PipelineIntegrationTest.BasicPlaybackOpus*TrimmingHashed* tests fail with ToT Clang |
||
Issue descriptionMy first guess is that some LLVM vectorizer or other FP change has changed the output of these hashing tests. I've reproduced on Linux and am going to dig into which objects trigger the problem. Example builds: https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinuxLLD%20tester/builds/124 https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinux%20tester/builds/1981 [ RUN ] PipelineIntegrationTest.BasicPlaybackOpusOggTrimmingHashed [8434:8434:0816/061820:17602218048:ERROR:ffmpeg_demuxer.cc(1595)] OnReadFrameDone result=-541478725 IsMaxMemoryUsageReached=0 ../../media/test/pipeline_integration_test.cc:1032: Failure Value of: GetAudioHash() Actual: "-12.96,-51.29,-32.11,-23.16,-72.73,17.14," Expected: kOpusEndTrimmingHash_1 Which is: "-4.56,-5.65,-6.51,-6.29,-4.36,-3.59," [8434:8434:0816/061820:17602278255:ERROR:ffmpeg_demuxer.cc(1595)] OnReadFrameDone result=-541478725 IsMaxMemoryUsageReached=0 ../../media/test/pipeline_integration_test.cc:1038: Failure Value of: GetAudioHash() Actual: "-47.65,-74.06,-78.57,-16.24,-85.82,-22.52," Expected: kOpusEndTrimmingHash_2 Which is: "-11.89,-11.09,-8.25,-7.11,-7.84,-9.97," [8434:8434:0816/061820:17602313240:ERROR:ffmpeg_demuxer.cc(1595)] OnReadFrameDone result=-541478725 IsMaxMemoryUsageReached=0 ../../media/test/pipeline_integration_test.cc:1045: Failure Value of: GetAudioHash() Actual: "-93.77,-86.38,-105.66,-57.83,-121.92,-44.28," Expected: kOpusEndTrimmingHash_3 Which is: "-13.28,-14.35,-13.67,-11.68,-10.18,-10.46," [ FAILED ] PipelineIntegrationTest.BasicPlaybackOpusOggTrimmingHashed (150 ms)
,
Aug 16 2016
Bisecting over bad: 278762 good: 278721 Based on Hans's roll revision and the last time https://build.chromium.org/p/chromium.fyi/builders/ClangToTLinux%20tester/ was green.
,
Aug 16 2016
This is the only change that stands out to me in that range: ------------------------------------------------------------------------ r278731 | dlkreitz | 2016-08-15 13:21:41 -0700 (Mon, 15 Aug 2016) | 6 lines Enhance SCEV to compute the trip count for some loops with unknown stride. Patch by Pankaj Chawla Differential Revision: https://reviews.llvm.org/D22377 ------------------------------------------------------------------------
,
Aug 16 2016
Agreed, testing at the revision before that now. I'm not sure how to reduce a test case for it though.
,
Aug 16 2016
... and testing before and after shows that r278731 is the culprit. Now to get the reduction.
,
Aug 16 2016
Reverted in r278853, running more tryjobs.
,
Aug 16 2016
|
||
►
Sign in to add a comment |
||
Comment 1 by r...@chromium.org
, Aug 16 2016Components: Internals>Media>Audio
Labels: -OS-Windows