LAV Filters - DirectShow Media Splitter and Decoders

[huhn]

if it eats shaders, how would it affect those of us that are pushing the limit with madvr?

obvious you can't use it than. hardware decoder are most useful in combination with a weak CPU.

[Elbart_]

Should be fixed in the next version.

Thanks!

I found another issue with VPx, this time in the video decoder:

When the height is not modulo 2, the decoder shows a green bar at the bottom. A non-mod2 width is no problem.

http://i.imgur.com/2IgLeoZ.jpg (image made from the vp9-file, but vp8 is also affected)
https://www.mediafire.com/folder/nl0gc7dq6z7e5/LAVVideoDecoder_VPx_modulo_2_%3D_1_green

[romulous]

Hi nev,

I added a comment to a closed bug report on your tracker (#479). Assuming you probably didn't get a notification of the addition, was just wondering if you had any assistance you could offer? Comments #5 and #6.

Thanks!

[nevcairiel]

I found another issue with VPx, this time in the video decoder:

When the height is not modulo 2, the decoder shows a green bar at the bottom. A non-mod2 width is no problem.

http://i.imgur.com/2IgLeoZ.jpg (image made from the vp9-file, but vp8 is also affected)
https://www.mediafire.com/folder/nl0gc7dq6z7e5/LAVVideoDecoder_VPx_modulo_2_%3D_1_green

They look fine here, might be the video renderer causing the problem.
Note that non-mod2 height with 4:2:0 chroma subsampling is technically an invalid format.

[jiayiming]

LAV doesn't support QHD with HSW/HD4600？
ex:http://www.libde265.org/hevc-bitstreams/tos-4096x1720-tiles.mkv
but,the latest pot can play this file..
http://i.imgur.com/IR7lnfv.jpg
--------------------
BTW,HSW also eat shader like Kepler cards？

[NikosD]

The clip above plays fine here with both LAV DXVA x86/x64

[jiayiming]

The clip above plays fine here with both LAV DXVA x86/x64
thanks,maybe my drivers is too old or some other problem.

[mindbomb]

I just heard arcsoft went out of business. This makes software dts-hd decoding more difficult in the future.

[nevcairiel]

The decoder still works, doesn't it. Maybe an open source alternative will be available soon as well.

[SeeMoreDigital]

I just heard arcsoft went out of business. Something has indeed changed over at Arcsoft... As their TotalMedia Theatre (http://www.arcsoft.com/totalmedia-theatre/?icn=foot&ici=click) media player is stated as being: "no longer maintained and updated"...

[vood007]

I have a problem when LAVAudio is connected to FFDShow Raw Audio Proc. Playback does not start on mp4 or mkv files. After a first seek forward it starts. Problem exists since LAV Builds around 16. Sep.

[NikosD]

Is it possible to fix HEVC DXVA2 decoder to fall back to software mode for 10bit HEVC ?

I have tried more than 60 HEVC samples and this little detail is the only incompatibility I have found out.

[huhn]

i guess this will work in the future.
nearly for sure 4K BD will be 10 bit.
these are currently hybrid decoder without fix function so they can be changed and optimized.

[NikosD]

Latest nightly (.47) is crashing using DXVAn with both Nvidia and Intel using HEVC DXVA2 decoder.

Copy-back and CUVID are fine.

Also in previous .46 build, when an incompatible GPU is used like GT610 (Fermi), it allows me to select DXVAn with HEVC and use that mode in HEVC files.

When you try to decode it, it uses CPU only.

I saw your fix for fall back in .47 build, maybe you can include this "strange" behavior for non-compatible GPUs too.

[wanezhiling]

The crash issue was fixed few minutes ago..

So wait for betaking to compile a new build again..:p

[nevcairiel]

Also in previous .46 build, when an incompatible GPU is used like GT610 (Fermi), it allows me to select DXVAn with HEVC and use that mode in HEVC files.

When you try to decode it, it uses CPU only.

I saw your fix for fall back in .47 build, maybe you can include this "strange" behavior for non-compatible GPUs too.

It already falls back to software decoding just fine if your GPU doesn't support HEVC.

Note that your version numbers are wrong. The latest now is 0.62-52, the one that crashed was -50.

[NikosD]

Yes I saw it falling back to sw and I wrote that above, but it was strange to be able to select it and use it on an incompatible GPU.

I was talking about the possibility of not selecting at all HEVC for incompatible GPUs.

But this is minor.

[nevcairiel]

No, the options do not know, and they never will.

[cyberbeing]

The latest LAV Filters git-1d591 build still seems to fail to software fallback with CUVID on every UHD 10bit HEVC TS sample I've tried.

It seems to actually attempt to decode them, but fails miserably with varying degrees of corruption or black output depending on the sample. Same problem reported earlier in the thread by wanezhiling.

http://i.imgur.com/H9v5WvY.jpg
Even 10-bit hevc materials can be also decoded with cuvid mode??:confused:

General
ID : 0 (0x0)
Complete name : D:\Videos\HD\Test\[三星专用.4K.演示片] UHD 4+3混搭.ts
Format : MPEG-TS
File size : 1.38 GiB
Duration : 3mn 51s
Overall bit rate mode : Constant
Overall bit rate : 51.4 Mbps

Video
ID : 257 (0x101)
Menu ID : 1 (0x1)
Format : HEVC
Format/Info : High Efficiency Video Coding
Format profile : Main 10@L5.2
Codec ID : 36
Duration : 3mn 50s
Width : 3 840 pixels
Height : 2 160 pixels
Display aspect ratio : 16:9
Frame rate : 23.976 fps
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 10 bits
Writing library : ATEME Titan KFE 3.5.1 (4.5.1.0)

Maybe just an accident, hehe..

[nevcairiel]

All the 10-bit samples I have tried fallback to software decoding perfectly. Note that it requires that the video profile is actually encoded properly in the SPS (and that a SPS is present in the media type), if either is not true, then it has no data to check against.
Provide one that doesn't work, or there is nothing I can do. Most of my samples are encoded by x265, I didn't take the grand tour through the net fishing for samples, sorry. :p

[cyberbeing]

The two 10bit UHD HEVC TS samples linked in the other thread both do this:
Astra-UHD@50fps-18Mbps (10bit)
https://www.dropbox.com/s/6rbqlrrhhvbz79l/Astra%20UHD.ts?dl=0

UHD_ENT_Transformer_Quad@24fps-51Mbps (10bit)
http://demo-uhd3d.com/files/uhd/Transformers_extinction_Trailer_4K.zip

FWIW, previously this fallback issue also occurred with DXVA2 Copyback (only Native unaffected), but you seem to have fixed that in the past few days. Only fallback for CUVID remains broken on samples like these in the latest build. Unless DXVA2 can somehow fallback much later than CUVID, I'd assume that means the information needed is somehow present.

[nevcairiel]

Both CUVID and Copyback can in theory fallback any time, they just need to check. Native needs to know from the media type, it cannot check any later.
I've fixed CUVID though, it should check more aggressively.

[vood007]

I have a problem when LAVAudio is connected to FFDShow Raw Audio Proc. Playback does not start on mp4 or mkv files. After a first seek forward it starts. Problem exists since LAV Builds around 16. Sep.

Seems fixed now, thanks!

[NikosD]

Latest nightly LAV HEVC DXVA is still crashing for Nvidia in all modes (DXVAn & cb, CUVID)

Intel plays fine - falls back to SW.

https://www.sendspace.com/file/tzzzqi

[nevcairiel]

Works perfectly here. Your setup must be broken in some way, the code doesn't care one bit if you use NVIDIA or Intel.
Note that when playing 10-bit on recent NVIDIA drivers, you need to disable P010 output in LAV Video, or you'll run into a NVIDIA driver bug, at least if you're using EVR.

[NikosD]

The reason I posted that specific clip was, that only for that I clip I get a crash with P010 output enabled.

For all the other 10 bit samples I have tried, it doesn't crash with P010 output enabled using Nvidia.

I get just a black screen with no decoding, which I think it's different than crashing.

So I thought it was just a different case.

Nvidia drivers are buggy lately, more than they used to be.

[nevcairiel]

Its the same bug, sometimes it crashes, sometimes it fails silently. I added another check to make it fail silently more often, but its still the same thing.

[NikosD]

Works perfectly here. .

After the last two fixes in LAV, the decoder is almost perfect for handling 10 bit clips using Nvidia.

The Gravity trailer and all other samples don't crash using P010 output, they give black screen for DXVA native and CUVID.

But DXVA copy-back broke and it doesn't work for any 10bit clips.

So, one last step left...

[nevcairiel]

Copy-back isn't meant to work on any 10-bit clips, it falls back to software decoding just like all the other modes, which it does just fine.

[NikosD]

Well, it doesn't after the last two fixes.

That's why I said it broke, because indeed it was working before the two fixes.

It's crashing during the enumeration, I think.

[nevcairiel]

Works perfectly for me. And the last two fixes are also impossible to influence anything there.

[NikosD]

I tried it many times to all 10 bit samples. Not even one worked in copy back.

Everything else (native, cuvid) is working fine for all clips.

Update:

Besides 740M, I tried latest nightly with a GT610.
Exactly the same behavior with 10 bit clips.

Fall back is working for DXVA native and CUVID, but not for DXVA copy-back (it's crashing during enumeration)

The only common of these different systems (Laptop w/740M and Desktop w/ GT610) is the Nvidia's driver version 344.11 and LAV filters.

[nevcairiel]

Still works just perfectly here, on two systems with NVIDIA. And like I mentioned before, the code just doesn't care if its run on NVIDIA or Intel, and the only changes made in that last version are impossible to cause issues like this. Noone else seemed to complain, either.

[NikosD]

OK.

For my systems, LAV 0.62.55 and LAV 0.62.57 have major differences specifically and only on 10 bit HEVC using Nvidia DXVA.

But maybe it's just me.

Don't expect though, a lot of people chasing latest nightly builds, checking compatibility issues with specific samples.

[NikosD]

I see no evidence that its 10-bit. But no, 10-bit doesn't really work.

I've just tried latest x64 version of PotPlayer using Intel's HEVC DXVA decoder and it can play fine most of my 10bit HEVC samples in HW.

Not all of them, because some they are just crashing the app.

The picture is not perfect, but how is even possible to decode 10 bit resolution in HW using HD 4600 ?

[nevcairiel]

Eh, the difference isn't that big, you can get away with most things by just throwing it at the same code, but anything thats not a perfect picture is absolutely not acceptable, and while a 8-bit decoder can potentially decode a partial picture of a 10-bit stream, its never going to be 100%. The decoder cannot even output a 10-bit picture, only 8-bit images, and as such its already impossible to get a 100% image out of it.

[cyberbeing]

Your latest nightly builds are no longer sending the subtitle styles from the correct segment on certain samples with mkv ordered chapters. This breakage appears to coincide with the recent FFMPEG library version jump, since the LAV 0.62.0 stable using the older libraries is unaffected.

[nevcairiel]

Fixed that just now.

[DragonQ]

Nev, is the source of your custom version of ffmpeg available anywhere? Most of the time when I see an interesting change in the LAV Filters tracker, it's just "update ffmpeg" so I can't see what actually has changed in the code.

[nevcairiel]

Its linked both in the readme or the first post of this thread.

[DragonQ]

Cheers.

[NikosD]

4K HEVC 8bit is crashing Intel's DXVA.

http://demo-uhd3d.com/files/uhd/Hispasat_UHD_NAB-2014_HEVC.zip

CPU plays fine.

Tell me if you want me to cut a small sample.

[cyberbeing]

Your latest nightly builds are no longer sending the subtitle styles from the correct segment on certain samples with mkv ordered chapters. This breakage appears to coincide with the recent FFMPEG library version jump, since the LAV 0.62.0 stable using the older libraries is unaffected.
Fixed that just now.

Thanks. I've now tested and confirmed your change yesterday fixed the issue. So it would seem that FFMPEG depreciating their internal AV_CODEC_ID_SSA format (http://git.1f0.de/gitweb?p=ffmpeg.git;a=commit;h=c7d8dbad14ed5fa3c217a4fc1790021d6c0b6416;js=1) was the underlying cause.

[DragonQ]

Hmm, I found a weird issue today. I wanted to add an audio delay on my HTPC (I used to do this using my AVR but I need that set to no delay for other purposes) so I added a 280 ms delay in the LAV Audio settings. However, now I get a regular ~1 second pause in video every ~7.5 seconds (very regular) in MediaPortal (using EVR). It's entirely reproducible and once I turn off the delay everything's fine. What could cause this?

[Motenai Yoda]

Just me or there are some nasty artefacts with 62.0 and lossless x264 encodes?

[sneaker_ger]

Old x264 lossless encodes are not spec-compliant and the ffmpeg decoder is now following the spec more strictly. It's recommended to re-encode with a current x264 version. The next LAV version will restore the old behavior, though, simply because 99% of the lossless H.264 files out there were encoded using x264. New x264 encodes are compatible with every LAV version.

/edit:
Try this version (or a newer nightly):
http://forum.doom9.org/showpost.php?p=1693385&postcount=18024

[nevcairiel]

LAV Filters 0.63

LAV Splitter
- NEW: Support for playing AES encrypted HLS streams
- NEW: Advanced Subtitle selection allows selecting subtitles by a string match on the stream title
- NEW: Support for rtspu, rtspm, rtspt and rtsph URLs to force the RTSP transport protocol
- NEW: Animated GIF image support
- Fixed: Improved timestamp handling of badly muxed/corrupted H.264 streams
- Fixed: 4K ProRes streams in MKV didn't play reliably
- Fixed: Some HEVC streams in MKV/MP4 didn't play properly
- Fixed: VobSubs in MP4 didn't properly export their color palette
- Fixed: Streaming MP3s through the Microsoft URL filter could result in the last audio frame to be partially repeated
- Fixed: The duration of MP3 files would be wrong if it contained long IDv3 tags
- Fixed: TrueHD streams with an Dolby Atmos sub-stream were not demuxed properly

LAV Video
- NEW: Experimental support for CUVID and DXVA2 HEVC acceleration
- Faster: HEVC decoding is up to 100% faster
- Fixed: DVD subtitle rendering could crash in 64-bit builds

LAV Audio
- Fixed: TrueHD streams with an Dolby Atmos sub-stream did not decode

Download: Installer (both x86/x64) (http://files.1f0.de/lavf/LAVFilters-0.63.exe) -- Zips: 32-bit (http://files.1f0.de/lavf/LAVFilters-0.63.zip) & 64-bit (http://files.1f0.de/lavf/LAVFilters-0.63-x64.zip)

Its been a long time since the last release, which had a multitude of reasons which prevented me from working on LAV as much as I wanted, but I finally finished most of the things I did have planned for this version, so here it is!

Subtitle selection on string match

This feature was contributed to LAV, and it seems at least mildly useful. You can now select subtitle streams based on their title, however its limited to single words, as spaces are a separator for a new token.

Example Syntax:
eng@signs - Select any english subtitle tracks with "signs" in their title.

Note that string matches are case-sensitive, now that I'm writing this, it may make sense to change that to case-insensitive for the next version, maybe?

HEVC HW Acceleration

This version introduces experimental support for CUVID and DXVA2 HEVC HW acceleration on Intel Haswell and recent NVIDIA cards. On the current hardware, this is only a "Hybrid" acceleration, which means it uses the GPUs 3D shaders to perform calculations, and not dedicated decoding hardware like H.264 or MPEG-2 decoding would.

Only 4:2:0 8-bit is supported for now, we'll have to see if the GPU vendors support 10-bit properly at a later time. DXVA2 defines a specification for 10-bit, but its not supported by the GPU drivers (yet?).

Note that I still consider this feature highly experiemental, and it is disabled by default, but it seems to work quite decently nevertheless.

TrueHD Atmos

Really just a bugfix, but maybe still worth mentioning, TrueHD streams with an Atmos substream decode perfectly in this version. Its important to note that the Atmos data is just discarded and not decoded in any way, but the lossless TrueHD stream is decoded perfectly now, while before the Atmos extension screwed up decoding.

Anyway, have fun, and if you find any new issues, please do report them!

[Boltron]

Thank you.

[ryrynz]

Back into the swing of things, feels like a fairly quiet year in software development for AV related things compared to last year. Great release.

[Elbart_]

They look fine here, might be the video renderer causing the problem.
Note that non-mod2 height with 4:2:0 chroma subsampling is technically an invalid format.
When the WebM-reference-decoder is used within mpchc, there's no green line.

I'll do another test-run today.