ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  TF_Tensor* result_tensor = TFE_TensorHandleResolve(result_t, status.get());

  // Copy Tensor data into an array.
  float result_data[4] = {0};
  memcpy(&result_data[0], TF_TensorData(result_tensor),
         TF_TensorByteSize(result_tensor));

  int data_len = 4;  // length of result_data
  for (int i = 0; i < data_len; i++) {

}
TEST(CAPI, Executor_MatMul_CPU) { Executor_MatMul_CPU(false); }
TEST(CAPI, Executor_MatMul_CPUAsync) { Executor_MatMul_CPU(true); }

void Deleter(void* data, size_t unused, void* tensor_handle) {
  TFE_DeleteTensorHandle(static_cast<TFE_TensorHandle*>(tensor_handle));
}

TEST(CAPI, TensorHandleOnDeviceMemory) {
  TF_Status* status = TF_NewStatus();
  TFE_ContextOptions* opts = TFE_NewContextOptions();

                                                   size_t proto_len,
                                                   TF_Status* status);

// A handle to a tensor on a device.
//
// Like a TF_Tensor, a TFE_TensorHandle refers to a tensor with a value, shape,
// type etc. Unlike a TF_Tensor, a TFE_TensorHandle may refer to such tensors
// placed in the memory of different devices or remote address spaces.

  DLManagedTensor* dlm_tensor = &tf_dlm_tensor_ctx->tensor;
  dlm_tensor->manager_ctx = tf_dlm_tensor_ctx;
  dlm_tensor->deleter = &DLManagedTensorDeleter;
  dlm_tensor->dl_tensor.device = tf_dlm_context;
  int ndim = tensor->dims();
  dlm_tensor->dl_tensor.ndim = ndim;
  dlm_tensor->dl_tensor.data = tf_dlm_data;
  dlm_tensor->dl_tensor.dtype = tf_dlm_type;

  std::vector<int64_t> tensor_ids(tensors.size());
  std::vector<tensorflow::DataType> tensor_dtypes(tensors.size());
  for (int i = 0; i < tensors.size(); i++) {
    tensor_ids[i] = ToId(tensors[i]);
    tensor_dtypes[i] = tensors[i]->DataType();
  }
  return GradientTape::ShouldRecord(tensor_ids, tensor_dtypes);
}
void Tape::DeleteTrace(const AbstractTensorHandle* t) {

                          const char* name) {
  unique_tensor_ptr tensor(Int32Tensor(v), TF_DeleteTensor);
  return Const(tensor.get(), graph, s, name);
}

TF_Operation* ScalarConst(double v, TF_Graph* graph, TF_Status* s,
                          const char* name) {
  unique_tensor_ptr tensor(DoubleTensor(v), TF_DeleteTensor);
  return Const(tensor.get(), graph, s, name);
}

        ],
    }),
)

tf_cuda_library(
    name = "tf_tensor_internal",
    hdrs = [
        "tf_tensor.h",
        "tf_tensor_helper.h",
        "tf_tensor_internal.h",
    ],
    visibility = ["//tensorflow:internal"],
    deps = [
        ":c_api_macros",
        ":tensor_interface",
        ":tf_datatype",
        ":tf_status",

  int version = TFE_CUSTOM_DEVICE_VERSION;
  // Method to copy a tensor to the custom device.
  TFE_TensorHandle* (*copy_tensor_to_device)(TFE_Context* context,
                                             TFE_TensorHandle* tensor,
                                             TF_Status* status,
                                             void* device_info);

  // Method to copy a tensor from the custom device to a target device.

  const std::vector<tensorflow::OpInfo::TensorProperties>& tensor_properties =
      reinterpret_cast<tensorflow::grappler::GraphProperties*>(graph_properties)
          ->GetInputProperties(name);
  const int len =
      std::min(num_values, static_cast<int>(tensor_properties.size()));
  for (int i = 0; i < len; ++i) {
    absl::Status s =
        tensorflow::MessageToBuffer(tensor_properties[i], properties[i]);
    if (!s.ok()) {

// This may cause incorrectness in graph analyses, but is useful for simulation
// or scheduling.
// If include_input_tensor_values is true, the values of constant
// tensors will included in the input properties.
// If include_output_tensor_values is true, the values of constant tensors will
// be included in the output properties.
TF_CAPI_EXPORT extern void TF_InferStatically(