jie.a.chen@intel.com: can you take a look?

hmm... Jie's small change in WebGL CTS is fair enough. Jie, I think you need to update the webgl cts expectations at first. Then investigate the failure when you have time. 

Make sense. I will update expectation first.

I am rolling WebGL conformance and suppressing it here:

https://chromium-review.googlesource.com/c/chromium/src/+/822914

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/0f7e3c6596e5c237c79bb39821e5d0b170307dbd

commit 0f7e3c6596e5c237c79bb39821e5d0b170307dbd
Author: Zhenyao Mo <zmo@chromium.org>
Date: Wed Dec 13 18:11:10 2017

Roll WebGL 05591bb..1a8ed15

https://chromium.googlesource.com/external/khronosgroup/webgl.git/+log/05591bb..1a8ed15

BUG=794335, 794339 , 794340 ,794341,794613
TEST=bots
TBR=kbr@chromium.org
NOTRY=true

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.win:win_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel;master.tryserver.chromium.angle:linux_angle_rel_ng;master.tryserver.chromium.angle:win_angle_rel_ng

Cq-Include-Trybots: master.tryserver.chromium.android:android_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel
Change-Id: I3d5567031523d5c1a8558b967d40835b9c1da77b
Reviewed-on: https://chromium-review.googlesource.com/822914
Commit-Queue: Zhenyao Mo <zmo@chromium.org>
Reviewed-by: Zhenyao Mo <zmo@chromium.org>
Cr-Commit-Position: refs/heads/master@{#523819}
[modify] https://crrev.com/0f7e3c6596e5c237c79bb39821e5d0b170307dbd/DEPS
[modify] https://crrev.com/0f7e3c6596e5c237c79bb39821e5d0b170307dbd/content/test/gpu/gpu_tests/webgl2_conformance_expectations.py
[modify] https://crrev.com/0f7e3c6596e5c237c79bb39821e5d0b170307dbd/content/test/gpu/gpu_tests/webgl_conformance_expectations.py
[modify] https://crrev.com/0f7e3c6596e5c237c79bb39821e5d0b170307dbd/content/test/gpu/gpu_tests/webgl_conformance_revision.txt

I have tested on several platforms including mac, linux, and windows for intel and nvdia against the 
 https://chromium.googlesource.com/external/khronosgroup/webgl.git/+/b7c786a9a8db12cea2faa3402c522ef6df430465.

It only caused regression on NV linux and windows.

Intel Kabylake and Skylake were all okay on linux, window, and mac.

Olli, can you take a look at this? According to Jie, this passes on Intel but fails on NVidia across platforms.

I ran on my Linux NVidia, below are the failing case:

Start testcase: texture_functions.texturelodoffset.sampler3d_float_vertex
Fuzzy image comparison failed: difference = 0.17093482316641254, threshold = 0.05

So it seems like a precision issue on NVidia's vertex shadering 3d float texture sampling.

What exact config did you use to repro the texture_functions.texturelodoffset.sampler3d_float_vertex failure (GPU, OS driver version)?

I tried to run the native ANGLE dEQP tests on GTX 1080 and Windows 10 with driver version 388.59, and the tests passed:

gyp\Debug_x64\angle_deqp_gles3_tests.exe --deqp-egl-display-type=angle-gl --deqp-case=dEQP-GLES3.functional.shaders.texture_functions.texturelodoffset.sampler3d*

Or does the failure happen only with the WebGL version of the test?

He's talking about the WebGL 2 test version. I haven't looked at this particular test.

https://www.khronos.org/registry/webgl/sdk/tests/deqp/functional/gles3/shadertexturefunction/texturelodoffset.html?webglVersion=2

Olli: I am running on Linux with NVidia Quadro 600, driver: 384.90

Jamie, how do I build the angle_deqp_gles3_tests target? So I can verify if the native version passes or not on my Linux machine.

Mo: you should be able to build it by setting build_angle_deqp_tests = true in your gn args, then using ninja to build angle_deqp_gles3_tests as per the usual way. you can use a command line similar to what Olli posted in comment #9 to run the specific test with the right ANGLE back-end.

angle_deqp_gles3_tests pass locally. So it seems there is a discrepancy between local test and WebGL2's version.

Unassign Olli.

If anyone with a machine with NVidia GPU can figure out what WebGL2 version deviates from native dEQP, I'd appreciate the help. Otherwise I'll work around to take a look -- but won't be anytime soon.

Flake seen on Mac with AMD GPU in WebglConformance_deqp_functional_gles3_shadertexturefunction_texturelod here:
https://ci.chromium.org/buildbot/tryserver.chromium.mac/mac_optional_gpu_tests_rel/20212

https://chromium-swarm.appspot.com/task?id=3b240a057d62de10&refresh=10&show_raw=1

Should we mark this flaky?

Log:

  Init testcase: texture_functions
  Init testcase: texture_functions.texturelod.sampler2d_fixed_vertex
  
  Init testcase: texture_functions.texturelod.sampler2d_fixed_fragment
  
  Init testcase: texture_functions.texturelod.sampler2d_float_vertex
  
  Init testcase: texture_functions.texturelod.sampler2d_float_fragment
  
  Init testcase: texture_functions.texturelod.isampler2d_vertex
  
  Init testcase: texture_functions.texturelod.isampler2d_fragment
  
  Init testcase: texture_functions.texturelod.usampler2d_vertex
  
  Init testcase: texture_functions.texturelod.usampler2d_fragment
  
  Init testcase: texture_functions.texturelod.samplercube_fixed_vertex
  
  Init testcase: texture_functions.texturelod.samplercube_fixed_fragment
  
  Init testcase: texture_functions.texturelod.samplercube_float_vertex
  
  Init testcase: texture_functions.texturelod.samplercube_float_fragment
  
  Init testcase: texture_functions.texturelod.isamplercube_vertex
  
  Init testcase: texture_functions.texturelod.isamplercube_fragment
  
  Init testcase: texture_functions.texturelod.usamplercube_vertex
  
  FAIL Fail
  Init testcase: texture_functions.texturelod.usamplercube_fragment
  
  texture_functions.texturelod.usamplercube_fragment: Fail
  FAIL Fail
  Init testcase: texture_functions.texturelod.sampler2darray_fixed_vertex
  
  Init testcase: texture_functions.texturelod.sampler2darray_fixed_fragment
  
  Init testcase: texture_functions.texturelod.sampler2darray_float_vertex
  
  Init testcase: texture_functions.texturelod.sampler2darray_float_fragment
  
  Init testcase: texture_functions.texturelod.isampler2darray_vertex
  
  Init testcase: texture_functions.texturelod.isampler2darray_fragment
  
  Init testcase: texture_functions.texturelod.usampler2darray_vertex
  
  Init testcase: texture_functions.texturelod.usampler2darray_fragment
  
  Init testcase: texture_functions.texturelod.sampler3d_fixed_vertex
  
  Init testcase: texture_functions.texturelod.sampler3d_fixed_fragment
  
  Init testcase: texture_functions.texturelod.sampler3d_float_vertex
  
  Init testcase: texture_functions.texturelod.sampler3d_float_fragment
  
  Init testcase: texture_functions.texturelod.isampler3d_vertex
  
  Init testcase: texture_functions.texturelod.isampler3d_fragment
  
  Init testcase: texture_functions.texturelod.usampler3d_vertex
  
  Init testcase: texture_functions.texturelod.usampler3d_fragment

Adding James Helferty (jhelferty@nvidia.com) to CC for WebGL conformance issues on Nvidia GPUs.

WebGL test case that fails on NVIDIA (GTX1080, Windows, driver 416.34, both D3D11 and Windows):
texture_functions.texturelodoffset.sampler3d_float_vertex
https://www.khronos.org/registry/webgl/sdk/tests/deqp/functional/gles3/shadertexturefunction/texturelodoffset.html?webglVersion=2


The command mentioned in the comment #9 does not exercise the native deqp test that fails here. This is filed as bug angleproject:3094.

This test should be added to the skip list. See issue angleproject:3094.

Thanks! I'll go ahead and do that.