Added a test that retrieved DPCPP compiled kernel and submits them via DPCTLQueue_SubmitRange

[oleksandr-pavlyk]

This does not presently work with coverage turned on due to LLVM version discrepancy between IGC and DPC++.

So the test is excluded for the build with coverage.

[coveralls]

Coverage increased (+1.2%) to 64.548% when pulling 0423551 on use-dpcpp-built-kernels into 521d277 on master.

[oleksandr-pavlyk]

Even though, logically, the sycl::kernel produced by dpcpp_kernel::get_fill_kernel has been given all the arguments that command group function object lambda would capture by reference, removing those from DPCTLQueue_SubmitRange results in an incorrect result (test fails).

Thus, duplicating these captured values via set_args appears needed. Perhaps this is to be expected, as the captured values are captured on the host side and must still be sent over to the device.

[oleksandr-pavlyk]

@diptorupd I have added tests for DCPTLQueue_SubmitNDRange. Expanded tests for int, unsigned int, float and double.

[oleksandr-pavlyk]

To fix the "generate-coverage" workflow it sufficies to remove linking to dpcpp_kernels static library.

The error is

[100%] Linking CXX executable dpctl_c_api_tests
error: Linking COMDATs named '__profd_dpcpp_kernels.cpp__ZN7__spirvL16initGlobalOffsetILi1EN2cl4sycl2idILi1EEEEET0_v': noduplicates has been violated!
dpcpp: error: sycl-link command failed with exit code 1 (use -v to see invocation)

@diptorupd Is there another way to fix it?

[oleksandr-pavlyk]

Failures on Linux are due to OpenCL runtime packages being unavailable on PyPI due to limitation on the size of published packages.

So this can be merged.

[diptorupd]

Merging to master as TeamCity CI fails due to dpcpp_cpp_rt package not being there on pypi.