[Cyberkid2002]

build I am currently using which is working fine is this one 20110711-bbdd73f. I actually had an old XBMC folder (renamed to XBMCJuly in my "program files" so just renamed it back to xbmc. It's how I do my upgrades incase I have any problems.)

[Doebe]

i`ll try this build thx....
see what happens

[Hitcher]

Play film -

XBMC dropped frame log

Play and pause film -

XBMC no dropped frame log

[xxrzdxx]

(Copied from my reply in an alternate thread)

XBMCSetup-20110803-1738a4f-master.exe: Constant Dropped Frames

XBMCSetup-20110727-33fcf32-master.exe: Zero Dropped Frames

[Doebe]

XBMCSetup 20110909 fa79f61 dx restore dxva yuv rt.exe

seems to be working fine....
i`ll stick with this one for now :-)

[WhiningKhan]

According to Isidrogar in http://forum.xbmc.org/showpost.php?p=816992&postcount=1 , the advanced deinterlacers require NV12 format.

I'll try to test this patch this evening (built yesterday's trunk, saw occasional stutter) ; just a note from what I've found during the summer, using Isidrogar's patch and which I think some people here may be suffering from: Having a GPU with marginal shader power results in occasional behavior where missed-counter increases continuously until stream is paused; the visible stutter may still be quite minor, however noticeable. For interlaced PAL SD MPEG2 streams, overclocking my HD3200 from 500MHz to 700MHz actually got rid of the issue.

[Voyager]

just confirming the exhibited behaviour: infrequent stuttering on dxva rendered playback (SD) with dxva deinterlacing on (and even without, but much less noticeable). Also, after a while the stream just briefly "freezes", then continues.

My CPU/GPU is Atom/ION 330, and using the latest 280 drivers on Win 7 32bit. I had to revert to use pixelshaders rendering and software deinterlacing (half) in order to have watchable results. I'll try the patch in the next couple of days and report back.

My observation is that also dxva decoded material (HD) is now stuttering a little bit, which is an annoyance (as before it did almost play without hiccup).

[WhiningKhan]

I'll try to test this patch this evening (built yesterday's trunk, saw occasional stutter) ; just a note from what I've found during the summer, using Isidrogar's patch and which I think some people here may be suffering from: Having a GPU with marginal shader power results in occasional behavior where missed-counter increases continuously until stream is paused; the visible stutter may still be quite minor, however noticeable. For interlaced PAL SD MPEG2 streams, overclocking my HD3200 from 500MHz to 700MHz actually got rid of the issue.

Well, the issue I described above appears with this fix applied to trunk even with GPU overclocked; the missed-counter increases steadily. However, I'm not seeing the same 'minor freeze - catchup to sync' issue as yesterday - I guess that was the original target.

So, for some reason, the trunk code for DXVA deinterlace seems to burden the GPU more than Isidrogar's patch applied to code from last summer. Went back to using that old build for now.

[CrystalP]

The results seem mixed regarding this patch. Maybe there actually are multiple problems and the patch only takes care of one.
For those who saw improvements, is that still the case after using it for a few days? Stuttering sometimes hides, only to reappear later...

And to clarify a few things:
- the missed counter "running away" has been here since Dharma, based on a few tests I did the other day. So no matter how annoying, unless there was a change in behavior lately, it's probably not relevant to the recent stuttering.
- the deinterlacing in master is not finished yet so yes you may have dropped frames, stuttering, different behavior than isidrogar's patch, ...

The problem to track down is that dxva decoding was working fine until recently and it seems to have been broken. That was before deinterlacing support but will most likely help playback with deinterlacing.

@Voyager-xbmc, I was hoping you could build and try older versions to get a better idea of when the issue started for your hardware. If this patch doesn't do the job ofc. I highlighted in this related thread a few important dates for dxva: http://forum.xbmc.org/showthread.php?tid=108546&page=7

edit: @WhiningKhan, the need for NV12 is at a different stage of the processing, there is no conflict here. I'd like to see a debug log with the patch of this thread and the HD3200, with some dxva decoding please.

[isidrogar]

Hi All,

In order to have full information to solve this problem, all those reporting can always include:

- Build version: Specify commit ID from GIT if possible, or indicate if you are using a custom build, patches applied (if any).
- Content playing: resolution (SD, HD), codec (H264, MPEG), container (MKV, TS), etc.
- Artifact seen: constant increasing dropped frames counter, random stuttering, etc.
- Renderer used: DXVA, Pixel Shader or Software.
- Deinterlacing method: Auto, None, DXVA Best, Temporal, etc.
- Decoder used: Software ffmpeg or DXVA.
- Video card manufacturer and model.
- CPU speed.

Please also specify if you have other build version which is working fine using same settings.

Thanks!

[Hack_kid]

CryshztalP why not just revert the change from aug24 you did you said boxee had the code that way but didn't know why, at this point lets just get it working again then play

[WhiningKhan]

And to clarify a few things:
- the missed counter "running away" has been here since Dharma, based on a few tests I did the other day. So no matter how annoying, unless there was a change in behavior lately, it's probably not relevant to the recent stuttering.
- the deinterlacing in master is not finished yet so yes you may have dropped frames, stuttering, different behavior than isidrogar's patch, ...

Well yes, my note about the runaway missed counter was inaccurate - in cases where insufficient shader capacity seemed to be the problem, the sync flickers constantly between 0 and +100%, in addition to missed counter increasing. I assumed myself this is probably not the issue you are after in this thread, but wanted to mention this because overclocking the IGP really does fix this issue on my system - I figured it could be useful info for some people seeing the same behavior.

edit: @WhiningKhan, the need for NV12 is at a different stage of the processing, there is no conflict here. I'd like to see a debug log with the patch of this thread and the HD3200, with some dxva decoding please.

Okay, here's one with the patch applied:
http://pastebin.com/4QdZc0xC

Here are the same files played with the normal trunk revision from two days back:
http://pastebin.com/mdvLZw9v

Didn't have much time for this, played two movies for a few minutes. In Hangover, I witnessed brief stutter and one green flash at least in the patched version; couldn't keep my eyes fixed on the screen for others all the time... Maybe more testing later.

[Voyager]

I did some initial testing of the YUV rt patch, and first impression is positive. When forcing DXVA Bob/Best, thereby doubling the fps rate, I get much less hiccups on the ION than without the patch. I'll need more time to provide some exact data.

[CrystalP]

@WhiningKhan, @Voyager-xbmc, seems you both applied the patch on top of current master, correct?
It's fine, I just want to make sure I understand correctly.

In unrelated news I found why the missed counter goes up but it's not easy to fix, and the rest of the code already compensates for it. So it's not a high priority.

[a11599]

This issue is weird. I also experience it, but mostly with DXVA deinterlacing only (it still happens with yadif or other high framerate (50p/60p) content, but it is very hard to pinpoint on my PCs). With vector adaptive deinterlacing it happens quite frequently and it is very visible on my 80-shader Mobility Radeon 5145 / Core i3 also with SD video.

What I see is that the video freezes for a few 100 msecs and then catches up. It is similar to the issue with subtitles on lower end machines. Like if the video thread were blocked. Anybody with a profiler could perhaps check it? (I only have free VC++.)

According to this thread: http://forum.xbmc.org/showthread.php?tid=108546&page=8, the problem started between 07.27 and 08.03. The biggest change regarding DXVA was the DXVA renderer PR, but it seems not to be the cause of the issue.

I have also seen the issue when I made the DXVA deinterlace patch, but had no idea what is causing it and then I didn't had time to investigate further. My local deinterlace branch is based on a8d22f68c77e3a68f7d530df9a807282d78cb0ef (2011-08-01 23:56:57) and since it also exhibits the problem, it should be something in between... There were lots of changes during this time period, if I see correctly some are dealing with threading implementation and others with the renderer (time smoother).

I will keep experimenting, but if somebody could profile it that would be great.