[SYCL] Added test-e2e for queue, device, and event interop.

[JackAKirk]

All CI backends (cuda, hip, l0, opencl) currently support these three types of interop, that are probably the most important cases, but a complete e2e test did not previously exist. opencl is also the only backend with an existing test-e2e that calls make_* for these three sycl objects.

This adds a short test-e2e corresponding to e.g. #10526.

[jinz2014]

Thank you for adding the test for my better understanding of interop. I have some questions about the program.

Could SYCL_EXT_ONEAPI_BACKEND_CUDA_EXPERIMENTAL become SYCL_EXT_ONEAPI_BACKEND_CUDA ?
Could "IDevice" become "InteropDevice" ?

Thanks

[JackAKirk]

Thank you for adding the test for my better understanding of interop. I have some questions about the program.

Could SYCL_EXT_ONEAPI_BACKEND_CUDA_EXPERIMENTAL become SYCL_EXT_ONEAPI_BACKEND_CUDA ? Could "IDevice" become "InteropDevice" ?

Thanks

• The reason it is SYCL_EXT_ONEAPI_BACKEND_CUDA_EXPERIMENTAL is because the cuda backend spec is in an experimental state. I don't think it is very important, but the EXPERIMENTAL in this macro is the most accurate naming IMO.
• Initially I using the naming InteropDevice, then switched to IDevice because the point is that this sycl device is created from a native device; "InteropDevice" didn't exactly sound like a correct naming since it referred to a sycl device. However I didn't think of a better name so I simply shortened it. If you prefer I will use InteropDevice etc instead.

[jinz2014]

I just read your reply. SyclDevice may be an alternative name, but your pushed your changes. Thanks.

[steffenlarsen]

Instead of using auto, could we consider making backend-dependent type aliases for these? I believe the backend specifications should be making some requirement for what they are, so it would be another level of checks we would do.

[JackAKirk]

Sure,
Do you mean having backend dependent alias' for e.g.

TYPE NativeDevice
such that TYPE == CUdevice for cuda etc?
Or do you mean using
backend_traits<backend::ext_oneapi_cuda>::return_type<device> NativeDevice

where backend_traits<backend::ext_oneapi_cuda>::return_type<device> ==CUdevice?

[steffenlarsen]

I was thinking the former, but we could do both while we are at it, just to make sure that the types adhere to both the backend traits and that the backend traits adhere to the expected type.

[JackAKirk]

Sounds good, how is the last commit for this?

[steffenlarsen]

LGTM!