TF_RETURN_IF_ERROR(status);
    graph->AddEdge(node, 0, main_node, i);
  }

  // Similarly with return values, create dummy _Retval nodes fed by `node`.
  for (int64_t i = 0, end = result_types.size(); i < end; ++i) {
    Node* node;
    string retval_name = absl::StrCat("_retval", i);
    Status status = NodeBuilder(retval_name, FunctionLibraryDefinition::kRetOp)
                        .Input(main_node, i)

      count: 4
    }
  }
  unclustered_op_histogram {
    op: "NoOp"
    count: 2
  }
  unclustered_op_histogram {
    op: "_Arg"
    count: 1
  }
  unclustered_op_histogram {
    op: "_Retval"
    count: 1
  }
}
)",
      &expected_auto_clustering_activity);
  EXPECT_EQ(listener()->auto_clustering_activity().DebugString(),
            expected_auto_clustering_activity.DebugString());

	}

	for _, res := range s.resources {
		if res == nil {
			continue
		}
		resVal := reflect.ValueOf(res)
		if resVal.Type().AssignableTo(targetT) {
			if refVal.Kind() == reflect.Slice {
				refVal.Set(reflect.Append(refVal, resVal))
			} else {
				refVal.Set(resVal)
				return nil
			}
		}
	}

	if s.parent != nil {
		// either didn't find the value or need to continue filling the slice

    for (auto input : inputs) {
      TF_RETURN_IF_ERROR(fn_op->AddInput(input));
    }
    int retvals = outputs.size() - null_indices.size();
    std::vector<AbstractTensorHandle*> fn_outputs(retvals);
    TF_RETURN_IF_ERROR(fn_op->Execute(
        absl::Span<AbstractTensorHandle*>(fn_outputs.data(), fn_outputs.size()),
        &retvals));
    int skipped_indices = 0;
    for (int i = 0; i < outputs.size(); i++) {

// CHECK-NEXT:    "tf.IfrtCall"(%arg0, [[ARRAYKEY]]) <{program_id = 6515870160938153680 : i64, variable_arg_indices = [1 : i32]}> {__tpu_compile_metadata_text = "retvals { sharding { } }"} : (tensor<1x3xf32>, tensor<!tf_type.string>) -> tensor<1x1xf32>
// CHECK-NEXT:    return
//
 func.func @serving_default(%arg0: tensor<1x3xf32>) -> tensor<1x1xf32> {

		if dst != nil && srcPatch != nil {
			retVal, errMerge := util.MergeAnyWithAny(dst, srcPatch)
			if errMerge != nil {
				return false, fmt.Errorf("function MergeAnyWithAny failed for ApplyClusterMerge: %v", errMerge)
			}

			// Merge the above result with the whole cluster
			merge.Merge(dst, retVal)
		}
	}
	return true, nil
}

  kAsync,
};

enum class DeviceCompileState {
  kUncompiled,
  kCompiling,
  kCompiled,
};

// Creates a single-node graph using the specified `node_def` as the only op
// apart from the arg and retval nodes corresponding to `args` and
// `result_types` respectively.
absl::StatusOr<std::unique_ptr<Graph>> CreateSingleOpGraph(
    const NodeDef& node_def, absl::Span<const XlaArgument> args,

Status TapeOperation::Execute(absl::Span<AbstractTensorHandle*> retvals,
                              int* num_retvals) {
  TF_RETURN_IF_ERROR(parent_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]);
  }

      auto metadata_arg = metadata_proto.add_args();
      metadata_arg->set_kind(
          tensorflow::tpu::TPUCompileMetadataProto::Arg::PARAMETER);
      metadata_arg->set_dtype(dtype);
    }
    // Set up one dummy retval.
    metadata_proto.add_retvals();
    bool use_tuple_args = true;
    std::vector<tensorflow::tpu::ShardingAndIndex> arg_core_mapping;
    std::vector<std::vector<xla::Shape>> per_core_arg_shapes;

	// universe scope
	_Append builtinId = iota
	_Cap
	_Clear
	_Close
	_Complex
	_Copy
	_Delete
	_Imag
	_Len
	_Make
	_Max
	_Min
	_New
	_Panic
	_Print
	_Println
	_Real
	_Recover

	// package unsafe
	_Add
	_Alignof
	_Offsetof
	_Sizeof
	_Slice
	_SliceData
	_String
	_StringData

	// testing support
	_Assert
	_Trace
)