Increase the sleep measurement tolerance to 100ms
authorPiers Daniell <pdaniell@nvidia.com>
Tue, 10 Sep 2019 22:33:58 +0000 (16:33 -0600)
committerAlexander Galazin <Alexander.Galazin@arm.com>
Mon, 16 Sep 2019 06:36:14 +0000 (02:36 -0400)
commitb7e9a31fe5687d303b026e4dfbac3998d2c06273
treeb2253db2a1d8080fd0ff7ba1eef5a442988a6d5b
parent4cd377357a171572ee268ff32df937bf0e162443
Increase the sleep measurement tolerance to 100ms

The calibration_test works by measuring the "before"
device time, sleeping the thread for 200ms, and then
measuring the "after" device time. It then checks
the "after"-"before" delta is not more than 200ms
plus/minus 10ms.

This 10ms is too short for many systems and may
result in false failures. If the system is heavily
loaded the thread may sleep for more than 200ms.

Increasing the tolerance to 100ms appears to
eliminate the false failure I've been seeing.

Affects:

dEQP-VK.pipeline.timestamp.calibrated.calibration_test

Components: Vulkan

VK-GL-CTS issue: 1990

Change-Id: Ifc7f877daf8e361e39b3cd9ee447445529fd01d7
external/vulkancts/modules/vulkan/pipeline/vktPipelineTimestampTests.cpp