TFE_TensorHandle* retvals[1];
  int num_retvals = 1;
  TFE_Execute(matmul, &retvals[0], &num_retvals, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TF_Tensor* t = TFE_TensorHandleResolve(retvals[0], status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  float product[4] = {0};
  EXPECT_EQ(sizeof(product), TF_TensorByteSize(t));

  auto* retval_task0 = TFE_TensorHandleCopyToDevice(
      retvals[0], ctx, "/job:localhost/replica:0/task:0/device:CPU:0", status);
  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);

  TF_Tensor* t = TFE_TensorHandleResolve(retval_task0, status);
  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  TFE_DeleteTensorHandle(retval_task0);
  float product[4] = {0};
  EXPECT_EQ(sizeof(product), TF_TensorByteSize(t));