> thanks!
> 
> The build undoes 775215 but leaves 767337, that means we shouldn't get the
> out of memory crash that we need to not return, but we still use the GDI
> compatible flags in other places, so this is a pretty obnoxious combination,
> but if it works is a possible temporary solution.

So like earlier in build created by Timothy Nikkel (:tn), your build with also backed out bug #775215 fixes this issue

Mozilla/5.0 (Windows NT 6.1; WOW64; rv:19.0) Gecko/19.0 Firefox/19.0
buildID: 20121105202554
Source: Built from http://hg.mozilla.org/try/rev/040112e997c4
I've sent an e-mail to NVidia.
Attached patch backout patch :-( β€” β€” Splinter Review 
[Approval Request Comment]
Bug caused by (feature/regressing bug #): 775215 
User impact if declined: scrolling artefacts for some users
Testing completed (on m-c, etc.): backout only
Risk to taking this patch (and alternatives if risky): possible decreased performance, extra memory use for some drivers
String or UUID changes made by this patch: none
Attachment #673809 - Attachment is obsolete: true
Attachment #679828 - Flags: review?(bas)
Attachment #679828 - Flags: approval-mozilla-beta?
Attachment #679828 - Flags: approval-mozilla-aurora?
The flags in this patch should totally be guarded on the driver type, but seeing as we don't have actual reports of bugs (as opposed to the same flag in the gradient/memory use bug), I hope that is not essential. I consider this a stop-gap measure until we solve this properly. I don't have time at the moment do this properly, sorry.
Attachment #679828 - Flags: review?(bas) → review+
Too bad that we need to completely undo path from bug #775215, if it prevents increased memory use and have slower performance for most D2D GPUs like you guys said. We can only hope that NVIDIA informed by Bas Schouten (:bas.schouten) will fix this bug in their drivers ASAP, so we can remove this second time without triggering any other issues.

P.S. FYI - Nick Cameron [:nrc] you could have use my patch, it's the same like yours ;d
(In reply to Virtual_ManPL [:Virtual] from comment #63)
> Too bad that we need to completely undo path from bug #775215, if it
> prevents increased memory use and have slower performance for most D2D GPUs
> like you guys said. We can only hope that NVIDIA informed by Bas Schouten
> (:bas.schouten) will fix this bug in their drivers ASAP, so we can remove
> this second time without triggering any other issues.

In a few weeks when things are a little less hectic (that might be optimistic), we can do better than currently by guarding on the driver, we will have to do this even if Nvidida fix their driver because of slow uptake :-( But yeah, hopefully they will fix 'em.

> 
> P.S. FYI - Nick Cameron [:nrc] you could have use my patch, it's the same
> like yours ;d

Yes, I only remembered about your patch after I made the one for the Try push, sorry. I had to overwrite the patch on the bug only because it needed a commit message.
Whiteboard: [leave open]
Comment on attachment 679828 [details] [diff] [review]
backout patch :-(

Please land to branches asap so we're ready for final beta on Monday, thank you.
Attachment #679828 - Flags: approval-mozilla-beta?
Attachment #679828 - Flags: approval-mozilla-beta+
Attachment #679828 - Flags: approval-mozilla-aurora?
Attachment #679828 - Flags: approval-mozilla-aurora+
Bas Schouten (:bas.schouten) - I send you email with all info that NVIDIA guys will need and also updated attachments here.
Guys which could confirm this issue, I need some more info about your PC specification.

Especially:
-motherboard and chipset info
-GPU working on which PCIe (3.0/2.1/2.0/1.1/1.0)
-monitor info (type, resolution, refresh rate, connected with HDMI/DVI/DisplayPort/VGA

CPU/memory/sound card/PSU/etc is optional



Thank you!
I had a similar problem but with an AMD graphics card on Firefox 19.
However, I could trace it down to the latest update of Win7 64bit (KB2670838 http://support.microsoft.com/kb/2670838/en-us) which changes the following:
    Direct2D
    DirectWrite
    Direct3D
    Windows Imaging Component (WIC)
    Windows Advanced Rasterization Platform (WARP)
    Windows Animation Manager (WAM)
    XPS Document API
    H.264 Video Decoder
    JPEG XR codec

I don't know if it's related to this bug, but the effects where the same, after scrolling down and up again parts of the content which weren't in view after scrolling up got blurred scrolling down again.
I will add an attachment later.
After reading the comments on bug #812695 I agree, you're right.
I wanted to retest this old issue with latest NVIDIA drivers 337.88 and see if it still happens on it and it seems that is fixed as I can't reproduce it on build which I could reproduce it (CSet-20db7c6d82cc), so we can try to push patch from bug #775215 once again.

What do you think guys?

As you said it will:

(In reply to Nick Cameron [:nrc] from comment #28)
> 775215 prevents increased memory use and slower performance for most D2D
> drivers (I think, Bas can confirm).

(In reply to Bas Schouten (:bas.schouten) from comment #42)
> ...It makes things slower and worse in all kinds of ways
> on all other devices.
Flags: needinfo?(ncameron)
Flags: needinfo?(bas)
Let's just assume this is fixed then if we have noone still reproducing it?
Flags: needinfo?(bas)
(In reply to Virtual_ManPL [:Virtual] from comment #82)
> Yep, so can we reland patch from bug #775215 once again?

Not relevant anymore, that's dead code on trunk and will fairly soon be killed off completely.
Flags: needinfo?(bas)
OK. Thanks for the info.
I'm closing this bug as FIXED.
Status: ASSIGNED → RESOLVED
Closed: 10 years ago
Flags: needinfo?(ncameron)
Resolution: --- → FIXED
Whiteboard: [leave open]
Target Milestone: --- → mozilla19
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: