webgl_conformance_tests on NVIDIA GPU on Windows (with patch) on Windows-2008ServerR2-SP1 fail on 32-bit clang builds |
|||
Issue descriptionhttps://codereview.chromium.org/2870543003/ reliably fails win_chromium_rel_ng, with webgl_conformance_tests failing every time. The output is a bit hard to read, but I think this is one of the errors: [4/859] gpu_tests.webgl_conformance_integration_test.WebGLConformanceIntegrationTest.WebglConformance_conformance_attribs_gl_disabled_vertex_attrib failed unexpectedly 13.9510s: Traceback (most recent call last): _RunGpuTest at content\test\gpu\gpu_tests\gpu_integration_test.py:73 self.RunActualGpuTest(url, *args) RunActualGpuTest at content\test\gpu\gpu_tests\webgl_conformance_integration_test.py:203 getattr(self, test_name)(test_path, *args[1:]) _RunConformanceTest at content\test\gpu\gpu_tests\webgl_conformance_integration_test.py:217 self._CheckTestCompletion() _CheckTestCompletion at content\test\gpu\gpu_tests\webgl_conformance_integration_test.py:213 self.fail(self._WebGLTestMessages(self.tab)) fail at c:\b\depot_tools\python276_bin\lib\unittest\case.py:412 raise self.failureException(msg) AssertionError: should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 FAIL should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 *** Error compiling VERTEX_SHADER '[object WebGLShader]': *** Error compiling FRAGMENT_SHADER '[object WebGLShader]': Error in compiling shader should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 FAIL should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 *** Error compiling VERTEX_SHADER '[object WebGLShader]': *** Error compiling FRAGMENT_SHADER '[object WebGLShader]': Error in compiling shader should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 FAIL should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 *** Error compiling VERTEX_SHADER '[object WebGLShader]': *** Error compiling FRAGMENT_SHADER '[object WebGLShader]': Error in compiling shader should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 FAIL should be green at (0, 0) expected: 0,255,0,255 was 0,0,0,0 *** Error compiling VERTEX_SHADER '[object WebGLShader]': *** Error compiling FRAGMENT_SHADER '[object WebGLShader]': Error in compiling shader should be green kbr@, can you describe briefly what this test tests for? Can you think of anything compiler-specific in these tests? (Do these tests happen to use deqp and could this be related to issue 722345 ? Since that one's on linux and this here is windows, probably not?) We don't run these tests on any current clang/win bots, could we just add them, or do they require special setup on a bot?
,
May 16 2017
Also: these tests require a real GPU for effectiveness, which is why they're run on physical hardware and triggered by win_chromium_rel_ng (and mac_, linux_).
,
May 16 2017
Is angle not used by any other tests? https://build.chromium.org/p/chromium.fyi/console?category=win%20clang has many many test suites passing (browser_tests, unit_tests, etc)
,
May 16 2017
Those tests are all running with software rendering. The fact that the VMs running those tests can't boot the GPU process was the main reason for bringing up the physical GPU hardware. Note that you can trigger the GPU tests (as defined by the test suites listed in src/content/test/gpu/generate_buildbot_json.py) from any waterfall builder that can invoke its tests on Swarming.
,
May 16 2017
I believe the second crash (gl::ComputeVertexAttributeStride) should be fixed by https://chromium-review.googlesource.com/c/505928/. This will be rolled into Chrome today.
,
May 16 2017
Is it ok if we add these tests to ~10 fairly slow-cycling bots? Compared to the CQ that should require negligible resources, right?
,
May 16 2017
Yes, that should be fine. It would be ideal if you could generate those bots' JSON files from that generate_buildbot_json.py script, like was recently done for the client.v8.fyi waterfall, so they don't get out of sync. Would you be willing to do that? Also, in for example https://cs.chromium.org/chromium/build/scripts/slave/recipe_modules/chromium_tests/chromium_gpu_fyi.py we use the 'serialize_tests' property to avoid triggering all of the GPU tests in parallel on a bunch of the waterfall bots to reduce load on the physical hardware. If cycle time isn't critical for your ~10 bots could we do the same on them?
,
May 16 2017
The first stack is probably here https://cs.chromium.org/chromium/src/gpu/command_buffer/service/gles2_cmd_decoder.cc?q=DoDrawElements+package:%5Echromium$&dr=CSs&l=10291 given that feature_info_ is a scoped_refptr. But from what I can see, that's always allocated here https://cs.chromium.org/chromium/src/gpu/ipc/service/gpu_command_buffer_stub.cc?dr=CSs&l=577 and then passed through, so I'm not sure how it'd end up corrupted.
,
May 16 2017
,
Jul 31 2017
Looks like https://chromium-review.googlesource.com/c/505928/ fixed this. |
|||
►
Sign in to add a comment |
|||
Comment 1 by kbr@chromium.org
, May 16 2017