New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 854100 link

Starred by 1 user

Issue metadata

Status: Assigned
Owner:
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Windows
Pri: 2
Type: Bug

Blocked on:
issue 838970

Blocking:
issue 854427



Sign in to add a comment

"gpu_tests.webgl_conformance_integration_test.WebGLConformanceIntegrationTest.WebglConformance_conformance_glsl_variables_gl_pointcoord" is flaky

Project Member Reported by chromium...@appspot.gserviceaccount.com, Jun 19 2018

Issue description

"gpu_tests.webgl_conformance_integration_test.WebGLConformanceIntegrationTest.WebglConformance_conformance_glsl_variables_gl_pointcoord" is flaky.

This issue was created automatically by the chromium-try-flakes app. Please find the right owner to fix the respective test/step and assign this issue to them. If the step/test is infrastructure-related, please add Infra-Troopers label and change issue status to Untriaged. When done, please remove the issue from Sheriff Bug Queue by removing the Sheriff-Chromium label.

We have detected 3 recent flakes. List of all flakes can be found at https://chromium-try-flakes.appspot.com/all_flake_occurrences?key=ahVzfmNocm9taXVtLXRyeS1mbGFrZXNykgELEgVGbGFrZSKGAWdwdV90ZXN0cy53ZWJnbF9jb25mb3JtYW5jZV9pbnRlZ3JhdGlvbl90ZXN0LldlYkdMQ29uZm9ybWFuY2VJbnRlZ3JhdGlvblRlc3QuV2ViZ2xDb25mb3JtYW5jZV9jb25mb3JtYW5jZV9nbHNsX3ZhcmlhYmxlc19nbF9wb2ludGNvb3JkDA.

Flaky tests should be disabled within 30 minutes unless culprit CL is found and reverted. Please see more details here: https://sites.google.com/a/chromium.org/dev/developers/tree-sheriffs/sheriffing-bug-queues#triaging-auto-filed-flakiness-bugs
 

Comment 1 by treib@chromium.org, Jun 19 2018

Cc: kbr@chromium.org zmo@chromium.org bajones@chromium.org
Components: Blink>WebGL

Comment 2 by treib@chromium.org, Jun 19 2018

Labels: -Sheriff-Chromium
Lots of these tests had some flakes but now they all look mostly green:
https://test-results.appspot.com/dashboards/flakiness_dashboard.html#testType=webgl_conformance_gl_passthrough_tests&tests=WebGLConformanceIntegrationTest.WebglConformance

Taking this out of the sheriff queue for now.

Comment 3 by kbr@chromium.org, Jun 20 2018

Blocking: 854427

Comment 4 by kbr@chromium.org, Jun 20 2018

Cc: yang...@intel.com yunchao...@intel.com geoffl...@chromium.org
Labels: -Pri-1 Pri-2
Owner: yang...@intel.com
Status: Assigned (was: Untriaged)
These failures are strange. When the test fails, incorrect pixel values are read back from the framebuffer. These three test flakes (one from a different bot) all have the same failure mode:
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/win_optional_gpu_tests_rel/3629
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/win_optional_gpu_tests_rel/3659
https://ci.chromium.org/p/chromium/builders/luci.chromium.ci/Win10%20FYI%20Exp%20Release%20%28Intel%20HD%20630%29/362

The logs start with:
  AssertionError: pixel 32,32 should be 1,253,0
  at (32, 32) expected: 1,253,0 was 2,2,0
  FAIL pixel 32,32 should be 1,253,0
  at (32, 32) expected: 1,253,0 was 2,2,0
  pixel 48,32 should be 65,253,0
  at (48, 32) expected: 65,253,0 was 66,2,0
  FAIL pixel 48,32 should be 65,253,0
  at (48, 32) expected: 65,253,0 was 66,2,0
...

Looking at the source code for the test:
https://github.com/KhronosGroup/WebGL/blob/master/sdk/tests/conformance/glsl/variables/gl-pointcoord.html

it's pretty straightforward. The test does run against a canvas element which is sized in the HTML to 256x256, and runs synchronously – all of the work is done before the page is loaded. But by definition I don't think layout can be flaky in this situation, causing the canvas to occasionally be incorrectly sized and the test to get a different size than it expected.

Even a failure from about 2 weeks ago has the same failure mode:
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/win_optional_gpu_tests_rel/3150

Yang, Yunchao: since this seems to be coming from the passthrough command decoder's OpenGL backend on Windows / Intel, I think this is likely a rare race condition in point rasterization in the driver. What do you think? Could you try to reproduce and/or ask the driver team to take a look?

The flakiness rate on this configuration is low. Downgrading to P2, but I'll also mark it flaky.

Project Member

Comment 5 by bugdroid1@chromium.org, Jun 20 2018

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/237deff6634f80dda15dab0d5f0fbf1ae2a23aac

commit 237deff6634f80dda15dab0d5f0fbf1ae2a23aac
Author: Kenneth Russell <kbr@chromium.org>
Date: Wed Jun 20 03:58:58 2018

Mark conformance/glsl/variables/gl-pointcoord.html flaky on Win/Intel/OpenGL.

Tbr: geofflang@chromium.org
Bug: 854100
Cq-Include-Trybots: luci.chromium.try:android_optional_gpu_tests_rel;luci.chromium.try:linux_optional_gpu_tests_rel;luci.chromium.try:mac_optional_gpu_tests_rel;luci.chromium.try:win_optional_gpu_tests_rel
Change-Id: I36ec64561f48b12d04269a3b3c031496902292ab
Reviewed-on: https://chromium-review.googlesource.com/1107275
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Commit-Queue: Kenneth Russell <kbr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#568719}
[modify] https://crrev.com/237deff6634f80dda15dab0d5f0fbf1ae2a23aac/content/test/gpu/gpu_tests/webgl_conformance_expectations.py

Labels: OS-Windows
Owner: yizhou.j...@intel.com
Yizhou, please take a look. And file a bug to Intel windows driver team if necessary. 
Hi, I've tested on latest driver on windows 10 angle gl path,but not able to reproduce the bug. 
driver version  GPU    windows version
24.20.100.6094  HD630  17134
24.20.100.6136  HD630  16299
Could you provide the driver version on trybots? I guess the driver is not up to date on trybots. 

Comment 8 by kbr@chromium.org, Jun 23 2018

Blockedon: 838970
Yizhou, thanks for testing. The test is flaky; how many times did you try to reproduce? May need many iterations, and may need to run surrounding tests (e.g. --test-filter=conformance_glsl ).

Here are the bots' configurations:

The bot running the "stable" Intel driver:
Win10 FYI Release (Intel HD 630)
https://cs.chromium.org/chromium/src/testing/buildbot/waterfalls.pyl?q=%22intel+hd+630%22&sq=package:chromium&dr=C&l=1913

The bot running the "experimental" Intel driver:
Win10 FYI Exp Release (Intel HD 630)
https://cs.chromium.org/chromium/src/testing/buildbot/waterfalls.pyl?q=%22intel+hd+630%22&sq=package:chromium&dr=C&l=1872

The stable driver version is:
23.20.16.4877

the experimental version is:
24.20.100.6025

We found significant problems with the experimental version per  Issue 838970 .

Should we try upgrading that bot to a more recent driver? Are the D3D9 DXVA issues fixed in any publicly released driver yet?

Comment 9 by yang...@intel.com, Jun 23 2018

The D3D9 issue hasn't been fixed internally. I will keep you updated on this. 
I also encounter this issue when running ANGLE bots, https://chromium-review.googlesource.com/c/angle/angle/+/1161754
Is it only a driver bug? Does it mean that we also need to update Intel GPU driver of ANGLE?

The error log is the same:
AssertionError: pixel 32,32 should be 1,253,0
  at (32, 32) expected: 1,253,0 was 2,2,0
  FAIL pixel 32,32 should be 1,253,0
  at (32, 32) expected: 1,253,0 was 2,2,0
  pixel 48,32 should be 65,253,0
  at (48, 32) expected: 65,253,0 was 66,2,0
  ....

Xinghua: I assume you're looking at this failure:
https://ci.chromium.org/p/chromium/builders/luci.chromium.try/win_angle_rel_ng/2627

It should have been marked flaky in the WebGL 2.0 conformance tests, not just the WebGL 1.0 tests. Suppressing in https://chromium-review.googlesource.com/1192582 .

Project Member

Comment 12 by bugdroid1@chromium.org, Aug 28

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/903c77e0a162c272e4de79165cae5eb1d149e7bc

commit 903c77e0a162c272e4de79165cae5eb1d149e7bc
Author: Kenneth Russell <kbr@chromium.org>
Date: Tue Aug 28 02:40:25 2018

Suppress gl-pointcoord.html Win/Intel/OpenGL in WebGL 2.0 tests.

conformance/glsl/variables/gl-pointcoord.html is marked flaky in the
WebGL 1.0 conformance tests when running on ANGLE's OpenGL backend on
Intel GPUs, and should be marked similarly in the WebGL 2.0 tests too.

Tbr: geofflang@chromium.org
Bug: 854100
Cq-Include-Trybots: luci.chromium.try:android_optional_gpu_tests_rel;luci.chromium.try:linux_optional_gpu_tests_rel;luci.chromium.try:mac_optional_gpu_tests_rel;luci.chromium.try:win_optional_gpu_tests_rel
Change-Id: I060a5974d68d6f943c3d785f5576b2d4fe8ac0f9
Reviewed-on: https://chromium-review.googlesource.com/1192582
Reviewed-by: Kenneth Russell <kbr@chromium.org>
Commit-Queue: Kenneth Russell <kbr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#586554}
[modify] https://crrev.com/903c77e0a162c272e4de79165cae5eb1d149e7bc/content/test/gpu/gpu_tests/webgl2_conformance_expectations.py

Sign in to add a comment