LAV Filters - DirectShow Media Splitter and Decoders

[nussman]

HW Deinterlacing settings is for CUVID (Nvidia) or QS (Intel) only, but hardware deinterlacing is done anyway if the videorenderer supports it.

[mzso]

HW Deinterlacing settings is for CUVID (Nvidia) or QS (Intel) only, but hardware deinterlacing is done anyway if the videorenderer supports it.

I remembered that it used to work with DXVA as well with AMD cards.
Anyway the renderer (madVR) fails to use HW deinterlacing. It shows enabled, but in fact it's still interlaced with a lot of combing.

[nussman]

With D3D11 HW deinterlacing is not supported by madVR.
Try DXVA2 (native) or DXVA2 (copyback).

[mzso]

With D3D11 HW deinterlacing is not supported by madVR.
Try DXVA2 (native) or DXVA2 (copyback).

I see. That is a change I made recently, because the player hangs very frequently with dx9.
So, I guess have to make peace with yadif, or seek some other solution?

"Try DXVA2 (native) or DXVA2 (copyback)"
Ummm. I did. Hence my original post of HW deinterlacing not being available for any of the HW decoder options.

[nussman]

But HW deinterlacing works with DXVA2 (native) or DXVA2 (copyback) and madVR.
HW deinterlacing setting is for CUVID or QS ...

[mzso]

But HW deinterlacing works with DXVA2 (native) or DXVA2 (copyback) and madVR.
HW deinterlacing setting is for CUVID or QS ...

Not sure what you're getting at. madVR's deinterlacing is independent of LAV, as far as I know.
It used to work with with SW decoding as well.

[nussman]

HW deinterlacing in LAV Filter works only for CUVID or QS.
In all other modes like DVXA2 nativ/cb, D3D11 nativ/cb or software decoding, hw deinterlacing is done by your videorenderer.
Only for D3D11 native it doesnt work with madVR, because its not supported (yet ... I hope so).
If hw deinterlacing doesnt work for you with these modes its AMD driver bug or wrong settings (i.e. disable deinterlacing in madVR or disable output formats in LAV).

[nevcairiel]

HW deinterlacing in LAV has never been available for anything but CUVID or QuickSync. For those its directly built into the decoding process, and not a separate step, so it was an easy addition.

[mzso]

HW deinterlacing in LAV has never been available for anything but CUVID or QuickSync. For those its directly built into the decoding process, and not a separate step, so it was an easy addition.

I guess I misremembered then. It has been a while since I used my 2600XT.

[VictorLS]

nevcairiel
It seems YADIF and both Weston Software Deinterlacing aren't working on H.264 4:2:2 1920x1080i50 ~15 Mbit/s MBAFF 10bit file Formula1.2019.Round06.Monaco.Race.Sat.Feed.1080i.H264.Multi.Language_fromMKVtsMuxerNotHDMVcompatible.ts (501 MB) https://yadi.sk/i/X1nPqEAqIAZXoQ recorded from SAT feed because madVR (namely madVR - not other known renderers which haven't colorsift issue as minimum on nVIDIA's videocards) always show two dxva rows in the bottom of Ctrl-J statistic, CPU load isn't increasing and colorshift issues best visible on red (i.e. at 44-46 seconds watch on CASINO CAFE DE PARIS or at 1min22s RedBull aside of same named F1 car) don't go away. While playing H.264 4:2:2 1920x1080i50 ~20 Mbit/s 8bit 20200216-181539_RUS-070 main +79199986019.ts (90 MB) https://disk.yandex.ru/d/TPRLVj_q06taeg (especially on ZDF badge on red jacket over woman's heart at 26-27s) activating any of three Software Deinterlacing solve colorshift issue with madVR (EVR and MPC-VR has not that colorshift issue with hardware deinterlacing) and CPU load dramatically increasing on my CPU Athlon II x3 360 and two dxva rows in the bottom of madVR's Ctrl+J statistic are absent.
Here's screenshot with madVR's Ctrl-J statistic where colorshift issue well visible as dark ghost letters lefter and upper of CASINO CAFE DE PARIS billboard
https://i.postimg.cc/3d85YpFB/4-2-2-mad-VR.png (https://postimg.cc/3d85YpFB)
Btw VLC can't fluently deinterlace Formula1.2019.Round06.Monaco.Race.Sat.Feed.1080i.H264.Multi.Language.mkv with activated Yadif (x2) deinterlacing - terrible stuttering present - but VLC try though ;)
One more thing: old and last ffdshow_rev4533_20140929_clsid does double framerate YADIF deinterlacing quiet well (without colorshift issue) while playing Formula1.2019.Round06.Monaco.Race.Sat.Feed.1080i.H264.Multi.Language.mkv (in the bottom of madVR's Ctrl+J statistic two dxva rows are absent).

About HW deinterlacing in AMD Ryzen 3 3200G with Vega 8 (tested in Win10x64 2004) - it's mostly bad quality (colorshift issue present while playing two 4:2:2 above files with all available in MPC-HC renderers (so not only with madVR as with nVIDIA's videocards) and some interlaced areas (interlaced "islands") still present near nose 01520-gamak.mkv (295 MB) (https://drive.google.com/file/d/12BPbVtatluYtdXKTFS2X7ShOv8pkOK6X/view?usp=sharing) - see frame-by-frame Screenshots.7z (65 MB) https://transfiles.ru/iflq3 - compared with nVIDIA doing almost ideal HW deinterlacing while playing 'gamak' VC-1 file) so for high quality deinterlacing one of Software Deinterlacing of LAV Video Decoder must be used (may be yet in case of AMD correct that bad deinterlacing issue in a future).

[VictorLS]

AMD VP9 hardware decoding works with 0.74.1.59 thanks.
After some research I've found AMD correct in their Adrenalin driver not working hardware acceleration of VP9 and VP9 HDR HLG (tried on Ryzen 3 3200G with Vega8) and with win10-...-20.5.1-june10 and newer driver hardware acceleration of VP9 works well with 0.74.1.58 and some older versions of LAV Video Decoder (with 20.4.2-may25 and 20.2.2-mar5 drivers hardware acceleration of VP9 works with 0.74.1.59 and newer only) so commonly it was unnecessary correction in LAV Video Decoder but not all frame decoding (i.e. decoding I-frames only present in ffmpeg) is necessary for me and other SAT amateurs still not implemented (

[v0lt]

VictorLS
You are boring. You are asking for what you have been told "no" several times. I advise you to stop, you only annoy people. Because of this, you will simply be included in the "Ignore List" (if not already).

[VictorLS]

v0lt
If your decoder (which has 'not all frame decoding' but hasn't I-frame decoding only) would be such good as LAV Video Decoder (i.e. support NVIDIA CUVID and stable 8K decoding) I'd use it and don't use LAV Video Decoder at all but we SAT-amateurs hasn't such alternative - LAV Filters the BEST and we want it'll still so in a futurewith new useful functions.
Besides I've heard 'no' from nevcairiel many times here on another my useful for SAT-amateurs asks like 10bit (may be 12bit too - I can't check by myself - my GTX750v2 supports only 10bit) NVIDIA CUVID decoding (it has been implemented but too late for me because XP hadn't supported by LAV Filters that time already) or possibility of deactivating DXVA processing while NVIDIA CUVID decoding to normal watch 'broken streams' on nVIDIA videocards with hardware acceleration...so I just don't loose hope with question of not all frame decoding because it's such normal ask as described above moreover it can be useful not only for SAT-amateurs but all people with not so powerful computers to decode all frames encoded with i.e. AV1, H.265 (H.266 and so on in a future) etc...

[nevcairiel]

Skipping frames will never happen. Its not an improvement in decoding, its a step backwards. If your system is too slow, get a faster one. LAV's primary goal is to decode as much as it can perfectly. It'll *never* get any feature that'll make it decode an image in a bad way just because its faster.

[VictorLS]

If your system is too slow, get a faster one.
You still don't understand there's many causes it's impossible, for example:
1. Not all people in the world have enough money to buy/upgrade their computers;
2. There isn't power enough computer to decode all frames at all (excluding many processors mainframe not available even for rich people).
I.e. can processor of your PC play 8K AV1 http://forum.doom9.org/showthread.php?p=1903584#post1903584 in 3 and half minutes (not more) and how many fps in renderer or DXVAChecker (screeenshot?) do you see? If yes (60fps) how much does it costs? I even don't know what processor need to decode H.266 VVC 8K - from the beginning of Digital Video Broadcasting since the end of 20th century I can say: modern middle-end processor always not enough for modern codecs (videocards doesn't hardware acceleration just appeared codecs for a long time by default as H.266 VVC now - even newest nVIDIA Ampere can't accelerate VVC).
If, in case of i.e. youtube, there's way to get video with lower quality to almost any computer can play video fluently with original fps (and audio/video synced) in case of SAT it's impossible ;)
New idea for you if you like quality (i.e. like in question of NVIDIA CUVID without DXVA processing) - now LAV Video Decoder gives terrible gray all frame artifacts after some normal frames if CPU don't so powerful to decode i.e. 8K stream from SAT live (and video can't be accelerated by videocard) leading to disappearing audio so repeat if you like quality can you change strategy of LAV Video Decoder's decoding (it's much harder to do than just only I-frame decoding done in professional Mainconcept Video Decoders long time ago so I haven't ask it before) to when audio becomes non-synced to video LAV Video Decoder drops decoding past frame and skips to decoding to nearest current (lifetime) I-frame to let audio be without silence and any other artifacts just skipping video without any additional settings in LAV Video Decoder?
I hope you understand me now.

[Asmodian]

I don't agree. If your computer cannot decode it properly then skipping frames is not higher quality. You might be more willing to watch it, and it is higher quality in that sense, but from a "quality video decoder" point of view it is a low quality hack. Also, playing games with the word "quality" to get nevcairiel to agree with you seems like a weird approach.

There is obviously no content in a format no one can decode in real time so you are asking for a hack to allow watching of test clips. Or, at least using test clips to try to argue for the slow computers without arguing for slow computers. :p

[VictorLS]

I don't agree. If your computer cannot decode it properly then skipping frames is not higher quality.
http://forum.doom9.org/showthread.php?p=1902471#post1902471 but ask you personally if you are i.e. watching socker from SAT (and no other way to get this match from - i.e. Internet absent) and your CPU (video don't accelerate this at all) power isn't enough for such quality do you prefer live (see current i.e. 0:1) with normal commentary with skipping frames or very slow motion with 0:0 and with silence and see goal in some minutes? I prefer live with most of other fans ;)
PS. Not watching match or buy new processor (now it almost always means additional buying new mainboard and memory so cost many money) are wrong answers (see my post above).

[nevcairiel]

I hope you understand me now.

I understand that you will never stop, and if you keep repeating it you will end up on ignore, permanently.
If your system is too slow for realtime playback, convert the video to a format that you can play in realtime. Problem solved.

[VictorLS]

If your system is too slow for realtime playback, convert the video to a format that you can play in realtime. Problem solved.
If you wrote so you don't understand (I was always talking about live watching in this thread) even my easy to understand question in the post (http://forum.doom9.org/showthread.php?p=1919411#post1919411) to Asmodian (((
I don't say about answer on my questions (http://forum.doom9.org/showthread.php?p=1919400#post1919400) to you about decoding 'old' AV1 (compared to new VVC) coded files on your PC...
PS. Can you confirm LAV Video Decoder can't activate any software deinterlacing for files like Formula1.2019.Round06.Monaco.Race.Sat.Feed.1080i.H264.Multi.Language_fromMKVtsMuxerNotHDMVcompatible.ts (501 MB) https://yadi.sk/i/X1nPqEAqIAZXoQ I'm written here http://forum.doom9.org/showthread.php?p=1919062#post1919062? Are you going to correct this? Ryzen 3 will go from me next week so I'd want to try new corrected version of LAV Video Decoder before ;)

[SamuriHL]

I've got a weird situation that I genuinely don't know where the problem lies, exactly. Hence I'm not sure this is the correct thread for it but I'll start here. I have an MKV with what it claims to be stereo LPCM audio. In reality, it's mono LPCM. During playback with JRiver set to use LAV Audio for bitstreaming, I indeed get one channel. However, it's only coming out of the left front speaker. I'm not entirely sure if it's a consequence of how it was ripped to MKV (this is the Bruce Lee blu-ray collection that was just released), a JRiver weird issue, or if LAV is doing something odd. Or it could be my receiver, too, I suppose. Like I said, I decided to start here to see what LAV is supposed to do in this case. I'd prefer a mono LPCM track hitting the center channel, but, is that even possible? I also don't have enough knowledge about how LPCM audio is stored in MKV and whether it's even possible to store a single channel...which probably explains why it's showing up as stereo. Thanks!

[mkver]

Matroska supports mono audio. You should be able to remove the silent channel and make the other channel center with FFmpeg's channelmap filter. It will also save space.

To find out whether this is a bug in the remuxing application or a defect/feature of the source you should play the source directly or remux the source with different tools and inspect the output.

[v0lt]

@SamuriHL
You can see the properties of the input pin for the audio renderer (or the output pin of the LAV Audio Decoder). In most cases, single-channel audio is not matched by the decoder to any speaker. This is done by the audio renderer or driver.

[clsid]

LAV Audio has an option to expand mono to stereo.

[SamuriHL]

Thanks all. That gives me a good place to start. When playing it on the shield I get nothing but a high pitched static so I'm guessing it wasn't ripped correctly to MKV. I'll mess around with my options.

[LigH]

I get nothing but a high pitched static

That sounds to me like you may receive a digitally compressed format which is not recognized to be decoded, or a sample format which is not supported (like IEEE float, and your device only supports integer PCM).

[SamuriHL]

That sounds to me like you may receive a digitally compressed format which is not recognized to be decoded, or a sample format which is not supported (like IEEE float, and your device only supports integer PCM).

Definitely not compressed as it's LPCM but I suspect the SHIELD is confused about the fact that it's a mono track defined as stereo in the MKV. I didn't get back around to messing with this over the weekend as I got distracted with the ability to finally play Dolby Vision + Dolby ATMOS MKV's on my SHIELD. I will look at this issue again tonight and see what I can find.

[el Filou]

@SamuriHL I've encountered this in the past with old movies with mono audio. I can't find my old posts quickly, but it's something to do with HDMI, 1-channel is not a standard layout, only 2;5.1;7.1 are. My solution is the option in LAV to upmix mono to stereo, and then use something like ProLogic in the receiver that will remap all sound to the center channel. It gets even more complicated with some Blu-rays that have stuff like 3.0 channels as you need to enable mixing only for those. I don't know how stand-alone Blu-ray players manage that, I guess they also upmix mono LPCM to 2-channel.
You still don't understand there's many causes it's impossible, for example:
1. Not all people in the world have enough money to buy/upgrade their computers;The recommended option if your computer is not powerful enough to play satellite broadcast feed and buying a more powerful computer isn't possible seems to me to buy a stand-alone dedicated satellite receiver. Surely all the people who want to watch satellite broadcasts are not using computers? I understand there are advantages to using a computer to watch DVB as I do it myself, but it's definitely an expensive way to do it and if it gets too expensive it's better to try other options than to watch a crippled stream with an underpowered computer.I-frame decoding done in professional Mainconcept Video DecodersJust my 2 cents, but that kind of option present in some video decoders is most probably not for normal playback, but for quicker decoding preview while doing things like editing.

[SamuriHL]

@SamuriHL I've encountered this in the past with old movies with mono audio. I can't find my old posts quickly, but it's something to do with HDMI, 1-channel is not a standard layout, only 2;5.1;7.1 are. My solution is the option in LAV to upmix mono to stereo, and then use something like ProLogic in the receiver that will remap all sound to the center channel. It gets even more complicated with some Blu-rays that have stuff like 3.0 channels as you need to enable mixing only for those. I don't know how stand-alone Blu-ray players manage that, I guess they also upmix mono LPCM to 2-channel.


Brilliant. I will definitely try this out. I haven't had time to get back to it yet. Work has been a little crazy. But I definitely want to get this resolved. Thanks!

[VictorLS]

I understand there are advantages to using a computer to watch DVB as I do it myself
My greetings
The recommended option if your computer is not powerful enough to play satellite broadcast feed and buying a more powerful computer isn't possible seems to me to buy a stand-alone dedicated satellite receiver.
Do you know any can play 4K 4:2:2 h265 50fps 20180428-175510_4K ENC 3 RMAD VS LEG.ts (563 MB) https://yadi.sk/d/m0s5ztpj3XnUyw ? ;)
Btw I don't 100% sure i.e. AzBox (I don't say about other even expensive but ordinary tuners) can 4:2:2 h264 (not h265) 1920x1080i50 (but your processor seems can't play this file Formula1.2019.Round06.Monaco.Race.Sat.Feed.1080i.H264.Multi.Language_fromMKVtsMuxerNotHDMVcompatible.ts (501 MB) https://yadi.sk/i/X1nPqEAqIAZXoQ fluently but due to nVIDIA videocard without color issue well visible on Vega8 with 50Hz native output as described above (http://forum.doom9.org/showthread.php?p=1919062#post1919062) - just ffdshow Video Decoder with yadif doubleframerate play it with high quality).
it's definitely an expensive way to do it and if it gets too expensive
Special standalone receivers too expensive too and they are don't upgradeable so much more expensive than computer in long term.
it's better to try other options than to watch a crippled stream with an underpowered computer.
Now I don't know other way to watch 4:2:2 live - just skipping frames - that's why ask implement this option in LAV Video Decoder. May be in nearest future Intel Xe videocard will hardware decode 4:2:2 (and even H.266 VVC) but now nothing!!!
Just my 2 cents, but that kind of option present in some video decoders is most probably not for normal playback, but for quicker decoding preview while doing things like editing.
We don't want normal playback - just normal preview of hard feeds (even H.266 VVC feeds - when standalone tuner with support of VVC in stock - in some years? - app SmartDVB can receive H.266 now - we just waiting for DirectShow H.266 decoder to try (if test translation still at https://www.flysat.com/astra28-uhd.php 11973v31000 that time) to watch live and skipping frames feature will be appreciated so VVC 4K must be much harder to decoding than HEVC 4K (as was before in line MPEG2 SD-MPEG2 HD-AVC HD-HEVC 4K middle-end processor can software decode all frames in 3-5 years after apperaing those standards - we don't wan't wait so long - we want to watch live slideshow now ;)

[Brazil2]

LAV Filters 0.74.1 (released 2019/03/19)
What about a new release ? ;) :)

[el Filou]

Do you know any can play 4K 4:2:2 h265 50fps [file] ? ;)
Btw I don't 100% sure i.e. AzBox (I don't say about other even expensive but ordinary tuners) can 4:2:2 h264 (not h265) 1920x1080i50I don't know about Sat as I'm using cable, but I don't understand why TV providers would use certain unusual codec profiles if hardly anybody would be able to receive them?

[VictorLS]

I'm using cable
So you receive and watch DVB-C - I'm DVB-S2(X).
I don't understand why TV providers would use certain unusual codec profiles if hardly anybody would be able to receive them?
On my own SAT transmitting providers use 4:2:2 (now more seldom than before) to achieve more quality picture after logos, tickers and so on and recoding to 4:2:0 for the end users like you but in most cases (except color issue (http://forum.doom9.org/showthread.php?p=1919870#post1919870) need to use software deinterlacing) it isn't problem for LAV Video Decoder

[ryrynz]

What about a new release ? ;) :)

He'll probably do one after the next round of updates, might have one or two things he wants to tick off for the next version.

[Rusty100]

Question regarding LAV Splitter's queue. Would any problems arise if I were to set the Maximum Queue Memory to, say, almost as much ram as I have?
I just got a new hard drive, and it's the loudest thing ever when at low read/write speeds, say the normal 8-14mbs while playing a film. My idea to solve this is to basically have it queue up as much as possible as it's playing, which forces the drive into ~250mbs speeds, which lo and behold, produces no hard drive noise at all.
And if I do do this, do I increase the queue packets too? And to what? I guess my main question is, am I damaging my hard drive or my ram? Haha. Or maybe if there's a better way to get higher read speeds during film playback.

[sneaker_ger]

To quote the author:
[...]it successfully buffers a full BD file at 20GB for me now.
One thing to note however is that the buffer is not "smart", if you seek it'll drop the entire buffer, do the seek in the actual file, and then re-fill the buffer from scratch. Due to that behavior, seeking is actually really slow, since freeing tens of thousands of small memory blocks is actually quite an intensive task to do. Hence why the default buffer sizes are only as big as I dare make them, since it can slow stuff down.

And yes: as I understand you have to increase both queue values.

[tebasuna51]

Posts deleted by Rule 6:

6) No warez, cracks, serials or illegally obtained copyrighted content! Links to content of a questionable nature (e.g. anything you don't own and/or have downloaded), asking for, offering, or asking for help/helping to process such content in any way or form is not tolerated.

[Fabulist]

Posts deleted by Rule 6:

6) No warez, cracks, serials or illegally obtained copyrighted content! Links to content of a questionable nature (e.g. anything you don't own and/or have downloaded), asking for, offering, or asking for help/helping to process such content in any way or form is not tolerated.

There was no warez mentioned on my posts, but rather from other members, but OK. May I ask why was nevcairiel's response also deleted, which clarified how LAV Audio works in comparison to other programs?

That clarification the developer wrote can stand on its own without any context, it could help someone who has the same issues in the future.

[videoh]

Fruit of the forbidden tree.

[richardpl]

Use some other help site instead.

[NikosD]

Fruit of the forbidden tree. You are one of a kind, but too strict.

All rules are made to be broken, sometimes.

[Rusty100]

To quote the author:


And yes: as I understand you have to increase both queue values.

Oh great. That's good info, thank you! Setting the queue to around 10gb of my available 16, seeking is still quite fast thankfully. For some reason, after the high speeds die down as I hit the maximum allowance, the noise doesn't return during playback. Like forcing it into prolonged high speeds to begin with negates eventual low-speed noise. Go figure. I can't believe hard drives are still like this sometimes. I hate seagate.

[LigH]

I just noticed that some hardware decoders (in my case, Nvidia PureVideo on GTX 1660 Super) may fail decoding AVC with 8 bpc in high resolutions (4K UHD and beyond). Using DXVA2 native or CUVID, I noticed heavy distortions (motion vector related, I guess?) which did not appear without hardware accelerated decoding (avcodec mode). Decoding AVC with 10 bpc or HEVC was fine.

Anyone interested in details? Any suggestions how to possibly configure LAV Filters to fall back automatically in this case?

[NikosD]

Anyone interested in details? A sample of the mentioned 4K H.264 8bit clip would help.
I could test it.

[NikosD]

... may fail decoding AVC with 8 bpc in high resolutions (4K UHD and beyond) Actually, there is no beyond for H.264 HW accelerated decoding.
All HW decoders of three (AMD, Intel, nVidia) stop at 4K regarding H.264 decoding.

[LigH]

Nightsky_2160p.mp4 (https://www.ligh.de/tmp/Nightsky_2160p.mp4) - watch the bottom left third at around 2 seconds for the most obvious glitches, there are a few more at the bottom.

It happens as well on a GTX 1050 Ti.

[nevcairiel]

Strictly speaking hardware decoders are (typically) limited to Level 5.1 or 5.2, your file is Level 6.0, so it might as well exceed the supported rates of the decoder. What part of it does actually exceed anything, I can't say without a much deeper investigation.
I suppose maybe I should limit the level to 5.1 (or 5.2). I have historically not added many level checks because the levels are often wrong - at least on the lower end, but I guess anyone encoding something at level 6.0 or higher will know what they are doing.

My first best guess is that the size of the DPB is being exceeded as Level 5.1/5.2 do not allow 8 ref frames for 4K, and level 6 allows up to 16. If you wanted to confirm, you could re-encode with at most 5 ref frames to stay within the limits of level 5.2 (or directly enforce level 5.2 limits, if possible)

The resolution alone wouldn't cause this, as we've been decoding 4K H264 content for years now, so the above assumption with the DPB seems most likely.

[NikosD]

I'll take a look at the sample when I return home, but since you have already tried Pascal and Turing decoders, I'll try Polaris and Haswell H.264 HW decoders.

The truth is that 8K H.264 is not commercial and H.264 HW decoders have left behind supporting only 4K, although x264 and other encoders can encode H.264 at 8K resolution.

As nevcairiel said, probably someone could encode 4K H.264 at higher level than L5.x which was the highest level at the 4K H.264 era of HW decoders and is mainly used for 4K encoding.

Haswell's and probably onwards Intel's HW H.264 is the fastest (even from Turing) and I have already tested 4K H.264 L5.2 Ref 16 with huge bandwidth (1Gbps) and with success.

However, I haven't tried before 4K H.264 at L6.x using any HW decoder.

Interesting.

[LigH]

The slow and very predictable motion of stars in a nightsky timelapse seems to benefit from long range references. Not much surprising. I guess a very verbose log could reveal interesting details.

I limited the references to 5 (and accordingly the level to 5.1); as expected, no glitches spotted in DXVA2 HW mode.

[NikosD]

I tried all HW modes of Intel iGPU (QuickSync, DXVA2 CB, DXVA2 Native, D3D11) and AMD RX 470 (DXVA2 CB, DXVA2 native and D3D11) that LAV Video (latest nightly) provides with EVR-CP and MPC-VR renderers.

Both HW decoders appear to have exactly the same issue with image distortion, as you mentioned using Pascal and Turing.

The only interesting point I want to report is QuickSync that falls back automatically to SW decoding.

I don't know if LAV or Intel's MediaSDK are responsible for this behavior, but it seems that it is what you asked for.

[LigH]

It may be the safe way to assume, according to Murphy's Law, that material flagged as "beyond hardware limits" regarding Profile@Level will probably cause issues, and fall back to software decoding. With or without a choice for the user in the UI. I don't expect any smarter solution than a P@L based decision. A decoding error that happened in a hardware decoder would probably not even be detectable by LAV Filters as the process using it, only by the human eyes.