auto* h0_task1 =
      TFE_TensorHandleCopyToDevice(h0_task0, ctx, remote_device_name, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  auto* h1_task1 =
      TFE_TensorHandleCopyToDevice(h1_task0, ctx, remote_device_name, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TFE_Op* matmul = MatMulOp(ctx, h0_task1, h1_task1);

  ASSERT_EQ(errors::OK, s.code()) << s.message();
  if (isa<TracingOperation>(check_numerics_op.get())) {
    s = dyn_cast<TracingOperation>(check_numerics_op.get())
            ->SetOpName("check_numerics");
    ASSERT_EQ(errors::OK, s.code()) << s.message();
  }
  s = AddInput(check_numerics_op.get(), t.get(), &forward_op);
  ASSERT_EQ(errors::OK, s.code()) << s.message();

    matmul = TFE_NewOp(ctx, "MatMulFunction", status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
    TFE_OpAddInput(matmul, h0_task0, status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
    TFE_OpAddInput(matmul, has_packed_input ? packed_handle : h1_task2, status);
    ASSERT_EQ(TF_GetCode(status), TF_OK) << TF_Message(status);
  } else {

  TFE_TensorHandle* h = TestScalarTensorHandle(ctx, 1.0f);
  TFE_TensorDebugInfo* debug_info = TFE_TensorHandleTensorDebugInfo(h, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  ASSERT_EQ(0, TFE_TensorDebugInfoOnDeviceNumDims(debug_info));

  TFE_DeleteTensorDebugInfo(debug_info);
  TFE_DeleteTensorHandle(h);
  TFE_DeleteContext(ctx);
  TF_DeleteStatus(status);
}

TEST(CApiDebug, 2DCPU) {

  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  TF_OutputListSetNumOutputs(mm_outputs, 1, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  TF_ExecuteOperation(fn_op, 1, &input_t, mm_outputs, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  ASSERT_EQ(1, TF_OutputListNumOutputs(mm_outputs));

  ASSERT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(status.get()));
  ASSERT_EQ(device_type, nullptr);
  ASSERT_EQ("Invalid handle", string(TF_Message(status.get())));

  TF_SetStatus(status.get(), TF_OK, "");

  int device_id = TFE_TensorHandleDeviceID(h, status.get());
  ASSERT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(status.get()));
  ASSERT_EQ(device_id, -1);

  TF_Operation* a = Placeholder(graph, s, "A");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* b = Placeholder(graph, s, "B");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* two = ScalarConst(2, graph, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* plus2 = Add(a, two, graph, s, "plus2");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  ASSERT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(status.get()));
  ASSERT_EQ(device_name, nullptr);
  ASSERT_EQ("Invalid handle", string(TF_Message(status.get())));

  TF_SetStatus(status.get(), TF_OK, "");

  int num_dims = TFE_TensorHandleNumDims(h, status.get());
  ASSERT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(status.get()));
  ASSERT_EQ(num_dims, -1);
  ASSERT_EQ("Invalid handle", string(TF_Message(status.get())));

  ASSERT_EQ(1, grads.size());
  ASSERT_EQ(func_name_, grads[0].first);
  ASSERT_EQ("MyGrad", grads[0].second);

  // These calls must be noops
  TF_GraphCopyFunction(host_graph_, func_, grad_func, s_);
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
  TF_GraphCopyFunction(host_graph_, func_, nullptr, s_);
  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);

  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;
  ASSERT_EQ(sizeof(sum), TF_TensorByteSize(t));
  memcpy(&sum, TF_TensorData(t), TF_TensorByteSize(t));