Skip to content
Browse files
Massive performance improvement on correctBlockNodeIds
correctBlockNodeIds does 2 lookups for each loaded node, one to translate DB ID to name and a second to translate name to real ID. Name to real ID is very consumming if done on every node. As mapblocks are in most cases composed of many identical adjacent nodes, cache previous source and destination id and use them on the next node to prevent any lookup on those maps.

This reduce the function load from 15% of my CPU usage to ~0.7%, on the test, calls was reduced from 2.5M lookups to 42k lookups, it's a huge performance gain
  • Loading branch information
nerzhul committed Jul 27, 2017
1 parent bc1654f commit 61e487719017511fdc37a944ea7321da46d28ee4
Showing 1 changed file with 23 additions and 1 deletion.
@@ -494,22 +494,44 @@ static void correctBlockNodeIds(const NameIdMapping *nimap, MapNode *nodes,
// correct ids.
std::unordered_set<content_t> unnamed_contents;
std::unordered_set<std::string> unallocatable_contents;

bool previous_was_found = false;
content_t previous_local_id = CONTENT_IGNORE;
content_t previous_global_id = CONTENT_IGNORE;

for (u32 i = 0; i < MapBlock::nodecount; i++) {
content_t local_id = nodes[i].getContent();
// If previous node local_id was found and same than before, don't lookup maps
// apply directly previous resolved id
// This permits to massively improve loading performance when nodes are similar
// example: default:air, default:stone are massively present
if (previous_was_found && local_id == previous_local_id) {

std::string name;
if (!nimap->getName(local_id, name)) {
previous_was_found = false;

content_t global_id;
if (!nodedef->getId(name, global_id)) {
global_id = gamedef->allocateUnknownNodeId(name);
if(global_id == CONTENT_IGNORE){
if (global_id == CONTENT_IGNORE) {
previous_was_found = false;

// Save previous node local_id & global_id result
previous_local_id = local_id;
previous_global_id = global_id;
previous_was_found = true;

for (const content_t c: unnamed_contents) {

0 comments on commit 61e4877

Please sign in to comment.