CCCP Project Forums

Please login or register.

Login with username, password and session length
Advanced search  

Author Topic: High-10-Profile Performance  (Read 11329 times)

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
High-10-Profile Performance
« on: October 31, 2011, 08:38:02 PM »

Generally, they play fine under my current settings. But I experience slowdown when encountering karaoke, certain episode titles, and other fancy subtitles. Can my machine do anything about this?

I already tried LAV Video with Overlay Mixer. Those do help, but the performance is still unsatisfactory.

http://img831.imageshack.us/img831/5128/filtert.png

(click to show/hide)
Logged
Bring me your planet sign / Faster than the seventh sound

JEEB

  • Hoser The Third
  • Administrator
  • *
  • Posts: 850
  • The Random Encoder
Re: High-10-Profile Performance
« Reply #1 on: October 31, 2011, 09:12:25 PM »

VSFilter is as slow as a turd, and you seem to have an under-2GHz dual core.

You could either try lachs0r's smplayer2 builds with mplayer2 (clicky, under the mplayer2 build listing there's text on smplayer2 and a link to the "easy package"), or the NT6+-only Threaded-VSFilter (as it seems to be generally more stable than the one that's not based on the NT6+-only stuff).

As you have to at least know basics on how to unregister the old VSFilter and register a new one, I do recommend the (s)mplayer2 route, not to mention that libass is generally very much faster as a subtitle renderer :3
Logged

gommorah

  • DirectShow Mage
  • *****
  • Posts: 137
Re: High-10-Profile Performance
« Reply #2 on: November 01, 2011, 06:20:49 AM »

If you have CCCP already installed, threaded-VSFilter can just be dropped into the Filters folder of the CCCP installation dir. I recommend that you back up the version that comes with CCCP in case threaded-VSFilter epically breaks something for you (and if it does, please report it in the dev thread :) ).
Logged

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
Re: High-10-Profile Performance
« Reply #3 on: November 01, 2011, 11:25:38 AM »

Unfortunately, smplayer2 has worse playback. And the threaded VSFilter made no visible difference. I tried both methods—installing through Run, and dropping it in the Filters folder. Thanks anyway.

Guess I should bear with LAV Video, Haali Media Splitter, the S key, and my not-shabby Japanese.
« Last Edit: November 01, 2011, 11:28:33 AM by aurabolt »
Logged
Bring me your planet sign / Faster than the seventh sound

gommorah

  • DirectShow Mage
  • *****
  • Posts: 137
Re: High-10-Profile Performance
« Reply #4 on: November 01, 2011, 01:14:09 PM »

Unfortunately, smplayer2 has worse playback. And the threaded VSFilter made no visible difference. I tried both methods—installing through Run, and dropping it in the Filters folder. Thanks anyway.

Guess I should bear with LAV Video, Haali Media Splitter, the S key, and my not-shabby Japanese.
Think you could demux the .ass script from the file and post it? I'm always interested in testing scripts that slow down threaded-VSFilter and I'm sure xy would be interested as well for his VSFilter mod. You can also try his VSFilter if you haven't already.
Logged

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
Re: High-10-Profile Performance
« Reply #5 on: November 01, 2011, 03:10:42 PM »

Here it is.

http://www.mediafire.com/?dnh198e3d2oephe

I actually missed xy's, and I will give it a try.
Logged
Bring me your planet sign / Faster than the seventh sound

gommorah

  • DirectShow Mage
  • *****
  • Posts: 137
Re: High-10-Profile Performance
« Reply #6 on: November 01, 2011, 07:54:23 PM »

I haven't had a chance to test the script out, but before I do, is there a time stamp when the lag starts? Also, I forgot to mention this earlier, but be sure to enable Subpicture Pre-buffering when using threaded-VSFilter. I believe CCCP defaults that to off so if you didn't change it, threaded-VSFilter will pick up the same setting. You can use whatever number of buffers you want and have system memory for (I personally use 30 since that's 1 second worth of subtitles for a 30FPS video and a little more than a second for 24FPS video), but subpicture pre-buffering helps a lot frame drops due to softsubs.
Logged

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
Re: High-10-Profile Performance
« Reply #7 on: November 01, 2011, 08:59:16 PM »

I ran more tests with LAV Video and Haali Media Splitter. And [SS-Eclipse] Shakugan no Shana Final - 04 (1280x720 Hi10P) [B3552EAF].mkv runs just fine. Oops, I noticed this has hard-subbed karaoke.

But here is the funny part. This is by the same group responsible for the script I provided earlier. [Chibiki]_THE_iDOLM@STER_-_14_[720p][132A89C6].mkv lags on karaoke and subtitles for on-screen text; this is only 8-bit. What the heck are they doing to those subtitles? D:

Quote
I haven't had a chance to test the script out, but before I do, is there a time stamp when the lag starts?
The karaoke/lag starts at 00:02:09.

Quote
Also, I forgot to mention this earlier, but be sure to enable Subpicture Pre-buffering when using threaded-VSFilter.
Oops, I have not. I will do this tomorrow along with testing one more 10-bit video.
« Last Edit: November 02, 2011, 08:55:55 AM by aurabolt »
Logged
Bring me your planet sign / Faster than the seventh sound

cyberbeing

  • DirectShow Mage
  • *****
  • Posts: 338
Re: High-10-Profile Performance
« Reply #8 on: November 02, 2011, 08:55:00 AM »

[SS-Eclipse] Shakugan no Shana Final - 04 (1280x720 Hi10P) [B3552EAF].mkv uses hardsubbed karaoke, so it's not surprising it runs well since VSFilter is doing nothing. [Edit: I see you noticed that with your ninja edit]

[Chibiki]_THE_iDOLM@STER_-_14_[720p][132A89C6].mkv you should be having no trouble with using xy-VSFilter (if you do try changing the sub-pixel positioning option and see if it helps), but the video decoding load is so low and karaoke very simple so it's surprising you are having trouble. For this sample, there is a good chance xy-VSFilter will be faster than threaded-VSFilter.

[UTW]_Fate_Zero_-_05_[h264-720p][0F358C5F].mkv does run a bit slow from use of both a medium strength \blur3.5 and as well as \be1 and \be2. I would expect this to run significantly faster on threaded-VSFilter since it has multi-threaded blur code, but it may look different than stock VSFilter.

If you unregister VSFilter (regsvr32 /u vsfilter.dll), what is your CPU load playing those samples? I also have to wonder if your CPU is overheating and throttling. Download ThrottleWatch and do a few tests. Press F5 to start logging and F5 again to stop logging. Once you stop logging rename journal.txt so you remember what you tested so it doesn't get overwritten when you log the next thing.

Test the following four things for each sample you test (attempting to start and stop logging during the same portions of each sample and noting down if you have playback problems or not).
1) Playback with VSFilter unregistered
2) Playback with CCCP VSFilter
3) Playback with threaded-VSFilter
4) Playback with xy-VSFilter

I'd start with the Fate/Zero OP karaoke since that's the most intensive of your samples. Pause the video, jump to the chapter at 01:59, hit F5 to start logging, un-pause the video, and hit F5 again to stop logging at 03:30. When finished, zip up your four journal.txt files and upload them to mediafire or something. I'm curious if this is just an issue of lack of CPU power, or something else is going on.
Logged

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
Re: High-10-Profile Performance
« Reply #9 on: November 02, 2011, 09:24:36 AM »

A quick post. I will get to the rest soon.

[Chibiki]_THE_iDOLM@STER_-_14_[720p][132A89C6].mkv
This does lag. But, yeah, not as much as other videos.

I actually wanted to address older episodes such as [Chibiki]_THE_iDOLM@STER_-_12_[720p][71ADE7A0].mkv yesterday, but didn't have it at hand. This has the "changing color effect" used in karaoke.

Also, regarding xy-VSFilter.
Quote
Remember to disable P010/P016 options in your decoder if you're not using madVR, see http://code.google.com/p/xy-vsfilter/issues/detail?id=41 .
How do I do this?
« Last Edit: November 02, 2011, 09:35:04 AM by aurabolt »
Logged
Bring me your planet sign / Faster than the seventh sound

cyberbeing

  • DirectShow Mage
  • *****
  • Posts: 338
Re: High-10-Profile Performance
« Reply #10 on: November 02, 2011, 09:45:48 AM »

Quote
Remember to disable P010/P016 options in your decoder if you're not using madVR, see http://code.google.com/p/xy-vsfilter/issues/detail?id=41 .
How do I do this?
If you are using FFDShow that doesn't apply to you, since it doesn't support P010/P016 output.

If you are using CoreAVC 3.x or LAV Video, you'll need to uncheck those formats (play a video, right-click->filters->click on CoreAVC/LAV_Video->disable P010 & P016) in the codec properties when not using madVR.
Logged

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
Re: High-10-Profile Performance
« Reply #11 on: November 02, 2011, 12:36:55 PM »

Okay, where should I start? Hope I do not miss answering anything. Well, the journals are here: http://www.mediafire.com/?1o47fa4a4ax5yf5

I did not notice improvements from CCCP's VSFilter to the threaded VSFilter. But xy-VSFilter gave the following results. [UTW]_Fate_Zero_-_05_[h264-720p][0F358C5F].mkv almost runs perfectly in the beginning. Then it starts desyncing here; you will see why: http://img231.imageshack.us/img231/4972/utwfatezero05h264720p0f.jpg

The following samples run without performance issues now.
[Chibiki]_THE_iDOLM@STER_-_14_[720p][132A89C6].mkv has the second opening.
[Chibiki]_THE_iDOLM@STER_-_12_[720p][71ADE7A0].mkv has the first opening.
[Chibiki]_THE_iDOLM@STER_-_06_[720p][D0D8A938].mkv has no karaoke due to a mistake. But I tested it since the video immediately begins with lots of signs.
[Coalgirls]_Yuru_Yuri_OP_(1280x720_Blu-Ray_FLAC)_[B6877357].mkv is a 10-bit, and runs well since it does not use fancy subtitles like the Fate/zero video.

So now I am using LAV Video, xy-VSFilter, and CCCP.
« Last Edit: November 02, 2011, 12:44:52 PM by aurabolt »
Logged
Bring me your planet sign / Faster than the seventh sound

cyberbeing

  • DirectShow Mage
  • *****
  • Posts: 338
Re: High-10-Profile Performance
« Reply #12 on: November 02, 2011, 02:39:52 PM »

Looking at your Journal.txt files, all I can say is ouch.

Even with VSFilter unregistered, your average CPU load is 87% with it hovering between 93-98% the majority of the time. Do you have any programs, anti-virus scans, or anything like that running in the background which you can close or turn off? It does seem like your biggest issues is lack of CPU power to decode Hi10P and leave ample room for rendering subtitles. The minor overhead of multi-threading in threaded-VSFilter must be making things worse since it starts competing with your video decoder, and even pre-buffering may end up harming things...

It's good to hear xy-VSFilter works for you mostly (btw a new version 20111103 is out). The place you experience a slowdown is right around where the cache fills to its default limit. As a test, try increasing all four xy-VSFilter caches to 2048 and make sure 8x8fast is enabled. Do things improve any? If that doesn't help, just keep your eye out for future updates.

x_xy_y2 (xy-VSFilter dev) still has some optimizing of the blur code on his to-do list, so with the Fate/Zero sample using a lot of blur, it may get better in the future. As for just plain h.264 decoding performance improvements, CoreAVC seems to be suggesting that 3.1 will finally beat LAV Video and FFDShow Hi10P decoding performance, which would be most helpful in your current situation, but we'll need to wait and see if they are actually successful in meeting that claim.
« Last Edit: November 02, 2011, 03:04:13 PM by cyberbeing »
Logged

aurabolt

  • CCCP Lieutenant
  • **
  • Posts: 20
  • Anime Metalhead
    • MyAnimeList
Re: High-10-Profile Performance
« Reply #13 on: November 02, 2011, 04:23:48 PM »

Quote
Do you have any programs, anti-virus scans, or anything like that running in the background which you can close or turn off?
No, I don't use any. mIRC is normally running, but I had it closed during the logging. This computer's already been tweaked down to run just twenty-three processes on startup, to use minimal visual effects, and etc. So this is probably the limit.

Quote
Do things improve any?
Nope.

Quote
x_xy_y2 (xy-VSFilter dev) still has some optimizing of the blur code on his to-do list, so with the Fate/Zero sample using a lot of blur, it may get better in the future. As for just plain h.264 decoding performance improvements, CoreAVC seems to be suggesting that 3.1 will finally beat LAV Video and FFDShow Hi10P decoding performance, which would be most helpful in your current situation, but we'll need to wait and see if they are actually successful in meeting that claim.
All right. Thanks for the help.

I normally wait and marathon rips from Coalgirls (who use sane subtitles), so I'll probably manage somehow for now.
« Last Edit: November 02, 2011, 08:47:58 PM by aurabolt »
Logged
Bring me your planet sign / Faster than the seventh sound

gommorah

  • DirectShow Mage
  • *****
  • Posts: 137
Re: High-10-Profile Performance
« Reply #14 on: November 03, 2011, 08:16:17 PM »

Ok so the reason karaoke lags is because every single line has a blurred border. Aside from the expense of the blurring itself, creating outline borders like that is very expensive and can't be multithreaded as far as I've been able to tell. The reason why xy-VSFilter might be running better for you is because he's much more aggressive about caching outlines than I am. For me, the additional caching systems I'm adding into threaded-VSFilter is meant to complement subpicture prebuffering, whereas xy needs more extensive and aggressive caching to compensate for the loss of subpicture prebuffering. The biggest advantage to enabling subpicture prebuffering is that it decouples the subtitle rendering from the video decoding (normally, VSFilter is run by the video decoder thread), so that even if VSFilter starts choking on subtitles, playback won't grind to a halt. The reason for that is subtitle rendering gets kicked off into a child thread, separate from the video decoder's thread, when subpicture prebuffering is enabled. The other advantage to subpicture prebuffering is that it allows threaded-VSFilter to render ahead and fill the buffer during periods where it has nothing to do.

Unfortunately, if you're averaging 87% CPU load just decoding the video, threaded-VSFilter might actually cause CPU contention, especially with subpicture prebuffering on. On my machine with a Phenom X4 9950, I don't get any lag, but I can see madVR's render queue dip every time a new line shows on screen, so I can imagine how it would be pretty bad for you. Seems xy-VSFilter might be better just because it won't cause CPU contention with the video decoder.

On the plus side, the script showed some additional functions that would benefit from the new SSE2 alpha blend I wrote a while back so those have been added in. Now the UTW logo should be less crippling to draw (gah, I can't believe that's softsubbed >.<). If you wouldn't mind testing the latest build with subpicture prebuffering enabled (any buffer value greater than 0), I'd appreciate it.

EDIT: So I tested UTW's Fate Zero episode 5 with the build I linked you on my laptop with the following configuration:

Core Duo T2500 2Ghz
2GB of RAM
ATI Mobility Radeon X1600 256MB
Windows 7 32-bit
CCCP 7-30-2011
madVR v0.77
LAV v0.37 set to output P010
Subpicture prebuffering set to 10

I didn't get any lag with the OP karaoke at all. In fact, using the VSFilter that came with CCCP didn't lag either. Our systems are pretty similar, although I have an older, but higher clocked CPU, so if you're still lagging, I'm not sure what's going on.

EDIT2: Just tested both CCCP VSFilter and threaded-VSFilter without subpicture prebuffering enabled and both of them were able to playback OP karaoke without any lag so I'm not sure what's going on with your system.
« Last Edit: November 04, 2011, 08:34:22 AM by gommorah »
Logged
 

Page created in 0.092 seconds with 18 queries.