Lower correlation threshold in flush-finish tests again
authorGurchetan Singh <gurchetansingh@chromium.org>
Fri, 6 Oct 2017 16:06:21 +0000 (09:06 -0700)
committerGurchetan Singh <gurchetansingh@google.com>
Tue, 10 Oct 2017 20:17:31 +0000 (13:17 -0700)
commit59f4fcde5ac1402ab3b911cf6a9f66b886c10464
tree26eb328a47829afecdabedf6b95047c6a005207b
parent3a1de9094d072324ef2c16a21c1893b881c2a8e7
Lower correlation threshold in flush-finish tests again

This test expects predictability from glFlush/glFinish timing.
In particular,  the code from FlushFinishCase::analyzeResults()
checks that the result falls within specific timing thresholds.

Notice that in the GLES3 specification, there is no requirement for
timing, just a guarantee that "commands [...] complete in finite time".
This is a problem because it makes the test very flaky, for example
if the CPU/GPU clocks are not ramping up fast enough.

BUG=b:67331095
TEST=Ran dEQP-GLES3.functional.flush_finish.finish on Elm for over 3+ hours
     Observed 1.5% to 2% flakiness without this patch. Observed no flakiness
     with this patch.
modules/gles3/functional/es3fFlushFinishTests.cpp