update data transfer initialization in TRT EP

Author: chilo-ms

plugin_execution_providers/tensorrt/tensorrt_provider_factory.cc

@@ -119,14 +119,6 @@ OrtStatus* ORT_API_CALL TensorrtExecutionProviderFactory::GetSupportedDevicesImp
 
       ep_devices[num_ep_devices++] = ep_device;
     }
-
-  // Create gpu data transfer
-  auto data_transfer_impl = std::make_unique<TRTEpDataTransfer>(
-      static_cast<const ApiPtrs&>(*factory),
-      cuda_gpu_mem_devices,    // device memory
-      cuda_pinned_mem_devices  // shared memory
-  );
-
 
   // C++ API equivalent. Throws on error.
   //{
@@ -142,6 +134,12 @@ OrtStatus* ORT_API_CALL TensorrtExecutionProviderFactory::GetSupportedDevicesImp
   //}
   }
 
+    // Create gpu data transfer
+  auto data_transfer_impl = std::make_unique<TRTEpDataTransfer>(static_cast<const ApiPtrs&>(*factory),
+                                                                cuda_gpu_mem_devices,    // device memory
+                                                                cuda_pinned_mem_devices  // shared memory
+                                                               );
+
   return nullptr;
 }