[SYCL][Graph] Support for native-command

[EwanC]

Support sycl_ext_codeplay_enqueue_native_command with SYCL-Graph for all of L0, CUDA, HIP, and OpenCL backends.

Introduces interop_handle::ext_oneapi_get_native_graph<backend>() to give the user access to the native graph object which native commands can be appended to. Implemented using new UR command-buffer entry-points urCommandBufferAppendNativeCommandExp and urCommandBufferGetNativeHandleExp.

To use CUDA as an example, code using ext_codeplay_enqueue_native_command eagerly can be updated from:

 CGH.ext_codeplay_enqueue_native_command([=](interop_handle IH) {
       auto NativeStream = IH.get_native_queue<cuda>();
       myNativeLibraryCall(NativeStream);
}

To

CGH.ext_codeplay_enqueue_native_command([=](interop_handle IH) {
     if (IH.ext_oneapi_has_graph())  {
       auto NativeGraph = IH.ext_oneapi_get_native_graph<cuda>();
       auto NativeStream = IH.get_native_queue<cuda>();

       // Start capture stream calls into graph
       cuStreamBeginCaptureToGraph(NativeStream, NativeGraph, nullptr,
                                             nullptr, 0,
                                             CU_STREAM_CAPTURE_MODE_GLOBAL);

       myNativeLibraryCall(NativeStream);

       // Stop capturing stream calls into graph
       cuStreamEndCapture(NativeStream, &NativeGraph);
     } else {
       auto NativeStream = IH.get_native_queue<cuda>();
       myNativeLibraryCall(NativeStream );
    }
}

Example of how this integration could work in GROMACS https://gitlab.com/gromacs/gromacs/-/merge_requests/4954

[npmiller]

CUDA/HIP LGTM

[Pennycook]

Specification changes LGTM. Thanks for adding the other examples!

[steffenlarsen]

Few small comments but otherwise LGTM!

[martygrant]

UR changes LGTM. The new CTS test checks the new urCommandBufferGetNativeHandleExp function, but AppendNativeCommandExp was also added, should there be a test for this too?

[kbenzie]

UR LGTM

[EwanC]

UR changes LGTM. The new CTS test checks the new urCommandBufferGetNativeHandleExp function, but AppendNativeCommandExp was also added, should there be a test for this too?

Good question, I briefly looked into this but the amount of work to get testing setup for all the L0,CUDA,HIP, OpenCL adapters was a fair bit due to the need for backend specific code, so I stuck with using E2E tests for verification. I agree this is something we should do, so I created #17448 as follow-on work for adding UR CTS testing.

[keyradical]

LGTM!

[hvdijk]

NativeCPU looks fine to mark this as unsupported.

[AerialMantis]

LGTM

[EwanC]

@intel/llvm-gatekeepers This is ready to merge, thanks