// Get number of elements and fill data.
  int num_elems = TF_TensorElementCount(theta_tensor);
  vector<float> theta_data(num_elems);
  memcpy(theta_data.data(), TF_TensorData(theta_tensor),
         TF_TensorByteSize(theta_tensor));

  // Initialize space for the numerical gradient.
  vector<float> dtheta_approx(num_elems);

  // Get theta shape and store in theta_dims.

TFE_TensorHandle* TestScalarTensorHandle(TFE_Context* ctx, float value) {
  float data[] = {value};
  TF_Status* status = TF_NewStatus();
  TF_Tensor* t = TFE_AllocateHostTensor(ctx, TF_FLOAT, nullptr, 0, status);
  memcpy(TF_TensorData(t), &data[0], TF_TensorByteSize(t));
  TFE_TensorHandle* th = TFE_NewTensorHandleFromTensor(ctx, t, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TF_DeleteTensor(t);
  TF_DeleteStatus(status);

    if (TF_GetCode(status.get()) != TF_OK) {
      ADD_FAILURE() << tag;
      continue;
    }
    EXPECT_EQ(TF_TensorByteSize(t), TF_TensorByteSize(tcopy)) << tag;
    EXPECT_EQ(
        0, memcmp(TF_TensorData(t), TF_TensorData(tcopy), TF_TensorByteSize(t)))
        << tag;
    TF_DeleteTensor(tcopy);
  }

  TF_DeleteDeviceList(devices);
  TF_DeleteTensor(t);
  TFE_DeleteTensorHandle(hcpu);
  TFE_DeleteContext(ctx);

  // Check that a write to one tensor shows up in the other.
  *(static_cast<int64_t*>(TF_TensorData(a))) = 4;
  EXPECT_EQ(4, *(static_cast<int64_t*>(TF_TensorData(b))));
  *(static_cast<int64_t*>(TF_TensorData(b))) = 6;
  EXPECT_EQ(6, *(static_cast<int64_t*>(TF_TensorData(a))));

  TF_DeleteTensor(a);
  TF_DeleteTensor(b);
}

TEST(CAPI, TestFromProto) {

    CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

    TF_Tensor* t =
        TF_AllocateTensor(TF_INT32, nullptr, 0, 1 * sizeof(tensorflow::int32));
    *reinterpret_cast<tensorflow::int32*>(TF_TensorData(t)) = 42;
    TFE_TensorHandle* h = TFE_NewTensorHandle(t, status);
    ASSERT_TRUE(TF_GetCode(status) == TF_OK) << TF_Message(status);
    TF_DeleteTensor(t);

    TFE_Op* op = TFE_NewOp(ctx, "ident", status);

  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  TF_Tensor* result_tensor = TFE_TensorHandleResolve(result_t, status.get());
  float* result_value = static_cast<float*>(TF_TensorData(result_tensor));
  EXPECT_EQ(*result_value, 4.0);

  TF_DeleteTensor(result_tensor);
  TF_DeleteAbstractTensor(result);
  TF_DeleteOutputList(o);
  TF_DeleteExecutionContext(ctx);
}

// MatMul Test

                                           const int64_t* dims, int num_dims) {
  TF_Status* status = TF_NewStatus();
  TF_Tensor* t = TFE_AllocateHostTensor(ctx, datatype, dims, num_dims, status);
  memcpy(TF_TensorData(t), data, TF_TensorByteSize(t));
  TFE_TensorHandle* th = TFE_NewTensorHandleFromTensor(ctx, t, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TF_DeleteTensor(t);
  TF_DeleteStatus(status);

  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  TFE_DeleteTensorHandle(retvals[0]);
  float sum = 0;
  EXPECT_EQ(sizeof(sum), TF_TensorByteSize(t));
  memcpy(&sum, TF_TensorData(t), TF_TensorByteSize(t));
  TF_DeleteTensor(t);
  EXPECT_EQ(sum, 6.0);

  TFE_DeleteTensorHandle(h0);
  TFE_DeleteTensorHandle(h1);
  TFE_DeleteTensorHandle(h2);

  auto num_elem_numerical = TF_TensorElementCount(numerical_tensor);
  ASSERT_EQ(num_elem_numerical, num_grad);

  float* dnumerical = new float[num_elem_numerical]{0};
  memcpy(&dnumerical[0], TF_TensorData(numerical_tensor),
         TF_TensorByteSize(numerical_tensor));

  for (int j = 0; j < num_grad; j++) {
    ASSERT_NEAR(dnumerical[j], expected_grad[j], abs_error);
  }
  delete[] dnumerical;

// completed. The memory returned might alias the internal memory used by
// TensorFlow. Hence, callers should not mutate this memory (for example by
// modifying the memory region pointed to by TF_TensorData() on the returned
// TF_Tensor).
TF_CAPI_EXPORT extern TF_Tensor* TFE_TensorHandleResolve(TFE_TensorHandle* h,
                                                         TF_Status* status);