clean up

Author: chilo-ms

plugin_execution_providers/tensorrt/cuda_allocator.h

@@ -4,8 +4,6 @@
 #pragma once
 #include <atomic>
 #include "onnxruntime_c_api.h"
-#define ORT_API_MANUAL_INIT
-#include "onnxruntime_cxx_api.h"
 
 using DeviceId = int16_t;
 

plugin_execution_providers/tensorrt/onnx_ctx_model_helper.cc

@@ -21,6 +21,7 @@ bool EPContextNodeHelper::GraphHasCtxNode(const OrtGraph* graph, const OrtApi& o
   RETURN_IF_ERROR(ort_api.Graph_GetNumNodes(graph, &num_nodes));
 
   std::vector<const OrtNode*> nodes(num_nodes);
+  RETURN_IF_ERROR(ort_api.Graph_GetNodes(graph, nodes.data(), nodes.size()));
 
   for (size_t i = 0; i < num_nodes; ++i) {
     auto node = nodes[i];

plugin_execution_providers/tensorrt/tensorrt_execution_provider.cc

@@ -5,9 +5,7 @@
 #include <iostream>
 #include <cuda_runtime.h>
 
-#define ORT_API_MANUAL_INIT
 #include "onnxruntime_cxx_api.h"
-#undef ORT_API_MANUAL_INIT
 
 #define ORT_EP_UTILS_ORT_GRAPH_TO_PROTO_IMPL
 #include "ort_graph_to_proto.h"

plugin_execution_providers/tensorrt/tensorrt_execution_provider.h

@@ -1,9 +1,5 @@
 #pragma once
 
-#define ORT_API_MANUAL_INIT
-#include "onnxruntime_cxx_api.h"
-#undef ORT_API_MANUAL_INIT
-
 #include "tensorrt_provider_factory.h"
 #include "utils/provider_options.h"
 #include "tensorrt_execution_provider_info.h"

plugin_execution_providers/tensorrt/tensorrt_execution_provider_utils.h

@@ -1,8 +1,6 @@
 #pragma once
 
-#define ORT_API_MANUAL_INIT
 #include "onnxruntime_cxx_api.h"
-#undef ORT_API_MANUAL_INIT
 
 #include "ep_utils.h"
 #include "flatbuffers/idl.h"

plugin_execution_providers/tensorrt/tensorrt_provider_factory.cc

@@ -1,6 +1,4 @@
-#define ORT_API_MANUAL_INIT
 #include "onnxruntime_cxx_api.h"
-#undef ORT_API_MANUAL_INIT
 #include "tensorrt_provider_factory.h"
 #include "tensorrt_execution_provider.h"
 #include "cuda_allocator.h"
@@ -145,7 +143,7 @@ OrtStatus* ORT_API_CALL TensorrtExecutionProviderFactory::GetSupportedDevicesImp
                                                                 cuda_gpu_mem_devices,    // device memory
                                                                 cuda_pinned_mem_devices  // shared memory
                                                                );
-
+  factory->SetGPUDataTransfer(std::move(data_transfer_impl));
   return nullptr;
 }
 
@@ -283,6 +281,10 @@ void TensorrtExecutionProviderFactory::SetHostAccessibleMemInfo(MemoryInfoUnique
   cuda_pinned_memory_infos_.push_back(std::move(mem_info));
 }
 
+void TensorrtExecutionProviderFactory::SetGPUDataTransfer(std::unique_ptr<TRTEpDataTransfer> gpu_data_transfer) {
+  data_transfer_impl_ = std::move(gpu_data_transfer);
+}
+
 // To make symbols visible on macOS/iOS
 #ifdef __APPLE__
 #define EXPORT_SYMBOL __attribute__((visibility("default")))

plugin_execution_providers/tensorrt/tensorrt_provider_factory.h

@@ -52,6 +52,8 @@ struct TensorrtExecutionProviderFactory : public OrtEpFactory, public ApiPtrs {
 
   void SetHostAccessibleMemInfo(MemoryInfoUniquePtr mem_info, uint32_t device_id);
 
+  void SetGPUDataTransfer(std::unique_ptr<TRTEpDataTransfer> gpu_data_transfer);
+
   const std::string ep_name_;           // EP name
   const std::string vendor_{"Nvidia"};  // EP vendor name
   const std::string ep_version_{"0.1.0"};  // EP version