[SYCL][CUDA] Fix LIT testing with CUDA devices

[bjoernknafla]

Build the get_device_count_by_type tool with the required preprocessor
definitions to enable CUDA, and link with CUDA.

Passed correct backend name to the get_device_count_by_type tool to
is required to run tests with PI OpenCL instead of PI CUDA.

Clean up the get_device_count_by_type tool code to make it harder to
call it with the wrong backend and get unexpected query results back.

Signed-off-by: Bjoern Knafla bjoern@codeplay.com

[bjoernknafla]

WARNING This PR fixes CUDA LIT testing and multiple LIT tests will unexpectedly pass while more will fail. A follow up PR will fix this as I didn't want to overload this PR.

ninja check-sycl-cuda
...
Unexpected Passing Tests (2):
    SYCL :: basic_tests/boolean.cpp
    SYCL :: hier_par/hier_par_basic.cpp
    SYCL :: kernel_from_file/hw.cpp

********************
Failing Tests (7):
    SYCL :: backend/cuda/primary_context.cpp

    SYCL :: basic_tests/handler/handler_copy_with_offset.cpp
    SYCL :: basic_tests/handler/handler_mem_op.cpp

    SYCL :: group-algorithm/all_of.cpp
    SYCL :: group-algorithm/broadcast.cpp
    SYCL :: group-algorithm/any_of.cpp
    SYCL :: group-algorithm/exclusive_scan.cpp
    SYCL :: group-algorithm/leader.cpp
    SYCL :: group-algorithm/inclusive_scan.cpp
    SYCL :: group-algorithm/none_of.cpp
    SYCL :: group-algorithm/reduce.cpp
    

    SYCL :: program_manager/env_vars.cpp
    SYCL :: multi_ptr/multi_ptr.cpp

    SYCL :: scheduler/DataMovement.cpp
    SYCL :: scheduler/MultipleDevices.cpp

    SYCL :: usm/allocator_vector.cpp
    SYCL :: usm/allocatorll.cpp
    SYCL :: usm/allocator_vector_fail.cpp
    SYCL :: usm/badmalloc.cpp
    SYCL :: usm/depends_on.cpp
    SYCL :: usm/dmemll.cpp
    SYCL :: usm/hmemll.cpp
    SYCL :: usm/memadvise.cpp
    SYCL :: usm/memcpy.cpp
    SYCL :: usm/memset.cpp
    SYCL :: usm/mixed.cpp
    SYCL :: usm/mixed2.cpp
    SYCL :: usm/mixed_queue.cpp
    SYCL :: usm/mixed2template.cpp
    SYCL :: usm/queue_wait.cpp
    SYCL :: usm/smemll.cpp

And the following two tests hang when run for CUDA:

basic_tests/device_event.cpp
regression/group.cpp

Follow up PR to rework: #1303

[romanovvlad]

@bjoernknafla ping

[bjoernknafla]

I have updated the list of failing and now also hanging tests. I will first rework the PR that fixes LIT tests before this PR can go in.

[bader]

@bjoernknafla, could you take a look at merge conflicts, please?

[bjoernknafla]

@bader Will do.

[bjoernknafla]

On a fail LIT prints out this information and eases running the test by hand. Though the user still needs to add the LD_LIBRARY_PATH to find the SYCL library, but that is not part of this change.

[bjoernknafla]

I removed this to avoid duplication and instead use the cudadrv target set up in the sycl/plugin/cuda/CMakeLists.txt CMake file.

[romanovvlad]

LGTM

[bader]

@romanovvlad, please, approve to unblock merge.

[bjoernknafla]

There is an open change request that GitHub cannot identify as the commit (due to rebasing) is gone...

[bader]

@bjoernknafla, basic_tests/buffer/buffer_dev_to_dev.cpp seems to hang. Do you know why?

[bjoernknafla]

I haven’t seen buffer_dev_to_dev.cpp hanging when running it locally (otherwise I wouldn’t have updated the PR). I will take a look (though not today).

[bader]

I've restarted the job and it passed.
I'll keep an eye on this check and if this issue repeats, I'll investigate.