Possible bug in ggml_backend_sched_split_graph()?

[CharlesJu1]

Is there a bug in the following code regarding "else if (cur_backend_id != -1) "?

When cur_backend_id is changed from -1 to other values in the if clause "if (*node_backend_id != -1)", the "else if (cur_backend_id != -1) " is not executed because the "previous" cur_backend_id is -1。The changed value of cur_backend_id in the if clause does not affect the decision of else if clause.

    // pass 2: expand current backend assignments
    // assign the same backend to adjacent nodes
    // expand gpu backends (i.e. non last prio) up and down, ignoring cpu (the lowest priority backend)
    // thus, cpu will never be used unless weights are on cpu, or there are no gpu ops between cpu ops
    // ops unsupported by the backend being expanded will be left unassigned so that they can be assigned later when the locations of its inputs are known
    // expand gpu down
    {
        int cur_backend_id = -1;
        for (int i = 0; i < graph->n_nodes; i++) {
            struct ggml_tensor * node = graph->nodes[i];
            if (ggml_is_view_op(node->op)) {
                continue;
            }
            int * node_backend_id = &tensor_backend_id(node);
            if (*node_backend_id != -1) {
                if (*node_backend_id == sched->n_backends - 1) {
                    // skip cpu (lowest prio backend)
                    cur_backend_id = -1;
                } else {
                    cur_backend_id = *node_backend_id;
                }
            } else if (cur_backend_id != -1) {
                ggml_backend_sched_set_if_supported(sched, node, cur_backend_id, node_backend_id);
            }
        }
    }

[CharlesJu1]

I figured it is intended.