TFE_DeleteTensorHandle(h0_task0);
  TFE_DeleteTensorHandle(h1_task0);
  TFE_DeleteTensorHandle(h0_task1);
  TFE_DeleteTensorHandle(h1_task1);
  TFE_DeleteTensorHandle(retvals[0]);

  TFE_DeleteOp(matmul);

  TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
  TFE_ExecutorWaitForAllPendingNodes(executor, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TFE_DeleteTensorHandle(retvals[0]);
  for (auto* h : handles_task0) {
    TFE_DeleteTensorHandle(h);
  }
  for (auto* h : handles_task2) {
    TFE_DeleteTensorHandle(h);
  }

  TFE_DeleteOp(matmul);

  TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
  TFE_ExecutorWaitForAllPendingNodes(executor, status);
  ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);

      TFE_OpAddInput(op, input, s);
    }
    if (TF_GetCode(s) != TF_OK) return;
  }
  std::vector<TFE_TensorHandle*> op_outputs(*num_outputs);
  TFE_Execute(op, op_outputs.data(), num_outputs, s);
  TFE_DeleteOp(op);
  if (TF_GetCode(s) != TF_OK) return;
  std::vector<TFE_TensorHandle*> unwrapped_outputs;
  unwrapped_outputs.reserve(op_outputs.size());
  for (auto* handle : op_outputs) {