LAV Filters - DirectShow Media Splitter and Decoders

[nevcairiel]

I couldn't get the video portion to play (it would remain on a fixed position as if paused) at all with LAV, but, interestingly works fine with Haali.


Wasn't your problem with having no audio just two hours ago...?
You should decide what your problem is. :p

There is a known issue with WMP being "stuck" like that when there is subtitles involved. I'm working on it.

[Della]

Wasn't your problem with having no audio just two hours ago...?
You should decide what your problem is. :p

Also, i cannot reproduce a single issue here. :) I'm looking at WMP playing a MKV with all 3 LAV components just now.
But then, if you use WMP, i leave to your own fate, unless someone actually gives me a problem i can reproduce.

Actually I should have been more clear. The audio not working on the first machine led me to trying it on a second computer, used solely as a HTPC, with a previous version of LAV installed. On this second machine, I ran into both the audio and video issues.

I believe the problem with the audio is limited to 2 channel AAC as everything else seemed OK. I'd be happy to provide the sample, I can assure you there is a problem in the lav splitter or audio.

Thanks for taking the time to respond.

I see you edited the post, yes there are sub titles involved.

Thank you.

Della

[nevcairiel]

The video stalling issue should be fixed for the next version.
If a specific file doesn't work, a sample would be good.

[CruNcher]

Writing library : mkv2rls x264-tv version built on 2011. febr. 17

isn't there something known about wrong sampling frequencies ;)

Della try to remux it with mkvmerge, does it fix the issue ?

[no1d]

Looks like i can reproduce the crash, i'll see what i can do.

Edit:
And should be fixed

Thanks.
Lav splitter works well with Lagarith decoder.
LAV Video seems to be very slow (~12fps on that sample)
Lav video + internal splitter = black video.

[ryrynz]

Nev, my HTPC was not splitting M4V videos (as well as not setting the splitter with git builds) I had to set MPC's internal MP4/MOV splitter or install Haali's splitter to get it to work.

I've fixed both of those issues by using a neat program called CodecTweakTool. Even after uninstalling Haali I still had references to it and there were numerous Cyberlink references also which were cleaned out.

I can play with nightlies now! I'm a happy chappy.

[Della]

Writing library : mkv2rls x264-tv version built on 2011. febr. 17

isn't there something known about wrong sampling frequencies ;)

Della try to remux it with mkvmerge, does it fix the issue ?

Will do, and will report back.
Thank you.

Della

[boyumeow]

what issues are you guys referring to? because I tried 295 on my 9800GTX and VMR9 becomes unstable after awhile and crashes. I reverted to 257 release drivers (which is also the last version that CUVID works right with interlaced VC-1 (havn't tested MPEG-2 or H.264)
Was this one of the known issues with 295 and 290.53?

I will assume 295.51 is having memory leak if I'm not wrong (eats up my ram). 290.53 works better in my machine than 290.36 (screen refresh after screen saving mode). Thanks.

[nevcairiel]

LAV Video seems to be very slow (~12fps on that sample)


I never claimed it was fast, but its Intra-only, so multi-threading it would be quite easy (and it should scale nearly linear)

[Aspeh]

Nev, my HTPC was not splitting M4V videos (as well as not setting the splitter with git builds) I had to set MPC's internal MP4/MOV splitter or install Haali's splitter to get it to work.

I've fixed both of those issues by using a neat program called CodecTweakTool. Even after uninstalling Haali I still had references to it and there were numerous Cyberlink references also which were cleaned out.

I can play with nightlies now! I'm a happy chappy.

Do you know if this tweak tool has been fully checked to be clean of adware and such and do they have an official webpage for it. ?

Thanks

[dlmh]

Hey nev,

Is there any PQ benefits from using the DXVA2 (copy back) instead of just the software decoder on an AMD GPU (HD 6870)? I'm assuming the onboard post-processors will kick in as soon it's hardware accelerated? It's still a bummer we can't user MadVR on XBMC ;)

[nevcairiel]

Is there any PQ benefits from using the DXVA2 (copy back) instead of just the software decoder on an AMD GPU (HD 6870)?

In general, quality is always the same no matter which decoder you use. The post-processor should not be used in this case.

In the end, you're the one with the hardware at hand, so just set the post-processing to maximum and check if you see a difference.

[ryrynz]

Do you know if this tweak tool has been fully checked to be clean of adware and such and do they have an official webpage for it. ?

Thanks

http://www.free-codecs.com/download/codec_tweak_tool.htm

No adware, very clean looking program from a reputable site.

[Aspeh]

Wonder if LAV filters will work with the new raspberry computer. :p

[nevcairiel]

Wonder if LAV filters will work with the new raspberry computer. :p

They are not x86 based, so obviously not.
ARM support might be of interest some day if Win8 is out and in widespread use on ARM (and i have such a device), but then again, i doubt those things can run Win8. ;)

[aufkrawall]

I found some strange behavior/maybe bug:
Somehow brightness of a x264 I444 10bit video doesn't look fully right without converting to RGB when decoding while playing.

This is how the original FRAPS RGB looks like:
http://www.ld-host.de/uploads/thumbnails/e494e8bd47438960d33d94b080e5ad46.png (http://www.ld-host.de/show/e494e8bd47438960d33d94b080e5ad46.png)

x264 I444 10bit with default LAV decoding settings (CUVID) (a very little too dark):
http://www.ld-host.de/uploads/thumbnails/cea7ebe79549ed195ece54bc87a69e2f.png (http://www.ld-host.de/show/cea7ebe79549ed195ece54bc87a69e2f.png)

The same video with forced RGB32 output (to me it looks exactly like the original FRAPS RGB video):
http://www.ld-host.de/uploads/thumbnails/a7ae5ad2f56418b1b46bd017bee44509.png (http://www.ld-host.de/show/a7ae5ad2f56418b1b46bd017bee44509.png)

Download a test sample (it's not crf 0 like the pictures but this shouldn't make any difference):
http://www.mediafire.com/?n5ymnpdasada4gg
Video has TV range (Rec709).

It would be nice if it looked right without having to force RGB32 output. :)
I don't know what's about 8bit or I420, didn't test it.
Tested with latest MPC HC, LAV nightly and EVR custom.

Btw: Memory consumption with huge x264 RGB file and LAV splitter is now reduced by a half (~1.2 GB), but that's still twice as much than with MPC's internal splitter.

[nevcairiel]

Somehow brightness of a x264 I444 10bit video doesn't look fully right without converting to RGB when decoding while playing.

I take the YUV 4:4:4 data and deliver it to the renderer (as untouched as possible), its most likely the renderer screws up something during its own RGB conversion.
I don't have any control over what the renderer does with it. Especially 4:4:4 data is usually very badly supported by "standard" renderers like EVR, so RGB is the recommended output format.

It might be a result of the dithering, afterall you're converting a 10-bit file to 8-bit, which will loose data, and then convert it to RGB in the renderer, which again is a lossy process.
The best course of action is to convert in LAV Video to RGB (which will retain the 10-bit all through the processing path up to the end), or use a renderer which understands 10-bit input (only madVR a this time)


Btw: Memory consumption with huge x264 RGB file and LAV splitter is now reduced by a half (~1.2 GB), but that's still twice as much than with MPC's internal splitter.

Especially with 10-bit 4:4:4 material, the decoder itself also needs quite a bit of memory (especially with multi-threading)
The splitter limits itself to 512MB now, i guess i could cut that down a bit still, however 1.2GB really seems to be well within acceptable limits for playback of such formats.

The queues that cache all those frames in memory serve a real purpose, so lowering them isn't always a good idea.

[aufkrawall]

It might be a result of the dithering, afterall you're converting a 10-bit file to 8-bit, which will loose data, and then convert it to RGB in the renderer, which again is a lossy process.
The best course of action is to convert in LAV Video to RGB (which will retain the 10-bit all through the processing path up to the end), or use a renderer which understands 10-bit input (only madVR a this time)

Is there any drawback when forcing RGB output (with EVR custom or VMR-9)?
Somebody stated some sites back in this thread that there would be a quality loss because dither noise is applied before upscaling or at least it would be better to apply dither noise after upscaling like madVR does.

[nevcairiel]

Is there any drawback when forcing RGB output (with EVR custom or VMR-9)?
Somebody stated some sites back in this thread that there would be a quality loss because dither noise is applied before upscaling or at least it would be better to apply dither noise after upscaling like madVR does.

This is generally true, however for 10-bit this doesn't apply, because 10-bit always needs to be dithered - EVR and VMR9 don't accept 10-bit input formats. So instead of dithering the YUV, i would claim you can get a better quality by using the full 10-bit for RGB creation and dithering at the end of that process.

[JarrettH]

I always noticed in the past that RGB32 output looked better when I used ffdshow. Is RGB32 preferred in certain setups?

[nevcairiel]

I always noticed in the past that RGB32 output looked better when I used ffdshow. Is RGB32 preferred in certain setups?

If you don't use madVR, you can get better results from RGB32 output, because alot of renderers use rather crappy upscaling algorithms for the chroma, and also have no access to the matrix or range flags.
However, using RGB32 output breaks renderer deinterlacing, so it should only be used if you have progressive content or using a decoder which can deinterlace (CUVID deinterlacing or YADIF, for example)

[aufkrawall]

This is generally true, however for 10-bit this doesn't apply, because 10-bit always needs to be dithered - EVR and VMR9 don't accept 10-bit input formats. So instead of dithering the YUV, i would claim you can get a better quality by using the full 10-bit for RGB creation and dithering at the end of that process.
Ok, thanks for the info. :)
But I find it strange that I have to disable AYUV in LAV to get RGB output, it's only listed for 8bit video.
Are those portrayal columns simplified and AYUV can also apply to 10bit video, or is it because of the renderer's 10bit incompatibility?

I always noticed in the past that RGB32 output looked better when I used ffdshow. Is RGB32 preferred in certain setups?
I got some strange artifacts when using RGB output for a Theora 4:2:0 video.
Didn't happen with LAV.

If you don't use madVR, you can get better results from RGB32 output, because alot of renderers use rather crappy upscaling algorithms for the chroma, and also have no access to the matrix or range flags.
However, using RGB32 output breaks renderer deinterlacing, so it should only be used if you have progressive content or using a decoder which can deinterlace (CUVID deinterlacing or YADIF, for example)
Can those of us who don't use madVR have a checkbox to simply output RGB with a corresponding mouse-tooltip that says what you just said? :D

[Xaurus]

Slightly off-topic, but does anyone know if it is possible to run the Intel 2000 IGP at the same time as a Nvidia 450 GTS?

I mean, connect the display to the 450 and just take the audio from the Intel 2000 IGP.

I've search everywhere but I can't really find an answer.

[BetA13]

I found some strange behavior/maybe bug:
Somehow brightness of a x264 I444 10bit video doesn't look fully right without converting to RGB when decoding while playing.

This is how the original FRAPS RGB looks like:
http://www.ld-host.de/uploads/thumbnails/e494e8bd47438960d33d94b080e5ad46.png (http://www.ld-host.de/show/e494e8bd47438960d33d94b080e5ad46.png)

x264 I444 10bit with default LAV decoding settings (CUVID) (a very little too dark):
http://www.ld-host.de/uploads/thumbnails/cea7ebe79549ed195ece54bc87a69e2f.png (http://www.ld-host.de/show/cea7ebe79549ed195ece54bc87a69e2f.png)

The same video with forced RGB32 output (to me it looks exactly like the original FRAPS RGB video):
http://www.ld-host.de/uploads/thumbnails/a7ae5ad2f56418b1b46bd017bee44509.png (http://www.ld-host.de/show/a7ae5ad2f56418b1b46bd017bee44509.png)

Download a test sample (it's not crf 0 like the pictures but this shouldn't make any difference):
http://www.mediafire.com/?n5ymnpdasada4gg
Video has TV range (Rec709).

It would be nice if it looked right without having to force RGB32 output. :)
I don't know what's about 8bit or I420, didn't test it.
Tested with latest MPC HC, LAV nightly and EVR custom.

Btw: Memory consumption with huge x264 RGB file and LAV splitter is now reduced by a half (~1.2 GB), but that's still twice as much than with MPC's internal splitter.



hy there mate.,..
i had teh same problem with FRAPS...i switched to MSI Afterburner..it has better features in video recording ingame..RTV!/mjpeg codec and one more..good quality..no diff to fraps, but its easyer to work with the files afterwards and the colors are better, well not better. lets say just gamma right...

heres an quote from the dev of afterburner:

o Added gamma corrected video capture support

its mostly not the CODEC from ur player causing this but the recording software..

give it a try if u want...u will love it, better then fraps..also u dont have 4 gb files in teh end..afterburner can record much bigger in a row..

http://www.abload.de/img/unbenanntnjuyg.jpg

http://forums.guru3d.com/showthread.php?t=358092

greetz BetA

[Thunderbolt8]

In general, quality is always the same no matter which decoder you use.even though this should be the case, sometimes I get more image corruption or in general a strange kind of flickerin of parts of the image when using coreavc. happens especially in case of documentaries. (its not there all the time, but there seems to be a certain starting point and from then on it doesnt seem to disappear any more)

[CruNcher]

@Nev
RV4 issue

http://www.mediafire.com/?dagppuv8p8wppnr

ffplay = ok
avplay = fail
vlc = ok
mpc hc splitter + lav video = ok
av splitter + lav video = fail
lav splitter + lav video = fail

[oddball]

I'm having trouble with a certain video using CUVID. It appears to put excessive strain on the CPU when high quality processing is enabled in LAV Video decoder. I have tested other videos and they play fine but this one can cause dropped frames on occasion and I am not sure why. I use MPC-HC latest with latest MadVR but I sometimes put ffdshow RAW filters in the chain for delogo or levels and with this clip it just drops frames all over the place when I do this. I can play other 720p and 1080p videos and nairy a dropped frame is in sight. So I don't know why this is different. I have an OC'd E7600 and Nvidia GTX560 so it should handle most stuff. If I use DXVA copy back or Quicksync with the dll (Not sure if it's doing anything on an E7600 to be honest) then everything plays fine but then it's not using high quality processing I guess since it goes greyed out.

http://www.mediafire.com/?ab69qo3wibfmb9j

[sexus]

well latest lav workin nicely , thou freezes if loading up a .avi file if i disable mpeg4 in lav videos format listing it works , please fix this id appreciate it as would others too id asume

[nevcairiel]

@Nev
RV4 issue

http://www.mediafire.com/?dagppuv8p8wppnr


Looks like the multi-thread freeze/hang issue i already reported to ffmpeg/libav. They have to fix it. Works fine if you play with 1 thread

[Aleksoid1978]

Looks like the multi-thread freeze/hang issue i already reported to ffmpeg/libav. They have to fix it. Works fine if you play with 1 thread

in MPC-HC play fine with multi-thread.

[sagematt]

I'm getting a nasty repeated frames issue when using QuickSync decoder. This capture (http://dl.dropbox.com/u/355632/20120207_035151.mkv) easily illustrates my problem.
I reproduced the problem on LAV Filters 0.43, 0.44 and 0.45. It happens with both EVR Custom and madVR.
CPU is Core i7-2630QM

[nevcairiel]

I'm getting a nasty repeated frames issue when using QuickSync decoder. This capture (http://dl.dropbox.com/u/355632/20120207_035151.mkv) easily illustrates my problem.
I reproduced the problem on LAV Filters 0.43, 0.44 and 0.45. It happens with both EVR Custom and madVR.
CPU is Core i7-2630QM

Install the latest Intel drivers directly from Intel.com, not from Windows Update.
It may also be caused by Lucid Virtu if you run that.

[nevcairiel]

in MPC-HC play fine with multi-thread.

Its a combination of splitter and video decoder bug. MPC-HC decoder also fails with LAV Splitter.
I'll try to hack together a temporary workaround for LAV Splitter until the decoder in libav can be fixed

[egur]

I'm getting a nasty repeated frames issue when using QuickSync decoder. This capture (http://dl.dropbox.com/u/355632/20120207_035151.mkv) easily illustrates my problem.
I reproduced the problem on LAV Filters 0.43, 0.44 and 0.45. It happens with both EVR Custom and madVR.
CPU is Core i7-2630QM

This clip is mpeg4 (MP4V fourcc). QS doesn't support mpeg4. Problem is somewhere else...

[nevcairiel]

This clip is mpeg4 (MP4V fourcc). QS doesn't support mpeg4. Problem is somewhere else...

I think he just recorded that to illustrate the original problem. :p
It seems like the old problem caused by odd drivers or Virtu

[ajp_anton]

This clip is mpeg4 (MP4V fourcc). QS doesn't support mpeg4. Problem is somewhere else...Isn't h.264 part of mpeg4? I always get confused when only "mpeg4" is mentioned... what exactly does it mean?

[DragonQ]

"MPEG4" usually refers to MPEG 4 Part 2 (e.g. DivX, Xvid, H.263). AVC/H.264 is MPEG 4 Part 10.

[nevcairiel]

Yeah, when people say MPEG4 they mean the old MPEG4, H264/AVC1 is usually always referred to by those designations.

[oddball]

Its a combination of splitter and video decoder bug. MPC-HC decoder also fails with LAV Splitter.
I'll try to hack together a temporary workaround for LAV Splitter until the decoder in libav can be fixed

Cheers. I look forward to it! :)

[ruinevil]

Can CUVID accelerate 10 bit h264?

I thought it couldn't; LAV splitter would just dump the video to the CPU.

[nevcairiel]

Can CUVID accelerate 10 bit h264?


It cannot. It'll fallback to software decoding automatically.

[Pat357]

I reverted to 257 release drivers (which is also the last version that CUVID works right with interlaced VC-1
What problem do you have with CUVID concerning VC-i ?
It stopped working if I understand this correct ? Only on your older 9xxx card or also on your newer ?

[nevcairiel]

What problem do you have with CUVID concerning VC-i ?
It stopped working if I understand this correct ? Only on your older 9xxx card or also on your newer ?

There seems to be an issue regarding partial acceleration with some drivers (so it would only affect older cards)

[dead_screem]

What problem do you have with CUVID concerning VC-i ?
It stopped working if I understand this correct ? Only on your older 9xxx card or also on your newer ?
horrible blocking on certain parts of video when hardware deinterlacing is active. Apparently only affects "partial" acceleration. VC-1 and MPEG-2, although I havn't tested it with MPEG-2 myself. 258.96 is the last version that is not bugged.
if you really wanna see, i can upload screenshots if i can find where i saved them.

[sexus]

well latest lav workin nicely , thou freezes if loading up a .avi file if i disable mpeg4 in lav videos format listing it works , please fix this id appreciate it as would others too id asume

[Thunderbolt8]

for some reason, dxva2 copy back works better with VC-1 for me than with AVC, while its the other way round with coreavc and MS decoder dmo. I only get few dropped and delayed frames with 30mbps VC-1, but a really non smooth picture with 18mbps AVC.

with coreavc, I can play stuff up to 35-40mbps without problems (mostly).

[kostasoft]

2 nevcairiel
It's possible to make support of multistream video in the container mkv that splitter divided both streams and allowed to reproduce them simultaneously? It is necessary for reproduction of three-dimensional films.

[roytam1]

@nev: access violation at the end of GraphStudioNext decoder performance test, testing RV40 decoder with a rmvb file.

eax=00000000 ebx=02b92180 ecx=02b4baec edx=02af8480 esi=02b920c8 edi=02b920c8
eip=02af848c esp=0012fa84 ebp=0012fa84 iopl=0 nv up ei pl zr na po nc
cs=001b ss=0023 ds=0023 es=0023 fs=003b gs=0000 efl=00000246

function: LAVSplitter!CLAVSplitter__AbortOperation
02af847b cc int 3
02af847c cc int 3
02af847d cc int 3
02af847e cc int 3
02af847f cc int 3
LAVSplitter!CLAVSplitter__AbortOperation:
02af8480 55 push ebp
02af8481 8bec mov ebp,esp
02af8483 8b4508 mov eax,[ebp+0x8]
02af8486 8b80c0000000 mov eax,[eax+0xc0]
fault ->02af848c 8b08 mov ecx,[eax] ds:0023:00000000=????????
02af848e 8b5118 mov edx,[ecx+0x18]
02af8491 6a01 push 0x1
02af8493 50 push eax
02af8494 ffd2 call edx
02af8496 5d pop ebp
02af8497 c20400 ret 0x4
02af849a cc int 3
02af849b cc int 3
02af849c cc int 3
02af849d cc int 3

*----> Stack trace <---*
*** ERROR: Module load completed but symbols could not be loaded for D:\_Programs\graphstudionext.exe
WARNING: Stack unwind information not available. Following frames may be wrong.
*** ERROR: Symbol file could not be found. Defaulted to export symbols for C:\WINDOWS\system32\USER32.dll -
*** ERROR: Symbol file could not be found. Defaulted to export symbols for C:\WINDOWS\system32\kernel32.dll -
ChildEBP RetAddr Args to Child
0012fa84 02af7442 02b9216c e8c999b5 02b920c8 LAVSplitter!CLAVSplitter__AbortOperation+0xc
0012faac 02af71b3 e8c999c5 02b920c8 02b920c8 LAVSplitter!CLAVSplitter__Close+0x52
0012fadc 02af6ebb 02b920d0 0012fafc 02b25253 LAVSplitter!CLAVSplitter__~CLAVSplitter+0xc3
0012fae8 02b25253 00000001 00ef9350 00f06fb8 LAVSplitter!CLAVSplitter__`scalar deleting destructor'+0xb
0012fafc 02af6661 02b920c8 0012fb40 005809c9 LAVSplitter!CUnknown__NonDelegatingRelease+0x23
0012fb08 005809c9 02b920d4 00f06fb8 00f06fb8 LAVSplitter!CLAVSplitter__Release+0x11
0012fb40 0040108b 00000001 0012fb64 0057cfee graphstudionext+0x1809c9
0012fb4c 0057cfee 00000001 00000005 00000000 graphstudionext+0x108b
0012fb64 00581f30 3db37c25 00f039c0 00ef9350 graphstudionext+0x17cfee
0012fba4 00582d58 00ef9350 00000000 00582e47 graphstudionext+0x181f30
0012fbf0 005755d9 00f066a8 00f1b4a8 00ef91e8 graphstudionext+0x182d58
0012fc10 0056f9f1 3db37b0d 00000000 00ef91e8 graphstudionext+0x1755d9
0012fc58 00577ad2 00000000 0056d326 3db37a59 graphstudionext+0x16f9f1
0012fd0c 0056daa2 00ef9ed8 00000770 00ef9ed8 graphstudionext+0x177ad2
0012fd34 00423dfd 00000770 00000000 00ef9350 graphstudionext+0x16daa2
0012fd54 004272b9 00000770 00000000 00ef9350 graphstudionext+0x23dfd
0012fdc8 00427348 00ef9ed8 0007076a 00000770 graphstudionext+0x272b9
0012fde8 77d18734 0007076a 00000770 00000000 graphstudionext+0x27348
0012fe14 77d18816 00427312 0007076a 00000770 USER32!GetDC+0x6d
0012fe7c 77d189cd 00000000 00427312 0007076a USER32!GetDC+0x14f
0012fedc 77d18a10 00177020 00000000 0012ff1c USER32!GetWindowLongW+0x127
0012feec 004204c3 00177020 00000000 00665ab0 USER32!DispatchMessageW+0xf
0012ff1c 005409c7 80000001 00000000 7ffd4000 graphstudionext+0x204c3
0012ff30 00527404 00400000 00000000 0002096e graphstudionext+0x1409c7
0012ffc0 7c817077 80000001 036aca00 7ffd4000 graphstudionext+0x127404
0012fff0 00000000 00527457 00000000 78746341 kernel32!RegisterWaitForInputIdle+0x49

EDIT: when I revert this commit (http://git.1f0.de/gitweb?p=lavfsplitter.git;a=commitdiff;h=d1db625b604f21eab896d5f9a79686c03005b62a), it works without crashing again

[Aleksoid1978]

Hi Nev
Can you modify you splitter, that for RealAudio set to output "normal" subtype:
MEDIASUBTYPE_COOK, MEDIASUBTYPE_ATRC and etc. This need for support AudioDecoder that don't support speciality WAVEFORMATEXFFMPEG.

[nevcairiel]

It already does output more then one media type. The LAV special one is always the second one in the list.
I can look at them and see if they are "standard", but they'll still not work with the RealAudio DLL based decoder.