View Issue Details

IDProjectCategoryView StatusLast Update
0000153madVRbugpublic2014-03-06 20:31
Reportermark007 Assigned Tomadshi  
PrioritynormalSeveritymajorReproducibilityalways
Status closedResolutionno change required 
Platformx64OSWindowsOS Version8.1
Summary0000153: Frame Drops On Initial PGS Subtitle Load After 0.87.4
DescriptionFrom 0.87.4 onwards, madVR has started to drop a bunch of frames when loading some PGS subtitles. The symptoms are, when cpu / gpu usage is relatively high (using SVP to interpolate to 60fps), 0000009:0000025% gpu and 0000052:0000060% cpu, madVR will drop frames when this section of PGS subtitles appears.

When mpc-hc is rewinded to a few seconds before the problem area and allowed play back over it, all is fine, it only happens the first time the subtitles appear.

This is using xySubFilter beta, mpc-hc or mpc-be, with LAV Filters / Splitter and SVP.

Something that changed in 0.87.4 made this occur as there are 0 drops in 0.87.3 through an entire blu-ray and 0 drops after the first block of PGS / frame drops appear, although version <=0.87.3 did show slight drops in render queue that never seems to fully fill back up at the initial PGS loads also, but most importantly it didn't drop any frames.
Steps To ReproduceLoad the sample file from here

https://www.mediafire.com/?0x4jr0ho6seup3z

Monitor cpu / gpu usage as the subtitle appears. It causes frame drops in 0.87.4 but not older versions on my system thats under relatively high load so something in the initial PGS subtitle load must be very intensive or blocking.

See screenshot of the problem area, where frame drops can be seen starting to occur, half way through the sample.
Additional InformationSee initial discussion with cyberbeing here on the xyVSFilter thread

http://forum.doom9.org/showthread.php?p=1669342#post1669342
TagsNo tags attached.
madVR Version0.87.4
Media Player (with version info)MPC-BE 1.3.1.0 Build 4395
Splitter (with version info)LAV 0.60.1.4
Decoder (with version info)LAV 0.60.1.4
DecodingSoftware
Deinterlacingnone (progressive)
DXVA2 Scaling Activeno
Aero / Desktop CompositionOn
Problem occurs with modeall modes
GPU ManufacturerNVidia
GPU ModelGTX 295
GPU Driver Version334.67

Activities

mark007

2014-02-19 23:45

reporter  

ztf21h.jpg (232,342 bytes)   
ztf21h.jpg (232,342 bytes)   

madshi

2014-03-04 15:30

administrator   ~0000432

I can't seem to be able to reproduce this here. Also, looking at the changes between v0.87.3 and v0.87.4, these changes should all improve things. I see no potential for lower performance or any other issues due to the changes done in v0.87.4, compared to v0.87.3.

Without being able to reproduce this there's probably not going to be much I can do about it. And to be honest, I'm not feeling like installing SVP. Any other way to reproduce this?

mark007

2014-03-04 19:52

reporter   ~0000448

Hi madshi, I don't suppose a debug log would help debug whats causing the queue to drop so quickly when load is high on my machine? I wonder can you reproduce if you have any other way of increasing load on your machine (ie simulate in some way what SVP is doing)

evr-cp also shows a noticible spike in the graph at the same time so perhaps it should go back to cyberbeing to see if its something that could be looked into from his end, although 0.87.4 drops where as v0.87.3 doesn't for me so I'm not sure where to go from here.

madshi

2014-03-04 20:11

administrator   ~0000449

Well, we can try a debug log with v0.87.4. Please try to keep it short. Alternatively we could also wait for v0.87.5 (due out soon).

mark007

2014-03-04 20:17

reporter  

madVR - log.rar (486,737 bytes)

mark007

2014-03-04 20:18

reporter   ~0000450

I have uploaded a short log, using the clip linked to above, using 0.87.4. I will also try with 0.87.5 and report back.

madshi

2014-03-04 21:11

administrator   ~0000451

I don't see anything which would be madVR's fault. There's a hole in the log where nothing happens for 29 milliseconds, which causes one of the frames to drop because madVR doesn't manage to present it fast enough, due to that 29 millisecond without CPU time. I think the problem will probably be similar for the other frames. Generally, the timing in windowed mode is quite sensitive. Is Aero/Desktop Composition enabled or disabled? Are you using Overlay mode? If not, try Overlay, it might help. But as I'm always recommending: You'll get the most reliable playback by using fullscreen exclusive mode.

madshi

2014-03-04 21:19

administrator   ~0000452

P.S: There is a chance that v0.87.4 might have been compiled with optimizations turned off, I'm not totally sure. Just to be safe, let's double check with v0.87.5, when it's released.

mark007

2014-03-05 19:49

reporter   ~0000459

madshi, your absolutely right, in fullscreen mode although queues drop slightly at the time of the first subtitle, they don't drop so much as to drop frames. This is the same behavior with 0.87.5. Only seeing frame drops in windowed mode with 0.87.4+. I doubt this is important enough to look into seeing as I can't reproduce in FSE but still strange how it only happens on the first subtitle and not when replaying back over the same section again.

Its of course up to you how you want to handle it. If you feel its something youd like to figure out, I'm willing to provide any more detail / logs as required.

madshi

2014-03-05 23:59

administrator   ~0000461

So it still does occur with v0.87.5? In any case, I don't really know what to do about this. I'm not sure if there is anything I can do. Have you tried overlay mode?

mark007

2014-03-06 20:28

reporter   ~0000470

Hi madshi, in windowed mode, with overlay enabled although the queues do reduce, I get 0 frame drops. :) So the only combination that shows the drops when subtitles appear is windowed non overlay.

madshi

2014-03-06 20:31

administrator   ~0000471

Well non-overlay windowed mode is the most sensitive, so it makes sense.

Anyway, I don't think there's anything I can do, so I'll close this bug report for now. I think you have enough alternatives (overlay, fse)... :-)

Issue History

Date Modified Username Field Change
2014-02-19 23:45 mark007 New Issue
2014-02-19 23:45 mark007 File Added: ztf21h.jpg
2014-03-04 15:30 madshi Note Added: 0000432
2014-03-04 15:30 madshi Assigned To => madshi
2014-03-04 15:30 madshi Status new => feedback
2014-03-04 19:52 mark007 Note Added: 0000448
2014-03-04 19:52 mark007 Status feedback => assigned
2014-03-04 20:11 madshi Note Added: 0000449
2014-03-04 20:12 madshi Status assigned => feedback
2014-03-04 20:17 mark007 File Added: madVR - log.rar
2014-03-04 20:18 mark007 Note Added: 0000450
2014-03-04 20:18 mark007 Status feedback => assigned
2014-03-04 21:11 madshi Note Added: 0000451
2014-03-04 21:13 madshi Status assigned => feedback
2014-03-04 21:19 madshi Note Added: 0000452
2014-03-05 19:49 mark007 Note Added: 0000459
2014-03-05 19:49 mark007 Status feedback => assigned
2014-03-05 23:59 madshi Note Added: 0000461
2014-03-05 23:59 madshi Status assigned => feedback
2014-03-06 20:28 mark007 Note Added: 0000470
2014-03-06 20:28 mark007 Status feedback => assigned
2014-03-06 20:31 madshi Note Added: 0000471
2014-03-06 20:31 madshi Status assigned => closed
2014-03-06 20:31 madshi Resolution open => no change required