Quote:
Originally Posted by clsid
I measured the total CPU time used by MPC while playing a 720x480 H.264 clip with subtitles that have karaoke effects. Clip duration 1m46s.
|
CPU time alone is not a good way to judge VSFilter performance. The real question is if there is a gain in FPS with extremely heavy scripts, which are unable or barely playback in real-time on slower computers.
I did some testing, and in most cases gommorah's fixed string manipulation patch results in a 0-2% increase in avg framerate on difficult samples when VSFilter is CPU limited and prebuffering is disabled (r3255 vs r3262 from xvidvideo.ru). Some subtitle samples and CPUs likely benefit from the patch more than others, but at least this resolved one of many causes of VSFilter slowdowns. Since the code hasn't really changed much, it's likely the optimizations from VS2010 vs VS2005 results in most of the gain over Guliverkli2.
AMD X2 4800+ (939) @2.64Ghz on WinXP SP3
VSFilter r3262 (xvidvideo.ru VS2010)
KaraokeSample1: ~22 fps
KaraokeSample2: ~20.5 fps
TypesettingSample1: ~20.5 fps
VSFilter r3255 (xvidvideo.ru VS2010)
KaraokeSample1: ~21.75 fps
KaraokeSample2: ~20 fps
TypesettingSample1: ~20 fps
VSFilter r3242 (VS2010 WPO+PGO)
KaraokeSample1: ~22.5 fps
KaraokeSample2: ~22.75 fps
TypesettingSample1: ~21.25 fps
VSFilter 2.39.5.2 (Guliverkli2 VS2005)
KaraokeSample1: ~20.75 fps
KaraokeSample2: ~16.75 fps
TypesettingSample1: ~19.25 fps
Note: The above are rough measurements, averaged over a few runs on 1000-2000 frame clips. They are accurate enough to see which builds are faster/slower with the samples tested, but not accurate enough to determine exact performance gains/losses in the way I tested. The r3242 WPO+PGO build I just added for comparisons sake. I haven't taken the time to make a new build, but I suspect a r3262 WPO+PGO build would have a similar 0-2% gain with the patch like the xvidvideo.ru builds do.