issue_2404_test

TEST(InferenceSessionTests, TestIssue2404) {
  string model_uri = "testdata/test_model.onnx";
  SessionOptions so;

  //so.session_logid = "InferenceSessionTests." + log_str;
  //so.session_log_verbosity_level = 1;  // change to 1 for detailed logging

  InferenceSession session_object{so, &DefaultLoggingManager()};
  CUDAExecutionProviderInfo epi;
  epi.device_id = 0;
  EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<CUDAExecutionProvider>(epi)).IsOK());
  ASSERT_TRUE(session_object.Load(model_uri).IsOK());
  ASSERT_TRUE(session_object.Initialize().IsOK());
  unique_ptr<IOBinding> io_binding;
  Status st = session_object.NewIOBinding(&io_binding);
  ASSERT_TRUE(st.IsOK());
  auto input_allocator = io_binding->GetCPUAllocator(0, kCudaExecutionProvider);

  std::vector<float> values_mul_x(32 * 3 * 112 * 112, 1.0f);
  OrtValue input_ml_value_A;
  std::vector<int64_t> dims_mul_x_A = {32, 3, 112, 112};
  CreateMLValue<float>(input_allocator, dims_mul_x_A, values_mul_x, &input_ml_value_A);

  io_binding->BindInput("input1", input_ml_value_A);

  std::vector<int64_t> expected_output_dims = {32, 1000, 1, 1};
  OrtValue output_ml_value;
  AllocateMLValue<float>(TestCudaExecutionProvider()->GetAllocator(0, OrtMemTypeDefault), expected_output_dims,
                         &output_ml_value);
  io_binding->BindOutput("output1", output_ml_value);
  ASSERT_TRUE(io_binding->SynchronizeInputs().IsOK());

  for (int i = 0; i < 10; ++i) {
    st = session_object.Run(*io_binding.get());
  }

  auto start = chrono::steady_clock::now();
  for (int i = 0; i < 1; ++i) {
    st = session_object.Run(*io_binding.get());
    if (!st.IsOK()) {
      std::cout << "Problem...\n";
      break;
    }
  }
  auto end = chrono::steady_clock::now();
  cout << "Elapsed time in microseconds : "
       << chrono::duration_cast<chrono::microseconds>(end - start).count()
       << " µs" << endl;

  cout << "Elapsed time in milliseconds : "
       << chrono::duration_cast<chrono::milliseconds>(end - start).count()
       << " ms" << endl;
}

You can run this as part of inference_session_test.cc.