gpu tests fail on the bots in a win GN build |
|||||
Issue descriptionIt looks like the GPU process is just flat-out not available in a Win GN build, either for x86 or x64. This was discovered as part of me attempting to flip the x86 CQ bots over to GN in https://codereview.chromium.org/1908733002/#ps40001 (see the failures on http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/216701 http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_dbg_ng/builds/1301 for starters) chrome://gpu reports that WebGL is unavailable and the GPU process was unable to boot. We don't run any GL tests on any of the existing GN bots, apparently, and so it's a bit unclear if this has ever worked :(. @jbauman - you offered to take a look at this, right? I think doing any GN build at all will show this issue; if not, we can look at it together to see what happens.
,
May 11 2016
In my local GN official 32-bit build and it seems to be working.
,
May 11 2016
Also seems to work for me with an x64 debug build.
,
May 11 2016
It does work on Linux -- Linux has been on GN for a long time now and GPU functionality is working.
,
May 11 2016
okay, apparently the GPU process is disabled if I RDP into my desktop; however, if I connect directly, it seems to work. So, maybe that's not the problem. But, the tests are still failing on the bots, so I still need help :). Let me update the description ...
,
May 11 2016
Some release build issues are GPU-related, but the issues on http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_dbg_ng/builds/1301 all seem to be nacl-related. The error message is, for example: [ RUN ] PPAPINaClNewlibTest.NetworkMonitor [3736:3436:0504/232612:WARNING:chrome_browser_main_win.cc(420)] Command line too long for RegisterApplicationRestart [3736:4448:0504/232614:WARNING:embedded_test_server.cc(190)] Request not handled. Returning 404: /favicon.ico [3736:4448:0504/232614:WARNING:embedded_test_server.cc(190)] Request not handled. Returning 404: /ppapi_nacl_tests_newlib_x64.nexe [3736:3436:0504/232614:INFO:CONSOLE(0)] "NativeClient: could not load nexe url.", source: http://127.0.0.1:59802/test_case.html?mode=nacl_newlib&testcase=NetworkMonitor_Basic,NetworkMonitor_2Monitors,NetworkMonitor_DeleteInCallback, (0) e:\b\build\slave\win\build\src\chrome\test\ppapi\ppapi_test.cc(259): error: Value of: handler.message().c_str() Actual: "Plugin did not load. 'NaCl module load failed: could not load nexe url.'" Expected: "PASS" [ FAILED ] PPAPINaClNewlibTest.NetworkMonitor, where TypeParam = and GetParam() = (3182 ms)
,
May 11 2016
sorry, I keep forgetting, the debug bots don't run the gpu tests either, only the release build matters.
,
May 11 2016
The ppapi thing is bug 609684
,
May 11 2016
Hmm, maybe the isolate is missing d3dcompiler_47.dll. Maybe :copy_compiler_dll should be in data_deps of //third_party/angle:libANGLE.
,
May 12 2016
jbauman: I'll try that and see ...
,
May 12 2016
,
May 12 2016
The following revision refers to this bug: https://chromium.googlesource.com/angle/angle/+/e271a7e4a15e5f58bd7c5a70e29e1f79803636c9 commit e271a7e4a15e5f58bd7c5a70e29e1f79803636c9 Author: John Bauman <jbauman@chromium.org> Date: Thu May 12 00:03:32 2016 Use data_deps to depend on :copy_compiler_dll This should ensure that d3dcompiler_47.dll is saved into an isolate in chromium, as it's used at runtime. BUG= 611235 Change-Id: I64801f00bf90f2be68c7a691724609bbf4d28dfa Reviewed-on: https://chromium-review.googlesource.com/344351 Tryjob-Request: John Bauman <jbauman@chromium.org> Reviewed-by: Kenneth Russell <kbr@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: John Bauman <jbauman@chromium.org> Reviewed-by: Dirk Pranke <dpranke@chromium.org> [modify] https://crrev.com/e271a7e4a15e5f58bd7c5a70e29e1f79803636c9/BUILD.gn
,
May 12 2016
It looks like the missing dll was the problem: https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/220415 (I manually added the data_dep in my patch while waiting for the roll). Does angle have an auto-roller, or do we need to roll this change into chromium manually?
,
May 12 2016
We roll manually, please feel free to run the roll_angle.py script to trigger a roll. I think in tools. If not one of us will get to it today.
,
May 12 2016
,
May 12 2016
Thanks!
,
May 12 2016
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/23ddbe77534f16e9babd8fe09870705458173c8f commit 23ddbe77534f16e9babd8fe09870705458173c8f Author: dpranke <dpranke@chromium.org> Date: Thu May 12 17:34:14 2016 Roll ANGLE 8c9e4d5..e271a7e https://chromium.googlesource.com/angle/angle.git/+log/8c9e4d5..e271a7e TBR=jbauman@chromium.org, jmadill@chromium.org BUG= 611235 , 540829 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/1976633002 Cr-Commit-Position: refs/heads/master@{#393287} [modify] https://crrev.com/23ddbe77534f16e9babd8fe09870705458173c8f/DEPS
,
May 12 2016
,
May 13 2016
|
|||||
►
Sign in to add a comment |
|||||
Comment 1 by thakis@chromium.org
, May 11 2016