webgl conformance tests rect_03.html and rect_04.html fail on Mac Intel |
||||||
Issue descriptionWhat steps will reproduce the problem? (1) Start Chromium with --enable-unsafe-es3-apis and run rect_03.html or rect_04.html What is the expected output? Pass What do you see instead? Fail Please use labels and text to provide additional information. Most of the cases for the same category(BlitNearestFilterConsistencyCase) are skipped because of driver issue or hw issue, we can find them in the skipList in deqp. However, it is very interesting that a few cases are left, and they can pass on some platforms. The test files and skipped tests are: Test files Skipped tests nearest_consistency_mag: no reverse, reverse src x, reverse src y, reverse dst x, reverse src and dst y, neares_consistency_min: no reverse, reverse src x, reverse dst x, reverse src and dst x nearest_consistency_out-of-bounds_mag: all tests are skipped. nearest_consistency_out-of-bounds_min: all tests are skipped. Only a few tests are left to test. Unfortunately, blit.rect.nearest_consistency_mag_reverse_src_dst_y fail in rect_03.html, blit.rect.nearest_consistency_min_reverse_src_dst_y fail in rect_04.html. Theoretically, when we reverse both src and dst y, the blitting result should be exactly the same with that of no reversion, so nearest_consistency_mag_reverse_src_dst_y should be exactly the same with nearest_consistency_mag. However, the latter is skipped because of hw issue, but the former is not skipped. I suppose that the latter can pass on some devices, but the former can not. It is strange.
,
Oct 24 2016
We should file all of the known driver bugs on macOS if we ever plan to pass the tests. If the tests are in the skip list for other reasons (such as bugs in graphics drivers on other platforms, or bugs in the tests) then it might not be that critical to file Radars about them.
,
Nov 8 2016
deqp/functional/gles3/framebufferblit/rect_03.html blit.rect.nearest_consistency_mag_reverse_src_dst_y deqp/functional/gles3/framebufferblit/rect_04.html blit.rect.nearest_consistency_min_reverse_src_dst_y These two test cases still fail on 10.12.2 Beta (16C32f) on Intel HD 6000 (likely other Intel GPUs) Ken, can you file a radar on it?
,
Sep 15 2017
Not reproducible on 10.12.6. The bug might have been fixed.
,
Sep 20 2017
Thanks for confirming. Closing as WontFix (no longer reproducible).
,
Nov 15
These tests were unskipped in https://github.com/KhronosGroup/WebGL/pull/2307 However, they failed in a WebGL conformance roll In this tryjob: https://ci.chromium.org/p/chromium/builders/luci.chromium.try/mac_optional_gpu_tests_rel/12209 These shards: https://chromium-swarm.appspot.com/task?id=41294762a5044710&refresh=10&show_raw=1 https://chromium-swarm.appspot.com/task?id=4129477d9be35e10&refresh=10&show_raw=1
,
Nov 15
,
Nov 16
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/f4891a30ea3f5b6222c42595c9be931e8e8d47db commit f4891a30ea3f5b6222c42595c9be931e8e8d47db Author: Austin Eng <enga@chromium.org> Date: Fri Nov 16 04:42:57 2018 Roll WebGL 0d55c88..5b6cbd7 https://chromium.googlesource.com/external/khronosgroup/webgl.git/+log/0d55c88..5b6cbd7 Include suppressions for the following failures: conformance2/textures/misc/copy-texture-image.html On Mac Intel: deqp/functional/gles3/framebufferblit/rect_03.html deqp/functional/gles3/framebufferblit/rect_04.html Bug: 905531 , 830046, 658724 Cq-Include-Trybots: luci.chromium.try:win_optional_gpu_tests_rel;luci.chromium.try:mac_optional_gpu_tests_rel;luci.chromium.try:linux_optional_gpu_tests_rel;luci.chromium.try:android_optional_gpu_tests_rel;luci.chromium.try:linux_angle_rel_ng;luci.chromium.try:win_angle_rel_ng Change-Id: I61d36e9ffab9461ee81cce80f320ab9df095f408 Reviewed-on: https://chromium-review.googlesource.com/c/1335270 Commit-Queue: Austin Eng <enga@chromium.org> Reviewed-by: Kai Ninomiya <kainino@chromium.org> Reviewed-by: Kenneth Russell <kbr@chromium.org> Cr-Commit-Position: refs/heads/master@{#608663} [modify] https://crrev.com/f4891a30ea3f5b6222c42595c9be931e8e8d47db/DEPS [modify] https://crrev.com/f4891a30ea3f5b6222c42595c9be931e8e8d47db/content/test/gpu/gpu_tests/webgl2_conformance_expectations.py [modify] https://crrev.com/f4891a30ea3f5b6222c42595c9be931e8e8d47db/content/test/gpu/gpu_tests/webgl_conformance_revision.txt
,
Nov 27
|
||||||
►
Sign in to add a comment |
||||||
Comment 1 by yunchao...@intel.com
, Oct 24 2016I looked into the test cases, I think it is a driver issue. The test case is simple and clear, and I didn't see any logic error in Chromium for blitFramebuffer for this case. (But I can not double-check this by running native deqp on Mac. Ken & Zhenyao, do you know how to build c++ deqp on Mac? ) IIRC, we didn't file radar for this one. BTW, do we file radar for the tests which are in the skip list for MacOSX? In addition, when I reverse the src/dst coordinates if both src and dst coordinates are reverse, the test case nearest_consistency_{mag|min}_reverse_src_dst_y in rect_03.html and rect_04.html should be exactly the same with the case nearest_consistency_{mag|min}. But all bots can pass with this change(see the result at https://codereview.chromium.org/2444803002/). So, I think maybe we should not skip nearest_consistency_{mag|min}? The skipped tests for this category: https://github.com/KhronosGroup/WebGL/blob/master/sdk/tests/deqp/framework/common/tcuSkipList.js#L99. You know, I have no access to the bug description(Bug 20682200) for C++ deqp: https://android.googlesource.com/platform/external/deqp/+/master/android/cts/master/src/gles3-hw-issues.txt#801 BTW, the tests for blitFramebuffer will blit a checkerboard from src to dst. The only difference of blit_consistency is that the width/height of a grid in the checkerboard is 1, but the other tests' width/height of a grid is 8.