LAV Filters - DirectShow Media Splitter and Decoders

[videonerd]

I have been testing the re-written asf/wmv demuxer from Libav, and it seems to seek OK in this sample.

I'll probably enable it in the next nightly build, so testing would be appreciated, ideally beyond seeking, as it is very new and may still have other short-comings that need fixing.

Seeking which used to have a 3-5 second lag has been reduced to 1-2 seconds. Improvement, but it is still a lag vis-a-vis other containers which has 0-100ms lag when seeking. Same bitrate, 64bti version.

[mogli]

1. Switch to the 4th audio stream then switch back to the first one.

2. Enable 'Enable Mixing' then disable it.

The audio volume will decrease in both cases.
Volume is consistent here. Any special setting or version I need to use?

[nevcairiel]

Seeking which used to have a 3-5 second lag has been reduced to 1-2 seconds. Improvement, but it is still a lag vis-a-vis other containers which has 0-100ms lag when seeking. Same bitrate, 64bti version.

No software can fix the fact that ASF is a terrible format. =p

As such, comparison to other containers is pointless.

[blaubart]

4k h.265 50/60fps still not playing really clean, stutters here, pixel storms there.. Reacting capricious to different PC conditions.. But now i know better workarounds.

..to further this - my test-PC is at the moment such a fancy hi-speed gaming thing and I did not have so much time to get used to. The energy levels of the Asus Maximus VII Hero board are not only controlled by BIOS and Windows but also an "AI Suite 3" which is buggy. Though set to Auto it mostly switches to Performance at PC boot.
But at this Performance mode the Lav Video Decoder playback of 50/60fps 4k HEVC is sometimes stuttering. Plus pixel storms if played by Lav-CopyBack. (SD, HD clean).

Please don't ask me why! But I've made 100drets tests meanwhile searching the bug so there is no doubt about that.
i.e. PDVD Generic playing clean in the Performance setting.

To get a clean Lav 4k HEVC simply deinstalling "AI Suite 3" didn't help. So I built some workaround to force the Suite to always start in Auto mode.

FINALLY patiently waiting 1-2 minutes after booting till this 'fight'-PC 'settled down to peace' - always smooth 4k Lav playback now.
Later if 4k will run in any simple office laptop such stress issues will never occur but may be others :)

[P.J]

Volume is consistent here. Any special setting or version I need to use?

Strange, I'm using LAV 0.65+MPC-HC 1.7.9 x64
It happens with some other video files too.


..to further this - my test-PC is at the moment such a fancy hi-speed gaming thing and I did not have so much time to get used to. The energy levels of the Asus Maximus VII Hero board are not only controlled by BIOS and Windows but also an "AI Suite 3" which is buggy. Though set to Auto it mostly switches to Performance at PC boot.
But at this Performance mode the Lav Video Decoder playback of 50/60fps 4k HEVC is sometimes stuttering. Plus pixel storms if played by Lav-CopyBack. (SD, HD clean).

Please don't ask me why! But I've made 100drets tests meanwhile searching the bug so there is no doubt about that.
i.e. PDVD Generic playing clean in the Performance setting.

To get a clean Lav 4k HEVC simply deinstalling "AI Suite 3" didn't help. So I built some workaround to force the Suite to always start in Auto mode.

FINALLY patiently waiting 1-2 minutes after booting till this 'fight'-PC 'settled down to peace' - always smooth 4k Lav playback now.
Later if 4k will run in any simple office laptop such stress issues will never occur but may be others :)

I told you there should be an issue there ;)

[nevcairiel]

LAV does never influence the volume of the audio being played, its more likely something else badly reacting to a change in channel count, which would neatly explain how switching on Mixing would also trigger it.
I do faintly remember someone else mentioning this before and maybe also finding a solution, but if I could remember... Might be worth trying to search through the thread, somehow.

[James Freeman]

Nev,

A few pages back you have noted that it is better to use DXVA2 instead of CUVID, why?

What is the advantage using copy-back vs native?
Isn't copying back the frame from the GPU memory back to main memory takes more time?

Thanks

[vivan]

CUVID:
- forces GPU into high power state.
- have problems/bugs that nev can't solve (they're in driver) / workaround requires too much effort. IIRC there were even crashes.
+ supports divx
And it was the first hardware decoder (with copyback).

DXVA2 native is fastest, and latency (time from you start seeking till you get frame) is very low. Difference in latency is very noticeable when you seek often, even on fast CPU native seeks faster (than s/w).
But it's less stable? I don't really remember any complains, tbh. The only problem that I have (and it not bothering me) - sometimes it produces teared frame right after seek.

[nevcairiel]

vivan outlined the drawbacks of CUVID already, so not much to add there.

For Copy-Back vs. Native, Native is a bit more efficient if you can use it. With madVR, using Copy-Back may give you slightly better quality, as madVR has a few problems getting access to the unprocessed image otherwise.
Copy-Back is also a bit more flexible, especially when it comes to clips which may change their format during playback (rare, but happens with some animes)

Personally, I prefer Copy-Back with madVR, but especially if you're trying to get the lowest resource consumption possible with say EVR on a low-power playback device, then Native is probably the best choice.
But Copy-Back is even efficient enough to handle 4k@60 on a decent system without any significant resource cost, so..

[James Freeman]

Thank you both.

[Thunderbolt8]

With madVR, using Copy-Back may give you slightly better quality, as madVR has a few problems getting access to the unprocessed image otherwise.does this problem also occur in software mode, without using any hardware acceleration?

[huhn]

in madVR only native can have an negative effect on picture quality. this has to do with DXVA scaling.

all other modes look like software to madVR.

[an3k]

For many years I used CoreAVC + Haali Splitter + VSFilter (https://nightly.mpc-hc.org/mpc-hc_apps/vsfilter/). I got subtitles (SRT, VOBSUB or PGS) shown in WMP12. Everything was fine but only for H.264 + MKV + AC3. I couldn't play eg. TrueHD/DTS/DTS-HD thus I was looking for a decoder and someone suggested LAVFilter. I have to admit I absolutely love it. Thanks for the great work!

Because it supports so many formats (and isn't overloaded with settings most people don't need) I bundled LAVFilters + Icaros + xy-VSFilter and created a batch script that a) installs the decoders/filters, b) writes some useful standard settings (eg. show icon) into the registry, c) writes the required data for native windows support of the file format / MIME type into the registry and d) disables the microsoft decoders (data taken from Win7DSFilterTweaker). The goal is to add support for the supported containers and codecs in Windows as native as possible.


However, I have a problem. I can't get subtitles shown with LAVFilter in WMP12. I tried it with VSFilter as well as xy-VSFilter. When I use graphstudionext the graph is correctly built and subtitles are shown but when using WMP12 there are no subtitles (sadly I can't load the graph used by WMP12). It worked with Haali so I think it's a problem with LAV Splitter. In Haali I have to enable "Autoload VSFilter" (When set to yes the splitter will load VSFilter and add it to the graph) to get subtitles working in WMP12. If that setting is disabled (default) I have the same problem as with LAVFilters.

[nevcairiel]

There is no autoloading of any filters in LAV.
Subtitles can work without that though, I've had it work before. Forgot if anything special had to be done though, I don't use WMP (nor should anyone in their right mind).

[an3k]

There is no autoloading of any filters in LAV.I know. That's why subtitles are not working :( I also saw the bugtracker ticket https://code.google.com/p/lavfilters/issues/detail?id=32
Subtitles can work without that though, I've had it work before. Forgot if anything special had to be done though, I don't use WMP (nor should anyone in their right mind).Yeah, everything works "somehow", the challenge is to know how. What's the problem with using WMP?

[nevcairiel]

What's the problem with using WMP?

You are here having a problem with WMP. Not me. My player allows me to do everything I want to do, because its flexible and gives me control .. yours doesn't. :D

[an3k]

Actually I have a problem with LAV Splitter because it works well with Haali's Splitter.

[huhn]

WMP isn't loading the subtitle stream (it officially doesn't support subtitles anyway). haali simply forced the vsfilter to load and not the player but it should be the other way a player should choice if and what subtitle should be loaded.

try mpc-hc/be or a ton of other players and it will work.

[an3k]

WMP isn't loading the subtitle stream (it officially doesn't support subtitles anyway). haali simply forced the vsfilter to load and not the player but it should be the other way a player should choice if and what subtitle should be loaded.Yes, I'm aware of that. Actually WMP supports subtitles but in the typical (weird) microsoft way. http://sourceforge.net/projects/wmpsub/

try mpc-hc/be or a ton of other players and it will work.Well, if I have to force my customers to use a different player than the one they want to use just to get subtitles working then I would force VLC. It would be the easiest solution (no work for me) but also the worst (no thumbnailing, no embebbed covers, no property sheet data, etc.). That's exactly how Microsoft works.

[nevcairiel]

Its possible to use VSFilter without auto-loading with WMP, like I mentioned in my earlier post. It just requires a bit more extra hacking around in the preferred filter and merit system. I forgot how to do it exactly and I cba to try, but its certainly possible - I have done so in the past to test some things.

[clsid]

Why are you trying to re-invent the wheel? You obviously fail in trying to do better than proven solutions. A batch script sounds horrible too. At least give your customers a proper (un)installer.

K-Lite Basic (= LAV Filters + VSFilter + Icaros) does all you need, including correctly configuring the preferred decoders for working subtitles in WMP12. It even allows custom silent installs. Disabling any Microsoft codecs is actually not needed at all for MKV playback.

[an3k]

It just requires a bit more extra hacking around in the preferred filter and merit system.That's the information I needed so I know where to start. I got it working now. Thanks.
I forgot how to do it exactly and I cba to try, but its certainly possible - I have done so in the past to test some things.Hehe yeah. I was going to donate a nice amount because of LAVFilters but because of how you replied I cba. But the money isn't lost, Icaros dev will get it.

[nevcairiel]

I did say that in my very first response that it can work in WMP without any auto-loading, but apparently calling your favorite player bad is a grave offense :rolleyes:

[an3k]

I did say that in my very first response that it can work in WMP without any auto-loading, but apparently calling your favorite player bad is a grave offense :rolleyes:

As I wrote it is HOW you replied and NOT what you replied. I know myself that WMP is not that good but most of my customers use it, thus I have to support it if I want to keep them.

[Akeno]

Hi everyone. MPC-HC uses the Microsoft DTV-DVD audio decoder whenever I run AC3 sources. I've checked to make sure that AC3 is checked under LAV Audio but the problem still persists. Any ideas on what could be causing this?

[Qaq]

untick AC3 in MPC internal source filters? AFAIK, for the sources MPC first check its own internal, so what you have checked in external is just doesnt matter.

[LigH]

Maybe a non-LAV splitter is being used for the container, so that it forces connection to MS decoders? Or do you mean raw AC3 audio files?

[nevcairiel]

The usual problem for LAV Audio not being used for AC3 or DTS is having bitstreaming activated, but the system not accepting a bitstreaming connection. So, make sure you have bitstreaming off, or if you want bitstreaming, that your audio renderer actually accepts it.

[clsid]

Sorry to ask this for a trillionth time, but why hasn't a fallback been implemented yet? The bug in MPC-HC audio switcher that prevented it in the past was fixed long ago.

[nevcairiel]

Its not a priority for me, as I consider bitstreaming somewhat obsolete, especially now that we can decode DTS-HD without any external help.
Time constraints don't allow me to work on everything at the same time, so....

[Akeno]

Maybe a non-LAV splitter is being used for the container, so that it forces connection to MS decoders? Or do you mean raw AC3 audio files?
Well, the source is an MKV file with AC3 audio. LAV Splitter still activates as it and LAV Video appear under the filters list but DTV-DVD is being used to decode audio rather than LAV Audio.

The usual problem for LAV Audio not being used for AC3 or DTS is having bitstreaming activated, but the system not accepting a bitstreaming connection. So, make sure you have bitstreaming off, or if you want bitstreaming, that your audio renderer actually accepts it.
I don't believe I have bitstreaming enabled. I even disabled all the internal renderers in MPC-HC and it still uses DTV-DVD.
-------
EDIT: Turns out I had bitstreaming enabled in LAV Audio. I use ReClock to render audio though. Doesn't that support bitstreaming?

[huhn]

reclock gets totally useless with bitstreaming.

[Kurtnoise]

@Nev : I'm trying to compile LAVFilters using Visual Studio 2015 but I've got an issue w/ LAVSplitter :

demuxers.lib(LAVFDemuxer.obj) : error LNK2001: unresolved external symbol "struct clpi_cl * __cdecl clpi_parse(char const *)" (?clpi_parse@@YAPEAUclpi_cl@@PEBD@Z)
demuxers.lib(LAVFDemuxer.obj) : error LNK2001: unresolved external symbol "void __cdecl clpi_free(struct clpi_cl *)" (?clpi_free@@YAXPEAUclpi_cl@@@Z)


Any idea ?

[ryrynz]

Sorry to ask this for a trillionth time, but why hasn't a fallback been implemented yet? The bug in MPC-HC audio switcher that prevented it in the past was fixed long ago.

Well I'm glad you did cos he totally ignored my question.

I consider bitstreaming somewhat obsolete

Typical HTPC setups a thing of the past huh? Wowww.

[NikosD]

@nevcairiel and all

I had involved as a "middle man" between Intel and Nevcairiel in order to fix the issue of P010/P016 output which was and still is crashing LAV filters when Intel GPU is used for 10bit clips (H.264/ H.265)

After months of occasional emails between me and Intel and reports here in private messages to nevcairiel , the situation seems stalled.

It seems that Microsoft has acknowledged a bug in EVR renderer which is going to fix in Windows 10 only and both parts (Intel, Nevcairiel) are not going to change anything in their software (drivers, LAV filters) in order to fix the issue.

Latest email from Intel was asking from Nevcairiel to change his code in order to "fix" Microsoft's bug.

The code of LAV filters sent to me by Intel (that I have already sent to nevcairiel) in order to be fixed by nevcairiel is this:

hr = m_pOutput->GetConnected()->ReceiveConnection(m_pOutput, &mt); // connect to EVR

if(SUCCEEDED(hr)) {
IMediaSample *pOut = nullptr;
if (SUCCEEDED(hr = m_pOutput->GetDeliveryBuffer(&pOut, nullptr, nullptr, 0)) && pOut) { // get buffers allocated by EVR

AM_MEDIA_TYPE *pmt = nullptr;
if(SUCCEEDED(pOut->GetMediaType(&pmt)) && pmt) {
CMediaType newmt = *pmt;
videoFormatTypeHandler(newmt.Format(), newmt.FormatType(), &pBIH);
DbgLog((LOG_TRACE, 10, L"-> New MediaType negotiated; actual width: %d - renderer requests: %ld", width, pBIH->biWidth));
if (pBIH->biWidth < width) { // found size is wrong, free buffers and return error.
DbgLog((LOG_ERROR, 10, L"-> Renderer requests width smaller than image width, failing.."));
DeleteMediaType(pmt);
pOut->Release();
return E_FAIL; // error return value will keep this decoder keep trying reconnect
}

I can't say for sure who is right or wrong in this situation, the only thing I know is that egoism doesn't help.

We, users, are the "victims" of this conflict between Intel and nevcairiel, which I'm sure would never exist if the other part was Nvidia and not Intel.

[Kurtnoise]

Intel guys should be smart enough to send proper patch...

[NikosD]

It wouldn't help, because nevcairiel doesn't want to change anything in his code

[nevcairiel]

It wouldn't help, because nevcairiel doesn't want to change anything in his code

Because you know what everyone wants and doesn't want, right?
Please keep out of my thread with such nonsense.

[NikosD]

Be more polite to someone who is trying for months to collaborate with you, a developer who is extremely rude and non cooperative in order to solve problems that he doesn't care.

You are not helping as much as you could and you are disappeared from the discussions, probably because you don't care as you often say.

Why don't you speak directly with Intel to solve the problem ?

Do you want them to come to you ?

Have you ever tried to get some help from them ?

You are sitting on your "perfect" code in your opinion and you are waiting from the others to change everything to adjust to your code.

Give me a break dude!

[nussman]

Why Nev or Intel should fix Microsoft Bugs?
Did you try to report this to MS?
It's a shame MS is not willing to fix this for pre Win10 OS!

[NikosD]

The problem is that it's not working even in latest Win 10 build 10240.

So, even if the bug is fixed, more work has to be done in order to have a complete solution of the issue.

[nevcairiel]

You are sitting on your "perfect" code in your opinion and you are waiting from the others to change everything to adjust to your code.

Noones code is perfect, but its simply not my bug, its Intels, or Microsofts, depending how you look at it.

I don't expect anyone to "adjust to my code", I expect them to fix their bugs, just like I fix mine. Intel enabled P010 input a couple months ago, after that it was broken. So now I should add a work-around to somehow handle this breakage, that Intel caused?

What about all other decoders that support 10-bit, they also need to add work-arounds? Or they don't even try to support P010 because they know its broken - which isn't any better either.
Why doesn't Intel fix it again, just like NVIDIA did after they broke it briefly?

LAV isn't behaving "wrongly", its just not trying to automatically work-around Intels or Microsofts bug.
If someone wants to send a patch to make it fallback to NV12 in that case, please do so, but my development system doesn't have a Intel GPU right now, so I would need to setup another one to even start working on that - which cuts into the time factor again. I might do it some time, but apparently NikosD expects results within days, a few of which I was on vacation, too =p

I'll be the first to fix my code if it does indeed behave badly or is buggy, but that doesn't include endless work-arounds for other softwares problems.
In fact, that was one of the premises that the LAV project started on. ffdshow and a few other things had so many hacks for so much broken software, and LAV was not going to have any of that. Something that served me well over the years - helps keeps ones sanity.

As for "speaking with them directly", I just don't have that kind of time to try to contact Intel, Microsoft, NVIDIA, AMD or whoever on every bug I encounter.
If I knew a good contact person, I might try, but contacts in a position to actually help are impossible to get, and I don't have the time or patience to try to get through first-tier support people, who don't have the expertise required to actually help.

LAV is project I do in my free time, excuse me for not being willing to spent even more time trying to cut through company politics and reach people that can actually help me.

[nussman]

Yeah no stupid hacks for driver bugs, because then we never get improvments at correct place and have to discuss this for every other decoder/software again.

[madshi]

You could send a short note to Eric Gur (egur). Maybe he can move something.

Adding a workaround is always problematic: What if Intel fixes the problem in a couple of months from now? Then some drivers will be broken and some work. Should LAV then contain a database of driver versions which work and which don't work, so that P010 support is enabled for some but not for others? That really doesn't make a lot of sense. It's easy enough for end users to uncheck P010 in the LAV settings if their Intel drivers are faulty.

[NikosD]

LAV isn't behaving "wrongly", its just not trying to automatically work-around Intels or Microsofts bug.
If someone wants to send a patch to make it fallback to NV12 in that case, please do so, but my development system doesn't have a Intel GPU right now, so I would need to setup another one to even start working on that - which cuts into the time factor again. I might do it some time, but apparently NikosD expects results within days, a few of which I was on vacation, too =p


I have sent you a private message on July 16th with the specific code above and what Intel said about the problem and you didn't reply at all.

That is what I say an attitude problem.

If you had replied that I'm busy or I don't have time to take a look - no problem.

So, I posted the issue here about Win 10 fix and huhn replied that is not working either.

You didn't comment on that too.

We all know that Intel's support has its own problems and they fix things after months or years sometimes depending on the priority they give.

It's a huge and heavy company with user communication issues.

But here in this forum, the same "heavy" attitude is inappropriate at least, between developers and users.

The direct communication between developer and the company would always be the fastest and most appropriate way of solving such specialised issues.

I will probably make a final contact with Intel asking them to send a patch for LAV filters, but I repeat that if it was Nvidia your attitude wouldn't be the same.

[nevcairiel]

I have sent you a private message on July 16th with the specific code above and what Intel said about the problem and you didn't reply at all.

I was on vacation until Monday (the 20th), only rarely checking forums and trying to ignore most mails or other communication to not spoil my vacation too much. I would've responded eventually, but I had lots of stuff piling up to work through.

If you think taking a vacation is attitude, then I'm sorry for you =p
Maybe you shouldn't jump to conclusions. Responding a couple days later isn't ignoring, I often delay a response until I had time to actually think on an issue instead of giving a pointless generic response.

PS:
I didn't add a work-around for NVIDIA either when they had the same problem. They fixed it though, so that will probably not convince you of anything. :p

[NikosD]

Do not go on on this subject because I will make it personal too.

You could have sent "I'm on vacations right now" instead of posting on other threads here in this forum, on the exact same day.

Get on the ground.

[NikosD]

Adding a workaround is always problematic: What if Intel fixes the problem in a couple of months from now? Then some drivers will be broken and some work. Should LAV then contain a database of driver versions which work and which don't work, so that P010 support is enabled for some but not for others? That really doesn't make a lot of sense. It's easy enough for end users to uncheck P010 in the LAV settings if their Intel drivers are faulty.

The problem is that Intel doesn't seem to acknowledge that it's their bug.

And I surely can't persuade them for that.
Only the developer of LAV filters could do that.

[madshi]

If they don't acknowledge their bugs then it's their fault, not nevcairiel's. You're barking up the wrong tree here.

Our responsibility as developers is to fix our own bugs. It's not our job to fix other people's/software's bugs, nor to convince other developers that their software is buggy. Still: Give us a link to an Intel GPU driver bug tracker, and we'll gladly report all their bugs there (I might have some to report, too). Or give us a contact person in the GPU driver department we can talk to, who actually has the power to get things fixed. And we'll gladly talk to him. But none of us devs has the resources to spend hours and hours on trying to talk to some support staff who isn't willing to acknowledge any bugs, nor has the power to get bugs fixed.

[nevcairiel]

You could have sent "I'm on vacations right now" instead of posting on other threads here in this forum, on the exact same day.


So now I am evil because I used my phone to post on the forum a few times but didn't respond to PMs, while being on vacation? Right...
Next time I won't access the forums at all, will that make you happier? :P