TFE_OpAddInput(op, var_handle, status);

  // Convert 'value' to a TF_Tensor then a TFE_TensorHandle.
  std::unique_ptr<TF_Tensor, decltype(&TF_DeleteTensor)> t(
      TF_AllocateTensor(TF_FLOAT, nullptr, 0, sizeof(value)), TF_DeleteTensor);
  memcpy(TF_TensorData(t.get()), &value, TF_TensorByteSize(t.get()));

  std::unique_ptr<TFE_TensorHandle, decltype(&TFE_DeleteTensorHandle)>

           TF_TensorByteSize(grad_tensor));
    TF_DeleteTensor(grad_tensor);
    dtheta_approx[i] = grad_data[0];
  }

  // Populate *numerical_grad with the data from dtheta_approx.
  TF_RETURN_IF_ERROR(TestTensorHandleWithDims<float, TF_FLOAT>(
      ctx, dtheta_approx.data(), theta_dims.data(), num_dims, numerical_grad));
  TF_DeleteTensor(theta_tensor);
  return absl::OkStatus();
}

                 1, nullptr, 0, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
  EXPECT_EQ(3, *(static_cast<int32*>(TF_TensorData(fetchValues1[0]))));
  TF_DeleteTensor(feedValues1[0]);
  TF_DeleteTensor(fetchValues1[0]);

  // Feed B and fetch (A + 2) + B.
  TF_Output feeds2[] = {TF_Output{b, 0}};
  TF_Output fetches2[] = {TF_Output{plusB, 0}};
  TF_Tensor* feedValues2[] = {Int32Tensor(4)};

  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
TEST_P(UnifiedCAPI, TestBasicEagerMatMul) {

// just nice to know that it compiles.
void compute(void* kernel, TF_OpKernelContext* ctx) {
  TF_Tensor* input;
  TF_Status* s = TF_NewStatus();
  TF_GetInput(ctx, 0, &input, s);
  TF_DeleteTensor(input);
  TF_DeleteStatus(s);
}

// Exercises tensorflow's C API.
int main(int argc, char** argv) {
  TF_InitMain(argv[0], &argc, &argv);

  struct TF_StringStream* s = TF_GetLocalTempDirectories();

  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);

  TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);

    std::unique_ptr<TF_Tensor, decltype(&TF_DeleteTensor)> tensor(
        TF_NewTensor(
            datatype_enum, /*dims=*/nullptr, /*num_dims=*/0,
            device_value.release(), sizeof(DataType),
            [](void* data, size_t, void* arg) {
              delete reinterpret_cast<DataType*>(data);
            },
            nullptr),
        TF_DeleteTensor);

         TF_TensorByteSize(numerical_tensor));

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

absl::Status MatMulModel(AbstractContext* ctx,
                         absl::Span<AbstractTensorHandle* const> inputs,
                         absl::Span<AbstractTensorHandle*> outputs) {

  std::unique_ptr<float[]> actual_values(new float[expected_values.size()]);
  EXPECT_EQ(sizeof(float) * expected_values.size(), TF_TensorByteSize(t));
  memcpy(actual_values.get(), TF_TensorData(t), TF_TensorByteSize(t));
  TF_DeleteTensor(t);

  for (int i = 0; i < expected_values.size(); i++) {
    EXPECT_EQ(expected_values[i], actual_values[i])
        << "Mismatch in expected values at (zero-based) index " << i;
  }
}