[SYCL][NFC] Remove unused host kernel member in CGExecKernel

[steffenlarsen]

The MHostKernel member in CGExecKernel was unused, so this commit removes it.

[steffenlarsen]

Looks like the host kernel copy is implicitly being used to keep the contents of it alive during scheduling. Converting this to draft for now.

[Pennycook]

Looks like the host kernel copy is implicitly being used to keep the contents of it alive during scheduling. Converting this to draft for now.

Which tests failed? If you revert my changes from #18387, does this work?

My fast path currently depends on the HostKernel member for lifetime extension -- I figured that since it was already there, I could make it do something useful! If my fast path is the only thing using this mechanism, that would be good to know, because it would be fairly straightforward to store the arguments somewhere else.

[steffenlarsen]

Which tests failed?

  SYCL :: Basic/empty_command.cpp
  SYCL :: Basic/out_of_order_queue_status_ext_oneapi_empty.cpp
  SYCL :: Basic/out_of_order_queue_status_khr_empty.cpp
  SYCL :: Graph/Explicit/add_nodes_after_finalize.cpp
  SYCL :: Graph/Explicit/basic_usm.cpp
  SYCL :: Graph/Explicit/basic_usm_host.cpp
  SYCL :: Graph/Explicit/basic_usm_mixed.cpp
  SYCL :: Graph/Explicit/basic_usm_shared.cpp
  SYCL :: Graph/Explicit/compile_time_local_memory.cpp
  SYCL :: Graph/Explicit/depends_on.cpp
  SYCL :: Graph/Explicit/empty_node.cpp
  SYCL :: Graph/Explicit/enqueue_ordering.cpp
  SYCL :: Graph/Explicit/host_task.cpp
  SYCL :: Graph/Explicit/host_task2.cpp
  SYCL :: Graph/Explicit/host_task_last.cpp
  SYCL :: Graph/Explicit/host_task_multiple_deps.cpp
  SYCL :: Graph/Explicit/host_task_multiple_roots.cpp
  SYCL :: Graph/Explicit/host_task_successive.cpp
  SYCL :: Graph/Explicit/local_accessor.cpp
  SYCL :: Graph/Explicit/local_accessor_multiple_accessors.cpp
  SYCL :: Graph/Explicit/local_accessor_multiple_nodes.cpp
  SYCL :: Graph/Explicit/memadvise.cpp
  SYCL :: Graph/Explicit/multiple_exec_graphs.cpp
  SYCL :: Graph/Explicit/node_ordering.cpp
  SYCL :: Graph/Explicit/prefetch.cpp
  SYCL :: Graph/Explicit/queue_constructor_usm.cpp
  SYCL :: Graph/Explicit/queue_shortcuts.cpp
  SYCL :: Graph/Explicit/repeated_exec.cpp
  SYCL :: Graph/Explicit/single_node.cpp
  SYCL :: Graph/Explicit/sub_graph.cpp
  SYCL :: Graph/Explicit/sub_graph_execute_without_parent.cpp
  SYCL :: Graph/Explicit/sub_graph_multiple_submission.cpp
  SYCL :: Graph/Explicit/sub_graph_nested.cpp
  SYCL :: Graph/Explicit/sub_graph_two_parent_graphs.cpp
  SYCL :: Graph/Explicit/usm_copy.cpp
  SYCL :: Graph/Explicit/work_group_memory.cpp
  SYCL :: Graph/NativeCommand/hip_explicit_usm.cpp
  SYCL :: Graph/NativeCommand/hip_record_usm.cpp
  SYCL :: Graph/RecordReplay/add_nodes_after_finalize.cpp
  SYCL :: Graph/RecordReplay/after_use.cpp
  SYCL :: Graph/RecordReplay/barrier_multi_graph.cpp
  SYCL :: Graph/RecordReplay/barrier_multi_queue.cpp
  SYCL :: Graph/RecordReplay/barrier_with_work.cpp
  SYCL :: Graph/RecordReplay/basic_usm.cpp
  SYCL :: Graph/RecordReplay/basic_usm_host.cpp
  SYCL :: Graph/RecordReplay/basic_usm_mixed.cpp
  SYCL :: Graph/RecordReplay/basic_usm_shared.cpp
  SYCL :: Graph/RecordReplay/compile_time_local_memory.cpp
  SYCL :: Graph/RecordReplay/dotp_in_order.cpp
  SYCL :: Graph/RecordReplay/dotp_in_order_pause.cpp
  SYCL :: Graph/RecordReplay/dotp_in_order_with_empty_nodes.cpp
  SYCL :: Graph/RecordReplay/dotp_multiple_queues.cpp
  SYCL :: Graph/RecordReplay/empty_node.cpp
  SYCL :: Graph/RecordReplay/ext_oneapi_enqueue_functions.cpp
  SYCL :: Graph/RecordReplay/ext_oneapi_enqueue_functions_submit_with_event.cpp
  SYCL :: Graph/RecordReplay/host_task.cpp
  SYCL :: Graph/RecordReplay/host_task2.cpp
  SYCL :: Graph/RecordReplay/host_task2_multiple_roots.cpp
  SYCL :: Graph/RecordReplay/host_task_in_order.cpp
  SYCL :: Graph/RecordReplay/host_task_last.cpp
  SYCL :: Graph/RecordReplay/host_task_multiple_deps.cpp
  SYCL :: Graph/RecordReplay/host_task_multiple_roots.cpp
  SYCL :: Graph/RecordReplay/host_task_successive.cpp
  SYCL :: Graph/RecordReplay/in_order_queue_with_host_managed_dependencies.cpp
  SYCL :: Graph/RecordReplay/in_order_queue_with_host_managed_dependencies_memcpy.cpp
  SYCL :: Graph/RecordReplay/in_order_queue_with_host_managed_dependencies_memset.cpp
  SYCL :: Graph/RecordReplay/local_accessor.cpp
  SYCL :: Graph/RecordReplay/local_accessor_multiple_accessors.cpp
  SYCL :: Graph/RecordReplay/local_accessor_multiple_nodes.cpp
  SYCL :: Graph/RecordReplay/memadvise.cpp
  SYCL :: Graph/RecordReplay/multiple_exec_graphs.cpp
  SYCL :: Graph/RecordReplay/prefetch.cpp
  SYCL :: Graph/RecordReplay/queue_constructor_usm.cpp
  SYCL :: Graph/RecordReplay/queue_shortcuts.cpp
  SYCL :: Graph/RecordReplay/repeated_exec.cpp
  SYCL :: Graph/RecordReplay/sub_graph.cpp
  SYCL :: Graph/RecordReplay/sub_graph_execute_without_parent.cpp
  SYCL :: Graph/RecordReplay/sub_graph_in_order.cpp
  SYCL :: Graph/RecordReplay/sub_graph_multiple_submission.cpp
  SYCL :: Graph/RecordReplay/sub_graph_nested.cpp
  SYCL :: Graph/RecordReplay/sub_graph_two_parent_graphs.cpp
  SYCL :: Graph/RecordReplay/temp_scope.cpp
  SYCL :: Graph/RecordReplay/transitive_queue.cpp
  SYCL :: Graph/RecordReplay/transitive_queue_barrier.cpp
  SYCL :: Graph/RecordReplay/usm_copy.cpp
  SYCL :: Graph/RecordReplay/usm_copy_in_order.cpp
  SYCL :: Graph/RecordReplay/usm_memset_shortcut.cpp
  SYCL :: Graph/RecordReplay/work_group_memory.cpp
  SYCL :: Graph/Threading/submit.cpp
  SYCL :: Graph/Update/Explicit/whole_update_host_task.cpp
  SYCL :: Graph/Update/Explicit/whole_update_local_acc.cpp
  SYCL :: Graph/Update/Explicit/whole_update_local_acc_multi.cpp
  SYCL :: Graph/Update/Explicit/whole_update_subgraph.cpp
  SYCL :: Graph/Update/Explicit/whole_update_usm.cpp
  SYCL :: Graph/Update/Explicit/whole_update_work_group_memory.cpp
  SYCL :: Graph/Update/RecordReplay/whole_update_host_task.cpp
  SYCL :: Graph/Update/RecordReplay/whole_update_local_acc.cpp
  SYCL :: Graph/Update/RecordReplay/whole_update_local_acc_multi.cpp
  SYCL :: Graph/Update/RecordReplay/whole_update_subgraph.cpp
  SYCL :: Graph/Update/RecordReplay/whole_update_usm.cpp
  SYCL :: Graph/Update/RecordReplay/whole_update_work_group_memory.cpp
  SYCL :: Graph/Update/dyn_cgf_accessor_deps.cpp
  SYCL :: Graph/Update/dyn_cgf_accessor_deps2.cpp
  SYCL :: Graph/Update/dyn_cgf_different_arg_nums.cpp
  SYCL :: Graph/Update/dyn_cgf_event_deps.cpp
  SYCL :: Graph/Update/dyn_cgf_get_active_index.cpp
  SYCL :: Graph/Update/dyn_cgf_host_task_usm.cpp
  SYCL :: Graph/Update/dyn_cgf_ndrange.cpp
  SYCL :: Graph/Update/dyn_cgf_ndrange_3D.cpp
  SYCL :: Graph/Update/dyn_cgf_overwrite_range.cpp
  SYCL :: Graph/Update/dyn_cgf_parameters.cpp
  SYCL :: Graph/Update/dyn_cgf_shared_nodes.cpp
  SYCL :: Graph/Update/dyn_cgf_update_before_finalize.cpp
  SYCL :: Graph/Update/dyn_cgf_usm.cpp
  SYCL :: Graph/Update/dyn_cgf_with_all_dyn_params.cpp
  SYCL :: Graph/Update/dyn_cgf_with_different_type_dyn_params.cpp
  SYCL :: Graph/Update/dyn_cgf_with_dyn_work_group_mem.cpp
  SYCL :: Graph/Update/dyn_cgf_with_some_dyn_params.cpp
  SYCL :: Graph/Update/dyn_work_group_memory_multiple.cpp
  SYCL :: Graph/Update/regression_padded_parameter.cpp
  SYCL :: Graph/Update/update_before_finalize.cpp
  SYCL :: Graph/Update/update_nd_range.cpp
  SYCL :: Graph/Update/update_ndrange_to_range.cpp
  SYCL :: Graph/Update/update_nullptr.cpp
  SYCL :: Graph/Update/update_range.cpp
  SYCL :: Graph/Update/update_range_to_ndrange.cpp
  SYCL :: Graph/Update/update_with_indices_multiple_exec_graphs.cpp
  SYCL :: Graph/Update/update_with_indices_ordering.cpp
  SYCL :: Graph/Update/update_with_indices_ptr.cpp
  SYCL :: Graph/Update/update_with_indices_ptr_3D.cpp
  SYCL :: Graph/Update/update_with_indices_ptr_double_update.cpp
  SYCL :: Graph/Update/update_with_indices_ptr_multiple_nodes.cpp
  SYCL :: Graph/Update/update_with_indices_ptr_multiple_nodes_different_indices.cpp
  SYCL :: Graph/Update/update_with_indices_ptr_multiple_params.cpp
  SYCL :: Graph/Update/update_with_indices_ptr_subgraph.cpp
  SYCL :: Graph/Update/update_with_indices_scalar.cpp
  SYCL :: Graph/Update/whole_update_barrier_node.cpp
  SYCL :: Graph/Update/whole_update_dynamic_cgf.cpp
  SYCL :: Graph/Update/whole_update_dynamic_param.cpp
  SYCL :: Graph/Update/whole_update_empty_node.cpp
  SYCL :: Graph/ValidUsage/linear_graph_copy.cpp
  SYCL :: Graph/ValidUsage/linear_graph_copy_D2H.cpp
  SYCL :: InorderQueue/in_order_ext_oneapi_submit_barrier.cpp

🫠

If you revert my changes from #18387, does this work?

Let me try. I haven't seen these failures before, so if things changed recently it might explain it.

[sarnex]

I think this PR caused a hang on Linux BMG so I cancelled the job. Please double check before rerunning CI, thanks!

[sarnex]

I think Linux PVC has a hang from this too :)

[aelovikov-intel]

Don't we need to preserve the kernel until enqueue? Because the arguments are part of it. If so, then the enqueue might be delayed if the kernel depends on the (yet unfinished) host task, and the scheduler has to pick up the ownership for it until it's enqueued.