CTS test profiling/write_image_uchar can fail with CL_PROFILING_COMMAND_START > CL_PROFILING_COMMAND_END #843
Description
This is similar to #783 which was likely fixed by #784. That issue was for SUBMIT > START. This is for START > END. :-)
It seems to be infrequent, but I hit it three times in the same internal build today.
I was using clvk commit 78aa08a which is the current top of tree.
Here's a part of a sample run:
Requesting Default device based on command line for platform index 0 and device index 0
Compute Device Name = Tesla V100-SXM2-16GB, Compute Device Vendor = NVIDIA Corporation, Compute Device Version = OpenCL 3.0 CLVK on Vulkan v1.3.242 driver 2246951296, CL C Version = OpenCL C 1.2 CLVK on Vulkan v1.3.242 driver 2246951296
Device latest conformance version passed: v2023-12-12-00
Supports single precision denormals: NO
sizeof( void*) = 8 (host)
sizeof( void*) = 4 (device)
write_image_uchar...
Image verified.
CL_PROFILING_COMMAND_QUEUED: 2225772481946636 CL_PROFILING_COMMAND_SUBMIT: 2225772481973894 CL_PROFILING_COMMAND_START: 2225772482950548 CL_PROFILING_COMMAND_END: 2225772482917843 CL_DEVICE_PROFILING_TIMER_RESOLUTION: 1
Profiling info:
Time from queue to submit : 0.027258ms
Time from submit to start : 0.976654ms
Time from start to end: 18446744073709.519531ms
CL_PROFILING_COMMAND_START > CL_PROFILING_COMMAND_END.
Time between queue and start is too big: 18446744073.709518s, test limit: 30.000000s.
write_image_uchar FAILED
PASSED sub-test.
FAILED test.
This is pretty low priority for us: a retries tend to work around it.