Function ggml_gallocr_alloc_graph_impl

external/ggml/src/ggml-alloc.c:717–822

}

static void ggml_gallocr_alloc_graph_impl(ggml_gallocr_t galloc, struct ggml_cgraph * graph, const int * node_buffer_ids, const int * leaf_buffer_ids) {
    // clear hash tables
    ggml_hash_set_reset(&galloc->hash_set);
    memset(galloc->hash_values, 0, sizeof(struct hash_node) * galloc->hash_set.size);

    // allocate leafs
    // these may be tensors that the application is not using in the graph, but may still want to allocate for other purposes
    for (int i = 0; i < graph->n_leafs; i++) {
        struct ggml_tensor * leaf = graph->leafs[i];
        ggml_gallocr_allocate_node(galloc, leaf, get_node_buffer_id(leaf_buffer_ids, i));
    }

    // count number of children and views
    // allocate other graph inputs and leafs first to avoid overwriting them
    for (int i = 0; i < graph->n_nodes; i++) {
        struct ggml_tensor * node = graph->nodes[i];

        // TODO: better way to add external dependencies
        // GGML_OP_NONE does not appear normally in the graph nodes, but is used by ggml-backend to add dependencies to
        // control when some tensors are allocated and freed. in this case, the dependencies are in `src`, but the node
        // itself is never used and should not be considered a dependency
        if (ggml_impl_is_view(node) && node->op != GGML_OP_NONE) {
            struct ggml_tensor * view_src = node->view_src;
            ggml_gallocr_hash_get(galloc, view_src)->n_views += 1;
        }

        if (node->flags & GGML_TENSOR_FLAG_INPUT) {
            ggml_gallocr_allocate_node(galloc, graph->nodes[i], get_node_buffer_id(node_buffer_ids, i));
        }

        for (int j = 0; j < GGML_MAX_SRC; j++) {
            struct ggml_tensor * src = node->src[j];
            if (src == NULL) {
                continue;
            }

            ggml_gallocr_hash_get(galloc, src)->n_children += 1;

            // allocate explicit inputs
            if (src->flags & GGML_TENSOR_FLAG_INPUT) {
                ggml_gallocr_allocate_node(galloc, src, get_node_buffer_id(node_buffer_ids, i));
            }
        }
    }

    // allocate tensors
    for (int i = 0; i < graph->n_nodes; i++) {
        struct ggml_tensor * node = graph->nodes[i];
        int buffer_id = get_node_buffer_id(node_buffer_ids, i);

        // allocate parents (only leafs need to be allocated at this point)
        for (int j = 0; j < GGML_MAX_SRC; j++) {
            struct ggml_tensor * parent = node->src[j];
            if (parent == NULL) {
                continue;
            }
            ggml_gallocr_allocate_node(galloc, parent, buffer_id);
        }