TFE_TensorHandle* retvals[1] = {nullptr};
  int num_retvals = 1;
  TFE_Execute(func, &retvals[0], &num_retvals, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  ASSERT_EQ(1, num_retvals);
  TF_Tensor* t = TFE_TensorHandleResolve(retvals[0], status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteTensorHandle(retvals[0]);
  float sum = 0;

// Execute the operation defined by 'op' and return handles to computed
// tensors in `retvals`.
//
// 'retvals' must point to a pre-allocated array of TFE_TensorHandle* and
// '*num_retvals' should be set to the size of this array. It is an error if
// the size of 'retvals' is less than the number of outputs. This call sets
// *num_retvals to the number of outputs.
//

    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
    TFE_TensorHandle* retvals[1];
    int num_retvals = 1;
    TFE_Execute(shape_op, &retvals[0], &num_retvals, status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());

    device_type = TFE_TensorHandleDeviceType(retvals[0], status.get());
    ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  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);

  auto* retval_task0 =
      TFE_TensorHandleCopyToDevice(retvals[0], ctx, local_device_name, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

                     int* num_retvals, ForwardOperation* forward_op_,
                     Tape* tape, const GradientRegistry& registry) {
  TF_RETURN_IF_ERROR(op_->Execute(retvals, num_retvals));
  for (int i = 0; i < *num_retvals; i++) {
    // TODO(srbs): Manage refcount of ForwardOperation's inputs/outputs.
    forward_op_->outputs.push_back(retvals[i]);
  }

                       ImmediateExecutionTensorHandle** retvals,
                       int* num_retvals) override {
    std::vector<TFE_TensorHandle*> outputs(*num_retvals);
    TF_Status status;
    device_.execute(tensorflow::wrap(op), num_retvals, outputs.data(), &status,
                    info_);
    if (status.status.ok()) {
      for (int i = 0; i < *num_retvals; ++i) {
        retvals[i] = tensorflow::unwrap(outputs[i]);

      }
      tf_outputs[i] = t->output_;
    }
    TF_AddInputList(op_.get(), tf_outputs.data(), tf_outputs.size());
    return absl::OkStatus();
  }
  absl::Status Execute(absl::Span<AbstractTensorHandle*> retvals,
                       int* num_retvals) override {
    auto* tf_opdesc = op_.release();
    if (tf_opdesc == nullptr) {
      return errors::InvalidArgument("AbstractOp is incomplete.");
    }

        @Override
        public int getOpnum() {
            return 0x01;
        }

        /** The return value of the operation */
        public int retval;
        /** The handle to be closed */
        public rpc.policy_handle handle;

        /**
         * Constructs a SamrCloseHandle request.
         *
         * @param handle The policy handle to close

        @Override
        public int getOpnum() {
            return 0x01;
        }

        /** The return value of the operation */
        public int retval;
        /** The handle to be closed */
        public rpc.policy_handle handle;

        /**
         * Constructs a SamrCloseHandle request.
         *
         * @param handle The policy handle to close

  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  TFE_TensorHandle* retval;
  int num_retvals = 1;
  TFE_Execute(matmul.get(), &retval, &num_retvals, status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  ASSERT_TRUE(executed);

  TFE_DeleteTensorHandle(retval);
  TFE_DeleteTensorHandle(hcpu);
  TFE_DeleteTensorHandle(hdevice);
  TFE_DeleteContext(context);
}