LAV Filters - DirectShow Media Splitter and Decoders

[e-t172]

Okay, I'll bite. You do realize that both OSes we are talking about were developed by the same company, right?

In addition, can you point me to at least one article from a serious website that claims that Windows XP is even comparable to Windows 10 in terms of security?

Let me suggest an experiment you could try: gather a list of the most reputed security specialists in the field (people like Bruce Schneider, Tavis Ormandy, Troy Hunt, and others) and tell them with a straight face that it's a "joke" that Win10 is more secure than WinXP. I would be curious to see how many facepalms you get.

I could go on and list all the security features that Win10 has that are not present in WinXP (UAC, Secure Boot, ASLR, BitLocker, TPM support, full DEP, service hardening, driver signing, etc.), but I have better things to do with my time. Come back when you have actual data proving that Win10 has more security vulnerabilities than WinXP (or even the same order of magnitude), then we'll talk. I would also be curious to see you explain how an OS that is not getting security updates anymore could possibly be more secure than one that is being actively patched today, because that just defies basic logic to me.

[el Filou]

My 2 cents regarding XP and LAV: I upgraded my HTPC (Core 2 Duo/2 GB RAM, so not a high end setup) directly from XP to Windows 10, and it is massively faster and smoother at absolutely everything, all while only consuming a few hundreds more MBs of RAM.
On my newer desktop PC, it's snappier than Windows 7 too.
Modern OSes give you all kinds of advantages when it comes to media playback, there just isn't any incentive for developers to keep supporting XP, it's a waste of time for everybody, sorry VictorLS.

[VictorLS]

Running a decades-old OS which has stopped getting security updates long ago...
That's offtop here but WinXP (namely embedded version) must be supported until 2020. Updates for WinXP Embedded are good for ordinary WinXP. There are good offline updatepacks for WinXP - I have limited mobile internet only and WinXP not download any byte I don't want, Win7x64 downloads few kilobytes, but Win10 seems can't leave without Internet and when I connect it begins neverending download and upload despite of turned off Windows Updates and others. There is even AntiWannaCry in KB4012598 (ms17-010) for WinXP because necessary for many old programs Windows 2003 Server is almost WinXP )
Besides, as in Win98 time when all virus writers had been written viruses for WinXP I used Win98 until 2006 and hadn't any virus. I came from Win98 to WinXP exlusively due to DirectX (DXVA). Now history repeats with H265, but thanks nVIDIA for CUVID (and LAV Video Decoder for support) - I don't need DXVA (not supported in any drivers for any videocard for H265 in WinXP) anymore.
sorry VictorLS.
Not for - I wrote my experiments with Win10 in this thread before and that's true - Win10x64 is fast on my PC too but boot from second time (after Power on my PC first boot hangs up) and check my HDDs for a long time after and not working my radioremote and Internet consumption and HDD using (head positioning all time) in any Windows after WindowsXP and many other small things )
Thanks.
You're welcome.

[P.J]

Windows 7 is far better than XP. I may use XP on old atom netbooks...

[mzso]

Windows 7 is far better than XP. I may use XP on old atom netbooks...

Since the conversation headed this way... I found 8/8.1 to be the best compromise, for now. It doesn't have all the crap of 10 (with sh*tty older hardware support), and is more efficient than 7.

[P.J]

Or at least disable 4:2:2 modes, that'll force 4:2:2 to 4:4:4/RGB but keep 4:2:0 as-is.

But there's a problem, de-interlacing doesn't work in RGB32 mode at all :(

[huhn]

you could output 4:2:0 this can still be deinterlanced if the source was 4:2:2 but it is generally not a good idea to subsample even more. on the other side 99% of all sources are 4:2:0 anyway.

can EVR-CP/madVR not handle this?

[P.J]

you could output 4:2:0 this can still be deinterlanced if the source was 4:2:2 but it is generally not a good idea to subsample even more. on the other side 99% of all sources are 4:2:0 anyway.

can EVR-CP/madVR not handle this?

Prores videos are mostly 4:2:2. How to output 4:2:2 as 4:2:0?
EVR-CP or plain EVR, same issue :( Haven't tried madVR yet.

Well, I tested Intel HD4600 and it has no issue at all, blaming Nvidia for outputting 4:2:2 badly xD

http://screenshotcomparison.com/comparison/214147

Interesting that Intel can de-interlace in RGB32 mode but Nvidia :scared:
I can upload a short sample if you're interested in :)

[P.J]

Now another topic about Inverse Telecine/IVTC pulldown, Nvidia GTX960 vs Intel HD4600

https://0-t.imgbox.com/wyuAnAkk.jpg (http://imgbox.com/wyuAnAkk) https://1-t.imgbox.com/1VpmnckY.jpg (http://imgbox.com/1VpmnckY)

http://screenshotcomparison.com/comparison/214157
http://screenshotcomparison.com/comparison/214159

Intel has very bad IVTC, sometimes works and sometimes not at all :scared:

[huhn]

it is technically not possible to deinterlance a 4:2:0 source that is transformed to RGB the image is scaled vertical so it can't be deinterlanced anymore. (well there is a very a technically way to do that but... complicated and unnecessary)

so RGB deinterlancing support is just bad.

if EVR-CP can handle 4:2:2 correct you should report this to MPC-HC/BE.
EVR normal has nothing to do with that.

you can force 4:2:0 by disabling everything that is not 4:2:0

[dude2012]

The Brave Bluray has a EAC3 track (French language). When I set Lav Audio to bitstream, my receiver displays Dolby Digital Plus +7.1. If I change Lav Audio to PCM, however, the receiver shows Multichannel 4.0 track. I know it's not an issue with my receiver or Lav Audio settings because with other Dolby Digital Plus 7.1 samples my receiver correctly shows Multichannel 7.1. Any ideas?

[sneaker_ger]

https://forum.doom9.org/showthread.php?p=1803228#post1803228

[nevcairiel]

LAV Filters 0.70.1

LAV Video
- NEW: DXVA2-CopyBack uses D3D9Ex to allow headless operation
- Changed: Increased the maximum number of decode threads to 32
- Changed: Rebalanced the "Auto" thread strategy to use the exact number of available CPU cores, instead of 1.5x the cores
- Fixed: Certain H264 streams could crash in 0.70 due to lack of buffer padding


Download: Installer (both x86/x64) (https://files.1f0.de/lavf/LAVFilters-0.70.1.exe) -- Zips: 32-bit (https://files.1f0.de/lavf/LAVFilters-0.70.1.zip) & 64-bit (https://files.1f0.de/lavf/LAVFilters-0.70.1-x64.zip)

Check the 0.70 release notes (https://forum.doom9.org/showthread.php?p=1809683#post1809683) for additional details.

DXVA2-CopyBack Headless

Using D3D9Ex allows using DXVA2-CopyBack decoding on the primary GPU of headless systems. Note that this only allows using the primary GPU on entirely headless systems, secondary GPUs without an attached screen are not available. This helps on media server systems to utilize DXVA2 for transcoding, for example.

Threading changes and benchmarking

LAV Video 0.70.1 rebalances the "Auto" threading strategy to use one thread for every available CPU "thread" (ie. real cores and virtual cores). Previously, LAV used 150% of the CPU threads to slightly boost benchmarking, however playback results have shown that overly taxing the CPU can cause playback glitches that wouldn't otherwise be present.

When you are benchmarking, this change should be kept in mind, and if needed the thread count adjusted manually to achieve comparable numbers.

In addition, the maximum number of threads (both in auto and manual) has been increased to 32 (from 16), to allow utilizing upcoming new CPUs with higher core counts.

PS:
It may be possible that this is the last version to work on Windows XP. I'm working on some new module that requires D3D functionality thats not available in the Windows SDK used when building with XP support, and using a newer SDK automatically removes XP support.
More details on this will follow as this feature is being developed.

[NikosD]

In addition, the maximum number of threads (both in auto and manual) has been increased to 32 (from 16), to allow utilizing upcoming new CPUs with higher core counts.

Knowing that you are an Intel guy, you should put max. threads to 36, according to the latest HEDT CPU Extreme Edition (XE) from Intel (18C/36T) and not 32, following AMD's HEDT CPUs - Threadripper 16C/32T

Unless of course you are like me and you don't believe that this thing - 18C/36T - can be delivered on time, later this year according to Intel ;)

[nevcairiel]

There is actually a 32 thread maximum limit in a few decoders, so going over that and introducing inconsistent behavior didn't seem very ideal.
Also, 32 is a much nicer number, being a power of 2 and all. :)

But no, I don't believe that CPU will likely launch this year (even though its basically just an existing CPU out of their Xeon lineup anyway, its not like they are making something new). Makes no huge difference to me either way, not something I would buy. :)

[legatinho]

has anyone been able to successfully output AAC 5.1 audio with LAV?

I'm running Kodi Dsplayer, with a 5.1 setup, and everything works properly (DTS, AAC, PCM, etc), but if the audio source is AAC, somehow it gets converted down to 2.0.

Windows is configured as 5.1, LAV is using default settings. I tried outputting to "default audio device", to "directsound: default audio device", to "Onkyo TX-NR709", all give the same result.

If I output to reclock, and use WASAPI, then it works fine (AAC 5.1 comes out as 5.1). If I use Sanear, same problem, 2.0 output.


Any ideas?

[nevcairiel]

LAV Filters 0.70.2

LAV Video
- Fixed: DXVA2 decoding could fail in 0.70 after a dynamic media type change


Download: Installer (both x86/x64) (https://files.1f0.de/lavf/LAVFilters-0.70.2.exe) -- Zips: 32-bit (https://files.1f0.de/lavf/LAVFilters-0.70.2.zip) & 64-bit (https://files.1f0.de/lavf/LAVFilters-0.70.2-x64.zip)

Check the 0.70 release notes (https://forum.doom9.org/showthread.php?p=1809683#post1809683) and 0.70.1 release notes (http://forum.doom9.org/showthread.php?p=1810861#post1810861) for additional details.

One regression fix for DXVA2 decoding in 0.70 which mostly occured with TV playback, as type changes are not common in file playback.

PS:
I originally posted this in the 0.70.1 notes, but to increase visibility I'll repost it here:
It may be possible that this is the last major version to work on Windows XP. I'm working on some new module that requires D3D functionality thats not available in the Windows SDK used when building with XP support, and using a newer SDK automatically removes XP support.
More details on this will follow as this feature is being developed.

Note that this is only meant as an early information. I do not plan to debate this. If new features turn out to really require this change, then it will happen.

[clsid]

has anyone been able to successfully output AAC 5.1 audio with LAV?

I'm running Kodi Dsplayer, with a 5.1 setup, and everything works properly (DTS, AAC, PCM, etc), but if the audio source is AAC, somehow it gets converted down to 2.0.

Windows is configured as 5.1, LAV is using default settings. I tried outputting to "default audio device", to "directsound: default audio device", to "Onkyo TX-NR709", all give the same result.

If I output to reclock, and use WASAPI, then it works fine (AAC 5.1 comes out as 5.1). If I use Sanear, same problem, 2.0 output
Are you using a SPDIF connection to your receiver? That is limited to 2.0 for uncompressed audio (PCM).
You either have to use HDMI connection or re-encode to AC3 audio (which is what reclock does).

[legatinho]

Are you using a SPDIF connection to your receiver? That is limited to 2.0 for uncompressed audio (PCM).
You either have to use HDMI connection or re-encode to AC3 audio (which is what reclock does).

I'm using HDMI, and I'm not re-encoding to AC3 with reclock, so the output is 5.1 uncompressed PCM, and it works fine, the only major difference being that Reclock uses Wasapi.

[huhn]

does it work properly with MPC-HC?
can you check the lav audio status or even make a screen when the problem happens?

[chros]

I'm using HDMI, and I'm not re-encoding to AC3 with reclock, so the output is 5.1 uncompressed PCM, and it works fine, the only major difference being that Reclock uses Wasapi.

You mentioned sanear: try out MPC-BE wasapi implementation (it's not the same as sanear) since sanear has a big bug regarding to the channel numbers.

[VictorLS]

nevcairiel
Not for discussing, but I believe it will be great before stop supporting WinXP release version of LAV Video Decoder (i.e. 0.70.3 or later) which can hardware decode h265 4K 10bit via NVIDIA CUVID in WinXP?

[nevcairiel]

Not for discussing, but I believe it will be great before stop supporting WinXP release version of LAV Video Decoder (i.e. 0.70.3 or later) which can hardware decode h265 4K 10bit via NVIDIA CUVID in WinXP?

I will not spend time on XP support. I have made this point quite clear in the past.

[max79]

Problem with "LAV Splitter". I can't playback these TV-streams:
http://event.radio-satisfaction.de:8550/;
http://81.169.229.251:9500/videopromo
Through players[VLC as example] which didn't using LAV i can hear only audio.

Any feedback would be appreciated.

[max79]

Problem with "LAV Splitter Source". From this radio-stream i got metadata("Now playing" tag) but only through RTMP-protocol, why HLS-protocol didn't show this metatag in Zoom Player? Examples:
rtmp://wse.planeta-online.tv:1935/live/channel_13
http://wse.planeta-online.tv:1935/live/channel_13/playlist.m3u8

And why RTMP-protocol not enabled by default in installer? It works without any trouble.

Any feedback would be appreciated.

[max79]

Problem with "LAV Splitter Source" metadata delivery for all Opus & Vorbis IceCast streams. "get_AuthorName" and "get_Description" functions work but the "get_Title" function fails[The error code is 0x80040216]. Examples:
http://209.126.66.166:10999/192.opus
http://209.126.66.166:10999/192.ogg

Any feedback would be appreciated.

[max79]

Problem with "LAV Splitter Source" metadata delivery. For this Vorbis IceCast stream "Artist" tag didn't delivered, only "Title":
http://ai-radio.org:8000/radio

Any feedback would be appreciated.

[max79]

Please add oportunity to deliver metadata tags from mmsT:// Radio-streams. Now it works only for mms:// streams. Example:
mmst://wm12.spacialnet.com/Julianradio

Any feedback would be appreciated.

[max79]

Problems with "LAVSplitter" & "LAVSource". Seeking didn't work for .DTSMA & .MLP-files and appear wrong mediafile duration[00:00:00] for that reason seeking is impossible. Examples:
https://www.dropbox.com/s/1mevg3htmr7zv7c/Blondie%20-%20Good%20Boys_%5B24bit_5.1_48kHz%5D.mlp?dl=0
https://www.dropbox.com/s/yh6ly2jw0k85tya/Sea%20Wolf%20-%20Old%20Friend_%5B24bit_7.1_96kHz%5D.dtsma?dl=0

Any feedback would be appreciated.

[max79]

Problem with "LAV Splitter Source", i can't play this SHOUTcast stream, source recognize it as <Subtitles>:
http://base-jam-music.com:8155/listen.pls?sid=1

p.s. Today this radio server is offline :(

Any feedback would be appreciated.

[nevcairiel]

I already told you that I'm not very interested in streaming stuff. Streaming is delivered "as is", either it works or it doesn't. If something doesn't work, test with and report to ffmpeg.

[LigH]

FONT="Comic Sans MS"

:o Somehow I am biased... :sly:

[max79]

I already told you that I'm not very interested in streaming stuff...
Okey. But through LAV Splitter_v0.69 i can see built in CUE marks(Chapters) for MP3 remote files(from my SR#15) and now in v0.70(and newer) it's broken, you did not answer why it happened...
And my SR#27 was about local file playback. You did not answer too.

[clsid]

Problems with "LAVSplitter" & "LAVSource". Seeking didn't work for .DTSMA & .MLP-files and appear wrong mediafile duration[00:00:00] for that reason seeking is impossible. Examples:
https://www.dropbox.com/s/1mevg3htmr7zv7c/Blondie%20-%20Good%20Boys_%5B24bit_5.1_48kHz%5D.mlp?dl=0
https://www.dropbox.com/s/yh6ly2jw0k85tya/Sea%20Wolf%20-%20Old%20Friend_%5B24bit_7.1_96kHz%5D.dtsma?dl=0
Any feedback would be appreciated.Raw audio and video files do not have a known duration, and hence seeking will not work. You need to put it into a container format if you want seeking. Duration can only be determined (accurately) by parsing the whole file.

[max79]

Raw audio and video files do not have a known duration, and hence seeking will not work. You need to put it into a container format if you want seeking. Duration can only be determined (accurately) by parsing the whole file.

Thanks again for quick response clsid. But you are wrong! Through other filters duration shows and seeking works too. Have a look:
https://forum.doom9.org/attachment.php?attachmentid=15962&stc=1&d=1499469771
https://forum.doom9.org/attachment.php?attachmentid=15961&stc=1&d=1499469744

[nevcairiel]

He is not wrong. raw vbr audio files have no duration information and you need to read the entire file from start to finish to figure it out, and LAV/ffmpeg does not do this (and will likely not start doing this, either).

[LigH]

VBR MP3 may have an additional non-audio frame with per-frame bitrate statistics (e.g. "Xing VBR frame").

Or not. Then the decoder would have to scan the whole file and build such statistics before playing.

Anyway, no decoder is forced to care about non-audio frames.

[P.J]

it is technically not possible to deinterlance a 4:2:0 source that is transformed to RGB the image is scaled vertical so it can't be deinterlanced anymore. (well there is a very a technically way to do that but... complicated and unnecessary)

so RGB deinterlancing support is just bad.

if EVR-CP can handle 4:2:2 correct you should report this to MPC-HC/BE.
EVR normal has nothing to do with that.

you can force 4:2:0 by disabling everything that is not 4:2:0

The source is 4:2:2 and the problem happens with both EVR/EVR-CP
Also deinterlacing doesn't work correctly when forcing 4:2:2 to play in NV12/YV12 mode, ghost colors...

[huhn]

when EVR CP is creating such artifacts you should contact the MPC-BE MPC-HC team. there is no need to use NN on chroma.

edit:
i got the chroma sub sampling wrong 4:2:2 is half horizontal not half vertical. you can't scale it to 4:2:0 because you can't scale an interlanced image vertical without "deinterlancing" it before.

[VictorLS]

Due to Aleksoid1978 I get hardware acceleration of h265 10bit via NVIDIA CUVID again - diff for 0.70.2.1-git https://yadi.sk/d/9xRWsX343KyoFB - and he said: "in WinXP CUVID not supported P016 output (even for HEVC 10/12bit)".
nevcairiel, before you stop support WinXP can you spend some minutes of coding LAV Video Decoder to force NV12 output in case of decoding h265 10bit via NVIDIA CUVID in WinXP like it done in diff in future releases of LAV Filters?

[TheShadowRunner]

nev did you implement TrackSet/TrackSetEx or am I dreaming!?

[nevcairiel]

There is no support for this, and unlikely to ever be.

[TheShadowRunner]

I'm confused as hell, using LAVSplitter in the graph for certain and getting this:
http://videoff7.free.fr/lav_editions.jpg
Going to explore how it's done..

[nevcairiel]

Editions have been supported for a couple years now. But thats unrelated to TrackSets.

[TheShadowRunner]

Oh right, bad memory on my end. What's not possible with Editions is to pre-define which given audio/sub tracks should be used.

[hubblec4]

nev did you implement TrackSet/TrackSetEx or am I dreaming!?

The TrackSet/TrackSetEx functionality is a Haali-Splitter feature and not a Matroska feature.
It is not possible to build a TrackSetEx with the Matroska-Elements.
Only with multiple editions and the ChapterTrackNumber(s) you could have a kind of TrackSetEx, but if you change the edition(to change the Tracks) always the movie start from beginning.

I would say, TrackSetEx are easy to implement. Its only a Tag like the edition-tags. LAV-Splitter could read this "new Tag".
All editions are shown in the TrayIcon-menu and now here could shown the TrackSetEx also.

When I change an audio or subtitle track, I make a click on the new track and LAV change it for me. Why can't LAV change two things at the same time.
A click on a "TrackSetEx" change first the audio and then the subtitle track.

[DragonQ]

Due to Aleksoid1978 I get hardware acceleration of h265 10bit via NVIDIA CUVID again - diff for 0.70.2.1-git https://yadi.sk/d/9xRWsX343KyoFB - and he said: "in WinXP CUVID not supported P016 output (even for HEVC 10/12bit)".
nevcairiel, before you stop support WinXP can you spend some minutes of coding LAV Video Decoder to force NV12 output in case of decoding h265 10bit via NVIDIA CUVID in WinXP like it done in diff in future releases of LAV Filters?
You know you can submit a pull request in GitHub, right?

I don't think nev is planning to "stop" Windows XP support, I'm pretty sure he stopped supporting it years ago (as did Microsoft).

[VictorLS]

You know you can submit a pull request in GitHub, right?
I don't know why nevcairiel don't want spend about 5 minutes to add h265 10bit hardware acceleration via CUVID in WinXP by implementing 100% working code in LAV Filters https://github.com/Nevcairiel/LAVFilters/issues/145 and then officially stop WinXP support in LAV Filters. All people using WinXP (I often see here in the bottom of posts too) will know i.e. 0.70.3 is last fully functional in WinXP version of LAV Filters and if they have (or buy) GM206-based videocard h265 10bit hardware acceleration via CUVID must work in their old sweet WinXP. I believe that's right.
PS. May be nevcairiel affraid National Security Agency (NSA) prohibit using WinXP to easier watch for people all over the world? ;)

[fijam]

Is the development of LAV Filters going to be impacted by MPC-HC ceasing development?

[LigH]

LAV filters can also be used via DirectShow API (which is used in MPC-HC when you disable internal decoders and splitters), in case LAV Filters might become incompatible to an older build of MPC-HC as substitute for internal filters.