[SYCL] Drain thread pool should not be called on Win

[KseniyaTikhomirova]

Signed-off-by: Tikhomirova, Kseniya [email protected]

PR #7908 accidentally introduced a behavior that causes lots of timeouts on the CI system on Windows.

[KseniyaTikhomirova]

/verify with intel/llvm-test-suite#1462

[KseniyaTikhomirova]

/verify with intel/llvm-test-suite#1462

[cperkinsintel]

I ran this PR against the llvm-test-suite on Windows. I see some failures, but none of the timeouts that this PR fixes. I think we should merge.

Oh, and the llvm-test-suite run just completed. There is a failure in Basic/event_profiling_info.cpp, but still that's far better than the timeouts which are dragging the whole CI system down. We should merge.

[KseniyaTikhomirova]

failing Basic/event_profiling_info is not related to drain & deferred buffers functionality since it is failing with the following error:
what(): Unable to get command group submission time: Device and/or backend does not support querying timestamp: OpenCL version for device and/or platform is less than 2.1 -59 (PI_ERROR_INVALID_OPERATION)

[bader]

failing Basic/event_profiling_info is not related to drain & deferred buffers functionality since it is failing with the following error: what(): Unable to get command group submission time: Device and/or backend does not support querying timestamp: OpenCL version for device and/or platform is less than 2.1 -59 (PI_ERROR_INVALID_OPERATION)

I'm not sure if current test behavior is friendly for developers. @intel/llvm-reviewers-runtime, do we want the test to fail, or skip is HW doesn't support profiling? Maybe running on OpenCL device with version >= 2.1 must be a prerequisite for launching pre-commit tests?

[againull]

It looks like this PR caused the following failure in post commit:

D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\.\SchedulerTests.exe --gtest_filter=SchedulerTest.HostTaskCleanup
--
D:\github\_work\llvm\llvm\src\sycl\unittests\scheduler\GraphCleanup.cpp(308): error: Expected equality of these values:
  EventImpl->getCommand()
    Which is: 00000[268](https://github.com/intel/llvm/actions/runs/3900784152/jobs/6661852328#step:10:269)560FC350
  nullptr
    Which is: (nullptr)

D:\github\_work\llvm\llvm\src\sycl\unittests\scheduler\GraphCleanup.cpp:308
Expected equality of these values:
  EventImpl->getCommand()
    Which is: 00000268560FC350
  nullptr
    Which is: (nullptr)


********************
Testing:  0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90.. 
********************
Failed Tests (1):
  SYCL-Unit :: scheduler/./SchedulerTests.exe/SchedulerTest/HostTaskCleanup

https://github.com/intel/llvm/actions/runs/3886587010/jobs/6631863893

@KseniyaTikhomirova will you be able to take a look?

[cperkinsintel]

@KseniyaTikhomirova - In reference to the last comment from Artur, about the failing unit test. In my observations, that test seems to pass on Linux, but sometimes fail on Win.

[KseniyaTikhomirova]

It looks like this PR caused the following failure in post commit:

D:\github\_work\llvm\llvm\build\tools\sycl\unittests\scheduler\.\SchedulerTests.exe --gtest_filter=SchedulerTest.HostTaskCleanup
--
D:\github\_work\llvm\llvm\src\sycl\unittests\scheduler\GraphCleanup.cpp(308): error: Expected equality of these values:
  EventImpl->getCommand()
    Which is: 00000[268](https://github.com/intel/llvm/actions/runs/3900784152/jobs/6661852328#step:10:269)560FC350
  nullptr
    Which is: (nullptr)

D:\github\_work\llvm\llvm\src\sycl\unittests\scheduler\GraphCleanup.cpp:308
Expected equality of these values:
  EventImpl->getCommand()
    Which is: 00000268560FC350
  nullptr
    Which is: (nullptr)


********************
Testing:  0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90.. 
********************
Failed Tests (1):
  SYCL-Unit :: scheduler/./SchedulerTests.exe/SchedulerTest/HostTaskCleanup

https://github.com/intel/llvm/actions/runs/3886587010/jobs/6631863893

@KseniyaTikhomirova will you be able to take a look?

My modifications broke attachScheduler() instrumentation for tests. SHould be fixed here #8006