LAV Filters - DirectShow Media Splitter and Decoders

[nevcairiel]

It doesn't occur with ffdshow, only with LAV.
So it may be more accurate to post this here, or not?


Its either an Intel driver bug or a bug in the QS decoder.
The QS decoder outputs 60 frames, but outputs the frames with timestamps for 30fps, and then occasionally there is a huge gap in the timestamps when it re-syncs to the actual time - which causes the drops.

Overall, i'm really not happy with how the QS DI performs so far, i'll probably disable it again soon until its working properly, or Intel finally decides to release a fixed driver.

[egur]

Its either an Intel driver bug or a bug in the QS decoder.
The QS decoder outputs 60 frames, but outputs the frames with timestamps for 30fps, and then occasionally there is a huge gap in the timestamps when it re-syncs to the actual time - which causes the drops.


I've committed r62 which fixes some (hopefully all) of stalls/crashes related to the DI.

As for the time stamps, per your request I don't calculate them, the new (extra) frames do not have time stamps. If you have an idea to improve this, let me know.

[jakmal]

At the outset, I am not sure whether this is the right thread to report this issue (where do I report avcodec decoding problems?), but here it is:

Sample file: Original MP4 version of this YouTube clip: www.youtube.com/watch?v=t56ooXC9VmY [ Size : ~252 MB ]

Playback system: Core i7-3770K + GT 640. Using LAV Filters 0.50.5 (from 1f0.de, downloaded couple of days back) and MPC-HC 1.6.2.4902 x86 version, display on to 1080p monitor with EVR-CP renderer.

With Video Decoder set to avcodec, there are blocking artifacts when the words 'Peter Jackson' swivels across the middle of the screen.

With Video Decoder set to cuvid or DXVA2 CB, playback is flawless.

[nussman]

I don't know what that option does, but since it works with it it should be a good starting point.
Then DVBViewer waits for more video format informations.
This slows down channel switching, but some decoders need this informations.

In Graphstudio you should see the differences.

P.S. For fileplayback its always on - so you have to test it with LiveTV!

[CiNcH]

I don't know what that option does, but since it works with it it should be a good starting point.
Check the connection format between DVBSource and the decoder with GraphStudio for example. In case the detection is disabled, the DVBSource will propagate some default values on the video pin and the decoder has to detect the correct values by parsing the bitstream.

I have no clue however, why it works with standard EVR, but not with the EVR custom presenter.

[meetajhu]

hi nevcairiel, will be adding support for Windows Media Audio 9.2?

[dansrfe]

@ nevcairiel

Is I/O and copy/transfer from HDD -> RAM -> CPU -> RAM -> VIDEO RAM streamed or buffered?

[nevcairiel]

Both. :)

You should ask more specific questions if you want specific answers. :p

[CruNcher]

@nev
Decodinerror.ts is now falling back from Quicksync to libavcodec what is the detection reason for the fallback or is it out of your hands ?

[nevcairiel]

@nev
Decodinerror.ts is now falling back from Quicksync to libavcodec what is the detection reason for the fallback, or is it out of your hands ?

Half my files are named "decodingerror.ts" and the other half is named "sample.ts", so either link to the file you are referring to, or don't expect any answers.

[CruNcher]

This one http://www.mediafire.com/?rld8gnlh52f03ud that intel wanted to look into fixing which works on the Nvidia DSP just fine, now the fix seems to be a fallback to Software Decoding which though doesn't fix the main problem which would be Microsofts DTV Decoder (Intels Driver 2761) there the issue is still apparent ;)

[nevcairiel]

The Media SDK just fails when decoding the SPS/PPS of that file. Maybe egur can look more into that, but it looks like they just blocked it - maybe temporarily until it can be fixed for realz.

[CruNcher]

Thx i see so they seem to block it entirely and fallback by that to libavcodec instead of their own Software Decoder which i think would produce the same problematic result in this case thus also most probably why they cant fix it for Microsofts Decoder they have no libavcodec fallback possibility from there only their Software Decoder which most probably is also affected then by this SPS/PPS Decoding issue so a very Generic parsing Bug that affects both of their Decoder currently :)

I see also nothing changed on the Lav and Videolan DXVA crash state of

720p.mpg
test-720p cuda.mkv

on the Intel Driver i still wonder how long it will take till someone exploits this.

[egur]

I've confirmed the regression for this clip in driver 2752 though 2778, using older MSDK DLLs returns the old behavior - decode header passes and decoding produces artifacts.
With the above MSDK DLLs, the H264 header decode fails, don't know why, and thus my compatibility check fails too.
I'll report this bug tomorrow.

[ipanema]

I'm just trying LAV for the first time today, and I'm getting occasional crashes when seeking. I'm using the Microsoft MPEG-2 Demultiplexer which works well for all of the decoder's I've tried in the past (Mainconcept, Elecard, ffdshow etc).

It's been ages since I've had to tinker with the Microsoft MPEG-2 Demultiplexer, but as far as I recall it passes samples with a maximum size of 8192 bytes, so frames may be split into several output samples with the last one being smaller than 8192. Only the first sample for a frame carries a time stamp, the continuation samples have no time stamp. I mention this in case it helps.

The decoder seems to start streaming OK, but when I try to seek I sometimes get a crash in avcodec-lav-54.dll, exception code c0000005, and it is always at offset 00218b4f.

This is just with the default (no hardware accel) settings of LAV.

[nevcairiel]

As far as i know, you can't just use the MS Demultiplexer, you need an application to properly configure it for usage, so i cannot really test it unless you tell me which application you're using specifically.
While you're at it, providing a sample file that causes such a crash will greatly help to analyse and fix the problem.

[wanezhiling]

@nev

For some WMP users (some of my friends), a tray is really useful when using lav splitter to play back mkv files which integrate several audio tracks.

I know you have no that plan, but don't close the door please.

[sneaker_ger]

ftp://ftp.heise.de/pub/ct/spezial/shuttle.mpg
Crashes LAV Splitter 0.50.5 stable.

[nevcairiel]

ftp://ftp.heise.de/pub/ct/spezial/shuttle.mpg
Crashes LAV Splitter 0.50.5 stable.

No crash on latest Git for me.

[sneaker_ger]

Yes, seems to be working. No more crash with this:
https://forum.doom9.org/showpost.php?p=1578054&postcount=11285

[ontherocks]

I have GeForce GT520.
mpc-hc.1.6.3.4992.x86
LAVFilters-0.50.5
Having problem bitstreaming DD+ tracks via HDMI. There is no audio.
Sample audio link below.
http://www.mediafire.com/?a3x0q52bg1av33m

[dansrfe]

@ nevcairiel and anyone else you can help me understand what exactly is going on

http://s13.postimage.org/u7erxxvwl/graphs.png

This is a screenshot of my CPU, System RAM, GPU, Video RAM, and I/O graphs along with madVR's OSD while playing a 1080p 10bit file and upscaling it to 2016 x 1134 in windowed mode (I couldn't take a screenshot in exclusive mode but in reality I upscale it to the full resolution of my monitor which is 2048 x 1152 @ 60Hz). Basically I'm dropping frames whether I upscale it to 1152p60 or play it at output 1080p24 on my other screen. It doesn't look like a decoding problem since the queue is full but I can't tell.

The only items in my filter chain are MPC-HC, madVR, LAV Filters. LAV has been set to CPU decoding. All of them are at the latest version. CPU: C2D @ 2.66Ghz. RAM: 4GB DDR2 800MHz.

I can't figure out why my GPU cannot upscale to 2048 x 1152 with anything other (higher quality) than bilinear. Currently I'm using SoftCubic 50 for chroma and Spline 3 taps for luma up/down scaling.

Thanks.

[kasper93]

@dansrfe
Reduce queue size in madVR options, because all your gpu memory is used and that may cause dropped frames.

[Pat357]

Nev,

I have a strange problem with lavfilters, probably related to the LAVsplitter.
I created a .m2v file from my .avs script using Hank's HC MPEG2 encoder.

The file has 165350 frames, but when I use LAVsplitter, I can not seek above frame 85907. The slider from MPC-HC is then completely to the right.

When I just change the splitter to MPC-HC's internal, I don't have this problem (ie I CAN seek to the end at frame 165350).

This is the info from LAVsplitter :

Filter : LAV Splitter Source - CLSID : {B98D13E7-55DB-4385-A33D-09FD1BA26338}

- Connected to:

CLSID: {EE30215D-164F-4A92-A4EB-9D4C13390F9F}
Filter: LAV Video Decoder
Pin: Input

- Connection media type:

Video: MPEG2 Video 720x576 (16:9) 25fps 9300kbps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_MPEG2_VIDEO {E06D8026-DB46-11CF-B4D1-00805F6CBBEA}
formattype: FORMAT_MPEG2_VIDEO {E06D80E3-DB46-11CF-B4D1-00805F6CBBEA}
bFixedSizeSamples: 0
bTemporalCompression: 1
lSampleSize: 1
cbFormat: 282

VIDEOINFOHEADER:
rcSource: (0,0)-(720,576)
rcTarget: (0,0)-(720,576)
dwBitRate: 9300000
dwBitErrorRate: 0
AvgTimePerFrame: 400000


I've a possible explanation for this : LAVsplitter calculates the length of the movie using the bitrate 9300.
This is the max. bit rate i had set (max. for DVD player compatibility), the video is 2 pass with an average from 4832 or something like that.
This would explain this behavior.

I guess MediaInfo and also the internal MPC-splitter somehow figure out the total nr. of frames and multiply this by the frame-time (file is CFR).

This is what MediaInfo says :

Algemeen
Volledige naam : K:\film\limitedsharpen_met_MC'd_FFT3dfilter_via_MVTools20.m2v
Formaat : MPEG Video
Formaatversie : Version 2
Bestandsgrootte : 3,72 GiB
Duur : 1h 50mn
Totale bitratemodus : Variabel
Totale bitrate : 4 832 Kbps

Video
Formaat : MPEG Video
Formaatversie : Version 2
Formaatprofiel : Main@Main
Formaatinstellingen, BVOP : Ja
Formaatinstellingen, Matrix : Aangepast
Duur : 1h 50mn
Bitrate-modus : Variabel
Bitrate : 4 832 Kbps
Maximale bitrate : 9 300 Kbps
Breedte : 720 pixels
Hoogte : 576 pixels
Beeldverhouding : 16:9
Framerate : 25,000 fps
Standaard : PAL
ColorSpace : YUV
ChromaSubsampling : 4:2:0
BitDepth/String : 8 bits
Scansoort : Progressief
Bits/(Pixel*Frame) : 0.466
Stroomgrootte : 3,72 GiB (100%)


Any idea how to fix this ?

PS : neither FFmpeg is able to determine the movie length : estimation based on bit rate (9300) : gives wrong duration.

[nevcairiel]

The file has 165350 frames, but when I use LAVsplitter, I can not seek above frame 85907. The slider from MPC-HC is then completely to the right

Can you create a small file that demonstrates the problem?
No sample, no fixings. :d

[ipanema]

As far as i know, you can't just use the MS Demultiplexer, you need an application to properly configure it for usage, so i cannot really test it unless you tell me which application you're using specifically.
While you're at it, providing a sample file that causes such a crash will greatly help to analyse and fix the problem.

The demux is part of a test environment here, so not possible to extract anything that would demonstrate the problem, without writing a new standalone program. So I guess we will have to park this for now.

Interestingly the example DS decode filter that is supplied with Intel Media SDK works fine, as do other decoders.

[nevcairiel]

Well if you cant give me anything to work with, i'm afraid i cannot help you.
For crashes, its crucial to be able to reproduce them, or fixing them is nearly impossible.

[ipanema]

Well if you cant give me anything to work with, i'm afraid i cannot help you.
For crashes, its crucial to be able to reproduce them, or fixing them is nearly impossible.

I understand and that's OK. We can carry on using other decoders as before. I wasn't sure whether the offset value alone would lead to the problem, so thought it worth mentioning.

[dansrfe]

@dansrfe
Reduce queue size in madVR options, because all your gpu memory is used and that may cause dropped frames.

Well that definitely worked. No frame drops or delays and even skipping back and forth is much snappier and smoother now. New queue sizes are pretty low but if that's what it takes then I guess I'll have to stick with it.

New queue sizes: 8/4/4/4

Thanks kasper93!

[Pat357]

Can you create a small file that demonstrates the problem?
No sample, no fixings. :d

Here's a small sample showing the problem :

http://www.mediafire.com/?uhpcg79jq7efzi7

[nevcairiel]

Here's a small sample showing the problem :

http://www.mediafire.com/?uhpcg79jq7efzi7

Which filter exactly you said produced the correct duration?
I can't seem to get the internal MPC-HC splitter to take over, and the fallback Microsoft splitter reports the same duration as LAV. (19 seconds for that file)

Sadly there is no functionality in place yet to actually measure the bitrate of a file and produce a duration based on that, but i suppose its a possiblity for the future.
I tried something quick and dirty, and it took reading until nearly half the file until the bitrate was accurate enough to calculate the duration properly (about 10MB in this case)

I would generally recommend to store it in an actual container format, instead of full raw video.

[Pat357]

Which filter exactly you said produced the correct duration?
I can't seem to get the internal MPC-HC splitter to take over, and the fallback Microsoft splitter reports the same duration as LAV. (19 seconds for that file)

Sadly there is no functionality in place yet to actually measure the bitrate of a file and produce a duration based on that, but i suppose its a possiblity for the future.
I tried something quick and dirty, and it took reading until nearly half the file until the bitrate was accurate enough to calculate the duration properly (about 10MB in this case)

I would generally recommend to store it in an actual container format, instead of full raw video.
What exactly should I store in the container ? Does an .MPG container (or .VOB) even have a field to store it ?
What about MKVmerge (MKV) or MP4Box(MP4) ? Where can I put what information ?

I must admit it was not the MPC-HC MPEG splitter that gave me correct durations : it's the MediaConcept MPEG demuxer that allows me to seek to the end...
Any idea how they do it ?

Also MediaInfo seems to be able to determine the movie length.
It displays both the average bit-rate and the max bitrate.

I gonna test with other mpeg-splitters too : cyberlink, arcsoft,.... and see what comes out.

[nevcairiel]

What exactly should I store in the container ? Does an .MPG container (or .VOB) even have a field to store it ?
What about MKVmerge (MKV) or MP4Box(MP4) ? Where can I put what information ?

Its simple MPEG2 video, you can store it in most containers available. MPEG-PS (.mpg/.vob) is one of the options, and it contains full timestamps which makes duration and seeking much easier. Of course you can also go with MKV or MP4 if you want.

The main problem with raw video (without a container around it) is that you can't know how many frames a file contains without counting them (or if you have CBR), so anything you try will be guesswork, accuracy depending on how much data/effort you put into it.

[ajp_anton]

For seeking in raw video, would it be possible to just go by filesize instead of actual duration?
Don't know if the media player can show the current location correctly, but at least when seeking, maybe you could just jump to x% of the filesize and return the next available frame.

[ney2x]

I just found out that there is a MPC-HC Lite here (http://xhmikosr.1f0.de/mpc-hc/lite/) , without Internal Filters. I hope someone will build a MPC-HC Lite with integrated LAV Filters only... xD

[DeadlyEmbrace]

I just found out that there is a MPC-HC Lite here (http://xhmikosr.1f0.de/mpc-hc/lite/) , without Internal Filters. I hope someone will build a MPC-HC Lite with integrated LAV Filters only... xD

That would be really cool, although I think I prefer it when the filters are external. Makes it a lot easier to update the separate pieces.

[kitame]

That would be really cool, although I think I prefer it when the filters are external. Makes it a lot easier to update the separate pieces.

how about having it work like an external plugin? where replacing some of the files updates it without issues.

i think its a good idea imho, with this the player can be fine-tuned to work perfectly with LAVs and the simplicity of updating it would be a plus.

[DeadlyEmbrace]

how about having it work like an external plugin? where replacing some of the files updates it without issues.

i think its a good idea imho, with this the player can be fine-tuned to work perfectly with LAVs and the simplicity of updating it would be a plus.


That would be awesome, but I fear it will require quite a bit of work. If I have some free time I might start looking into it, although I can't promise anything.

[kasper93]

how about having it work like an external plugin? where replacing some of the files updates it without issues.

How about you make patch for MPC-HC? I guess many people likes idea of integrating LAVF into MPC-HC, but till someone make a patch we will not have it. And it's wrong thread for that discussion.

[Keiyakusha]

Idea about integrating LAV and MPC-HC is around for a long time. AFAIR even before LAV was released to the public. Year? Or more already? While its a not bad idea, we still need someone to actually code it. Right now there is no developers that can do it.
Personally I like it how it is now. MPC-HC lite and separate LAV. Instead I still have hope that nev will find some time/motivation to add avisynth support to LAV, which uses avisynth api to load scripts. I still think this is the best middle-ground between those who wants filters in LAV, and those who against it.

[SamuriHL]

Aren't the settings stored either in the registry or ini depending on what the user chooses? So, couldn't someone in theory take the filterless version, set it up with lav as a sub-directory under the mpc-hc install, configure it to use those external filters, and package that? Granted you couldn't save the lav configuration as that's stored elsewhere, but, it's a hack around for now. Ideally they'd use the lav API to configure them on the fly but that requires coding.

[hoborg]

Question about MP3 in *.ISO file playback.

Hi.
I was thinking about a way how to store a mp3 album into one file and still be able to play it by using DS player.
Firstly, i was trying RARsource filter, but after open the RAR file, it will ask what file to play instead to play first one.

So i put them to iso file and by accident i noticed, that such ISO file can be played in Graphstudio - what a surpise for me. But it will play as one 40min file.

My question - is it possible to somehow support jump to next/prev. MP3 track ?

[Keiyakusha]

You can put all mp3 in mkv as separate tracks. You will be able to chose track then and it will be one file. Not sure about navigating... Or you can make one single mp3, again put it in mkv or mp4 but this time add chapters. This will work as you want. mkvtoolnix even supports some cues. Don't know what it means but maybe it even can make chapters from .cue file? Who knows...

[hoborg]

You can put all mp3 in mkv as separate tracks. You will be able to chose track then band it will be one file. Ot you can make one single mp3, again put it in mkv or mp4 but this time add chapters. This will work as you want.

Well, i already tryed that. But this will "destroy" original files, and that is not what i like to do.

[Keiyakusha]

Well, i already tryed that. But this will "destroy" original files, and that is not what i like to.

Not sure if I understand how it destroys anything... you will be able to extract mp3 back if needed.

[hoborg]

Not sure if I understand how it destroys anything... you will be able to extract mp3 back if needed.

I mean tags in MP3 will be lost, or not?

[nevcairiel]

In case anyone is interested, i replaced the GCC compiler used for LAV by another build.
It still uses the same basic MSYS environment as assembled by XhmikosR, but a different build of the GCC compiler. The main difference is that i am now using the same edition of the compiler for both 32-bit and 64-bit, the old environment used the legacy mingw setup for 32-bit and the new mingw-w64 setup for 64-bit. Now both builds use the same mingw-w64 runtime, which is more up2date and much better maintained these days.

Additionally, the compiler has been upgraded to GCC 4.7.1 stable.

You can get the compiler i use from now on here:
http://files.1f0.de/mingw/

To use it, i would suggest to download XhmikosR's MSYS bundle from http://xhmikosr.1f0.de/tools/, remove/rename the "mingw" directory included in that bundle, and replace it with the one from my .zip. Maybe i'll provide a complete bundle as well some day...
Of course using his bundle without any change should continue working (mostly) fine.

I do not know if the new GCC setup will work building MPC-HC or ffdshow, because i don't do that, so feel free to try. ;)

[SamuriHL]

Woa man! Do we need the static libs in therre, too, or just the compiler? This will be fun.

[nevcairiel]

Woa man! Do we need the static libs in therre, too, or just the compiler? This will be fun.

My zip contains GCC as well as static libs for zlib and the bz2 lib, which should be everything thats needed for building. Everything else is included in LAVs repository.

[SamuriHL]

I mean the libspeex stuff but I remember that from May. I don't think we need that for this. So I'll give it a try and see what happens. :D