dpctl doesn't support the cl::sycl::kernel in pybind11

[chengjunlu]

I met issue when integrate the cl::sycl::kernel into python.

The error log like this:

TypeError: Unable to convert function return value to a Python type! The signature was
	(arg0: str, arg1: str, arg2: int, arg3: dpctl.SyclDevice) -> Tuple[int, cl::sycl::kernel, int, int]

It seems the sycl::device has been integrated successfully. But the cl::sycl::kernel is not supported.

Is there any plan to support the cl::sycl::kernel?

[chengjunlu]

@oleksandr-pavlyk

Would you have any suggestion about the my issue?

[oleksandr-pavlyk]

For it to be added, we need to make Cython definitions of dpctl.program.SyclProgram and dpctl.program.SyclKernel public, add respective API and implement integration with pybind11. This definitely can be done.

Could you please share why this would be useful?

[chengjunlu]

I am working on integrating the Triton to SYCL for experimental.

There is some sycl::kernel generated by Triton and pass to other python package to run.

This feature would be great help in AI compiler component integration.

[oleksandr-pavlyk]

@chengjunlu Can you please look at #970 and see if it meets your needs?

[chengjunlu]

@oleksandr-pavlyk
I quick reviewed the code changes. And looks good to me.

Do you want to me verify the branch before you merge it to main trunk?

[oleksandr-pavlyk]

@oleksandr-pavlyk I quick reviewed the code changes. And looks good to me.

Do you want to me verify the branch before you merge it to main trunk?

Yes, that would be most helpful. Thank you

[chengjunlu]

@oleksandr-pavlyk I quick reviewed the code changes. And looks good to me.
Do you want to me verify the branch before you merge it to main trunk?

Yes, that would be most helpful. Thank you

I have tested it. And it works properly.