LAV Filters - DirectShow Media Splitter and Decoders

[noee]

Anybody try regular old DVDs yet? Stuttering playback of discs and ripped DVDs with MPC-HC/JRiver w/madVr/EVR. Okay when I go back to the release....

[STaRGaZeR]

I'm not sure could be slower, however the Fraps decoder right now does not output "repeat" frames, i still have to redesign that patch after some significant ffmpeg changes, so it might appear to produce less fps.

Yeah, I saw that. However I have a Fraps RGB source here with no repeated frames that decodes painfully slow, around 30fps for a 60fps file, and CPU usage is much lower than 0.55.2, so there's something up there. Changing threads from auto to 1 slows it a bit to around 28fps or so.

EDIT: I just recorded something in 4:2:0 at 120fps and it results in the same bad perfomance, 0.55.2 decodes at steady 120fps/55-60% usage and the newer builds at around 68fps/30-35% usage. Samples on demand but they're huge :p

[Im2bz2p345]

Yes, i can bitstream it as DTS over SPDIF.

You should check in your windows audio device properties that both the checkboxes for DTS and Dolby Digital are checked, as well as at least 48000 Hz and 44100 Hz, otherwise it may not work.
Like this:
http://images.gammatester.com/pics/b9c163aa993c096781abb9e6ef862608.png

DTS-WAV tracks typically use 44100 Hz, and DTS tracks in movies usually are 48000 Hz, so both need to be enabled.

My Windows Audio Device settings are setup just like yours:
http://img.photobucket.com/albums/v36/Im2bz2p345/spdifsettings_zps76f9946b.png

When searching around more online, apparently I'm not the only one in this situation: http://forum.team-mediaportal.com/threads/cant-play-wav-files-with-dts-sound-only-whitenoise.93753/

If I install AC3Filter, I'm sure that it will start working fine when I play the file through regular Windows Media Player (not MPC-HC).

Just trying to get to the bottom of why I'm having trouble with LAV Audio passing uncompressed DTS audio (through a .wav file).

Thanks for your continued help nevcairiel,

~ Im2bz2p345 :)

[wanezhiling]

I casually compared the 4k dxva decoding performance between mpc-hc and lav.

mpc dxva + vanilla EVR (http://i.minus.com/i2UVUVOGUyeBN.png)

lav dxva + vanilla EVR (http://i.minus.com/ibaxD1d9MxXCK0.png)

:scared:

Tested sample is Ducks.Take.Off.2160p.QHD.CRF25.x264-CtrlHD.mkv (a 30fps version), can be downloaded from Google Search.

[nevcairiel]

I casually compared the 4k dxva decoding performance between mpc-hc and lav.

Are you sure these results are accurate?
I tried both mpc-hc and mpc-be. mpc-hc never uses dxva on that 4k sample (on my GTX 680), even with skip all checks set, and mpc-be is equally slow as lav, dropping plenty frames.

Its weird, if i play it in GraphStudio with EVR it seems to work much better with LAV too.

[dansrfe]

I casually compared the 4k dxva decoding performance between mpc-hc and lav.

mpc dxva + vanilla EVR (http://i.minus.com/i2UVUVOGUyeBN.png)

lav dxva + vanilla EVR (http://i.minus.com/ibaxD1d9MxXCK0.png)

:scared:

Tested sample is Ducks.Take.Off.2160p.QHD.CRF25.x264-CtrlHD.mkv (a 30fps version), can be downloaded from Google Search.

Is that the frame rate in the left window? 24.59 vs 29.97?

[Aleksoid1978]

I test this clip on GTX 660Ti. On both decoder - MPC-BE & LAV with DXVA unable to normal playback. Drop frames, jitter.

[nevcairiel]

Here is a test build, which may or may not increase the DXVA2 Native performance on that clip:

32-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-31-g1571c9c.zip
64-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-31-g1571c9c-x64.zip

Some testing would be appreciated to ensure no new errors are caused by this.

While it increases the fps number that EVR shows, the video still looks like its running too slow, in fact, it looks very much the same as before, maybe those numbers are just meaningless?

[wanezhiling]

Are you sure these results are accurate?
mpc-hc never uses dxva on that 4k sample (on my GTX 680), even with skip all checks set
:D MPC-HC ignored your GTX680?

mpc-hc dxva on slow 520 (http://i.minus.com/ictDQG1Fs3idi.png)

mpc-be dxva on slow 520 (http://i.minus.com/ibxquPe6r0je84.png)
I test this clip on GTX 660Ti. On both decoder - MPC-BE & LAV with DXVA unable to normal playback. Drop frames, jitter.
Yes, both them cannot have a normal playback, but at least with vanilla EVR, mpc-hc/be can touch full 30fps while lav is only 24...
Here is a test build, which may or may not increase the DXVA2 Native performance on that clip:

32-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-31-g1571c9c.zip
64-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-31-g1571c9c-x64.zip

Tested, now has a same fps number with mpc-hc/be.:)
http://i.minus.com/i5REC56Otle45.png

While it increases the fps number that EVR shows, the video still looks like its running too slow, in fact, it looks very much the same as before, maybe those numbers are just meaningless?
Yes, same experience. you can try software decoding, just same too.
I dont know what those numbers of EVR means.

[Aleksoid1978]

Here is a test build, which may or may not increase the DXVA2 Native performance on that clip:

32-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-31-g1571c9c.zip
64-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-31-g1571c9c-x64.zip

Some testing would be appreciated to ensure no new errors are caused by this.

While it increases the fps number that EVR shows, the video still looks like its running too slow, in fact, it looks very much the same as before, maybe those numbers are just meaningless?

Nothing is change.

[nevcairiel]

I didn't commit that DXVA2 Native change and reverted it locally because it doesn't really help, current GPUs are just not well equipped for 4K decoding, 30p is borderline, and even with the patch it looks very unsmooth, pretty much the same as it looks without the patch, so not taking any risks.

[wanezhiling]

Yes, AMD cheat us, Nvidia is very slow, only IVB GPUs can handle 4K decoding well.

So the 4K industry Sony announced is still a dream in 2013. :scared:

[nevcairiel]

Yeah, I saw that. However I have a Fraps RGB source here with no repeated frames that decodes painfully slow, around 30fps for a 60fps file, and CPU usage is much lower than 0.55.2, so there's something up there. Changing threads from auto to 1 slows it a bit to around 28fps or so.

I found the reason, the threading code in ffmpeg was still suffering from some bugs after the recent changes, but it should benefit from threading properly again now. :)

[dukey]

Are frames supposed to arrive at the renderer in order ? Some formats .. vc1, look like they are arriving in decoding order, and not presentation order.

[nevcairiel]

Are frames supposed to arrive at the renderer in order ? Some formats .. vc1, look like they are arriving in decoding order, and not presentation order.

Frames always arrive in order.
What can happen is that timestamps are not in order, that usually means the source does not provide correct timestamps.

[dukey]

okay thanks

[Mercury_22]

EDIT : NVM was a failed build on my end of the x86 version

[nevcairiel]

After last update(s) I can't play BDs from index.bdmv with lav (working OK with MPC-BE's internal splitter)

Seems to be working fine here.

[Mercury_22]

Seems to be working fine here.

Sorry was a problem at my end

FFmpeg failed to build for x86 and since I have everything "automatized" I didn't realized I was missing some dlls from lav folder:o

[STaRGaZeR]

I found the reason, the threading code in ffmpeg was still suffering from some bugs after the recent changes, but it should benefit from threading properly again now. :)

Nice, it works fine now :)

I found another issue however. If you start seeking like mad on a Fraps file the memory consumption goes thru the roof, like if there was a memory leak somewhere. I found an avutil crash after seeking a lot too. Maybe they're related.

[nevcairiel]

Maybe it crashed because of OOM?
I may see why its leaking memory, i'll take a look when i'm back at my dev box.

[STaRGaZeR]

It's possible, it seems to top around ~3,6GB here, but it may be a coincidence. When it reaches that it doesn't crash however, video freezes and audio keeps playing. avutil crash is quite random.

[nevcairiel]

At least the leak is now fixed, i'm just going to hope your crash was related to this. :)

[NikosD]

Yes, AMD cheat us, Nvidia is very slow, only IVB GPUs can handle 4K decoding well.

So the 4K industry Sony announced is still a dream in 2013. :scared:

Do you have an Ivy to run a benchmark of the clip with DXVAChecker ?

I'm interested in full potential of QS performance min/max and average, not just playback.

[nevcairiel]

Ok here is another set of test builds which should hopefully now close the chapter on the ffmpeg update, and allow me to get back to the other features i was working on:

32-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-34-gf1f44df.zip
64-bit: http://files.1f0.de/lavf/LAVFilters-0.55.3-34-gf1f44df-x64.zip

Please let me know if you still experience any issues with this build that were not present in the last release.

[STaRGaZeR]

At least the leak is now fixed, i'm just going to hope your crash was related to this. :)

Sweet, the leak is fixed and I haven't found any avutil crashes either :)

Any chance of reworking the repeat frame patch? I've been trying to do it myself and I got it working, but it's not doing what it's supposed to do. I'm a bit lost with the new thread functions :D
Here's what I have now in case you want to give it a shot: http://www.mediafire.com/view/?q3xt0inxog3z8o4
I think the problem is in update_thread_context, but I don't know what to do to make it work as before.

[tahaa7]

I have a problem. When I play an mkv file with a lossy audio stream (aac, ac3) using LAV Splitter in MPC-HC, I do not get "bit-exact" mark in ReClock. When playing a lossless format, it is there. Does anyone know what causes this issue? Is it supposed to be this way? I have also installed ffdshow Audio Decoder, but nothing changes...

[nevcairiel]

"Lossy" formats are output as floating-point audio, which audio hardware typically does not support, so ReClock needs to convert it to an integer format, which obviously is not "bit exact" anymore.

[XPC]

Hi Nev,

I just reported an issue I found each time I try to see any IP channel (rtp) from my local TV provider (Imagenio). It takes few seconds to start but then I can see the channel perfectly during about 40 secs. After that time, the image is frozen and after a while mpc-hc gets paused.

Using VLC works OK.

(Please refer to issue 337 for a log output)

Any ideas on what could be happening?
Thanks

[nevcairiel]

Streaming support is considered experimental, since i have no way to test it in real-world environments properly, so i can't really help with any debugging of it.

[XPC]

May be you could take a look to the debug output I attached to the issue. It could give us an idea on what is happening.

If you want, I can do the debugging, but I would need some support.

[doskabouter]

Hi,

I was just experimenting a bit (seems that I still have some SAF parts installed) with LAV and h264 video, and I get a lot of framedrops using LAV.

When I choose ffdshow, the video is playing smoothly, so hardware is definitely capable of displaying it.

Screenshots of different config/movie in mediaportal here:
ftp://dump@88.159.164.124/LAV/LAV Experiments.zip

Os is vista32, AMD Athlon 4850e with onboard Radeon HD 3200.

Video:
Video
ID : 1
Format : AVC
Format/Info : Advanced Video Codec
Format profile : High@L4.0
Format settings, CABAC : Yes
Format settings, ReFrames : 4 frames
Muxing mode : Container profile=Unknown@4.0
Codec ID : V_MPEG4/ISO/AVC
Duration : 1h 48mn
Bit rate : 8 515 Kbps
Nominal bit rate : 8 968 Kbps
Width : 1 920 pixels
Height : 1 080 pixels
Display aspect ratio : 16/9
Frame rate : 23.976 fps
Resolution : 24 bits
Colorimetry : 4:2:0
Scan type : Progressive
Bits/(Pixel*Frame) : 0.171
Writing library : x264 core 94 r1583 7608d73
Encoding settings : cabac=1 / ref=4 / deblock=1:0:0 / analyse=0x3:0x113 / me=umh / subme=7 / psy=1 / psy_rd=1.00:0.20 / mixed_ref=1 / me_range=16 / chroma_me=1 / trellis=1 / 8x8dct=1 / cqm=0 / deadzone=21,11 / fast_pskip=1 / chroma_qp_offset=-3 / threads=12 / sliced_threads=0 / nr=0 / decimate=1 / interlaced=0 / constrained_intra=0 / bframes=3 / b_pyramid=2 / b_adapt=1 / b_bias=0 / direct=3 / weightb=1 / weightp=0 / keyint=250 / keyint_min=25 / scenecut=40 / intra_refresh=0 / rc_lookahead=40 / rc=2pass / mbtree=1 / bitrate=8968 / ratetol=1.0 / qcomp=0.60 / qpmin=10 / qpmax=51 / qpstep=4 / cplxblur=20.0 / qblur=0.5 / ip_ratio=1.40 / aq=1:1.00
Language : English


Can you take a look at it, or do you need more info?

[itsonlyjustincase]

Hi guys,

I have worked a lot and understood a lot. But that i don't manage to get it :http://www.imagup.com/npic/1178472500.html

Left part of the screen is the 2 videos playing at the sametime and being mixed (that is why there is only 1 output windows)

I had enabled the lav tray icon. As you can see the DXVA2 native is choosen. You can see that it is not in use. It is written "avcodec".

But on the right part is the same video launched with MPC-HC forced with lav (same configuration with DXVA2 native). And that time it's being used well and i can confirm it with GPU-Z graph which shows that the video engine load increases only when i start the video with MPC-HC

My goal would be to ensure my application is using DXVA2

File used is 720p H264 AAC (taken from youtube)

http://npic.imagup.com/1/1178472500.png (http://www.imagup.com/npic/1178472500.html)

[itsonlyjustincase]

My Serato Video app is OpenGL perhaps it's the reason ? Is there any LAV Filters OpenGL version ?

[Konrad Klar]

@itsonlyjustincase
Try DXVA2 (copy-back). Native requires a compatible renderer (not sure what Serato Video app is exactly doing).

[itsonlyjustincase]

@itsonlyjustincase
Try DXVA2 (copy-back). Native requires a compatible renderer (not sure what Serato Video app is exactly doing).

The thing i don't understand is why it's not the lav video decoder that manages the hardware acceleration :s.

I want to use DXVA2 native because i have the best video engine load with it and my GPU is the most used with that setting so my CPU is relaxed the most and allow me to mix in better condition and as well record my mix with 3rd part soft like msi after burner.

Here is a kind of stress test with 6 H264 720p videos loaded at the same time with MPC-HC using LAV and the different settings :

http://npic.imagup.com/2/1178484933.png (http://www.imagup.com/npic/1178484934.html)
http://npic.imagup.com/2/1178484657.png (http://www.imagup.com/npic/1178484657.html)
http://npic.imagup.com/1/1178484660.png (http://www.imagup.com/npic/1178484660.html)

[itsonlyjustincase]

Same test with 12 videos loaded at the same time (still H264 720p)
In the order of the screeshot :
-CUVID
-DXVA2CB
-DXVA2N

http://npic.imagup.com/1/1178485908.png (http://www.imagup.com/npic/1178485908.html)
http://npic.imagup.com/2/1178485807.png (http://www.imagup.com/npic/1178485807.html)
http://npic.imagup.com/1/1178485811.png (http://www.imagup.com/npic/1178485812.html)

[itsonlyjustincase]

In case of some would be curious i also did the test with Intel Quicksync as my asus UX32VD has also intel HD 4000. In that case i forced MPC-HC to use the intel GPU. I couldn't do anything more as the CPU was 100% :

http://npic.imagup.com/2/1178486593.png (http://www.imagup.com/npic/1178486593.html)

[Konrad Klar]

If the codec is used for playback (i.e. video is not decoded faster than in its supposed FPS) what a virtue in higher GPU load?

Native decodes frames to the graphics card memory and passes them directly to the renderer.
Copy-back and CUVID use more CPU time just because they copy frames from GPU to main memory and back, but they do not require compatible renderer and allow for processing a frames before sending them to renderer.

[itsonlyjustincase]

If the codec is used for playback (i.e. video is not decoded faster than in its supposed FPS) what a virtue in higher GPU load?

Native decodes frames to the graphics card memory and passes them directly to the renderer.
Copy-back and CUVID use more CPU time just because they copy frames from GPU to main memory and back, but they do not require compatible renderer and allow for processing a frames before sending them to renderer.

Yes I understand but it is just that visually with the 6 vids for example and dxva2 native, it was playing fluid. It was not the case with CUVID. And the major point is that my my Serato Video software, when I play 1 video the fps is around 40 (the pipe fps) then drops to 20fps with two videos loaded at the same time so it's not fluid anymore (less than 25-30 fps). If i had msi after burner recording with vfw x264 recording optimized to its best performance settings (fast decode, zero latency, etc...) it drops to something like 10 fps and this even if the CPU and GPU are OK :s (yes so freaking weird :s). That is why i thought perhaps using dxva2 native it would stress the GPU a bit more cause i don't understand why it's slow regarding the fact the CPU is around 60% of load, the GPU around 40%, the video engine around 30%, the RAM around 50%, I have a samsung 830 SSD on which the msi afterburner output file is being copied and the output don't manage to stay at a reasonnable 30 fps :s. I thought of a bottleneck in a first time but everything seems normal

[itsonlyjustincase]

It would be awesome to have lav using CUVID+QUICKSYNC at the same time to decode

[itsonlyjustincase]

Or option to force framerate

[Asmodian]

Please! smaller screen shots. :(

The thing i don't understand is why it's not the lav video decoder that manages the hardware acceleration :s.

LAV does manage the hardware acceleration but to leave the video frame in the GPU's memory you need a renderer that knows how to deal with it. If your renderer doesn't you need copy back.

It would be awesome to have lav using CUVID+QUICKSYNC at the same time to decode

HUH? no it wouldn't, that doesn't even make sense. Or do you mean for different videos? Mulitple instances of LAV running at the same time some using CUVID and some using Quicksync?

This is about trying to play multiple videos all at the same time? None of the hardware acceleration options have good support for multiple streams at once as far as I understand it. This is why you start getting dropped frames even with both the GPU and CPU at less than 100%. It isn't something LAV can influence.

[itsonlyjustincase]

Please! smaller screen shots. :(



LAV does manage the hardware acceleration but to leave the video frame in the GPU's memory you need a renderer that knows how to deal with it. If your renderer doesn't you need copy back.



HUH? no it wouldn't, that doesn't even make sense. Or do you mean for different videos? Mulitple instances of LAV running at the same time some using CUVID and some using Quicksync?

This is about trying to play multiple videos all at the same time? None of the hardware acceleration options have good support for multiple streams at once as far as I understand it. This is why you start getting dropped frames even with both the GPU and CPU at less than 100%. It isn't something LAV can influence.

Sorry for the XL screenshots.

Do you agree that using CPU+GPU to encode or decode videos does make sense to have better performance ? That's what the good video application do nowadays to have better conversion times etc...

So why using CPU+GPU+GPU wouldn't make sense ? It's just about being able to take power from everywhere to avoid bottleneck of one of the elements. It's like an application which uses all the cores of your CPU will work with better performance than one using only a single core.

But analyzing all of this just make me realize that i should just admin that it's the Serato Video software (www.serato.com) which was developped like shit. I can open 6 videos using 6 lav instance at the same time and playing fluid, but 2 with the Serato app can't well :s. The rendering from what i understand is crappy.

I just hopped i could be able to find a solution thanks to a kind of magical video codec

[Mercury_22]

Latest versions of LAV DXVA2 (native) are crashing when trying to play Blu-ray menus with MediaPortal and DSLibBluray (http://dslibbluray.svn.sourceforge.net/viewvc/dslibbluray/) there are also black screens instead of video for some menu videos too (this one tested only in DSLibBluray)

Everything it's playing fine with LAV DXVA2 (native) "official" version (from the first page)

P.S. With CB the only problem in all version are the black screens in the menu

[moliko]

I have a issue with the Dolby Pro Logic II.
Enable Mixing is set to 5.1.
Matrix Encoding: Dolby Pro Logic II
Setttings: Don´t mix Stereo sources unchecked
The Output shows 6 Channels but its plays only Stereo.

Is it possible in future version to get LFE Crossover (Bass Redirection) and a Profil Manager for Lav Audio ?

Thanks for your work

[nevcairiel]

I have a issue with the Dolby Pro Logic II.
Enable Mixing is set to 5.1.
Matrix Encoding: Dolby Pro Logic II
Setttings: Don´t mix Stereo sources unchecked
The Output shows 6 Channels but its plays only Stereo.


DPLII is only for downmixing, so that a DPLII-compatible receiver can reconstruct a 5.1 signal (useful in case you only have a stereo SPDIF link to your receiver, for example)
LAV does not do any upmixing right now, because proper upmixing requires a lot of filtering, and simple upmixing sounds terrible for movies.


Is it possible in future version to get LFE Crossover (Bass Redirection) and a Profil Manager for Lav Audio ?

LFE crossover will most likely not happen, and its doubtful a profile manager will ever be added, because it goes against my design goal of simple configuration, and focusing the effort on perfect playback instead, not post-processing.

[wanezhiling]

http://www.sendspace.com/file/h18xmn
jitter in sw/dxva mode, cuvid is fine.

latest lav s + lav v

[nevcairiel]

http://www.sendspace.com/file/h18xmn
jitter in sw/dxva mode, cuvid is fine.

latest lav s + lav v

Ah fun another ffmpeg bug from the recent changes. Its fixed locally, will push when i'm back home.

[wanezhiling]

:thanks:

a question: sw and 2 dxva modes are based on ffmpeg? cuvid and qs are independent? am i right?