[SYCL] Mark sycl::marray device copyable

[vladimirlaz]

sycl::marray is device copyable if the element type is device copyable.
Also test was added to verify that the following types are device copyable:

• sycl::marray when T is device copyable;
• sycl::vec (all supported element types are device copyable);
• sycl::id;
• sycl::range.

[vladimirlaz]

@intel/llvm-reviewers-runtime, could you please review the PR?

[alexbatashev]

Overall LGTM, but I think we need a better test. On Windows the compiler can (possibly) produce types with different alignment on host and device. We either need a runtime test or a static test, that checks the alignment.

[vladimirlaz]

we need a better test. On Windows the compiler can (possibly) produce types with
@alexbatashev, my expectation here is that if a type is device copyable it should have the same alignment on host and device. Otherwise scalar variable of element type will go into trouble as well.

[alexbatashev]

@alexbatashev, my expectation here is that if a type is device copyable it should have the same alignment on host and device. Otherwise scalar variable of element type will go into trouble as well.

I share these expectations, I'm just not sure if compiler has the same point of view. Since device and host compilation are two separate processes, with two different ABIs. For SYCL-defined types we must ensure ABI is the same on host and device, at least we should update this test: https://github.com/intel/llvm/blob/sycl/sycl/test/abi/symbol_size_alignment.cpp

[vladimirlaz]

@alexbatashev, my expectation here is that if a type is device copyable it should have the same alignment on host and device. Otherwise scalar variable of element type will go into trouble as well.

I share these expectations, I'm just not sure if compiler has the same point of view. Since device and host compilation are two separate processes, with two different ABIs. For SYCL-defined types we must ensure ABI is the same on host and device, at least we should update this test: https://github.com/intel/llvm/blob/sycl/sycl/test/abi/symbol_size_alignment.cpp

The requested test change does not look like connected to the PR. If there are gaps in testing it should be covered as separate issue/PR.