madVR - high quality video renderer (GPU assisted)

[huhn]

here are my first observations from chroma reconstruction.

judged on a screen that isn't subsampling.

soft is working great so far.
but sharp and placebo have some problems.
they boost colors in this case red:

soft: http://abload.de/img/softreconstructbxxsw.png
sharp: http://abload.de/img/sharpreconstruct22x5d.png
placebo: http://abload.de/img/plareconstruct6bucz.png
bilinear: http://abload.de/img/bilineararreconstruct4nubb.png

sharp and placebo are ringing even with AR and placebo is ringing more than sharp i don't see a real difference in ringing with AR or without:

placebo AR: http://abload.de/img/plaarreconstructtgurb.png
sharp AR: http://abload.de/img/sharparreconstruct3au35.png

there are other issues to like the E in RED and again soft is working great there.
performence:
soft isn't taking a lot of performance about the same as nnedi3 16.

my r9 270 can barely do sharp and can't do placebo for a 1080p23 source at 1080p which is pretty heavy...

edit:
found the first major issue with soft:
here are some screens. the issue is most obvious at the D.

soft: http://abload.de/img/softrdgtmswp.png
bilinear: http://abload.de/img/bilinearrdg24sfi.png
bilateral:http://abload.de/img/bilateralrdgf9sca.png
sharp: http://abload.de/img/sharprdgyes45.png
sharp AR: http://abload.de/img/sharparrdgn4ssg.png

[MistahBonzai]

To those running the v0.90 update with MPC-HC x64. Please check that the MadVR refinement settings (like sharpen and such) appear on the screen upon selection. Mine only display their effects after selecting OK and continuing with the video. Upon rechecking I see that this glitch also occurs in my MadVR x86. It seems to be connected to the v0.90 upgrade - at least this is when it first occured for me. MPC-BE/MadVR v0.90 works fine. If someone joins me here I'll provide specifics. As it is I'm running Win 7 64-bit/sp1, HD 7850 (15.7.1) and all the latest related plug-ins from the nightlies.

[Ver Greeneyes]

It depends a whole lot on the sample you're testing with. In some samples Bilateral wins, in some Reconstruction "sharp" wins, in some NNEDI3 wins, in some Reconstruction "soft" wins. Both Bilateral and Reconstruction take the luma channel into account, while NNEDI3 does not. So if the luma channel is helpful for chroma upscaling, Bilateral and Reconstruction may beat NNEDI3, sometimes significantly. But the luma channel might be flat, or even detrimental. In the latter case Bilateral and Reconstruction "sharp" can produce bad artifacts, while Reconstruction "soft" usually at least manages to not make things worse compared to e.g. Jinc or super-xbr upsampling.

Of course it should be said that differences in chroma upscaling quality are often near to invisible. You need to pick specific scenes to really see a difference. E.g. red stuff on black background makes for a great test.

If you want to look at the luma and chroma planes separately, you can toggle between them by pressing Ctrl+Alt+Shift+F12 multiple times.
Thanks for the explanation. Man, that makes it hard to choose though :P I wish there was a way to augment NNEDI3 for chroma to be the best of both worlds - some sort of local structural similarity comparison between chroma and luma to see if bilateral/reconstruction will be effective *handwaves furiously*.

[ryrynz]

the creator must be fucking drunk.
wtf are you doing ?
arnt you supposed to make the renderer better?
why go back to the 80s for graphics and 16 color system?


Thanks for the laugh Patrik. You should really just state your specs and settings though. Likely you'll find you could deselect one of the sharpeners and that might come right. Alternatively you could do a fresh install of your drivers.. The amount of times I've had issues after upgrading drivers makes me always start there first.. Reinstall and then try reverting and see if it persists!

I don't think madshi would appreciate being called drunk.


Well at least not while he's coding anyway.. ;)

Madshi, I think you might need to update your visual comparison graph for Reconstruction, all options for it are identical to Bilateral.

[chros]

Probably you remember that I had/have huge performance difference using D3D9 FSE Old Path on my laptop. I managed to screw up things in the last 2-3 weeks and it works only with 32bit player (luckily found out) and filters now but not with x64 chain!!!
So maybe this is the perfect occasion to figure it out what is going on since I can replicate both now.

What I did in the past 2-3 weeks (I wanted to try out 4:4:4 output on my system but I didn't manage to do it :) ):
- uninstalling/installing intel graphic driver and Management Engine as well (it responsible for HDMI handshake if I understand it correctly)
- upgrading nvidia driver to 361.xx
- updating players, filters
Now I tried to "revert" everything to the state that it was before., you can see it in my signature.

What I tried is using different versions of MPC-HC , madvr , lavfilters (external) but nothing helped.

I have created logs with madvr, but all I can see it in them that it using the igsimsum*.dll -s successfully. I checked them in:
- c:\windows\system32 and c:\windows\system32\NV
- c:\windows\wow*\ and c:\windows\wow*\NV (I forgot the full path for that)
but the correct version are there for both 32 and 64 bit.

So, how the heck can I debug it what's the bottleneck?

MadVR x32 and x64 logs (~ 20 sec playback): http://www1.zippyshare.com/v/YdKiA4Zc/file.html

This is the stat of nvidia inspector (GPU-Z provides the same results) http://www.imagebam.com/image/6c7fde458360826
Rows are from top to bottom: GPU usage, VPU usage, memory, GPU Clock, memory clock, sahders (zero all the time), GPU voltage. x64 chain is on the left and x32 is on the right.

Thanks

[Aleksoid1978]

3D in new version don't work for me. System in signature, Win10 Ent x64, try open test 3DMKV(MediaInfo show that video is Stereo High@L4.1 / High@L4.1) - only 2D. I use PowerDVD/Arcsoft Total Media for 3D BD - all perfect.
That a see - LAV Video Decoder don't accept MEDIASUBTYPE_MVC1 - only MEDIASUBTYPE_AVC1.

P.S. Run MPC-BE on LG 47LM620T desktop. I don't use 3D in Nvidia drivers.

[Shiandow]

here are my first observations from chroma reconstruction.

judged on a screen that isn't subsampling.

soft is working great so far.
but sharp and placebo have some problems.
they boost colors in this case red:


Keep in mind that sub-sampling has a tendency to diminish colours, especially with small letters such as in that image. The sharp and placebo methods might be overcorrecting it a bit, but the soft method might not be strong enough.

[Manni]

3D in new version don't work for me. System in signature, Win10 Ent x64, try open test 3DMKV(MediaInfo show that video is Stereo High@L4.1 / High@L4.1) - only 2D. I use PowerDVD/Arcsoft Total Media for 3D BD - all perfect.
That a see - LAV Video Decoder don't accept MEDIASUBTYPE_MVC1 - only MEDIASUBTYPE_AVC1.

P.S. Run MPC-BE on LG 47LM620T desktop. I don't use 3D in Nvidia drivers.

Exactly the same for me with AMD HD7870 and Catalyst 14.12, but I'm waiting for more info on the Intel dll name and location as an incomplete install of LAV might be the issue in my case.

[Nullack]

Hi Madshi. Thankyou for the new 0.9 version some really fantastic features there. I've done some testing on it. I can say for sure, my system Win 10 x64 with all the latest updates (thats threshold 2+ with win10 now being a service not a SP release method) and my GTX 960 2GB with my UHD display, Im still crashing MPC-BE with the tick box enabled for enable stereoscopic mode even when Im still in a 2D desktop but just have the tickbox enabled. Thats the same as before when I reported it in my old UHD madvr thread. The workaround is the same as what it was before too - users can simply disable that mode and all 2D content plays fine. Im on NVIDIAs latest prod drivers being revision 361.43. If I can provide any help with further diagnostic information just ask, thanks.

EDIT: Auto FSE in D11 results in MPC-BE crashes with 3d stereoscopic nvidia on. Auto FSE in D9 mode results in MPC-BE playing audio OK, but the video is all just a blank screen with 3d stereoscopic nvidia on.

[sheppaul]

Exactly the same for me with AMD HD7870 and Catalyst 14.12, but I'm waiting for more info on the Intel dll name and location as an incomplete install of LAV might be the issue in my case.

It works well here with HD7850 and the latest Crimson driver. You need to install the nightly build of LAV filters to get the 3d MVC decoding libraries (libmfxsw32.dll, libmfxsw64.dll).

PS. Should use lav source + lav decoder + madVR combinations to get 3d MVC output.

[Manni]

It works well here with HD7850 and the latest Crimson driver. You need to install the nightly build of LAV filters to get the 3d MVC decoding libraries (libmfxsw32.dll, libmfxsw64.dll).

PS. Should use lav source + lav decoder + madVR combinations to get 3d MVC output.

Yes, I have cleared this up with Nev in the LAV thread. I had all this installed as confirmed above, but with KCP, which doesn't allow a full install of the 3D libraries in LAV. I'm going to try to install everything separately tonight, and I'll report.

Good to know it works well on AMD.

[Aleksoid1978]

For see 3D BD movie on PowerDVD/Arcsoft Total Media i don't install 3D Vision from Nvidia drivers - and 3D working. But not working with latest LAV Source + Video Decode + madVR. What's wrong, maybe some logs or debug info help ??
Or it's requires install 3d Vision from Nvidia ??

[madshi]

@Aleksoid1978, please see the bottom of this post for 3D requirements:

http://forum.doom9.org/showthread.php?p=1752682#post1752682

Do you have 3D enabled in madVR? If so, what does LAV say what the "active decoder" is in the LAV settings page, while playing such a 3D MKV file? (3D Vision should not be required.)

[Sunset1982]

What are the keyboard shortcuts for madvr's 3dlut splittscreen?

Is there an overview for all madvr keyboard shortcuts?

[MokrySedeS]

madshi, I think there's a deinterlacing regression in the latest version.

http://i.imgur.com/ePy3ry0.png

Here's a sample: http://www39.zippyshare.com/v/x6jBv8gv/file.html
It's an interlaced ProRes file.
With 0.89.19 I get lots of repeated frames but at least it looks ok.

[Aleksoid1978]

@Aleksoid1978, please see the bottom of this post for 3D requirements:

http://forum.doom9.org/showthread.php?p=1752682#post1752682

Do you have 3D enabled in madVR? If so, what does LAV say what the "active decoder" is in the LAV settings page, while playing such a 3D MKV file? (3D Vision should not be required.)

Yes - 3d enabled in madVR. Problem that LAV Video Decoder don't accept MEDIASUBTYPE_MVC1 from LAV Source, but accept standart MEDIASUBTYPE_AVC1 and play as normal H264 stream.

[nevcairiel]

Yes - 3d enabled in madVR. Problem that LAV Video Decoder don't accept MEDIASUBTYPE_MVC1 from LAV Source, but accept standart MEDIASUBTYPE_AVC1 and play as normal H264 stream.

Make sure the Intel DLL is present then, and LAV was re-installed to accept the new type.
If you use it as an external filter in MPC-HC/BE, you may need to refresh its type list in there as well.

[RainyDog]

So short summary: You sound like you hate all of those algos? :p

FWIW, I've noticed myself that "sharpen edges" might be slightly too dark. It will be ever so slightly brighter in the next build. However, I'm not sure if anything will change for artificial test patterns. Due to the way "sharpen edges" works internally, it might not look too great on such test patterns.

I do like artificial test patterns to get an idea of what an algorithm does exactly. But final judgement of an algo's quality should be made with real life content. After all, if you sit down at an evening to use madVR for your entertainment, you're going to watch movies and not look at test patterns. That said, a good test can be movies with hard coded subtitles. They make a good compromise between artificial test patterns and filmed content. They're usually very sharp, but still often have a bit of anti-aliasing in them, which makes them more suitable to sharpening tests than 1 pixel on/off patterns.

No, not at all :p I’m a big fan of HQ sharpening and actually use it all the time, not just to sharpen upscaling softening. These new sharpeners are my favourite addition to madVR in ages.

I prefer sharpening to have a noticeable, pronounced effect and personally think it can subjectively enhance good quality content tenfold. Currently really happy with the combination of Sharpen Edges at 1.4, Thin Edges at 0.6 and Enhance Detail at the default 1.0.

With the test pattern observations, was just pointing out how it lays bare each sharpeners flaws and what areas they generally target on the make up of an image. Agree that it’s not the best way to assess whether you like what they do or judge which one you prefer though. Always fall back on actual viewing content for deciding that and tweaking and tuning.

In fact, Image Enhancement and Upscaling Refinement shortcut toggles would be handy if you wouldn’t mind sometime please :)

[Aleksoid1978]

I do full reinstall LAV.
libmfxsw32.dll/libmfxsw64.dll is present.
In MPC-BE remove all LAV entries and add LAV Souce/LAV Video Decoder again. Don't help - don't accept MEDIASUBTYPE_MVC1(it's present in PinInfo but active is MEDIASUBTYPE_AVC1).

[Aleksoid1978]

nevcairiel
Maybe you do debug build for testing ??

[nevcairiel]

I do full reinstall LAV.
libmfxsw32.dll/libmfxsw64.dll is present.
In MPC-BE remove all LAV entries and add LAV Souce/LAV Video Decoder again. Don't help - don't accept MEDIASUBTYPE_MVC1(it's present in PinInfo but active is MEDIASUBTYPE_AVC1).

If everything is setup properly, there is only two reasons why LAV Video would refuse MEDIASUBTYPE_MVC1

(1) The Intel MediaSDK decoder cannot be loaded for some reason (I doubt it works on XP, for example)
(2) The MVC1 media type signals an incompatible stream or is otherwise damaged

LAV Source doesn't verify case (2), it just sends the data that is contained in the MKV file, so its possible that there is something in there the decoder doesn't like.

[sheppaul]

I've just tried with MPC-BE. It works as expected without any problems. 3D MKV file is created using MakeMKV though.

[chros]

What are the keyboard shortcuts for madvr's 3dlut splittscreen?
CTRL+SHIFT+ALT+3 (I think).
Is there an overview for all madvr keyboard shortcuts?
There's a keyboard shortcut section on the left part of the config interface of madvr.

[clsid]

If you are using CUVID you should switch to DXVA2 Copy-Back. Nvidia drivers are broken on Win10 when multiple things use openCL.

[nevcairiel]

nevcairiel
Maybe you do debug build for testing ??

Here is a debug version, it should leave a log on your desktop:
http://files.1f0.de/lavf/LAVFilters-0.67-57-debug.zip

[huhn]

Keep in mind that sub-sampling has a tendency to diminish colours, especially with small letters such as in that image. The sharp and placebo methods might be overcorrecting it a bit, but the soft method might not be strong enough.

i totally understand that a small line that is subsampled get's dimmer but not a huge solid latter. no other resizer does this. the soft looks like every other resizer on huge solid latter.

your biliteral resizer doesn't do things like that.

[Razoola]

I have a couple of questions related to the New 3D support.

You mention ' Make sure your OS has 3D enabled in the control panel'. By this are you talking about simply having 3D vision enabled in the NVidia control panel or something else?

Also is there any way 3D support will be added for windows 7?

[Balthazar2k4]

I do full reinstall LAV.
libmfxsw32.dll/libmfxsw64.dll is present.
In MPC-BE remove all LAV entries and add LAV Souce/LAV Video Decoder again. Don't help - don't accept MEDIASUBTYPE_MVC1(it's present in PinInfo but active is MEDIASUBTYPE_AVC1).

I have the exact same problem. On my Nvidia based rig it will not pickup the MVC and only plays the AVC content. On my Intel GPU machine it plays the same MVC content (albeit with some stuttering, but it is a Haswell NUC so that could be the issue). Seems to be an Nvidia issue somehow.

[huhn]

newer operation systems have an option like this: http://abload.de/img/3doptionx4sj5.png

[madshi]

I've had this weird thing happen where in the middle of a movie, it suddenly froze up for several seconds while the seek bar indicator kept moving. Then it stuttered a little bit as though I was watching something trying to stream and not getting enough bandwidth. Then the movie caught up to the audio in a time lapse. Look at the queue.

Edit: Happened again in an old MPEG4. I don't think this is decoder or video related.

h264 movie, madVR 0.90 x64, MPC-HC x64, internal LAV 0.66
That does look pretty weird. Is that a new problem with v0.90.0? Maybe a debug log could help. It would have to capture the problem, though. Not sure if can create such a debug log because the log tends to get very big very fast. So if you can't reliably reproduce the problem, it's going to be hard to capture it.

I can't select DXVA option for "Image Upscaling" section with v0.90.0
Oooops. That's a bug, sorry. I'll fix that for the next build.

0.9 reintroduced an old'ish bug with playback freezing on D3D11 ( 1 (https://www.doom9.org/showthread.php?p=1721013#post1721013) 2 (https://www.doom9.org/showthread.php?p=1721676#post1721676)).
Can you double check with v0.89.19 to make sure it's really only occurring with v0.90.0, but not with v0.89.19?

found the first major issue with soft:
here are some screens. the issue is most obvious at the D.

soft: http://abload.de/img/softrdgtmswp.png
bilinear: http://abload.de/img/bilinearrdg24sfi.png
bilateral:http://abload.de/img/bilateralrdgf9sca.png
sharp: http://abload.de/img/sharprdgyes45.png
sharp AR: http://abload.de/img/sharparrdgn4ssg.png
Can I have a sample that shows this problem, please?

To those running the v0.90 update with MPC-HC x64. Please check that the MadVR refinement settings (like sharpen and such) appear on the screen upon selection. Mine only display their effects after selecting OK and continuing with the video. Upon rechecking I see that this glitch also occurs in my MadVR x86. It seems to be connected to the v0.90 upgrade - at least this is when it first occured for me. MPC-BE/MadVR v0.90 works fine. If someone joins me here I'll provide specifics. As it is I'm running Win 7 64-bit/sp1, HD 7850 (15.7.1) and all the latest related plug-ins from the nightlies.
Seems to work fine on my PC.

Thanks for the explanation. Man, that makes it hard to choose though :P I wish there was a way to augment NNEDI3 for chroma to be the best of both worlds - some sort of local structural similarity comparison between chroma and luma to see if bilateral/reconstruction will be effective *handwaves furiously*.
I don't think it's possible to use NNEDI3 that way.

Madshi, I think you might need to update your visual comparison graph for Reconstruction, all options for it are identical to Bilateral.
Oh well, which values to use if the quality so greatly depends on the video sample? Maybe I should remove those graphs altogether... :(

Probably you remember that I had/have huge performance difference using D3D9 FSE Old Path on my laptop. I managed to screw up things in the last 2-3 weeks and it works only with 32bit player (luckily found out) and filters now but not with x64 chain!!!
So maybe this is the perfect occasion to figure it out what is going on since I can replicate both now.

What I did in the past 2-3 weeks (I wanted to try out 4:4:4 output on my system but I didn't manage to do it :) ):
- uninstalling/installing intel graphic driver and Management Engine as well (it responsible for HDMI handshake if I understand it correctly)
- upgrading nvidia driver to 361.xx
- updating players, filters
Now I tried to "revert" everything to the state that it was before., you can see it in my signature.

What I tried is using different versions of MPC-HC , madvr , lavfilters (external) but nothing helped.

I have created logs with madvr, but all I can see it in them that it using the igsimsum*.dll -s successfully. I checked them in:
- c:\windows\system32 and c:\windows\system32\NV
- c:\windows\wow*\ and c:\windows\wow*\NV (I forgot the full path for that)
but the correct version are there for both 32 and 64 bit.

So, how the heck can I debug it what's the bottleneck?

MadVR x32 and x64 logs (~ 20 sec playback): http://www1.zippyshare.com/v/YdKiA4Zc/file.html

This is the stat of nvidia inspector (GPU-Z provides the same results) http://www.imagebam.com/image/6c7fde458360826
Rows are from top to bottom: GPU usage, VPU usage, memory, GPU Clock, memory clock, sahders (zero all the time), GPU voltage. x64 chain is on the left and x32 is on the right.
I don't remember any details. I'm not sure I fully understand your post. Are you saying you have the problem only in 32bit now but not in 64bit? Or are you saying that 64bit doesn't work at all, anymore, now? I'm not really sure. In any case, an Optimus laptop behaving strangely is not really new. If you find a way to make it work well enough for you, then maybe it doesn't make too much sense spending hours on trying to figure out why it sometimes behaves strangely?

Hi Madshi. Thankyou for the new 0.9 version some really fantastic features there. I've done some testing on it. I can say for sure, my system Win 10 x64 with all the latest updates (thats threshold 2+ with win10 now being a service not a SP release method) and my GTX 960 2GB with my UHD display, Im still crashing MPC-BE with the tick box enabled for enable stereoscopic mode even when Im still in a 2D desktop but just have the tickbox enabled. Thats the same as before when I reported it in my old UHD madvr thread. The workaround is the same as what it was before too - users can simply disable that mode and all 2D content plays fine. Im on NVIDIAs latest prod drivers being revision 361.43. If I can provide any help with further diagnostic information just ask, thanks.
Well, I'm not sure why it crashes. On my PC I could reproduce crashes when using D3D11 presentation with 3D enabled, when using v0.89.19, and it no longer crashes in v0.90.0 on my PC. But the crash was always in an NVidia driver DLL. It's not really that I fixed a bug or something, just some refresh rate related changes seem to have worked around the crash on my PC. If it still crashes for you, I've no idea what to do about it. It's most probably NVidia's bug and only they might be able to properly fix it, unfortunately.

EDIT: Auto FSE in D11 results in MPC-BE crashes with 3d stereoscopic nvidia on. Auto FSE in D9 mode results in MPC-BE playing audio OK, but the video is all just a blank screen with 3d stereoscopic nvidia on.
D3D9 always worked for fine me, no blank screen and no crashes, with either v0.89.19 or v0.90.0.

madshi, I think there's a deinterlacing regression in the latest version.

http://i.imgur.com/ePy3ry0.png

Here's a sample: http://www39.zippyshare.com/v/x6jBv8gv/file.html
It's an interlaced ProRes file.
With 0.89.19 I get lots of repeated frames but at least it looks ok.
Hmmmm... I just get a black screen here. But it works with v0.89.19, so there must be something wrong. Will have a look...

No, not at all :p I’m a big fan of HQ sharpening and actually use it all the time, not just to sharpen upscaling softening. These new sharpeners are my favourite addition to madVR in ages.

I prefer sharpening to have a noticeable, pronounced effect and personally think it can subjectively enhance good quality content tenfold. Currently really happy with the combination of Sharpen Edges at 1.4, Thin Edges at 0.6 and Enhance Detail at the default 1.0.

With the test pattern observations, was just pointing out how it lays bare each sharpeners flaws and what areas they generally target on the make up of an image. Agree that it’s not the best way to assess whether you like what they do or judge which one you prefer though. Always fall back on actual viewing content for deciding that and tweaking and tuning.
Glad to hear that!

In fact, Image Enhancement and Upscaling Refinement shortcut toggles would be handy if you wouldn’t mind sometime please :)
Just all enhancements on/off? Or per algorithm or something? Neither is something I can implement in 5 minutes...

I have the exact same problem. On my Nvidia based rig it will not pickup the MVC and only plays the AVC content. On my Intel GPU machine it plays the same MVC content (albeit with some stuttering, but it is a Haswell NUC so that could be the issue). Seems to be an Nvidia issue somehow.
Then please use the debug build nevcairiel linked to a few posts back, create a log file and upload it somewhere for nevcairiel to look at.

[GCRaistlin]

That's not my fault. I'm switching refresh rates based on the framerate information the decoder provides to me. If that isn't correct, then that's the decoder's fault, not mine.
I've reported (https://github.com/Nevcairiel/LAVFilters/issues/62) the issue in LAV Video, and nevcairiel fixed it, at least he said he did; but for me, there's no change in latest LAV+madVR+ReClock behavior for PAL video if it slowed down by ReClock and its framerate is doubled by YADIF: madVR still switches the display to 24 Hz (though ReClock does see that video stream is 50 fps).
With ffdshow, madVR behaves the same way if "automatically activate deinterlacing when needed" is disabled; if it is enabled (default) madVR switches the display to 50 Hz regardless of ffdshow's "Double framerate" option state.


It's not as simple as that. YADIF is a video mode deinterlacer. Forced film mode in madVR is a pure IVTC algorithm, which only works properly with fully telecined sources.
A telecine source is always NTSC, therefore it's incorrect to force film mode in madVR for PALs, right? How to explain the comparison (http://forum.doom9.org/showthread.php?p=1752369#post1752369) result then?

For telecined sources, usually madVR's forced film mode is the best deinterlacer out of those 3
Which 3? YADIF, madVR's forced film mode and DXVA deinterlacing (= madVR's auto mode)?

(Thanks for your patience)

[Asmodian]

A telecine source is always NTSC, therefore it's incorrect to force film mode in madVR for PALs, right?

madVR's film mode does any sort of field matching and duplication deletion, not purely NTSC 3:2 telecine. It can also decimate duplicates in progressive streams.

Shouldn't 24fps video doubled by yadiff be 48fps? Why are you getting 50?

[NanoBot]

Hi,

* added support for 3D frame packed playback via HDMI 1.4+ (requires win8.1+)

could anybody please explain why this feature needs Win 8.1 or higher ?

Thx in advance, NanoBot

[nevcairiel]

could anybody please explain why this feature needs Win 8.1 or higher ?


Because it uses D3D11 to support 3D output on all GPUs, a feature which is only available in Windows 8 or newer.

[iakoboss7]

I've had this weird thing happen where in the middle of a movie, it suddenly froze up for several seconds while the seek bar indicator kept moving. Then it stuttered a little bit as though I was watching something trying to stream and not getting enough bandwidth. Then the movie caught up to the audio in a time lapse. Look at the queue.

Edit: Happened again in an old MPEG4. I don't think this is decoder or video related.

h264 movie, madVR 0.90 x64, MPC-HC x64, internal LAV 0.66

http://i.imgur.com/lhzVW06.png


i can confrim this also. it happened right after i updated.

also this happens all the time

Yes. MPC-HC's window needs to be fully covered by another window or minimized, then it freezes after ~30 seconds being like that. Sometimes it doesn't appear frozen after getting back to\unminimizing it and instead very quickly shows all the frames it supposed to show while being stuck, OSD reports 900+ dropped frames and the playback continues. Pretty weird. I PM'd you a freeze report.

thanks for your amazing work!! :D

[NanoBot]

Because it uses D3D11 to support 3D output on all GPUs, a feature which is only available in Windows 8 or newer.

Thanks for the information.

To clarify that, if I understand it correctly, the new feature uses / needs DirectX 11.2, which in fact is only available with Windows 8.1 or Windows 10.

Thx, NanoBot

[GCRaistlin]

Shouldn't 24fps video doubled by yadiff be 48fps? Why are you getting 50?
I believe that YADIF doubles the source framerate (25 fps) first, doesn't it?

[Asmodian]

Ah, so madVR sees a 25 fps source and thinks Reclock will slow it to 24 fps but LAV doubles the frame rate so Reclock leaves it alone. madVR sets the refresh rate to 24Hz but then has 50 fps video to play?

[fairchild]

Thanks for the latest version of MadVR! I guess now I have a reason to upgrade from Windows7? I guess I am forced to get Win10 or Win8 if I want to use MadVR for 3d blurays when they are supported? I guess it's not that big a deal for me as I have a standalone bluray player.

[GCRaistlin]

Asmodian
I think ReClock slows the source down to 48 fps. At least it wants me to set the display to 48 Hz.

[madshi]

2) 3D is enabled in MadVR settings. If I try to use the options to force the OS 3D enabling, although it shouldn't be necessary, I get a crash (I sent the bug report, please let me know if you need anything else).
For some reason I didn't get the crash report. Can you please upload it somewhere for me to look at?

I've reported (https://github.com/Nevcairiel/LAVFilters/issues/62) the issue in LAV Video, and nevcairiel fixed it, at least he said he did; but for me, there's no change in latest LAV+madVR+ReClock behavior for PAL video if it slowed down by ReClock and its framerate is doubled by YADIF: madVR still switches the display to 24 Hz (though ReClock does see that video stream is 50 fps).
What does the madVR Ctrl+J OSD say about "movie xx.xxx fps (says source filter)"? If that fps information is not correct, then it's not madVR's fault.

A telecine source is always NTSC, therefore it's incorrect to force film mode in madVR for PALs, right?
Different people have different "definitions" of what telecined sources are. My personal opinion is that any interlaced source which can be turned back to progressive by weaving the correct fields together can be considered a telecined source. For any such source madVR's film mode should work very well.

Which 3? YADIF, madVR's forced film mode and DXVA deinterlacing (= madVR's auto mode)?
Yes.

i can confrim this also. it happened right after i updated.

also this happens all the time
If it happens all the time then please to capture it in a debug log. Also, while playback is frozen, try pressing Ctrl+Alt+Shift+Pause/Break, which should create a freeze report text file on your desktop. Please upload the zipped debug log and the freeze report somewhere (don't attach to this forum) to look at. Thanks.

To clarify that, if I understand it correctly, the new feature uses / needs DirectX 11.2, which in fact is only available with Windows 8.1 or Windows 10.
The key problem is that Windows 8 introduced WDDM 1.2 drivers which add OS support for stereoscopic 3D output via HDMI 1.4+. The Windows 7 Platform Update retrofits some Direct3D11 features back into Windows 7, but the Windows 7 GPU drivers stay WDDM 1.1, so stereoscopic 3D output is only possible in Windows 8+.

There are custom APIs available from each GPU manufacturer for stereoscopic output in Windows 7. But I'm not going to implement support for that, because it would require me to spend many hours trying to make 3D work for each GPU manufacturer, while in Windows 8 it's built into the OS.

Thanks for the latest version of MadVR! I guess now I have a reason to upgrade from Windows7? I guess I am forced to get Win10 or Win8 if I want to use MadVR for 3d blurays when they are supported? I guess it's not that big a deal for me as I have a standalone bluray player.
It's your decision. You can stick with Windows 7, madVR will continue to support it, probably for many years. However, it's very unlikely that 3D output will ever be supported in Windows 7. So if you want to use madVR to render 3D Blu-Rays, you probably have no other choice than to upgrade to either Windows 8.1 or Windows 10.

[Nullack]

Well, I'm not sure why it crashes. On my PC I could reproduce crashes when using D3D11 presentation with 3D enabled, when using v0.89.19, and it no longer crashes in v0.90.0 on my PC. But the crash was always in an NVidia driver DLL. It's not really that I fixed a bug or something, just some refresh rate related changes seem to have worked around the crash on my PC. If it still crashes for you, I've no idea what to do about it. It's most probably NVidia's bug and only they might be able to properly fix it, unfortunately.

D3D9 always worked for fine me, no blank screen and no crashes, with either v0.89.19 or v0.90.0.

Thanks Ill see what I can drum up with Nvidia on their support forums. I tried already being in the 3D mode and executing MPC-BE there directly in that mode so there is no 2D desktop shift (I normally run 2D desktop and just full screen swap it when doing say PowerDVD 3D blurays into 3D mode) but unfortunately already being in 3D doesnt workaround the crashing. This is an exciting feature and hopefully overtime the graphics stack will mature.

[madshi]

Thanks Ill see what I can drum up with Nvidia on their support forums. I tried already being in the 3D mode and executing MPC-BE there directly in that mode so there is no 2D desktop shift (I normally run 2D desktop and just full screen swap it when doing say PowerDVD 3D blurays into 3D mode) but unfortunately already being in 3D doesnt workaround the crashing. This is an exciting feature and hopefully overtime the graphics stack will mature.
Are you saying that you're getting a crash even when doing actual 3D playback with MVC MKV files created with MakeMKV, using madVR v0.90.0 and the special LAV test build?

[Nullack]

That works OK for that narrow use case in DX11 FSE with 3D stereoscopic on, as long as Im already in 3D mode and not at a 2D desktop mode and require the app to enter into 3D full screen - that crashes. I have to be already in a 3D desktop for working 3D MVC playback.

The problems are with the rest of my regression test video library which is 2D as far as LAV is concerned (its a combination of non MVC 3D, and normal 2D). Even if I set madVR to disable 3D for 2D and also including to reset OS settings on exit, MPC-BE crashes. The workaround is either turn off 3d in madVR, or turn off 3d stereoscopic in Nvidia when trying to do 2D only. Im trying now to really nail down the exact situation so I can give Nvidia the best possible info.

[madshi]

Do you ever get a crash box looking like the following one when any of these crashes occur?

http://madshi.net/exc-ss3.gif

If so, please press "show bug report", then press Ctrl+C, then the bug report is in your clipboard. Please upload it somewhere for me to look at. If you don't get that crash box, then it's likely the NVidia driver DLL crashing again.

[iakoboss7]

If it happens all the time then please to capture it in a debug log. Also, while playback is frozen, try pressing Ctrl+Alt+Shift+Pause/Break, which should create a freeze report text file on your desktop. Please upload the zipped debug log and the freeze report somewhere (don't attach to this forum) to look at. Thanks.

how do i get the debug log? i tried running "activate debug mode.bat" and it says "file madvr.ax not found".

the other one is here and as i said it happens if i minimize or another windows cover it! (btw if i go full screen the video starts playing again forwarding what it missed FAST until it gets to where its supposed to be and plays fine, cursor keeps moving and audio also keeps playing fine when video freezes).

https://drive.google.com/file/d/0Bz9-okPmTFK7VUpMUTBST0I0NTA/view?usp=sharing

[GCRaistlin]

What does the madVR Ctrl+J OSD say about "movie xx.xxx fps (says source filter)"? If that fps information is not correct, then it's not madVR's fault.
"25 fps". With ffdshow it reports the same, though madVR switches to 50 Hz, not 24.
My personal opinion is that any interlaced source which can be turned back to progressive by weaving the correct fields together can be considered a telecined source. For any such source madVR's film mode should work very well.
On the bugtracker, huhn said (http://bugs.madshi.net/view.php?id=198&PHPSESSID=db64394b0525b07e9cb259d270c66b09#c999) a time ago:
forced film mode can't handle interlaced sources is for telecine sources like most movies.
You said that he's right. Now you're saying that it should work with any source that can be turned back to progressive. I suppose that interlaced NTSC is that kind of source (at least YADIF can turn it back to progressive!). I'm confused.

[Nullack]

Do you ever get a crash box looking like the following one when any of these crashes occur?

http://madshi.net/exc-ss3.gif


Mate its not those, I would have already sent those to ya :) Its MPC-BE crashing as a symptom with no madVR specific error message. I deeply suspect your right about it being an Nvidia driver root cause - its got all the hallmarks of what happened a few months back when I couldn't get madVR working at UHD resolutions on Windows 10. Last time I ended up getting PMs from Nvidia customer care staff on their forums after making a bit of noise over there and I emailed them with various kernel memory dumps (some cases nv processes would BSOD) and EDID data and so on. I know with Nvidia Im going to have to nail down all the facts and have a clear bug report as otherwise they wont even look at it, and even when you do have all that, it usually takes a fair bit of noise on their forums to get them to be serious about it and debug it. But Im a stubborn tester so I wont give up :)

[madshi]

how do i get the debug log? i tried running "activate debug mode.bat" and it says "file madvr.ax not found".
Then manually rename "madVR.ax" to "madVR [release].ax", and then rename "madVR [debug].ax" to "madVR.ax". Do the same for madVR64.ax. Doing this is basically the same what the batch file is supposed to do. After having done this, reproduce the problem. There will be a VERY big log file on your desktop afterwards. Don't forget to rename the files back to their original names when you're done.

Please try to keep the log file "clean". Which means: After you've renamed the files, start a video, reproduce the problem, let the problem remain for maybe 20 seconds, then close the media player and upload the zipped log file. Done. Try to avoid changing the video player window size, or switching between fullscreen and windowed mode, if you can avoid it.

the other one is here and as i said it happens if i minimize or another windows cover it! (btw if i go full screen the video starts playing again forwarding what it missed FAST until it gets to where its supposed to be and plays fine, cursor keeps moving and audio also keeps playing fine when video freezes).

https://drive.google.com/file/d/0Bz9-okPmTFK7VUpMUTBST0I0NTA/view?usp=sharing
Unfortunately it doesn't help. Will need the debug log.

"25 fps". With ffdshow it reports the same, though madVR switches to 50 Hz, not 24.
And this is when using YADIF frame doubling? In that case madVR is obviously receiving incorrect information from the decoder. Which is not madVR's fault, but the fault of the decoder. So please follow up on this with the decoder developer(s). Will be hard for ffdshow, though, because nobody is developing it, anymore.

On the bugtracker, huhn said (http://bugs.madshi.net/view.php?id=198&PHPSESSID=db64394b0525b07e9cb259d270c66b09#c999) a time ago:

You said that he's right. Now you're saying that it should work with any source that can be turned back to progressive. I suppose that interlaced NTSC is that kind of source (at least YADIF can turn it back to progressive!). I'm confused.
There are 2 different types of interlaced videos:

1) Natively interlaced videos, like sports or music concerts. In these files every field comes from a different point in time. So 50 fields were recorded at 50 different time points. You cannot turn this into progressive by weaving the fields together. You would get heavy combing artifacts if you tried. You need to use video mode deinterlacing for this, like YADIF or DXVA.

2) Telecined film content. This type of content was originally progressive, with either 24fps and 25fps, and then the encoder split it into interlaced fields. The original frames are just from 24 or 25 different points in time. This type of content can be converted back into progressive by simply weaving the right fields together.

huhn was talking about 1). madVR's forced film mode can only handle 2), but not 1). For 1) you need to use DXVA deinterlacing. YADIF is also mainly aimed at 1). Both YADIF and DXVA can also handle 2), but often madVR's forced film mode will give you better results than either DXVA or YADIF, but only for 2).

There are also files that have a mixture of 1) and 2) content. Often music videos switch back and forth between 1) and 2). In that case madVR's film mode won't work well, either. Forced film mode is strictly only for 2).

Mate its not those, I would have already sent those to ya :) Its MPC-BE crashing as a symptom with no madVR specific error message. I deeply suspect your right about it being an Nvidia driver root cause - its got all the hallmarks of what happened a few months back when I couldn't get madVR working at UHD resolutions on Windows 10. Last time I ended up getting PMs from Nvidia customer care staff on their forums after making a bit of noise over there and I emailed them with various kernel memory dumps (some cases nv processes would BSOD) and EDID data and so on. I know with Nvidia Im going to have to nail down all the facts and have a clear bug report as otherwise they wont even look at it, and even when you do have all that, it usually takes a fair bit of noise on their forums to get them to be serious about it and debug it. But Im a stubborn tester so I wont give up :)
Too bad. Well, in that case there's probably not much I can do. I wish you good luck with NVidia support, in any case! :D If you hear back from them, let me know, would be good to hear if they were able to find/fix the issue.

[nevcairiel]

And this is when using YADIF frame doubling? In that case madVR is obviously receiving incorrect information from the decoder. Which is not madVR's fault, but the fault of the decoder. So please follow up on this with the decoder developer(s). Will be hard for ffdshow, though, because nobody is developing it, anymore.

You would think so, but I have confirmed that LAV sends 50 fps in the initial media type (and not even only in a new type send later) and yet madVR still shows 25 in its OSD. You don't cheat around the decoder again and ask the splitter, do you?