Quote:
Originally Posted by cyberbeing
Thanks for pointing out that perf regression from 3.1.0.705. It seems this is one instance where our original xy tag parsing/caching was significantly faster than MPC-HC's implementation, which I had merged is since I thought it was always faster. My original test case back then must have been too limited.
Here is a test build which reverts the MPC-HC implementation and restores the original xy tag parsing/caching implementation:
https://www69.zippyshare.com/v/dBNj6Fpg/file.html
Please compare it against 3.1.0.352 on extremely line/tag heavy scripts and see which you find overall faster. I'll go back to our original implementation in the next release if you don't find any instances where it is significantly slower than 3.1.0.352.
As for VobSub placement, I'm unsure what that would refer to. If you come across a sample which reproduces that issue again (as a regression from 3.1.0.705), please send it to me. Did you have Black Bar detection enabled in madVR? There is a chance that could have been a madVR bug which has since been fixed. |
For the vobsub placement problem, I remembered an obvious example (my memory isn't as bad as I expected) :
http://dl.free.fr/vrbBwaBoi
after 19s, with the jpn vobsub track. It works as expected with 705, not after. I don't have "black bar detection" enabled in madvr.
For the performance thing, I don't have another example I can remember than the one I already pointed so I can't test more.