ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    TF_OutputListPushBack(func_outputs, add_output2, s);
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    TF_OutputListPushBack(func_outputs, mm_output, s);
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    func = TF_FinalizeFunction(graph_ctx, func_outputs, s);
    ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    TF_DeleteOutputList(func_outputs);
  }

  ASSERT_TRUE(TF_GetCode(status) == TF_OK) << TF_Message(status);
  TF_DeleteGraph(function_graph);
  TFE_ContextOptions* opts = TFE_NewContextOptions();
  TFE_Context* ctx = TFE_NewContext(opts, status);
  ASSERT_TRUE(TF_GetCode(status) == TF_OK) << TF_Message(status);
  TFE_DeleteContextOptions(opts);
  TFE_ContextAddFunction(ctx, fn, status);
  ASSERT_TRUE(TF_GetCode(status) == TF_OK) << TF_Message(status);
  TF_DeleteFunction(fn);

  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  bool arrived = false;
  bool executed = false;
  const char* name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  RegisterLoggingDevice(context, name, /*strict_scope_placement=*/true,
                        &arrived, &executed, status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());

  TF_OperationDescription* desc = TF_NewOperation(graph, "Const", name);
  TF_SetAttrTensor(desc, "value", t, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
  TF_SetAttrType(desc, "dtype", TF_TensorType(t));
  *op = TF_FinishOperation(desc, s);
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
  ASSERT_NE(*op, nullptr);
}

TF_Operation* Const(TF_Tensor* t, TF_Graph* graph, TF_Status* s,
                    const char* name) {

  TFE_Op* op = TFE_NewOp(ctx, "AddV2", status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInput(op, a, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInput(op, b, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TF_DeleteStatus(status);
  TFE_OpSetAttrType(op, "T", TFE_TensorHandleDataType(a));

  return op;
}

  TF_Tensor* tensor_1X6 = Int32Tensor({1, 6});
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);
  TF_Tensor* tensor_1X1X6 = Int32Tensor({1, 1, 6});
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);

  TFE_Op* reshape_op = TFE_NewOp(tfe_context_, "Reshape", status_);
  CHECK_EQ(TF_OK, TF_GetCode(status_)) << TF_Message(status_);
  TFE_OpSetAttrType(reshape_op, "T", TF_FLOAT);

    status_ = StatusFromTF_Status(status.get());
    ASSERT_EQ(errors::OK, status_.code()) << status_.message();

    {
      AbstractContext* ctx_raw = nullptr;
      status_ =
          BuildImmediateExecutionContext(std::get<1>(GetParam()), &ctx_raw);
      ASSERT_EQ(errors::OK, status_.code()) << status_.message();
      immediate_execution_ctx_.reset(ctx_raw);
    }

  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);

  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  TFE_OpSetAttrType(handle_op.get(), "dtype", TF_FLOAT);
  TFE_OpSetAttrShape(handle_op.get(), "shape", /*dims=*/nullptr, /*num_dims=*/0,
                     status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  auto outputs =

  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());
  TFE_OpSetDevice(op.get(), device_name, status.get());
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());
  TFE_OpAddInput(op.get(), parallel_value.get(), status.get());
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

  TFE_TensorHandle* raw_result_handle;
  int num_retvals = 1;