Issue metadata
Sign in to add a comment
|
Canvas clipping interacting strangely
Reported by
p...@lucidchart.com,
Mar 22 2016
|
|||||||||||||||||||||||||
Issue descriptionUserAgent: Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/49.0.2623.87 Safari/537.36 Example URL: https://jsfiddle.net/461mtck2/5/ Steps to reproduce the problem: var canvas = document.getElementById('canvas'); var context = canvas.getContext('2d'); context.beginPath(); context.rect(0, 0, 500, 500); context.rect(0, 0, 500, 500); context.clip(); context.save(); context.beginPath(); context.rect(0, 0, 200, 200); context.clip(); context.fillRect(25, 25, 150, 150); context.restore(); context.fillRect(200, 25, 150, 150); What is the expected behavior? The canvas shows two black squares. What went wrong? The canvas shows only the left black square. Does it occur on multiple sites: Yes Is it a problem with a plugin? No Did this work before? Yes I think it worked recently, perhaps before Chrome 49. Does this work in other browsers? Yes Chrome version: 49.0.2623.87 Channel: stable OS Version: 10.0 Flash Version: Shockwave Flash 21.0 r0 We've seen large increases in reports of canvas rendering artifices in the last week or two. This is one reproducible case. I can't infer what might be happening, so perhaps this appears in other ways. It is machine specific. I have attached chrome://version and chrome://gpu.
,
Mar 23 2016
,
Mar 23 2016
,
Mar 23 2016
,
Mar 23 2016
I was unable to reproduce this in Chrome 51. @paul: Could you install the Chrome Canary (https://www.google.com/chrome/browser/canary.html) and confirm whether this is fixed in the very latest builds? Thanks.
,
Mar 23 2016
I can't repro at all actually. Could be an issue specific to this particular GPU or driver. jbauman, jmadill and geofflang: Do you know about bugs with AMD GPUs, the person reporting this bug has a switchable GPU on windows. And do you have access to hardware that could be used to repro this? bsalomon: I am not sure how clipping works internally in ganesh, AFAIK there are many paths. Anyways, it looks like whatever state is used for storing the current clip (could it be a stencil buffer?) is not getting reset correctly.
,
Mar 23 2016
There is code in ANGLE to work around a bug with stencil buffers on AMD: https://code.google.com/p/chromium/codesearch#chromium/src/third_party/angle/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp&l=803 maybe the addition of dirty bits caused that workaround to break.
,
Mar 24 2016
+senorblanco who was also seeing AMD D3D9 issues. Good find John, I was able to repro this on our test bench, and confirm that commenting out that AMD workaround fixes the bug. I'm a little overloaded right now but might have time to look at this in the next couple weeks if it isn't taken up by then. Justin let me know if you need help reproing, I can help you with that.
,
Mar 24 2016
Sorry I missed the initial question directed at me. Yes, we sometimes store the current clip in the stencil buffer.
,
Mar 24 2016
Going to take this but I might not have time for a week or so.
,
Mar 28 2016
Unable to reproduce the issue on Windows 10, 7, Mac 10.10.5, Ubuntu 14.04 using 49.0.2623.87, latest stable 49.0.2623.108, canary 51.0.2692.0 with below steps: 1.Enabled "Accelerated 2D Canvas" in chrome://flags. 2.Opened URL: https://jsfiddle.net/461mtck2/5/ 3.Observed that the canvas shown two black squares. Please find attached screencast and update if anything missed here in triaging the issue. paul@Could you please check the issue on clean profile and update the thread if issue still persists.
,
Mar 28 2016
This only happens on AMD (GPU Vendor) with D3D9 (--disable-d3d11 or --use-angle=d3d9). You won't be able to repro on other vendors or with the default D3D11 rendering AFAIK.
,
Mar 28 2016
,
Mar 29 2016
GPU Vendor-AMD with D3D9 is not available to test the issue.Hence, assigning to MTV to check the issue if specified gpu driver available. Attached the GPU details. MTV@Could anyone from MTV look into this issue please.
,
Apr 1 2016
I have a computer that reproduces locally, I will look into this soon.
,
Apr 15 2016
Hello. The problem got worse with the release of Chrome 50. In our company (browser games developer) we now have 4 machines (all with AMD graphics cards) which exhibit this issue. Is there anything we can do to work around it? Anything we can provide in order for the problem to get fixed?
,
Apr 18 2016
Can confirm Comment 16 from Tobias. We have a lot of players who use Chrome and experience this issue. It is so bad that they can't play our game anymore and their only workaround is to use another browser or deactivate hardware acceleration. But many will not even try and just stop, which will decrease our revenue in the future.
,
Apr 18 2016
Can confirm the same as described by fellow Bytro members above. It is very frustrating for our users. Either, they will switch to another browser or stop playing our game. Both should not be acceptable for Chrome or us, since Chrome also loses customers through this issue.
,
Apr 18 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9 commit ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9 Author: Geoff Lang <geofflang@chromium.org> Date: Fri Apr 15 19:04:14 2016 Fix incorrect dirty bit state tracking on D3D9 AMD with color masks. When using the zero color mask workaround in StateManager9, the blend state and blend equation were being modified. This caused inconsistant state if the workaround stopped being used in subsequent draw calls. To deal with this issue, dirty the BLEND_ENABLED and BLEND_EQUATION states when the COLOR_MASK changes and dirty the COLOR_MASK state wheneither BLEND_ENABLED or BLEND_EQUATION changes. BUG= 597107 Change-Id: I03de934b419b7593e4863838720ced1e5773c092 Reviewed-on: https://chromium-review.googlesource.com/339280 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Tryjob-Request: Geoff Lang <geofflang@chromium.org> [modify] https://crrev.com/ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9/src/libANGLE/renderer/d3d/d3d9/StateManager9.h [add] https://crrev.com/ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9/src/tests/gl_tests/ColorMaskTest.cpp [modify] https://crrev.com/ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9/src/tests/angle_end2end_tests.gypi [modify] https://crrev.com/ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp [modify] https://crrev.com/ae2d0a0e865545fbc5f55bf2014d00bb8f4669c9/src/libANGLE/renderer/d3d/d3d9/Renderer9.cpp
,
Apr 18 2016
Issue 604127 has been merged into this issue.
,
Apr 18 2016
We're seeing problems with stencil-fill text that I reported to WebGL Dev. The triangles don't seem to have discarded pixels in concave areas, and fill the entire triangle. The issues occur primarily with AMD switchable graphics devices, and I was able to reproduce them on an HP Envy 14. It's unclear yet as to whether the AMD or Intel part is at fault, but errors are shown when running on Intel HD 2500 and Intel HD 4000 portions of rendering. Firefox/Edge/IE11 rendering properly on the same devices. Firefox uses the D3D11 ANGLE path, but Chrome disables the D3D11 path on switchable graphics devices and uses the DX9 ANGLE path. It's my suspicion that the DX9 path has the bug.
,
Apr 18 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/4cca4afeb2e2b74a716c82917d9578938b17756f commit 4cca4afeb2e2b74a716c82917d9578938b17756f Author: geofflang <geofflang@chromium.org> Date: Mon Apr 18 20:48:01 2016 Roll ANGLE 6684007..ae2d0a0 https://chromium.googlesource.com/angle/angle.git/+log/6684007..ae2d0a0 BUG= 597107 TBR=cwallez@chromium.org TEST=bots CQ_INCLUDE_TRYBOTS=tryserver.chromium.win:win_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.linux:linux_optional_gpu_tests_rel Review URL: https://codereview.chromium.org/1901793002 Cr-Commit-Position: refs/heads/master@{#388007} [modify] https://crrev.com/4cca4afeb2e2b74a716c82917d9578938b17756f/DEPS
,
Apr 19 2016
Hi all, The fix should be in the latest version of Chrome canary (52.0.2712.0). I'd appreciate it if those who experienced this issue could verify that it is fixed and I'll try to merge it back into stable Chrome.
,
Apr 19 2016
FYI, you can get the Canary (nightly build) here: https://www.google.com/chrome/browser/canary.html You can install it alongside an existing Chrome installation without any conflicts (no need to uninstall anything). Doing the following steps will greatly help us determine whether the fix should be pulled into Chrome 51 and possibly into a Chrome 50 update (so you can get it faster): 1. Install Canary. 2. Try to reproduce the bug. 3. Come back here to confirm the fix. Thanks for your help.
,
Apr 19 2016
I just tested it on the test case and on Agar.io (which was one of the affected websites) and the issue no longer exists there.
,
Apr 19 2016
I tested it in our game "Call of War", and the issue is solved there. Thanks!
,
Apr 19 2016
Can confirm. On all our affected machines games are rendering nicely again. Pulling the fix into Chrome 51 and 50 would of course be greatly appreciated! Thanks for fixing the issue so quickly.
,
Apr 20 2016
,
Apr 20 2016
[Automated comment] DEPS changes referenced in bugdroid comments, needs manual review.
,
Apr 21 2016
hi guys. unfortunately we have to report that on current Chrome canary (52.0.2714.0) clipping now seems to have issues again. What seems to happen is that the clipping regions (sometimes) is not respected and things can overlap to the outside. Happens on all our previously affected machines, but not on any other. Anybody can confirm that?
,
Apr 21 2016
Still seems to be working normally in here
,
Apr 22 2016
52.0.2714.0 works for us, on the machine in bug description.
,
Apr 22 2016
I can verify that current canary build (52.0.2715.0) still has clipping issues. On canary image you can detect them easily around the coastal area, while the same does not happen on stable version (with hardware acceleration disabled of course).
,
Apr 22 2016
Kostas, do you have a link to a repro case? This is may be a different issue than the one originally described. Does the jsfiddle linked the the first post repro on these machines? Also, can you save the about:gpu page and upload it?
,
Apr 22 2016
Here's a file showing the stencil failure on AMD 5650 with switchable graphics. The canvas version renders a teardrop shape using the four Bezier handles. The WebGL version uses samples of that in a triangle fan, and stencil-cover to generate the winding number. The same file renders fine in Firefox, Edge, and Safari, but not in Chrome. Chrome and Canary 52.0.2715.0 (64-bit) both fail on this test case. I suspect that the state manager has another bug that wasn't fixed related to overriding color write enable. Somehow the stencil is thrown off, and the entire region is filled with the cover pass.
,
Apr 22 2016
I ran the "stencil-fail.html" on my machine, and those are the results I got: 50.0.2661.87 m: http://i.imgur.com/hX53IXf.png 52.0.2715.0 canary: http://i.imgur.com/CLIHzOM.png Firefox: http://i.imgur.com/Rypue1G.png
,
Apr 23 2016
I got the Canary result that you did. Thanks for the png images.
,
Apr 25 2016
We have a repro case now about the clipping issues in Canary 52.0.2715.0. Please check this fiddle: https://jsfiddle.net/qud2nkbn/2/ You should be able to spot the issue right away on Canary. This fiddle works fine with other browsers or with disabled hardware acceleration on Canary. @geoff yes, the jsfiddle from the first post did reproduce on our machines.
,
Apr 25 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/250062bc7fd3f4229332f2dd6381be6576120505 commit 250062bc7fd3f4229332f2dd6381be6576120505 Author: Geoff Lang <geofflang@chromium.org> Date: Mon Apr 25 16:33:44 2016 Make sure D3DRS_TWOSIDEDSTENCILMODE is enabled if the stencil test is. This state-setting code was not ported when adding the dirty bits for D3D9 depth stencil state in https://chromium-review.googlesource.com/#/c/316449. BUG= 597107 Change-Id: I0a1bdae18dd22c9b01f12a73267040157a866654 Reviewed-on: https://chromium-review.googlesource.com/340436 Tryjob-Request: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> [modify] https://crrev.com/250062bc7fd3f4229332f2dd6381be6576120505/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp
,
Apr 25 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/1d0ecd382887571a8bb00e1289d1b67eaa6a1396 commit 1d0ecd382887571a8bb00e1289d1b67eaa6a1396 Author: geofflang <geofflang@chromium.org> Date: Mon Apr 25 22:48:06 2016 Roll ANGLE 82c47ad..250062b https://chromium.googlesource.com/angle/angle.git/+log/82c47ad..250062b BUG= 597107 TBR=jmadill@chromium.org TEST=bots CQ_INCLUDE_TRYBOTS=tryserver.chromium.win:win_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.linux:linux_optional_gpu_tests_rel Review URL: https://codereview.chromium.org/1915293002 Cr-Commit-Position: refs/heads/master@{#389595} [modify] https://crrev.com/1d0ecd382887571a8bb00e1289d1b67eaa6a1396/DEPS
,
Apr 26 2016
Clipping issue seems to be fixed in the current version of canary (52.0.2717.0). Thanks a lot for fixing it so fast! Is there a chance of adding this fix to chrome 50 or 51?
,
Apr 26 2016
Thanks for providing a good repro case. The second issue affected all D3D9, not just AMD. I've created a merge request for chrome 50, waiting on a reply. 51 should be an easier target.
,
Apr 26 2016
[Automated comment] DEPS changes referenced in bugdroid comments, needs manual review.
,
Apr 26 2016
Before we approve merge to M50, Could you please confirm whether this bug is baked/verified in Canary and safe to merge?
,
Apr 26 2016
govind: Both fixes are now merged into Canary, one for about a week, the other for a day. Risk is low, both fixes are very small.
,
Apr 26 2016
Merge approved for M51 (branch 2704)
,
Apr 26 2016
Please merge your change before 5:00 PM PST today to M51 branch 2704, so we can take it for this week beta. Thank you.
,
Apr 26 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/707b2d540088ec2683aac026277e294d667319aa commit 707b2d540088ec2683aac026277e294d667319aa Author: Geoff Lang <geofflang@chromium.org> Date: Fri Apr 15 19:04:14 2016 Fix incorrect dirty bit state tracking on D3D9 AMD with color masks. When using the zero color mask workaround in StateManager9, the blend state and blend equation were being modified. This caused inconsistant state if the workaround stopped being used in subsequent draw calls. To deal with this issue, dirty the BLEND_ENABLED and BLEND_EQUATION states when the COLOR_MASK changes and dirty the COLOR_MASK state wheneither BLEND_ENABLED or BLEND_EQUATION changes. BUG= 597107 Change-Id: I689d9cf3c1eb20206a2abc7beff481fc7842f85e Reviewed-on: https://chromium-review.googlesource.com/340763 Reviewed-by: Geoff Lang <geofflang@chromium.org> [modify] https://crrev.com/707b2d540088ec2683aac026277e294d667319aa/src/libANGLE/renderer/d3d/d3d9/StateManager9.h [add] https://crrev.com/707b2d540088ec2683aac026277e294d667319aa/src/tests/gl_tests/ColorMaskTest.cpp [modify] https://crrev.com/707b2d540088ec2683aac026277e294d667319aa/src/tests/angle_end2end_tests.gypi [modify] https://crrev.com/707b2d540088ec2683aac026277e294d667319aa/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp [modify] https://crrev.com/707b2d540088ec2683aac026277e294d667319aa/src/libANGLE/renderer/d3d/d3d9/Renderer9.cpp
,
Apr 26 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/1a1b30c37e1316234280c00fd7f8a01758b1a437 commit 1a1b30c37e1316234280c00fd7f8a01758b1a437 Author: Geoff Lang <geofflang@chromium.org> Date: Mon Apr 25 16:33:44 2016 Make sure D3DRS_TWOSIDEDSTENCILMODE is enabled if the stencil test is. This state-setting code was not ported when adding the dirty bits for D3D9 depth stencil state in https://chromium-review.googlesource.com/#/c/316449. BUG= 597107 Change-Id: I1b12fa24f7b6421453656355df52a3c2cb89d31d Reviewed-on: https://chromium-review.googlesource.com/340764 Reviewed-by: Geoff Lang <geofflang@chromium.org> [modify] https://crrev.com/1a1b30c37e1316234280c00fd7f8a01758b1a437/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp
,
Apr 26 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/d13db894dcf4860e68ca309fa10b5277d6a0b918 commit d13db894dcf4860e68ca309fa10b5277d6a0b918 Author: geofflang <geofflang@chromium.org> Date: Tue Apr 26 22:01:02 2016 Roll ANGLE d4102f0..1a1b30c https://chromium.googlesource.com/angle/angle.git/+log/d4102f0..1a1b30c BUG= 597107 TBR=jmadill@chromium.org NOTRY=true NOPRESUBMIT=true Review URL: https://codereview.chromium.org/1922263002 Cr-Commit-Position: refs/branch-heads/2704@{#254} Cr-Branched-From: 6e53600def8f60d8c632fadc70d7c1939ccea347-refs/heads/master@{#386251} [modify] https://crrev.com/d13db894dcf4860e68ca309fa10b5277d6a0b918/DEPS
,
May 5 2016
Has anyone found any workarounds for this until it gets merged in? We are seeing an increasing number of players on our game affected by this and we are losing customers.
,
May 5 2016
Our workaround for the two-sided stencil state manager bug is to tell users to use IE/Edge/Firefox/Safari. Firefox can still hit the same ANGLE DX9 path, but seems to take that path less often (f.e. on AMD and AMD-switchable).
,
May 5 2016
https://www.chromium.org/developers/calendar signals that Chrome M51 should be out around the end of the month. The fix is merged in there, so should be available very soon.
,
May 5 2016
Note you can also suggest your users to use Chrome Beta/Dev or Canary, the fix is available in all of those. Canary is nice because it installs along side stable Chrome without replacing it.
,
May 5 2016
The users should deactivate the hardware acceleration in the advanced settings Von: james@goldfirestudios.com via Monorail [mailto:monorail@chromium.org] Gesendet: Donnerstag, 5. Mai 2016 23:42 An: info@heinz-woerner.de Betreff: Issue 597107 in chromium: Canvas clipping interacting strangely Comment #52 on issue 597107 by ja...@goldfirestudios.com <mailto:ja...@goldfirestudios.com> : Canvas clipping interacting strangely https://bugs.chromium.org/p/chromium/issues/detail?id=597107#c52 Has anyone found any workarounds for this until it gets merged in? We are seeing an increasing number of players on our game affected by this and we are losing customers.
,
May 5 2016
That disables WebGL entirely. Not really an option for our title.
,
May 8 2016
Before we approve merge to M50, Could you please confirm whether this bug is baked/verified in Canary/Beta and safe to merge?
,
May 9 2016
I confirm that the bug is fixed in both canary (52.0.2726.0 ) and beta (51.0.2704.36). Merge to M50 would be great. Thanks!
,
May 9 2016
govind: The fix has been in Canary for two weeks and merged into M51 for 1 week.
,
May 9 2016
We've had reports of the stencil render state bug in Opera and Firefox on Windows, so you might make sure that they update ANGLE as well. The fix in Canary works for us.
,
May 9 2016
Approving Merge to M50 based on comment # 59 and #60. Please merge to M50 branch 2661 ASAP.
,
May 9 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/706cdd81b757c3f5dd1d5a8269274d07384a4201 commit 706cdd81b757c3f5dd1d5a8269274d07384a4201 Author: Geoff Lang <geofflang@chromium.org> Date: Fri Apr 15 19:04:14 2016 Fix incorrect dirty bit state tracking on D3D9 AMD with color masks. When using the zero color mask workaround in StateManager9, the blend state and blend equation were being modified. This caused inconsistant state if the workaround stopped being used in subsequent draw calls. To deal with this issue, dirty the BLEND_ENABLED and BLEND_EQUATION states when the COLOR_MASK changes and dirty the COLOR_MASK state wheneither BLEND_ENABLED or BLEND_EQUATION changes. BUG= 597107 Change-Id: I50ed119c31d40b1c19fc5276b1407342a8a0bd16 Reviewed-on: https://chromium-review.googlesource.com/343540 Reviewed-by: Geoff Lang <geofflang@chromium.org> [modify] https://crrev.com/706cdd81b757c3f5dd1d5a8269274d07384a4201/src/libANGLE/renderer/d3d/d3d9/StateManager9.h [add] https://crrev.com/706cdd81b757c3f5dd1d5a8269274d07384a4201/src/tests/gl_tests/ColorMaskTest.cpp [modify] https://crrev.com/706cdd81b757c3f5dd1d5a8269274d07384a4201/src/tests/angle_end2end_tests.gypi [modify] https://crrev.com/706cdd81b757c3f5dd1d5a8269274d07384a4201/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp [modify] https://crrev.com/706cdd81b757c3f5dd1d5a8269274d07384a4201/src/libANGLE/renderer/d3d/d3d9/Renderer9.cpp
,
May 9 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/4d76061e2be5b8b005afba8e0c145592c3b04527 commit 4d76061e2be5b8b005afba8e0c145592c3b04527 Author: Geoff Lang <geofflang@chromium.org> Date: Mon Apr 25 16:33:44 2016 Make sure D3DRS_TWOSIDEDSTENCILMODE is enabled if the stencil test is. This state-setting code was not ported when adding the dirty bits for D3D9 depth stencil state in https://chromium-review.googlesource.com/#/c/316449. BUG= 597107 Change-Id: I1c7d3f6188d69c4891a5744f16bc8b8852d2bb92 Reviewed-on: https://chromium-review.googlesource.com/343541 Reviewed-by: Geoff Lang <geofflang@chromium.org> [modify] https://crrev.com/4d76061e2be5b8b005afba8e0c145592c3b04527/src/libANGLE/renderer/d3d/d3d9/StateManager9.cpp
,
May 9 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/242de4fdebc941d5ad39b70cfc790b5aaddb741d commit 242de4fdebc941d5ad39b70cfc790b5aaddb741d Author: Geoff Lang <geofflang@chromium.org> Date: Mon May 09 18:09:49 2016 Roll ANGLE 1150560..1a1b30c https://chromium.googlesource.com/angle/angle.git/+log/1150560..4d76061 BUG= 597107 TBR=jmadill@chromium.org NOTRY=true NOPRESUBMIT=true Review URL: https://codereview.chromium.org/1962783002 . Cr-Commit-Position: refs/branch-heads/2661@{#672} Cr-Branched-From: ef6f6ae5e4c96622286b563658d5cd62a6cf1197-refs/heads/master@{#378081} [modify] https://crrev.com/242de4fdebc941d5ad39b70cfc790b5aaddb741d/DEPS
,
May 12 2016
Unfortunately the bug still persists on 50.0.2661.102m. We can provide screenshots if needed.
,
Jun 17 2016
Ping. Should this be Pri-1? If so please assign a milestone.
,
Jun 20 2016
Hey Kostas, sorry I missed your comment. I assume this is fixed now since there hasn't been any other activity on the bug. Going to mark this bug as fixed. Let's open a new one if you're still seeing issues, they're probably a different issue. |
||||||||||||||||||||||||||
►
Sign in to add a comment |
||||||||||||||||||||||||||
Comment 1 by p...@lucidchart.com
, Mar 22 2016