[WIP] Out-Tree EP feature

include/onnxruntime/core/framework/ort_type_constraints.h

@@ -0,0 +1,15 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#pragma once
+#include "core/session/onnxruntime_c_api.h"
+#include <unordered_map>
+#include <string>
+#include <set>
+
+struct OrtTypeConstraints {
+    bool AddTypeConstraint(const char* type_symbol, ONNXTensorElementDataType type);
+    inline const std::unordered_map<std::string, std::set<ONNXTensorElementDataType>>& GetTypeConstraints() const { return type_constraints_; };
+private:
+    std::unordered_map<std::string, std::set<ONNXTensorElementDataType>> type_constraints_;
+};

include/onnxruntime/core/session/environment.h

@@ -88,6 +88,10 @@
    */
   Status CreateAndRegisterAllocatorV2(const std::string& provider_type, const OrtMemoryInfo& mem_info, const std::unordered_map<std::string, std::string>& options, const OrtArenaCfg* arena_cfg = nullptr);
 
+  void InsertCustomEp(const char* ep_name, OrtExecutionProviderFactory* ep_factory);
+
+  OrtExecutionProviderFactory* GetOrtExecutionProviderFactory(const std::string& ep_name);
+
  private:
   ORT_DISALLOW_COPY_ASSIGNMENT_AND_MOVE(Environment);
   Status Initialize(std::unique_ptr<logging::LoggingManager> logging_manager,
@@ -99,5 +103,6 @@
   std::unique_ptr<onnxruntime::concurrency::ThreadPool> inter_op_thread_pool_;
   bool create_global_thread_pools_{false};
   std::vector<AllocatorPtr> shared_allocators_;
+  std::unordered_map<std::string, std::unique_ptr<OrtExecutionProviderFactory>> custom_ep_factories_;
 };
 }  // namespace onnxruntime

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -304,6 +304,15 @@
 ORT_RUNTIME_CLASS(OpAttr);
 ORT_RUNTIME_CLASS(Logger);
 ORT_RUNTIME_CLASS(ShapeInferContext);
+ORT_RUNTIME_CLASS(ExecutionProvider);
+ORT_RUNTIME_CLASS(ExecutionProviderFactory);
+ORT_RUNTIME_CLASS(Node);
+ORT_RUNTIME_CLASS(Model);
+ORT_RUNTIME_CLASS(Graph);
+ORT_RUNTIME_CLASS(GraphViewer);
+ORT_RUNTIME_CLASS(KernelRegistry);
+ORT_RUNTIME_CLASS(TypeConstraints);
+ORT_RUNTIME_CLASS(Device);
 
 #ifdef _WIN32
 typedef _Return_type_success_(return == 0) OrtStatus* OrtStatusPtr;
@@ -395,6 +404,13 @@
   OrtMemoryInfoDeviceType_FPGA = 2
 } OrtMemoryInfoDeviceType;
 
+typedef enum OrtMemoryType {
+  OrtMemoryType_Default = 0,
+  OrtMemoryType_CUDA_PINNED = 1,
+  OrtMemoryType_HIP_PINNED = 2,
+  OrtMemoryType_CANN_PINNED = 3,
+} OrtMemoryType;
+
 /** \brief Algorithm to use for cuDNN Convolution Op
  */
 typedef enum OrtCudnnConvAlgoSearch {
@@ -689,6 +705,66 @@
  */
 ORT_EXPORT const OrtApiBase* ORT_API_CALL OrtGetApiBase(void) NO_EXCEPTION;
 
+typedef struct OrtCreateStream {
+  int device_type;
+  void*(ORT_API_CALL* CreateStreamFunc)(const OrtDevice*);
+} OrtCreateStream;
+
+typedef struct OrtMetaDef {
+  const char* name;
+  const char* domain;
+  int since_version;
+
+  const char** inputs;
+  size_t input_len;
+  const char** outputs;
+  size_t output_len;
+  const char** constant_initializers;
+  size_t initializer_len;
+
+  const char* doc_string;
+} OrtMetaDef;
+
+typedef struct OrtIndexedSubGraph {
+  OrtMetaDef* meta_def; // TODO(leca): how to define a nested structure pointer?
+  size_t* node_index;
+  size_t node_index_len;
+} OrtIndexedSubGraph;
+
+typedef struct OrtComputeContext {
+  void*(ORT_API_CALL* AllocateFunc)(void*, size_t, size_t);
+  void(ORT_API_CALL* DestroyFunc)(void*, void*);
+  void* allocator_handle;
+  const char* node_name;
+} OrtComputeContext;
+
+typedef struct OrtNodeComputeInfo {
+  int(ORT_API_CALL* CreateFunctionStateFunc)(OrtComputeContext*, void*, void**);
+  OrtStatusPtr(ORT_API_CALL* ComputeFunc)(void*, void*, const OrtApi*, OrtKernelContext*);
+  void(ORT_API_CALL* DestroyFunctionStateFunc)(void*);
+} OrtNodeComputeInfo;
+
+typedef struct OrtExecutionProvider {
+#ifdef __cplusplus
+  OrtExecutionProvider() : GetCapability{nullptr}, Compile{nullptr}, RegisterKernels{nullptr}, CanCopy{nullptr}, CopyTensor{nullptr}, type{nullptr}, create_stream{nullptr}, default_device{nullptr},
+                           extra_param_for_create_state_func{nullptr}, extra_param_for_compute_func{nullptr} {}
+#endif
+  void(ORT_API_CALL* GetCapability)(const OrtExecutionProvider* this_, const OrtGraphViewer* graph, size_t* cnt, OrtIndexedSubGraph***);
+  OrtStatusPtr(ORT_API_CALL* Compile)(OrtExecutionProvider* this_, const OrtGraphViewer** graph, const OrtNode** node, size_t cnt, OrtNodeComputeInfo** node_compute_info);
+  void(ORT_API_CALL* RegisterKernels)(OrtKernelRegistry* kernel_registry);
+  bool(ORT_API_CALL* CanCopy)(const OrtDevice* source, const OrtDevice* target);
+  OrtStatusPtr(ORT_API_CALL* CopyTensor)(const void* src, OrtMemoryInfoDeviceType source_device_type, OrtMemoryType source_mem_type, void* dst, OrtMemoryInfoDeviceType target_device_type, size_t count, void* stream);
+  const char* type;
+  OrtCreateStream* create_stream;
+  const OrtDevice* default_device;
+  void* extra_param_for_create_state_func;
+  void* extra_param_for_compute_func;
+} OrtExecutionProvider;
+
+typedef struct OrtExecutionProviderFactory {
+  OrtExecutionProvider*(ORT_API_CALL* CreateExecutionProvider)(OrtExecutionProviderFactory* this_, const char* const* ep_option_keys, const char* const* ep_option_values, size_t option_size);
+} OrtExecutionProviderFactory;
+
 /** \brief Thread work loop function
  *
  * Onnxruntime will provide the working loop on custom thread creation
@@ -4665,7 +4741,112 @@
                   _In_reads_(num_external_initializer_files) char* const* external_initializer_file_buffer_array,
                   _In_reads_(num_external_initializer_files) const size_t* external_initializer_file_lengths,
                   size_t num_external_initializer_files);
-};
+
+  ORT_API2_STATUS(CreateDevice, _In_ enum OrtMemoryInfoDeviceType device_type, _In_ enum OrtMemoryType memory_type, _In_ int16_t device_id, _Outptr_ const OrtDevice** out);
+
+  ORT_API2_STATUS(DeviceGetDeviceType, _In_ const OrtDevice* device, _Out_ OrtMemoryInfoDeviceType* out);
+
+  ORT_API2_STATUS(DeviceGetMemoryType, _In_ const OrtDevice* device, _Out_ OrtMemoryType* out);
+
+  ORT_API2_STATUS(DeviceGetDeviceId, _In_ const OrtDevice* device, _Out_ int16_t* out);
+
+  ORT_CLASS_RELEASE(Device);
+
+  ORT_API2_STATUS(RegisterOrtExecutionProviderLibrary, _In_ const ORTCHAR_T* lib_path, _In_ OrtEnv* env, _In_ const char* ep_name);
+
+  ORT_API2_STATUS(SessionOptionsAppendOrtExecutionProvider, _In_ OrtSessionOptions* options, _In_ const char* ep_name, _In_ OrtEnv* env,
+                   _In_reads_(num_keys) const char* const* provider_options_keys, _In_reads_(num_keys) const char* const* provider_options_values, _In_ size_t num_keys);
+
+  ORT_API2_STATUS(OrtGraph_IsConstantInitializer, const OrtGraphViewer* graph, const char* name, bool check_outer_scope, _Out_ bool* ret);
+
+  ORT_API2_STATUS(OrtGraph_GetNodesIndexInTopologicalOrder, const OrtGraphViewer* graph, int execution_order, _Out_ size_t* len, _Out_ const size_t** nodes_index_in_topological_order);
+
+  ORT_API2_STATUS(OrtGraph_IsSubgraph, const OrtGraph* graph, _Out_ bool* ret);
+
+  ORT_API2_STATUS(OrtGraph_GetParentGraph, const OrtGraph* graph, _Outptr_ const OrtGraph** parent_graph);
+
+  ORT_API2_STATUS(OrtGraph_GetParenNode, const OrtGraphViewer* graph, _Outptr_ const OrtNode** parent_node);
+
+  ORT_API2_STATUS(OrtGraph_GetModelPath, const OrtGraphViewer* graph, _Outptr_ const void** path);
+
+  ORT_API2_STATUS(OrtGraph_GetOrtGraph, const OrtGraphViewer* graph_viewer, _Outptr_ const OrtGraph** graph);
+
+  ORT_API2_STATUS(OrtGraph_GetInputsIncludingInitializers, const OrtGraphViewer* graph, _Out_ size_t* num_inputs, _Outptr_ const char*** input_names);
+
+  ORT_API2_STATUS(OrtGraph_GetOrtNode, const OrtGraphViewer* graph, size_t node_index, _Outptr_ const OrtNode** node);
+
+  ORT_API2_STATUS(OrtGraph_GetNodesConsumingInput, const OrtGraphViewer* graph, const char* input_name, _Out_ size_t* len, _Outptr_ const OrtNode*** consumers); // TODO(leca): ValueConsumers::comprehensive ?
+
+  ORT_API2_STATUS(OrtGraph_GetNodeProducingOutput, const OrtGraphViewer* graph, const char* output_name, _Outptr_ const OrtNode** producer);
+
+  int(ORT_API_CALL* OrtGraph_NumberOfNodes)(const OrtGraphViewer*)NO_EXCEPTION ORT_ALL_ARGS_NONNULL;
+
+  ORT_API2_STATUS(OrtGraph_MaxNodeIndex, const OrtGraphViewer* graph, _Out_ int* out);
+
+  size_t(ORT_API_CALL* OrtGraph_GetOutputSize)(const OrtGraphViewer*)NO_EXCEPTION ORT_ALL_ARGS_NONNULL;
+
+  const char*(ORT_API_CALL* OrtGraph_GetIthOutputName)(const OrtGraphViewer*, size_t i)NO_EXCEPTION ORT_ALL_ARGS_NONNULL;
+
+  int32_t(ORT_API_CALL* OrtGraph_GetIthOutputElemType)(const OrtGraphViewer*, size_t i)NO_EXCEPTION ORT_ALL_ARGS_NONNULL;
+
+  size_t(ORT_API_CALL* OrtGraph_SerializeToArray)(const OrtGraphViewer*, _Out_ void** data)NO_EXCEPTION;
+
+  ORT_API2_STATUS(OrtNode_GetName, const OrtNode* node, _Out_ const char** name);
+
+  ORT_API2_STATUS(OrtNode_GetDescription, const OrtNode* node, _Out_ const char** description);
+
+  ORT_API2_STATUS(OrtNode_GetDomain, const OrtNode* node, _Out_ const char** domain);
+
+  ORT_API2_STATUS(OrtNode_SinceVersion, const OrtNode* node, _Out_ int* since_version);
+
+  ORT_API2_STATUS(OrtNode_GetExecutionProviderType, const OrtNode* node, _Out_ const char** ep_type);
+
+  ORT_API2_STATUS(OrtNode_GetOpType, const OrtNode* node, _Out_ const char** op_type);
+
+  ORT_API2_STATUS(OrtNode_GetImplicitInputSize, const OrtNode* node, _Out_ size_t* input_size);
+
+  ORT_API2_STATUS(OrtNode_GetIthImplicitInputName, const OrtNode* node, size_t i, _Out_ const char** ith_input_name);
+
+  ORT_API2_STATUS(OrtNode_GetInputSize, const OrtNode* node, _Out_ size_t* input_size);
+
+  ORT_API2_STATUS(OrtNode_GetIthInputName, const OrtNode* node, size_t i, _Out_ const char** ith_input_name);
+
+  ORT_API2_STATUS(OrtNode_GetOutputSize, const OrtNode* node, _Out_ size_t* output_size);
+
+  ORT_API2_STATUS(OrtNode_GetIthOutputName, const OrtNode* node, size_t i, _Out_ const char** ith_output_name);
+
+  ORT_API2_STATUS(OrtNode_GetIndex, const OrtNode* node, _Out_ size_t* index);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeSize, const OrtNode* node, _Out_ size_t* attr_size);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeKeyCount, const OrtNode* node, const char* key, _Out_ size_t* count);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeIntSize, const OrtNode* node, const char* key, _Out_ int* int_size);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeFloatSize, const OrtNode* node, const char* key, _Out_ int* float_size);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeStringSize, const OrtNode* node, const char* key, _Out_ int* str_size);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeIthInt, const OrtNode* node, const char* key, int i, _Out_ int64_t* ints);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeIthFloat, const OrtNode* node, const char* key, int i, _Out_ float* floats);
+
+  ORT_API2_STATUS(OrtNode_GetAttributeIthStr, const OrtNode* node, const char* key, int i, _Out_ const char** strs);
+
+  const char*(ORT_API_CALL* OrtNode_GetAttributeStr)(const OrtNode*, const char* key)NO_EXCEPTION ORT_ALL_ARGS_NONNULL;
+
+  int64_t(ORT_API_CALL* OrtNode_GetAttributeInt)(const OrtNode*, const char* key)NO_EXCEPTION ORT_ALL_ARGS_NONNULL;
+
+  ORT_API2_STATUS(OrtNode_GetSubgraphs, const OrtNode* node, _Out_ size_t* len, _Outptr_ const OrtGraphViewer*** subgraphs);
+
+  ORT_API2_STATUS(OrtKernelRegistry_RegisterKernel, OrtKernelRegistry* kernel_registry, OrtCustomOp* custom_op, OrtTypeConstraints* type_constraints);
+
+  ORT_API2_STATUS(CreateOrtTypeConstraints, _Outptr_ OrtTypeConstraints** type_constraints);
+
+  ORT_API2_STATUS(AddTypeConstraint, _In_ OrtTypeConstraints* type_constraints, _In_ const char* type_symbol, ONNXTensorElementDataType type);
+
+  ORT_CLASS_RELEASE(TypeConstraints);
+};  // struct OrtApi
 
 /*
  * Steps to use a custom op:

onnxruntime/core/framework/ort_type_constraints.cc

@@ -0,0 +1,14 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#include "core/framework/ort_type_constraints.h"
+
+bool OrtTypeConstraints::AddTypeConstraint(const char* type_symbol, ONNXTensorElementDataType type) {
+    std::unordered_map<std::string, std::set<ONNXTensorElementDataType>>::iterator iter = type_constraints_.find(type_symbol);
+    if (iter == type_constraints_.end()) {
+        std::set<ONNXTensorElementDataType> types{type};
+        type_constraints_[type_symbol] = types;
+        return true;
+    }
+    return (iter->second).insert(type).second;
+}

onnxruntime/core/framework/provider_adapter.h

@@ -0,0 +1,133 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#pragma once
+#include "core/session/onnxruntime_c_api.h"
+#include "core/framework/compute_capability.h"
+
+namespace onnxruntime {
+
+class DataTransferAdapter : public IDataTransfer {
+public:
+  DataTransferAdapter(OrtExecutionProvider* ep) : ep_impl_(ep) {}
+  virtual bool CanCopy(const OrtDevice& src_device, const OrtDevice& dst_device) const override {
+    return ep_impl_->CanCopy(&src_device, &dst_device);
+  }
+
+  virtual common::Status CopyTensor(const Tensor& src, Tensor& dst) const override {
+    OrtMemoryInfoDeviceType source_device_type = static_cast<OrtMemoryInfoDeviceType>(src.Location().device.Type());
+    OrtMemoryInfoDeviceType target_device_type = static_cast<OrtMemoryInfoDeviceType>(dst.Location().device.Type());
+    OrtMemoryType source_mem_type = static_cast<OrtMemoryType>(src.Location().device.MemType());
+    return ToStatus(ep_impl_->CopyTensor(src.DataRaw(), source_device_type, source_mem_type, dst.MutableDataRaw(), target_device_type, src.SizeInBytes(), nullptr));
+  }
+
+  virtual common::Status CopyTensorAsync(const Tensor& src, Tensor& dst, Stream& stream) const override {
+    OrtMemoryInfoDeviceType source_device_type = static_cast<OrtMemoryInfoDeviceType>(src.Location().device.Type());
+    OrtMemoryInfoDeviceType target_device_type = static_cast<OrtMemoryInfoDeviceType>(dst.Location().device.Type());
+    OrtMemoryType source_mem_type = static_cast<OrtMemoryType>(src.Location().device.MemType());
+    return ToStatus(ep_impl_->CopyTensor(src.DataRaw(), source_device_type, source_mem_type, dst.MutableDataRaw(), target_device_type, src.SizeInBytes(), stream.GetHandle()));
+  }
+private:
+  OrtExecutionProvider* ep_impl_;
+};
+
+class ExecutionProviderAdapter : public IExecutionProvider {
+public:
+  ExecutionProviderAdapter(OrtExecutionProvider* ep) : IExecutionProvider(ep->type, ep->default_device ? *(ep->default_device) : OrtDevice()), ep_impl_(ep) {
+    if (ep_impl_->RegisterKernels) {
+      kernel_registry_ = std::make_shared<KernelRegistry>();
+      ep_impl_->RegisterKernels(reinterpret_cast<OrtKernelRegistry*>(kernel_registry_.get()));
+    }
+  }
+  virtual std::vector<std::unique_ptr<ComputeCapability>> GetCapability(const GraphViewer& graph_viewer, const IKernelLookup& kernel_lookup) const override {
+    size_t cnt = 0;
+    OrtIndexedSubGraph** indexed_subgraph = nullptr;
+    if (ep_impl_->GetCapability) ep_impl_->GetCapability(ep_impl_, reinterpret_cast<const OrtGraphViewer*>(&graph_viewer), &cnt, &indexed_subgraph);
+
+    if (cnt == 0) return IExecutionProvider::GetCapability(graph_viewer, kernel_lookup);
+
+    std::vector<std::unique_ptr<ComputeCapability>> ret;
+    for (size_t i = 0; i < cnt; i++) {
+        std::unique_ptr<IndexedSubGraph> sb = std::make_unique<IndexedSubGraph>();
+        sb->nodes.reserve(indexed_subgraph[i]->node_index_len);
+        for (size_t j = 0; j < indexed_subgraph[i]->node_index_len; j++) sb->nodes.push_back((indexed_subgraph[i]->node_index)[j]);
+        if (indexed_subgraph[i]->meta_def != nullptr) {
+            std::unique_ptr<IndexedSubGraph::MetaDef> meta_def = std::make_unique<IndexedSubGraph::MetaDef>();
+            meta_def->name = indexed_subgraph[i]->meta_def->name ? indexed_subgraph[i]->meta_def->name : "";
+            meta_def->doc_string = indexed_subgraph[i]->meta_def->doc_string ? indexed_subgraph[i]->meta_def->doc_string : "";
+            meta_def->domain = indexed_subgraph[i]->meta_def->domain ? indexed_subgraph[i]->meta_def->domain : "";
+            meta_def->since_version = indexed_subgraph[i]->meta_def->since_version;
+
+            meta_def->inputs.reserve(indexed_subgraph[i]->meta_def->input_len);
+            for (size_t j = 0; j < indexed_subgraph[i]->meta_def->input_len; j++) meta_def->inputs.push_back(indexed_subgraph[i]->meta_def->inputs[j]);
+
+            meta_def->outputs.reserve(indexed_subgraph[i]->meta_def->output_len);
+            for (size_t j = 0; j < indexed_subgraph[i]->meta_def->output_len; j++) meta_def->outputs.push_back(indexed_subgraph[i]->meta_def->outputs[j]);
+
+            meta_def->constant_initializers.reserve(indexed_subgraph[i]->meta_def->initializer_len);
+            for (size_t j = 0; j < indexed_subgraph[i]->meta_def->initializer_len; j++) meta_def->constant_initializers.push_back(indexed_subgraph[i]->meta_def->constant_initializers[j]);
+
+            sb->SetMetaDef(std::move(meta_def));
+        }
+
+        ret.push_back(std::make_unique<ComputeCapability>(std::move(sb)));
+    }
+    return ret;
+  }
+
+  virtual common::Status Compile(const std::vector<FusedNodeAndGraph>& fused_nodes_and_graphs, std::vector<NodeComputeInfo>& node_compute_funcs) override {
+    std::vector<const OrtGraphViewer*> ortGraphs;
+    std::vector<const OrtNode*> ortNodes;
+    for (auto& fused_node_graph : fused_nodes_and_graphs) {
+      const GraphViewer& graph_viewer = fused_node_graph.filtered_graph;
+      const Node& fused_node = fused_node_graph.fused_node;
+      ortGraphs.push_back(reinterpret_cast<const OrtGraphViewer*>(&graph_viewer));
+      ortNodes.push_back(reinterpret_cast<const OrtNode*>(&fused_node));
+    }
+    size_t count = fused_nodes_and_graphs.size();
+    std::vector<OrtNodeComputeInfo> cache;
+    cache.resize(count);
+    OrtNodeComputeInfo* cache_data = cache.data();
+    OrtStatus* ret = ep_impl_->Compile(ep_impl_, ortGraphs.data(), ortNodes.data(), count, &cache_data);
+    if (ret != nullptr) return ToStatus(ret);
+    node_compute_funcs.reserve(count);
+    for (size_t i = 0; i < count; i++) {
+        NodeComputeInfo compute_info;
+        compute_info.create_state_func = [&, cache, i](ComputeContext* context, void** state) {
+            if (cache[i].CreateFunctionStateFunc) return cache[i].CreateFunctionStateFunc(reinterpret_cast<OrtComputeContext*>(context), ep_impl_->extra_param_for_create_state_func, state);
+            return 0;
+        };
+        compute_info.compute_func = [&, cache, i](void* state, const OrtApi* api, OrtKernelContext* context) {
+            return ToStatus(cache[i].ComputeFunc(state, ep_impl_->extra_param_for_compute_func, api, context));
+        };
+        compute_info.release_state_func = [&, cache, i](void* state) {
+            if (cache[i].DestroyFunctionStateFunc) {
+                cache[i].DestroyFunctionStateFunc(state);
+            }
+        };
+        node_compute_funcs.emplace_back(std::move(compute_info));
+    }
+
+    return Status::OK();
+  }
+
+  virtual void RegisterStreamHandlers(IStreamCommandHandleRegistry& stream_handle_registry, AllocatorMap&) const override {
+    if (ep_impl_->create_stream) {
+      CreateStreamFn csf = [&](const OrtDevice& device) -> std::unique_ptr<Stream> {
+        void* stream = ep_impl_->create_stream->CreateStreamFunc(&device);
+        return std::make_unique<Stream>(stream, device);
+      };
+      stream_handle_registry.RegisterCreateStreamFn(static_cast<OrtDevice::DeviceType>(ep_impl_->create_stream->device_type), csf);
+    }
+  }
+
+  virtual std::unique_ptr<onnxruntime::IDataTransfer> GetDataTransfer() const override {
+    return std::make_unique<DataTransferAdapter>(ep_impl_);
+  }
+
+  virtual std::shared_ptr<KernelRegistry> GetKernelRegistry() const override { return kernel_registry_; }
+private:
+  OrtExecutionProvider* ep_impl_;
+  std::shared_ptr<KernelRegistry> kernel_registry_; // TODO(leca): should be static local
+};
+}