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
static constexpr deUint64 kBatchTimeLimitNanos = 1000000000u; // 1 sec.
static constexpr deUint64 kDeviationErrorLimitNanos = 100000000u; // 100 ms.
static constexpr deUint64 kDeviationWarningLimitNanos = 50000000u; // 50 ms.
- static constexpr deUint64 kDefaultToleranceNanos = 10000000u; // 10 ms.
+ static constexpr deUint64 kDefaultToleranceNanos = 100000000u; // 100 ms.
#if (DE_OS == DE_OS_WIN32)
// Preprocessor used to avoid warning about unused variable.