![]() ![]() Message_pump_thread is still spinning even after the official patch. I did say I removed most fixes because QLOC did not identify one of the CPU hogs I found. Loading times will not be longer anymore! With this change done officially, I was able to remove most of the CPU usage fixes SilentPatch introduced – this means that with an updated SilentPatch, I wanted to wait and see if it was going to be fixed officially. If I could reasonably hook up such invasive changes via SP… Well OK, I could have done it regardless – but since it would have required a lot of effort, ![]() Great job (pun unintended)! That’s exactly how I would have solved this problem Whereas a pre-patch version periodically checked if it has any jobs to serve and slept otherwise. Pxd::server_job thread is now reasonable with its CPU usage! A quick dive into the disassembly reveals that QLOC made this thread wait for jobs on a condition variable, Throttling that thread affected performance negatively, and a proper fix was invasive.Ī quick test run with Special K (and without CPU usage fixes from SilentPatch) shows that the CPU usage is indeed significantly lower: I was not able to easily patch the biggest offender, pxd::server_job.Loading times got significantly longer.My approach had some side effects, though: In the last blog post, I identified several hot pointsĬausing the needlessly high CPU usage and fixed some of them. While I’m at it, let’s also take a look at how This means that the fixes I bring need to be re-visited, as they might be obsolete. That changelog is… very similar to what SilentPatch brings! Random crash when ending fights with Heat Move – we managed to fix a crash occurring occasionally after finishing battles with a Heat Action.High CPU usage thread – CPU usage has been cut down by ~30%.Setting ‘Auto’ as default framerate settings. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |