LAV Filters - DirectShow Media Splitter and Decoders

[pirlouy]

(If I'm not wrong) Notice that you can't use QuickSync if you use a ATI or NVidia card. I don't know if it's an Intel imitation, or if it's Egur's one.

[DragonQ]

You can if you use Lucid Virtu with a Z68 motherboard.

[pirlouy]

Lucid Virtu is for crazy people who search problems, isn't it ? :D

[CruNcher]

you don't even need Virtu, Virtu is just the interface to alot new features in NT6 ;)

Test version time!

http://files.1f0.de/lavf/LAVFilters-0.43-52-g1ec5766.zip

This is the latest build, and amongst mostly fixes and tweaking, it includes egur's QuickSync decoder for Intel GPUs.
The QuickSync decoder still has some glitches and rough edges, but its mostly ready for some action, so get nuts on it.

It does not yet support hardware deinterlacing, its a pure decoder so far - but because the Intel hardware is so fast, its still worth it.
In contrast to the CUVID decoder which probably even costs extra power, this one has the potential to actually save a bit. ;)

I only tested it with Sandy Bridge, but it supposedly also works with previous Intel GPU series, as long as they have SSE 4.1 - which means Penryn or newer. It will however only be really fast on QuickSync enabled CPUs - Sandy Bridge and newer.

PS:
Also supported now is YADIF in conjunction with hardware decoding, so that the Intel decoding isn't left interlaced. :p

PPS:
The code on google code isn't up2date right now because pushing results in some odd error... my personal git host is updated, however.

Perfect :) especially having Yadif right there for EVR Custom though makes i guess more sense to use MadVRs Render Deinterlacer (DXVA2) with it then going to the CPU (For the Realtime Playback Case) but the more options the better :)

Btw Im not sure what happened but Arcsofts DXVA Decoder works better for VC-1 AP and MP (WMV3) now with Lav Splitter (it doesn't goes out of sync anymore, though it glitches heavily @ the start for the film test (WMV-HD Encode) until it locks the framerate right (over 10sec) and so gets the sync right, happens visibly and hearably faster with Arcsofts own Splitter only 3 seconds or so) and EVR Custom :) it's not as good as with Arcsofts own splitter (less glitches generally for a lot of files that even with the WMV-ASF reader show heavy peak issues in the middle of the stream somewhere and cause double paint time issues for a small time) but Arcsofts Splitter cant parse as much files from the test database even crashes MPC-HC experimental on the Microsoft Build 2011 Official WMV Keynote Download (which is strange, tough largest non AP test with 5.6 GB)

[STaRGaZeR]

You need to have your monitor connected to the IGP in order to use it right? Or is just having it activated enough?

[pirlouy]

With my Radeon plugged, I have no image on motherboard connectors. Maybe it's different with NVidia (remember Nevcairiel mentioning something about it).
But I remember having read on Anandetch that IGP should be disabled as soon as a third party GPU is detected...

[STaRGaZeR]

No, the IGP can be active and functional at the same time, check your BIOS. You have to be on W7 thou. But I'm afraid that in order to use this the IGP has to drive the monitors. I can run Virtu, but it doesn't work properly. Plus dual link DVI is just a no-no with Intel, so I'm out.

[RBG]

(If I'm not wrong) Notice that you can't use QuickSync if you use a ATI or NVidia card. I don't know if it's an Intel imitation, or if it's Egur's one.

In fact you can. You have to create virtual VGA display connected to Intel HD graphics, then extend your primary desktop there, and after that you can use QS decoder even if you have Nvidia/Ati card as a main device.

http://forum.doom9.org/showthread.php?p=1532786#post1532786 - full instruction here.

[asasadad_1]

here (http://uploading.com/files/3dcc8ba9/rmvb.mkv/) is a mkv file contain realvideo&cooker, LAV splitter + LAV decoder(A+V), audio stutter;AV splitter +LAV decoder(A+V) ok.

[Superb]

I've started recommending using LAV Filters in my codec installation instructions on a Hebrew site.
Few things people reported:
a) When enabling CUVID on a system w/o a NVIDIA card, opening a (H.264/mkv) file seem to be "much" slower. Probably it figuring out CUDA isn't available, right? (is this supposed to take a while)
Proposed solution: make the installer check for the availability of CUDA (and now QS as well) and choose it automatically on such systems.

b) People reported a much lower volume levels (than ffdshow w/ mixer enabled) when playing 5.1 audio streams (ac3 or dts) on a 2.0 hardware setup.
Probable cause: windows mixer? sound drivers?
Workaround I give such people: enabling (http://i.imgur.com/Bv2rz.png) system-wide volume normalization.
I'm really not sure about a proper/proposed solution here. I mean, how will a built-in mixer in LAV Audio differ from the Windows mixer...?

[clsid]

I am not a fan of the Loudness Equalization feature of Win7. I prefer controlling volume just for video playback. Everything else is already loud enough.

A mixer is already on Nev's ToDo list. Hopefully that will include an option to apply a gain to the center channel when downmixing. That is one way to solve the volume issue. Another would be a simple normalization option like in ffdshow.

[Superb]

I'm not a fan of it either, but it's a temporary workaround.
Does ffdshow apply a gain to the center channel automatically (when using its mixer w/ the 2.0 preset)? That would explain why people say they had a higher volume w/ it (than the Windows built-in mixer).
Note: they didn't turn on normalization in ffdshow. The higher volume was "out-of-the-box" after choosing 2.0 channel mixer in the installer.

EDIT:
Until a custom built-in mixer will be available in LAV Audio... Maybe it would be possible to control the Windows mixer a bit:
http://msdn.microsoft.com/en-us/library/ms678838.aspx (the SetLevel method inherited from IPerChannelDbLevel)

[Midzuki]

Another possible "temporary workaround": AC3Filter :)

( I'm assuming it works properly on Windows 7, on my Vista-64 notebook it works very well :cool: )

[Superb]

Another possible "temporary workaround": AC3Filter :)

( I'm assuming it works properly on Windows 7, on my Vista-64 notebook it works very well :cool: )It does work. And so does ffdshow. But I rather have the least number of filters in my chain, don't you?

[dead_screem]

Its a bug in the driver, seems to have appeared a while ago. You could try different drivers, otherwise there isn't anything that can be done.

258.96 is the last ver to work right. And I think its actually hardware deinterlacing related. cause if i set CUVID to weave with newer drivers the blocking goes away.

[nevcairiel]

258.96 is the last ver to work right. And I think its actually hardware deinterlacing related. cause if i set CUVID to weave with newer drivers the blocking goes away.

It still is somehow related to partial acceleration, because it doesn't happen on cards with full acceleration capabilities. It might just be coincidence and depend on the series of the card, but somewhere is another factor in there. :)

[nevcairiel]

(If I'm not wrong) Notice that you can't use QuickSync if you use a ATI or NVidia card. I don't know if it's an Intel imitation, or if it's Egur's one.

Like others already said, you need the GPU to be active.
There are several ways to fake that, the most useful being the "fake" VGA monitor which you can just trick Windows into using.

I think it was also linked before, but so everyone sees it, here its again:

http://forum.doom9.org/showthread.php?p=1532786#post1532786

Another general limitation is that the decoder only works on Vista/7, but thats a limitation in Intels drivers and i doubt it'll ever change. So, no XP love for you!

[STaRGaZeR]

Eh, the fake screen trick seems to work. Will report my findings later. For now, it looks very promising: same power consumption as DXVA :D

[magilvia]

Hi folks, I'd like to report a possible bug with lav splitter and the lord of the ring italian blu-ray extended edition. All my others blu-ray plays flawlessly but with these ones the audio tracks are not recognized correctly. LAV recognizes 6 tracks in the stream, see the image attached: the first (english eac3 mono) according to the package specs, should be DTS-HD 6.1 and crashes MPC if selected. The second (italian ac3 stereo) should be DTS-HD 6.1 too but outputs no audio if selected. The others are commentary and work good.
Power dvd recognizes 6 audio tracks too, english,italian,4x english and works good with each one.
Media info report attached (doesn't make much sense to me too)

Thank you very much for your hard work.

[betaking]

to:nevcairiel i Compiling last lav but i can not see IntelQuickSyncDecoder.dll ?
i not install intel media sdk !but comp ffdshow i can Comp IntelQuickSyncDecoder.dll! do i need install last intel media sdk beta?

[cyberbeing]

If I load LAVSplitter.ax in MPC-HC as an external filter (using Browse) without it actually being registered, I've occasionally been getting crashes in MPC-HC citing avformat-lav-53.dll when opening a new video. No idea if this is a MPC-HC bug or a LAV bug, but even though this is non-standard usage, I'd be nice if it worked reliably.

[nevcairiel]

That mode is somewhat unstable in MPC-HC for filters that rely on external dlls, i tried to fix it some time ago, but it never worked 100%. Its a MPC-HC bug. Register the filters already. :)

[nevcairiel]

Hi folks, I'd like to report a possible bug with lav splitter and the lord of the ring italian blu-ray extended edition. All my others blu-ray plays flawlessly but with these ones the audio tracks are not recognized correctly. LAV recognizes 6 tracks in the stream, see the image attached: the first (english eac3 mono) according to the package specs, should be DTS-HD 6.1 and crashes MPC if selected. The second (italian ac3 stereo) should be DTS-HD 6.1 too but outputs no audio if selected. The others are commentary and work good.
Power dvd recognizes 6 audio tracks too, english,italian,4x english and works good with each one.
Media info report attached (doesn't make much sense to me too)

Thank you very much for your hard work.

I would need a short sample from the m2ts file on the disc (maybe 20-30mb), otherwise there isn't anything i can do.
Use a tool like DGSplit to cut the first few MB off, and then upload it to mediafire or multiupload.

[cyberbeing]

Register the filters already. :)
Does LAV Filters backup and revert the Media Type registry settings to their previous values on unreg, uninstall, and when unselecting a format?

[Qaq]

AFAIK, DTS-HD, DTS-ES 6.1/6.0 requires ArcSoft DTS Decoder v.1.1.0.0. 7 & 8 have troubles with 6.1.

[nevcairiel]

Does LAV Filters backup and revert the Media Type registry settings to their previous values on unreg, uninstall, or when unselecting a format?

It reverts them to defaults on uninstall, not the previous value. This only applies to the installer, unregistering does no magic.
Unselecting through the filter properties does nothing to media types, because its not running with elevated access.

I've been pondering on removing all sorts of media type handling from the DLL and let only the installer do it, that way it would seem more consistent (and only registering the filter would not do "harm", because it wouldn't touch anything)

AFAIK, DTS-HD, DTS-ES 6.1/6.0 requires ArcSoft DTS Decoder v.1.1.0.0. 7 & 8 have troubles with 6.1.

.8 at least works perfectly fine with 6.1, only 6.0 is broken (by broken i mean you only get 5.0 instead of 6.0). I've never seen anything use a 6.0 track, though.
Also, that only applies to decoding of course, the issue reported above seems to be an issue with identifying the streams in LAV Splitter, long before decoding happens. :)

[nevcairiel]

Proposed solution: make the installer check for the availability of CUDA (and now QS as well) and choose it automatically on such systems.


Somehow i expect users to at least know what kind of video card they have. Hey, even if not, when selecting the option in the GUI it checks for availability and tells the user - if they leave the option on anyway, not my fault. :p
I can probably make the software fallback process a bit faster, though.

One of these days i will need to make the installer smarter, however duplicating the compat checks in the installer is somewhat annoying.
If the auto-select process is fast enough i might add an "Auto" option that probes and selects the appropriate HW decoder.

to:nevcairiel i Compiling last lav but i can not see IntelQuickSyncDecoder.dll ?
I do not see a point in duplicating the code in my repository. If you want to build it, get the code from the official repository at http://sourceforge.net/projects/qsdecoder/

[cyberbeing]

It reverts them to defaults on uninstall, not the previous value. This only applies to the installer, unregistering does no magic.
Unselecting through the filter properties does nothing to media types, because its not running with elevated access.

I've been pondering on removing all sorts of media type handling from the DLL and let only the installer do it, that way it would seem more consistent (and only registering the filter would not do "harm", because it wouldn't touch anything)

If you removed the media type handling from the DLL, that may be helpful. Isn't it possible to add optional install switches to be used with regsvr32? That may let you have it both ways.

My main concern with installing splitters which support a lot of formats like LAV, is their tendency to break the Media Type registry settings for most other installed splitters/source-filters, including windows default splitters/source-filters which use quartz.dll and are unable to be easily re-registered, and even when you don't want to use the splitter you just installed for those formats.

Currently I'm forced to backup my registry key each time I install/test a new splitter so my system doesn't get left in a completely broken state if I choose the disable/uninstall at some point. This gets rather annoying, which is why the MPC-HC feature to use external filters without installing/registering is handy, it it was working reliably.

[betaking]

Somehow i expect users to at least know what kind of video card they have. Hey, even if not, when selecting the option in the GUI it checks for availability and tells the user - if they leave the option on anyway, not my fault. :p
I can probably make the software fallback process a bit faster, though.

One of these days i will need to make the installer smarter, however duplicating the compat checks in the installer is somewhat annoying.
If the auto-select process is fast enough i might add an "Auto" option that probes and selects the appropriate HW decoder.


I do not see a point in duplicating the code in my repository. If you want to build it, get the code from the official repository at http://sourceforge.net/projects/qsdecoder/

OK!big thanks!

[Budtz]

I am using lav audio and also reclock. If I play a dolby digital 5.1 file or a dts 5.1 file, I get dolby digital 5.1 wich is fine. If I play dts-es file however, I dont get digital surround but just dolby pro logic. Is there anything I can do to make my setup work with dts-es or 6.1 surround files?

[Superb]

Somehow i expect users to at least know what kind of video card they have.You would be surprised. People don't even know if they have a x64 system or not. (or "what is x64?")

If the auto-select process is fast enough i might add an "Auto" option that probes and selects the appropriate HW decoder.
That would be the ideal solution IMO.

Thanks for the answer.

[fastplayer]

MPC-HC freezes with this video (http://video.ch9.ms/ch9/08a6/26175b1f-8fb7-4d85-b1ce-9fb9011108a6/RefreshandresetyourPC_high_ch9.mp4) [36MB, MP4]. LAV Video (from build g1ec5766) together with madVR seem to be the cause as I couldn't reproduce the issue when using different combinations of MPC's internal filters, EVR-CP and LAV Filters. Disabling Multi-threading in LAV Video helps a bit, allowing to at least properly close the player.

If I use the release version of LAV 0.43, the video plays just fine with madVR as the renderer.

Setup: MPC-HC rev3936, LAV Filters (QS test build: g1ec5766), madVR 0.80

Edit: Builds 76ad254, 7ac7d48, 6fca483 and 8cc6a5a are affected as well. Build 49c4c15 is the last known working one that I could find.

[magilvia]

Hi Nevcairiel, I splitted the stream in chunks of 25M, the first chunk is processed with haali media source which exposes just the 4 commentary tracks. The successive chunks are instead processed by lav splitter and all 6 tracks are correctly recognised, two as DTS-HD 7 channels, and work good. There must be something wrong at the very beginning of the stream. BTW it's just "the two towers", disk 1 and 2 which presents this problem.
Here are the first chunks:
1st chunk 25M (http://www.megaupload.com/?d=BD5C3F8Q)
2nd chunk 25M (http://www.megaupload.com/?d=1MJ8MNC4)

[STaRGaZeR]

Here are some results from my limited QS tests:

- DXVA-like power consumption.
- Seeking creates a bit of corruption around moving stuff in the first frame after the seek. It doesn't happen with all H.264 files, x264 encodes are fine, but some BDs suffer it like this one (http://img269.imageshack.us/img269/3601/88596576.png).
- Seeking does something funny. If you seek to, for example, 00:20, audio starts playing from 00:20 as it should. But video most of the time starts from 00:15 or so. It plays normally, out of sync, from 00:15 until it decides to jump to the proper frame. This jump can be a simple skip, which is perceived as a 2-3 second jump, or it plays those 2-3 seconds very fast to catch the audio. Here's a graph that shows it (http://img849.imageshack.us/img849/9899/71750009.png). The spikes are the seek, and a bit later the fast movement to the proper position.
- It seems 16 ref frames are not properly supported. Sometimes you can get some seconds of proper playback, but it hangs the player shortly afterwards. Or it hangs the player without showing a single frame.
- Yadif double framerate mode results in 60fps for a 30i source as it should, but those frames are delivered at 30fps, which result in half video speed while the audio keeps going at normal speed.

[DragonQ]

I tried the QuickSync version but unfortunately it doesn't work with a pre-Sandy Bridge i5 (Arrandale). It just says "not supported" and even if I do select it, CPU usage is unaffected.

[bjd]

.8 at least works perfectly fine with 6.1, only 6.0 is broken (by broken i mean you only get 5.0 instead of 6.0).

Works here too providing the 'expand 6.1->7.1 flag' is ticked

[nevcairiel]

- Seeking creates a bit of corruption around moving stuff in the first frame after the seek. It doesn't happen with all H.264 files, x264 encodes are fine, but some BDs suffer it like this one (http://img269.imageshack.us/img269/3601/88596576.png).

It doesn't have a flag yet to tell me about corrupted frames, but i hope that can be added. Otherwise i'll try to hack something up.


- Seeking does something funny. If you seek to, for example, 00:20, audio starts playing from 00:20 as it should. But video most of the time starts from 00:15 or so. It plays normally, out of sync, from 00:15 until it decides to jump to the proper frame. This jump can be a simple skip, which is perceived as a 2-3 second jump, or it plays those 2-3 seconds very fast to catch the audio. Here's a graph that shows it (http://img849.imageshack.us/img849/9899/71750009.png). The spikes are the seek, and a bit later the fast movement to the proper position.

Managed to reproduce it as well as figure out why it happens, not sure how to fix yet.
Stupid Intel Media SDK uses unsigned for timestamps, and negative timestamps don't make it all too happy.....



- Yadif double framerate mode results in 60fps for a 30i source as it should, but those frames are delivered at 30fps, which result in half video speed while the audio keeps going at normal speed.

That seems odd, even impossible, YADIF will always take the timestamps of frame 1 and frame 2, make a new frame at exactly 1.5, in the center of the two, which will result in perfect doubling of the fps.
The only thing that doesn't change right now is the reported frame rate, which will remain at 30 fps.
I'll try to test it, but i doubt it.

Edit:
Seems to work just beautifully here.

[mark0077]

Its a bug in the driver, seems to have appeared a while ago. You could try different drivers, otherwise there isn't anything that can be done.

ty, you're right. If anyones interested, heres some nvidia drivers I tried, showing vc1 decoding was broken somewhere between 280.36 and 285.27 on these cards at least. (I tried using the working cuda related dlls with the newer drivers but that doesn't seem to work very well :P )

Nvidia Driver - Cuda dll version - Status
285.62 - 8.17.12.8562 - broken
285.38 - 8.17.12.8538 - broken
285.27 - 8.17.12.8527 - broken
280.36 - 8.17.12.8036 - working
280.26 - 8.17.12.8026 - working

Nev, regarding cuvid decoding in lav video. I am trying to push the quality of avisynth / svp frame interpolation settings on my machine again and so am trying to make sure all load during playback is spread as evenly across my machine to make best use of it.

I notice that lav video only seems to use one of the two gpus on my machine though. Is this by design or is this something that could be implemented in a future release of lav video?

PS: I'm using gpu-z to check the load on each of the gpus during playback.

[nevcairiel]

I notice that lav video only seems to use one of the two gpus on my machine though. Is this by design or is this something that could be implemented in a future release of lav video?


It'll always use the first compatible GPU it finds, and it cannot use two at the same time to spread the load, thats just plain impossible.

[kalston]

Works here too providing the 'expand 6.1->7.1 flag' is ticked

Works for me too without even ticking the 6.1 to 7.1 flag, using the Arcsoft 1.1.0.7 and not 8 though.

[mark0077]

ah I understand, no problem.

Would it possible to allow the user to choose which gpu to use in such cases, even via registry setting, like gpu_to_use=0 or gpu_to_use=1? My goal is to use svp on one gpu and lav video decoder on the other, to ensure both arn't maxing out one gpu, leaving the other idle... If its a big job to make it configurable no worries at all. I'm just greedy :D

[nevcairiel]

Pure decoding only uses the hardware decoder, a part of the GPU that SVP doesn't even touch. Don't worry about it.
(And interlaced material already gets deinterlaced to 50/60 fps, no need for SVP then)

[mark0077]

I know but I'm thinking more about heat than performance. I can see temperatures on one gpu reaching 80's and the other is 20c lower. When the gpu gets to ~90 I think it is, it starts throttling down and frames start getting dropped obviously because of that. Do you think putting cuda on the other gpu would bring this down a few degrees... I'd love to try if you're ever interested in making it configurable.

[STaRGaZeR]

That seems odd, even impossible, YADIF will always take the timestamps of frame 1 and frame 2, make a new frame at exactly 1.5, in the center of the two, which will result in perfect doubling of the fps.
The only thing that doesn't change right now is the reported frame rate, which will remain at 30 fps.
I'll try to test it, but i doubt it.

Edit:
Seems to work just beautifully here.

Okay, just found the cause. As you say, LAV works fine. That crap only happens with ffdshow (YV12 input to the normal video decoder, not that raw filter stuff) between LAV and the renderer doing some postprocessing. However it doesn't happen in software mode, so there's something in there to look after. I'm using ffdshow r3978, just in case.

Good luck on the other issues :)

[Pat357]

It seems that Lav-video can not play dirac video.
The lav-splitter connects and lav-audio is also working (I hear the audio), but the screen stays black.
In the filter list I see only Lav-spitter and lav-audio : lav-video is missing.

This is on the video-out pin from Lav-splitter :
Filter : LAV Splitter - CLSID : {171252A0-8820-4AFE-9DF8-5C92B2D66B04}

- Enumerated media type 0:

Video: DRAC 640x360 (160:99) 24.00fps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: Unknown GUID Name {63617264-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo2 {F72A76A0-EB0A-11D0-ACE4-0000C0CC16BA}
bFixedSizeSamples: 0
bTemporalCompression: 1
lSampleSize: 1
cbFormat: 112

VIDEOINFOHEADER:
rcSource: (0,0)-(640,360)
rcTarget: (0,0)-(640,360)
dwBitRate: 0
dwBitErrorRate: 0
AvgTimePerFrame: 416667

VIDEOINFOHEADER2:
dwInterlaceFlags: 0x00000000
dwCopyProtectFlags: 0x00000000
dwPictAspectRatioX: 160
dwPictAspectRatioY: 99
dwControlFlags: 0x00000000
dwReserved2: 0x00000000

BITMAPINFOHEADER:
biSize: 40
biWidth: 640
biHeight: 360
biPlanes: 1
biBitCount: 12
biCompression: drac
biSizeImage: 345600
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 80 02 00 00 68 01 00 00 ........€...h...
0010: 00 00 00 00 00 00 00 00 80 02 00 00 68 01 00 00 ........€...h...
0020: 00 00 00 00 00 00 00 00 9b 5b 06 00 00 00 00 00 ........›[......
0030: 00 00 00 00 00 00 00 00 a0 00 00 00 63 00 00 00 ........*...c...
0040: 00 00 00 00 00 00 00 00 28 00 00 00 80 02 00 00 ........(...€...
0050: 68 01 00 00 01 00 0c 00 64 72 61 63 00 46 05 00 h.......drac.F..
0060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................


The files play fine in FFmpeg's Fplay and mplayer (uses the FFmpeg's codec too) :
Video: dirac (drac / 0x63617264)

Is it possible to at support for Dirac ? (is open format, sources are available and free to use)

[nevcairiel]

MPC-HC freezes with this video (http://video.ch9.ms/ch9/08a6/26175b1f-8fb7-4d85-b1ce-9fb9011108a6/RefreshandresetyourPC_high_ch9.mp4) [36MB, MP4]. LAV Video (from build g1ec5766) together with madVR seem to be the cause as I couldn't reproduce the issue when using different combinations of MPC's internal filters, EVR-CP and LAV Filters. Disabling Multi-threading in LAV Video helps a bit, allowing to at least properly close the player.

That was an odd bug, i can't explain why it didn't happen with 0.43, but anyhow, i fixed it.

[Boltron]

Lately I have been having problems seeking (by chapters usually) in many of my mkv movies. I finally decided to investigate and it seems to be caused by Lav 43. I went back to Lav 42 and it works fine. I have not had these kinds of seeking issues in a very long time. It doesn't seem to happen on all movies but I can tell you that it happens on Avatar EE and LotR The Two Towers both rips from Bluray into mkv created my MKVmerge.

I did try with MPC-HC 3898 and 3933 but it made no difference, only reverting to LAV 42 fixes the problem. I use the standard official releases of 42 and 43 btw and not any of the interim builds.

I use LAV, madVR and MPC-HC.

[Boltron]

Forgot to add, when I seek ahead by chapters and the seek problem occurs, my disk i/o reads jumps up to max until eventually the point is reached and playback resumes, it may take 20 seconds or longer.

[nevcairiel]

I have been over my library and all seek fine, no issues whatsoever. I even remuxed by Avatar EE Blu-ray to MKV and tested that, all fine!
In fact, the only changes to mkv seeking were to ignore any tracks that are currently inactive, because reading those is kind of a waste. :)

In any case, i reverted one change that also was in 0.43, although for this to cause any issues you would probably have to have a very underpowered system.
Anyway, try this, if you can:

http://files.1f0.de/lavf/LAVFilters-0.43-58-g2596e21.zip

[fastplayer]

That was an odd bug, i can't explain why it didn't happen with 0.43, but anyhow, i fixed it.
No issues with your g2596e21 build. :thanks: