Add cl_khr_kernel_clock tests #1960
Conversation
…er defines for extension subtest
test_conformance/compiler/test_compiler_defines_for_extensions.cpp
Outdated
Show resolved
Hide resolved
…xtensions array. Add newline to CMakeLists.txt Wrap everything under an anonymous namespace, remove extern keyword, use c++ raw strings for the ocl kernels, and remove underscore from CL_KHR_KERNEL_CLOCK_PROCS_H
| if (buf == 1) | ||
| { | ||
| log_error( | ||
| "Sampling the clock returned bad values, time1 > time2.\n"); |
There was a problem hiding this comment.
Not sure if the extension provides a guarantee that time2 > time1? The description for clock_read_* says
an observer may see sampled values wrap around zero.
which would make the test fail spuriously if a wraparound happens between the time1 and time2 sampling points.
I tried to think of a more robust way to test the returned values, but I'm not even sure if we have a guarantee that time1 != time2? What do other reviewers think?
There was a problem hiding this comment.
There was a bit of discussion on GitLab about this also. I agree there could be a problem here in theory, but these tests are currently matching the behavior for the Vulkan tests, see:
Since there doesn't seem to have been a problem with the Vulkan tests in practice so far, I'm inclined to keep these tests as-is, and if we end up needing something fancier we can cross that bridge when we come to it.
There was a problem hiding this comment.
Thanks for cross-checking, that sounds reasonable to me then.
|
Reviewing |
|
Merging as discussed in the May 21st teleconference. |
Adds cl_khr_kernel_clock test.
Also fixes failure in the compiler defines for extension compiler subtest when cl_khr_kernel_clock is supported.