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_t*>(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)};

    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);
}

TEST(CAPI, TensorHandleCopyBetweenDevices) {
  TensorHandleCopyBetweenDevices(false);
}

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

TF_Operation* ScalarConst(int32_t v, TF_Graph* graph, TF_Status* s,
                          const char* name) {
  unique_tensor_ptr tensor(Int32Tensor(v), TF_DeleteTensor);
  return Const(tensor.get(), graph, s, name);
}

  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  float product[4] = {0};
  EXPECT_EQ(sizeof(product), TF_TensorByteSize(t));
  memcpy(&product[0], TF_TensorData(t), TF_TensorByteSize(t));
  TF_DeleteTensor(t);
  EXPECT_EQ(7, product[0]);
  EXPECT_EQ(10, product[1]);
  EXPECT_EQ(15, product[2]);
  EXPECT_EQ(22, product[3]);

  TFE_DeleteTensorHandle(h0_task0);
  TFE_DeleteTensorHandle(h1_task0);

  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) {

                    /*expected_shape*/ make_shape({1, 1, 6}));
  TFE_DeleteOp(fill_op);
  fill_op = nullptr;

  TFE_DeleteTensorHandle(scalar);
  TF_DeleteTensor(scalarTensor);
  TF_DeleteTensor(tensor_1X1X6);
  TF_DeleteTensor(tensor_1X6);
}

TEST(CAPI_EXPERIMENTAL, LibraryPluggableDeviceLoadFunctions) {
  // TODO(penpornk): Enable this test on Windows.
#if !defined(PLATFORM_WINDOWS)

    EXPECT_EQ(0, memcmp(m_float, resolved_data, 4 * sizeof(float)));
    TF_DeleteTensor(resolved);
    TFE_DeleteTensorHandle(copy_aliased);  // Note that this will delete copy.
    TFE_DeleteTensorHandle(on_host);
  }
  TF_DeleteDeviceList(devices);
  TF_DeleteTensor(m_data);
  TFE_DeleteTensorHandle(m);
  TFE_DeleteContext(ctx);
  TF_DeleteStatus(status);
}

// 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();