TFE_Executor* old_executor = TFE_ContextGetExecutorForThread(ctx);
  TFE_Executor* executor = TFE_NewExecutor(
      /*is_async=*/async, /*enable_streaming_enqueue=*/true,
      /*in_flight_nodes_limit=*/0);
  TFE_ContextSetExecutorForThread(ctx, executor);

  TFE_TensorHandle* m = TestMatrixTensorHandle(ctx);
  TFE_Op* matmul = MatMulOp(ctx, m, m);
  TFE_TensorHandle* retvals[2] = {nullptr, nullptr};
  int num_retvals = 2;