LAV Filters - DirectShow Media Splitter and Decoders

[NikosD]

@Andy,
I think it's obvious that I don't want LAV x86 to behave differently than x64.

Just to have different settings between the two versions, when used on the same system.

You are playing with words in the first part of your answer, although I liked some other parts, they were very informative.

@Pankov I don't necessarily need profiles for every app.

The separation could only happen between architectures (x86/x64), not every single app, it's enough for me.

I just want all my x86 apps to use DXVA2 native and all my x64 apps to use DXVA copy-back of LAV video.

----------------------------------------------------------------------------------------------------------------------------

@andy and nev

I couldn't believe that would be so difficult to implement.

But, on the other hand, it's different when we say "that is difficult to implement" from what Andy and nev say "it's designed to be like that"

I don't get it.

[nevcairiel]

Like I said earlier, its meant to share the same settings, and thats the end of the discussion.

[NikosD]

The way you answer means, that you could make it work differently, but you decided to make it work like that.

Because I find it constrained and inflexible, I would like to know the reason of making it work like that.

I think it's a simple question and maybe from your reply, I can see things that I don't see right now.

[andyvt]

I think it's obvious that I don't want LAV x86 to behave differently than x64.


TBH, I don't really care either way. The apps that I use configure the filters via the runtime mechanism.


You are playing with words in the first part of your answer, although I liked some other parts, they were very informative.


Words are fun, but that doesn't necessarily mean that I'm playing with them. You asked for the logic, I explained (as a software developer with many years of full life-cycle software development (gather requirements, writing code, testing, & support)). TBC, it's not my goal to silence your opinion (by all means continue), just explain the why.


I couldn't believe that would be so difficult to implement.

But, on the other hand, it's different when we say "that is difficult to implement" from what Andy and nev say "it's designed to be like that"

I don't get it.

I don't remember saying it would be difficult to implement the feature. In fact the feature you want might already exist if you write the values stored in HKCU to the right places in HKLM.

[theoneofgod]

The way you answer means, that you could make it work differently, but you decided to make it work like that.

Because I find it constrained and inflexible, I would like to know the reason of making it work like that.

I think it's a simple question and maybe from your reply, I can see things that I don't see right now.

From a programmers perspective, it would be simple enough to have each version run with separate settings. I don't see the need for this though as what is the purpose? Do you have an example?

[Superb]

I personally see no purpose in separating settings based on architecture.
Once you find the optimal settings for your machine, it shouldn't matter if you use x64 or x86.
Making users configure LAV twice doesn't make any sense.
I guess that power users could benefit from some sort of profiles feature, but that's only for really special cases and doesn't necessarily justify the time to code the functionality.

[vood007]

Just use Sandboxie for such things. You may lose a bit of performance but you really get isolated envs for any apps you want.

http://www.sandboxie.com/

[mysterix]

Does an "Normalize matrix" option apply any dynamic range compresion?
Should I use an "Normalize matrix" option in the LAV audio filter to prevent clipping? Does it prevent from clipping 100%? How it works, I mean, how this function chooses attenuation level to prevent clipping without preliminary analysing entire track (like foobar2000's ReplayGain does)? Does this function make any distortion or sound worse somehow?Please anybody, answer my question.

[nevcairiel]

It'll ensure that no channel is ever above 100%, if the source file had no clipping, then the decoded stream will also never have any.
It doesn't distort the sound or make it sound any worse, but it will make quite a bit quieter.

[mysterix]

It'll ensure that no channel is ever above 100%, if the source file had no clipping, then the decoded stream will also never have any.
It doesn't distort the sound or make it sound any worse, but it will make quite a bit quieter.What is 100% channel level in this case? Will matrix normalising help if the source file had clipping?

[mysterix]

Seems, I have understood that while reading earlier posts. Thanks!) It is downmixing-based option, so, do not refers to clipping directly.

[sebas_led]

Hi Nev:

Thanks for the new version 0.62. It looks great.

I've noticed a problem with window render size since version 0.60.
When playing a video, the window is adjusted to my desired W/H size, but shortly after that it get's increased to the original video size.
It doesn't happen in all videos.

Sample code:
hr = pGraph->RenderFile(wideFilename, NULL);
pGraph->QueryInterface(IID_IVideoWindow, (void **)&pVideoWindow);
if (pVideoWindow) {
pVideoWindow->put_Top(videoYDesktop);
pVideoWindow->put_Left(videoXDesktop);
pVideoWindow->put_Width(videoWDesktop);
pVideoWindow->put_Height(videoHDesktop);
}
...
if (pControl) {
HRESULT hr = pControl->Run();
}
...


Version 0.59.1 works fine.

I can provide you a sample .mp4 file if you wish.

[nevcairiel]

How in the world would my code ever affect your video window?
I just deliver images and metadata. How the thing reacts to that is entirely up to your code.

[Liandri]

Question about LAV Splitter and its Advanced subtitle mode.

I have a file with 2 subtitles of the same language but different quality, they appear as:
- light [rus] [default]
- full [rus]
- no subtitles

I want to auto-select full [rus].

If I set the filter to "!d" it goes with no subtitles. Tried setting it to "rus!d" and "*:rus!d" but then it always selects the default one. I remember it worked for me once, but I don't understand why it doesn't work now.

It would help to have some way to force selection of some subtitles based on their number in menu, perhaps.

[sebas_led]

Hi Nev:

Frankly I don't know :)
But the result is reproducible in several PC's, with several version of Windows (7, 8, etc.)
I'm using the default video render.
When using 0.59.1 I can sucesfully put the width and height of the render window.
When switching to 0.60/0.62 the video render window changes from the programmed size to the original video size.

[DarkSpace]

Tried setting it to "rus!d" and "*:rus!d" but then it always selects the default one. I remember it worked for me once, but I don't understand why it doesn't work now.
Try "*:rus|!d", with "|" as the separator between language code and track property.

[turbojet]

What are the chances of an option to not use hardware acceleration with high framerate (48+) video? On my GTS250 with 720p59 madvr+cuvid or dxva = some dropped frames. Switching to software decoding fixes it.

[Liandri]

Try "*:rus|!d", with "|" as the separator between language code and track property.
That worked, thank you!

However I still think that Advanced mode could be more flexible. For example, it likely won't work for the case with 3 subtitles options of the same language. For difficult cases where user doesn't mind to change settings specifically for some set of episodes, choosing by number or by matched phrase would be a good thing to have.

[P.J]

We're waiting for LAV HEVC DXVA decoder :)

[mysterix]

...and OpenCL-based decoder or DirectCompute-based decoder as an alternative to DXVA-CUVID-QuickSync for all formats, please.

[nevcairiel]

That worked, thank you!

However I still think that Advanced mode could be more flexible. For example, it likely won't work for the case with 3 subtitles options of the same language. For difficult cases where user doesn't mind to change settings specifically for some set of episodes, choosing by number or by matched phrase would be a good thing to have.

By number seems rather useless, its hardly an identifying attribute in which position in the file a stream is, and it doesn't really tell you anything about that stream.

Someone posted a patch for matching by the stream title, but i'm not entirely happy with the patch yet - and i'm out of the country so I don't have much time to look at it myself.

[Liandri]

By number seems rather useless, its hardly an identifying attribute in which position in the file a stream is, and it doesn't really tell you anything about that stream.

Someone posted a patch for matching by the stream title, but i'm not entirely happy with the patch yet - and i'm out of the country so I don't have much time to look at it myself.
It IS useless if you use it all the time. But I found myself in a situation where I wanted to set it just for one 24-episodes show and be done with it, but even after readme I found it rather confusing.

If I could just tell it to always choose 2nd subtitles item, it would save me some headache, and I wouldn't mind reverting settings back after watching that specific show - it's easier than choosing that manually for each episode.

Anyway, it would help if real examples with "!" get included in the readme, for others to spend less time on playing with settings. Thanks for paying attention!

[Mangix]

...and OpenCL-based decoder or DirectCompute-based decoder as an alternative to DXVA-CUVID-QuickSync for all formats, please.

This is not a good idea. Except if you're after increased power usage.

[Asmodian]

...and OpenCL-based decoder or DirectCompute-based decoder as an alternative to DXVA-CUVID-QuickSync for all formats, please.

I think you are missing the point that all those use dedicated hardware to do the decoding very efficiently, OpenCL/DC would be very different and would require a good video card. Writing a highly optimized OpenCL decoder for everything LAV supports would also be a .. big .. project. Why would someone want this anyway? Use software decoding on the CPU if you don't want hardware acceleration, if your CPU cannot do it your GPU wouldn't be able to either using OpenCL. ;)

[mysterix]

I think you are missing the point that all those use dedicated hardware to do the decoding very efficiently, OpenCL/DC would be very different and would require a good video card. Writing a highly optimized OpenCL decoder for everything LAV supports would also be a .. big .. project. Why would someone want this anyway? Use software decoding on the CPU if you don't want hardware acceleration, if your CPU cannot do it your GPU wouldn't be able to either using OpenCL. ;) My CPU is Pentium D 925 and it can't process FullHD itself on middle to high profiles. My GPU is HD5450 and it can by DXVA, but i prefer use MadVR render on lower resolutions and FullHD with low bitrates also and according to my low-end videocard, DXVA is expensive for it if using with MadVR together. But it is not reason for OpenCL since my CPU still can decode all that. However, there is some reasons to realise that idea: many people have had a troubles with decoding some nontypical videoprofiles on several formats. Take for example 10-bit video and trouble with 60 fps decoding on some models of videocards that can do 60 fps seemingly, but... And yes, my card supports OpenCL 1.2 and DC nevertheless. But on the other side, it can't do WebM and DivX accelerated and several other formats and profiles. There are another reasons, still. As I know, there is unexplained occasional frame drops on supported formats and profiles on DXVA (maybe on CUVID also) sometimes while cpu and gpu usage is far from 100%. Maybe it appears not on all cards and not always, don't know. And one more: my gpu has not separate DXVA block and uses main GPU block for DXVA processing as I know. Contrary to middle and high-end cards which have separate HA-module. So, that's one more reason: to process more than one videostream by using HA and main blocks simultaneously for translating different streams to several monitors in a house, e.g. But that depends on render abilities, also.

[ryrynz]

My CPU is Pentium D 925 and it can't process FullHD itself on middle to high profiles. My GPU is HD5450 and it can by DXVA

The solution is to just upgrade your old hardware. What you're asking for isn't going to happen.

[foxyshadis]

There's no way a 5450 could handle the OpenCL processing you're asking for, you'd have to upgrade anyway. I'm not going to knock them, they're great silent cards; on the other hand, I know for a fact that it works perfectly smooth at 60fps for H.264 on my HTPC (I had a passive-cooled one before upgrading to 6450 in a trade), so the problem is your memory or motherboard, unless your video card is going bad. I don't know where you get the idea that it doesn't have a DXVA module, it has UVD 2.2 unless your card is broken. (6450 has UVD 3.)

6-8 years is a good run for a system, but if it's the money you should look at an A8-6600K. Less than $200 gets you CPU, motherboard, and 8GB memory, and even the CPU's integrated graphics is twice as powerful as the HD5450.

[mysterix]

foxyshadis, thanks! I believe I'll upgrade my system as soon as possible.

nevcairiel, would you realise ability to LAV Video filter of choosing which GPU installed in the system to use for hardware decoding, please? I mean configuration where monitor is connected to one of them only.

[huhn]

what will this bring the end user? using DXVA native on current hardware has no real affect on the gpu load. a ASIC is used for decoding anyway(at least at the moment).

and if you use a decoder from a different GPU that is not displaying the image later on you have to use a copyback.

QS and CUVID are both copyback decoder and you can do this already if you have a intel and a nvidia gpu you can display the image on the nvidia gpu and let it decode from QS. same goes for intel and AMD.
this normally only wastes power.

if you have multiple monitors Lavfilter will use the decoder from the current monitor the player is active on.

[jkauff]

if you have multiple monitors Lavfilter will use the decoder from the current monitor the player is active on.
You can use QS decoding even if the iGPU is not connected to a monitor. You just set up a "dummy" monitor for the iGPU via Screen Resolution in Windows. It doesn't need to be set as the active monitor for QS to become enabled in LAV. You must have the Intel iGPU drivers installed, though.

[P.J]

How to change the de-interlacing method and switching between HW/SW in LAV like PowerDVD/PotPlayer?
It's only possible to change the decoder :(

http://t.imgbox.com/EsRwsvrQ.jpg (http://imgbox.com/EsRwsvrQ)

[clsid]

LAV has an option for that called "High-quality processing".

[jkauff]

How to change the de-interlacing method and switching between HW/SW in LAV like PowerDVD/PotPlayer?
It's only possible to change the decoder :(
Set "Hardware Decoder" to None to use software decoding, then select "YADIF software de-interlacing" if you want LAV to do the de-interlacing.

[Deihmos]

The last version of lav splitter that worked for me is version 59.1. On windows 8.1 using windows media center when I resume from sleep lav is unable to play the video until I restart wmc. I don't know what the problem could be but it does the same thing on all my htpcs. Using version 59.1 works without issues.

[pihug12]

Hello!

This M4A audio file seems to cannot be played with LAV Splitter : http://www.criticalmusic.com/podcast/episodes/CriticalPodcast32Emperor.m4a

General
Complete name : C:\Users\Pierre\Downloads\CriticalPodcast32Emperor.m4a
Format : MPEG-4
Format profile : Apple audio with iTunes info
Codec ID : M4A
File size : 72.1 MiB
Duration : 59mn 15s
Overall bit rate mode : Variable
Overall bit rate : 170 Kbps
Movie name : Critical Music Podcast Vol.32 - Hosted by Emperor
Encoded date : UTC 2014-02-11 11:19:44
Tagged date : UTC 2014-02-11 11:19:55
Writing application : GarageBand 6.0.5
Cover : Yes
Comment : 1. Ed:it - I Would (SGN:LTD) / 2. Rregular & Dementia ft. Disphonia - Doom Loop...

Video
ID : 4
Format : JPEG
Codec ID : jpeg
Duration : 59mn 15s
Source duration : 59mn 15s
Bit rate mode : Variable
Bit rate : 3 684 bps
Width : 500 pixels
Height : 500 pixels
Display aspect ratio : 1.000
Frame rate mode : Variable
Frame rate : 0.004 fps
Minimum frame rate : 0.001 fps
Maximum frame rate : 0.010 fps
Color space : YUV
Chroma subsampling : 4:4:4
Bit depth : 8 bits
Compression mode : Lossy
Bits/(Pixel*Frame) : 3.684
Stream size : 1.43 MiB (2%)
Source stream size : 1.43 MiB (2%)
Language : English
Encoded date : UTC 2014-02-11 11:19:42
Tagged date : UTC 2014-02-11 11:19:55

Audio
ID : 3
Format : AAC
Format/Info : Advanced Audio Codec
Format profile : LC
Codec ID : 40
Duration : 59mn 15s
Bit rate mode : Constant
Bit rate : 160 Kbps
Channel(s) : 2 channels
Channel positions : Front: L R
Sampling rate : 44.1 KHz
Compression mode : Lossy
Stream size : 69.7 MiB (97%)
Language : English
Encoded date : UTC 2014-02-11 11:19:42
Tagged date : UTC 2014-02-11 11:19:55

Text #1
ID : 1
Format : Timed Text
Codec ID : tx3g
Duration : 59mn 15s
Bit rate mode : Variable
Bit rate : 2 bps
Stream size : 899 Bytes (0%)
Language : English
Encoded date : UTC 2014-02-11 11:19:42
Tagged date : UTC 2014-02-11 11:19:55

Text #2
ID : 2
Format : Timed Text
Codec ID : tx3g
Duration : 59mn 15s
Bit rate mode : Variable
Bit rate : 2 bps
Stream size : 775 Bytes (0%)
Language : English
Encoded date : UTC 2014-02-11 11:19:42
Tagged date : UTC 2014-02-11 11:19:55

The MPC-HC window allocates 500px * 500px for the artwork, but the screen stays black.
The "Play" button is pushed, but nothing is played (stays at 00:00).

If I switch the splitter to Haali, I can read this file (see more infos here (http://codecs.forumotion.net/t1523-m4a-audio-file-doesn-t-play-with-mpc-hc)).

Sorry if this problem is not only LAV Splitter related, I'm no expert.

Sorry to dig up this old post.
Did you have some time to take a look at it? Is it LAV Splitter related? Is there a possible fix?

Here is another file if you want to make some tests : http://www.criticalmusic.com/podcast/episodes/CriticalPodcast33Enei.m4a

Thanks!

[NikosD]

It seems it's a problem with LAV filters and JPEG files.

It's not the audio part, it's probably a splitter problem in MP4 container (or a video issue)

e.g this JPEG file has no audio.
https://www.sendspace.com/file/k7q9em

It can't be played by MPC-HC (it shows just a black screen in pause mode or even moving slider displaying black screen)

MPC-HC old v1.6.8 and PotPlayer can play it fine.

The above file from pihug12 plays fine by PotPlayer, BTW.

[CiNcH]

Hi nev,

I am currently developing the video renderer for the DVBViewer and hoped you could help me out with something...

The EVR presenter provides the IDirect3DDeviceManager9 over which we provide the IDirect3DDevice9 for you to create your D3D resources (surfaces) for the DXVA2 decoder.

Problem now is that when playback is running, a call to IDirect3DDevice9::Reset fails with D3DERR_INVALIDCALL. Reset is called for example when the graphics adapter changes by moving the video window to a different monitor. I release all resources in the renderer before the reset, but I have no clue how to tell the decoder to release them. Only after the reset do I call IDirect3DDeviceManager9::ResetDevice to give you the new device. But I could not find anything to tell you that I am about to reset and that you should release everything.

Do you have any idea? Do I have to kill the graph before doing the reset? I also tried to send an EC_ERRORABORT with D3DERR_DEVICELOST into the graph before the reset. But it didn't help either.

[Aleksoid1978]

it's probably a splitter problem in MP4 container (or a video issue)


Yes - bug in LAVSource/Splitter. It's output incorrect subtype for JPEG stream in MP4/MOV

[Deihmos]

The last version of lav splitter that worked for me is version 59.1. On windows 8.1 using windows media center when I resume from sleep lav is unable to play the video until I restart wmc. I don't know what the problem could be but it does the same thing on all my htpcs. Using version 59.1 works without issues.

Anyone know why the new versions are doing this.?

[TheDarkTemplar]

I can't seem to get the Advanced Subtitle Selection mode to work. I read the readme section that talked about it, so I am fairly sure I am understanding the code correctly...

I put in eng:eng|f eng:eng|d eng:off *:eng so that it will only enable English forced or default subtitles if the audio is English, and enable any English subtitles if the audio is not English. It doesn't load default, forced English subtitles no matter what the audio language is though. In fact, even if I simply put *:eng in and nothing else then it still won't load English subtitles. The readme specifically says that command will play them no matter what though...

Is it just broken or something?

[nevcairiel]

Sounds like your player overwrites the choice.

[TheDarkTemplar]

If I change the selection mode from advanced to default, then it does load default subtitles fine. So, it seems it's definitely capable of switching them on if it wants. Hmm...

[theoneofgod]

Any chance of a function to pass-through AAC or similar to how ffdshow handles it?

[wanezhiling]

With lav splitter, the color of the subtitle (https://www.sendspace.com/file/edpg3v) becomes green: http://i1.tietuku.com/66a3005df2638fd1.png

[Farfie]

Hey nev,
Do you have any intention to support flash in an swf container? Or is it not possible, or does it already? Because I see Flash in the supported types for LAV Video. I just want to apologize in advance if this is a silly question.

[LigH]

What exactly do you mean by "Flash"? ActionScript directed vector/bitmap-sprite animations of "Shockwave Flash" (*.swf)? — They are not video streams (H.263, VP6, H.264) like in "Flash Video" (*.flv).

It is a pity that two completely different concepts share the same name part, causing misunderstanding and confusion.

[Farfie]

What exactly do you mean by "Flash"? ActionScript directed vector/bitmap-sprite animations of "Shockwave Flash" (*.swf)? — They are not video streams (H.263, VP6, H.264) like in "Flash Video" (*.flv).

It is a pity that two completely different concepts share the same name part, causing misunderstanding and confusion.

Unfortunately, I'm not sure. From trying to deal with these files, I'm pretty sure they're definitely a pain... all I can offer is what mediainfo gives me.

http://puu.sh/9UKHC/8c9c1b8fe6.png

So maybe it's VP6? Or maybe mediainfo is confused? Because I don't see why there would be 2 video streams either, but hey, what do I know! :confused:

[LigH]

Well, maybe it is ActionScript directed VP6 video inside a ShockWave Flash container. The presence of ActionScript is probably not detected by MediaInfo; but this interactivity (even if it is just playback controls or cue points) is what makes it rather impossible to implement completely without a dependency to Adobe libraries. I doubt there are freely usable AS interpreters not by Adobe. And as you see yourself, supporting only the few SWF containing a video, but not the many SWF containing vector animations, would make "SWF support" a rather bold statement.

Are there any samples of SWF files with ... G-rated video content?

[hoborg]

Hey nev,
Do you have any intention to support flash in an swf container? Or is it not possible, or does it already? Because I see Flash in the supported types for LAV Video. I just want to apologize in advance if this is a silly question.

Hi.
Did you tryed to rename *.swf to *.flv ?

[LigH]

That will not remultiplex the file with a different container format. SWF and FLV are different.