madVR - high quality video renderer (GPU assisted)

[psymed]

Does 0.79 lessen cpu usage by any chance?

[Thunderbolt8]

A clip of the material would help. Deinterlacing methods have not changed in many years. I'm using amd 5870. Basically the renderer only commands the GPU to deinterlace the material, the problem being before that MadVR did not for some reason understand bobbing hence only showed duplicates.I tried to make a clip, but that problem was gone then in that clip. but it still exists within the original video.

it looks like the camera or the whole picture becomes very shaky and flickering.

edit: that problem seems to be related to directvobsub. it didnt appear in the sample, because I didnt mux subs in it this time. when deactivating directvobsub, the shaking is gone.
would be nice if this interaction could be improved.

[pankov]

Hi madshi, sadly this bug is back with 0.79 final.
madshi,
I confirm this again

[sToN3d]

That's a known problem for some users. Unfortunately it doesn't occur on my PC, so I don't know right now what to do about it. I've implemented a little change in v0.79, though, which *may* help (but I rather doubt it). So you can retest this with v0.79.

Your change did indeed help a little. My CPU load is almost back to normal. It was ~6-7% before and now I'm at ~2.5-3.5%.

[Thunderbolt8]

A clip of the material would help.heres a sample http://www.mediafire.com/?jypobakil48bhnd

could you please check if you can see traces of interlaced leftovers as well in there?

[Gser]

I am curious as to how MadVR commands the use of DXVA deinterlacing. What mode does it choose, seeing as different GPU's need different flags.

[Gser]

heres a sample http://www.mediafire.com/?jypobakil48bhnd

could you please check if you can see traces of interlaced leftovers as well in there?

Why exactly are you trying to deinterlace film material? : D Just use tfm. I got similar weaving artifacts with different software deinterlacers as well.

[Thunderbolt8]

dunno what kind of material that is. I just want it to look normal during playback (without reencoding the movie) -.-

how can I use tfm during playback?

[Gaius]

I get no video picture when using LAV video, audio and splitters when using my MadVR settings or when restoring the default settings.

Also, clicking "Ok" on the MadVR properties window doesn't close the window and you have to close the entire player to get rid of the MadVR properties window.

This is on 32 bit Windows 7 with MPC-HC 1.5.3.3816 and latest stable Nvidia drivers.

[Gser]

dunno what kind of material that is. I just want it to look normal during playback (without reencoding the movie) -.-

how can I use tfm during playback?

Well if you want proper IVTC use ffdshow with avisynth plugin. Or if you don't mind the extra frames just turn off deinterlacing.

I have a rather interesting problem with Dscaler IVTC mod. For some reason it reports to madvr that the fps is film speed and deinterlacing is turned off. I don't know if the problem is with Dscaler or MadVR, perhaps i will try the non ivtc mod version since its ivtc function never worked for me anyways.

[Anima123]

Madshi, could you please add a force deinterlace option with the Ctrl+Alt+Shift D shortcut in case the source doesn't report the interlace status correctly?

[Thunderbolt8]

Well if you want proper IVTC use ffdshow with avisynth plugin. Or if you don't mind the extra frames just turn off deinterlacing.ignoring it looks horrible, really distracting when watching.

and my PC is too slow to do tcm() via avisynth on the fly during playback :S

[Gser]

Madshi, could you please add a force deinterlace option with the Ctrl+Alt+Shift D shortcut in case the source doesn't report the interlace status correctly?

Yes this would be handy. I have a source here that for some reason is reported as interlaced with 0.78 and not reported as interlaced with 0.79. Still trying to figure out the fps issue. If I put FFDShow inbetween DScaler and MadVR, it reports the correct fps but for some reason MadVR is forcing film speed.

[Gser]

ignoring it looks horrible, really distracting when watching.

and my PC is too slow to do tcm() via avisynth on the fly during playback :S
http://forum.doom9.org/showthread.php?t=141441
Try using Phase filter with FFDShow, its what i generally use.

[Thunderbolt8]

http://forum.doom9.org/showthread.php?t=141441
Try using Phase filter with FFDShow, its what i generally use.is it possible that it doesnt work so well when when I also use sharpening with ffdshow?

is it generally possible to fix an order in which all ffdshow filters can be used? in that case, I could try to run sharpening after applying phase.

[Andy o]

Thanks madshi, deinterlacing is now working at 50p for my material instead of at 25p.

The other thing I reported where 29i material with a "24p" tag in the file name would switch to 50p or 60p has also been fixed, but then deinterlacing gets turned off. The way I used to do it with CUVID was the decoder outputting 30p and then decimating with ffdshow processor/avisynth. I suppose it's not possible to decimate like this with madVR's deinterlacing, right?

[Gser]

is it possible that it doesnt work so well when when I also use sharpening with ffdshow?

is it generally possible to fix an order in which all ffdshow filters can be used? in that case, I could try to run sharpening after applying phase.

Just drag the sharpening filter after the dscaler filter.

[dansrfe]

madshi,
I confirm this again

Confirmed again as well.

Also deinterlacing gives rise to noticeable jerkiness and aliasing.

Here's the log: http://www.mediafire.com/?t7jzm59ln9wa6dp

[Andy o]

madshi, one other thing, would it be possible to add the option to force 29i/30i whenever I add a "29i" or "30i" to the file name, even when there's a "1080p60" in madVR options? Right now I have both 1080i29 and 1080p59, and it always switches to the latter. The reason is that the only way IVTC is working fully for me is if I let madVR IVTC to 60p and then convert that to a 30i output so my display's IVTC processor can detect proper cadence and apply IVTC again to 72 Hz. A bit convoluted, but works. I was using this method when I was using EVR.

[dansrfe]

I have a question:

How does madVR deal with the refresh rate reported by the source filter or splitter and ffdshow's video decoder and then if deinterlacing is activated within ffdshow?

For example:

I play a VOB file hard-telecined/3:2 pulldown at 29.970 reported by the source filter (navigator) and ffdshow, but then I enable deinterlacing within ffdshow. Now pure deinterlacement would keep the same fps of 29.970 and I realize that but what happens if I use AviSynth internally within ffdshow and inverse telecine it using tfm().tdecimate()? How does madVR deal with that? Does ffdshow report 23.976 as the new fps? Does madVR still use the rate reported by the source filter?

I feel like the problem gets more complicated with the introduction of user-added refresh rates...but I like it nevertheless.

[cyberbeing]

madVR 0.79's GPU RAM IN USE number is either inaccurate on WinXP SP3 or something is wrong with the queues. The number madVR shows is up to 3x that GPU-Z or NVIDIA Inspector shows.

12 CPU and 8 GPU queues w/ 1280x720 video
madVR reports 181/512MB
GPU-Z & NVIDIA Inspector reports 111/512MB

12 CPU and 8 GPU queues w/ 1920x1080 video
madVR reports 374/512MB
GPU-Z & NVIDIA Inspector reports 189/512MB


24 CPU and 24 GPU queues w/ 1280x720 video
madVR reports 519/512MB
GPU-Z & NVIDIA Inspector reports 228/512MB

24 CPU and 24 GPU queues w/ 1920x1080 video
madVR reports 1070/512MB
GPU-Z & NVIDIA Inspector reports 390/512MB

[6233638]

Just had a quick try of 0.79, mostly in Windowed mode, and haven't had a chance to test deinterlacing yet. (I suspect it won't have changed for me as I need it to work well at 24p)

However I have noticed two things: When setting the queue sizes to their maximum, they now stay almost maxed out:31–32/32
23–24/24
22–24/24
15–16/16
Previously they were close to, but not reaching the maximum with the old default values, so to fill that many extra queues, it seems like performance must have gone up quite a bit.

GPU memory usage, and estimates based on it seem inaccurate. By default, it sets the queues to: 20/16/16 and reports memory usage of 709/1280MB.
Turning that up to 32/24/24 reports 989/1280MB, when Nvidia Inspector and MSI Afterburner both report a maximum usage of 555MB.

[dansrfe]

I maxed out all the queues and I don't see any problem. GPU memory usage is 50MB below the max with all queues filled to the max with very slight fluctuations.

[Boltron]

Very good tests, thanks! Please retest with v0.79, which has significantly reduced the deinterlacing GPU RAM consumption, and also a bit the progressive GPU RAM consumption.

With v0.79 I think a 512MB GPU should be good enough for madVR. However, having 1024MB will allow you to double the queue sizes, when using v0.79.


Thanks! I will however not till Saturday, I am away from home.

[madshi]

Deinterlacing is now smooth!!
AWESOME! Isn't it. Madshi is a genius.
Seriously?? Not sure why, I didn't change much in terms of deinterlacing compared to the last test build I sent you. Anyway, I'm happy to hear that!!

PS: I really hope you have written down my request of adding a function that displays the refresh rate or the CTRL-J screen automatically only for a limited time when a video is opened (like 10 seconds). I am sometimes paranoid so I want to keep the refresh rate where it should be. ;)
I have not, your wish falls under "minor cosmetical wishes" and I'm not writing these down at this point in time. Please ask me about this again when madVR has reached v1.0.

anyway, that change of brightness in exclusive mode when activating deinterlacing is still there.

edit: that brightness shift seems to be related to the disable desktop composition box. when unticked, the change of brightness or colour is only very minor (probably thats normal?) but when ticked, then it really looks like a PC -> TV levels change. I'd like to be able to have that box ticked, because performance increases a little that way.
That is really weird. Does the black level change everytime you turn deinterlacing on/off via Ctrl+Alt+Shift+D? Does it help if you turn on the "Dynamic Range" option in the GPU "advanced video" options and set it to 16-235?

Hi madshi, sadly this bug is back with 0.79 final. (1)

2. A small cosmetic bug : the settings window name changes from "madVR settings" to "pure power curve" when "color & gamma" page is selected. But it never goes back to being "madVR settings" when other settings page are selected afterwards.

3. The ZP/slow paint/freeze on seek bug after madVR frequency switch is still there.
Ok, noted.

with v0.79 you've managed to keep (or even improve for some) the quality and still lower the GPU Load by 3-4%. Now it's 75-76% with the same 1080i60 sample that was using 79-80% before.
Nice! I did a small improvement where with NVidia I skipped one step that is needed for ATI. That must have made the difference you're seeing. So I guess using CUDA could eventually drop it by another 2-3%.

Does 0.79 lessen cpu usage by any chance?
Maybe a little bit, yes.

I tried to make a clip, but that problem was gone then in that clip. but it still exists within the original video.

it looks like the camera or the whole picture becomes very shaky and flickering.

edit: that problem seems to be related to directvobsub. it didnt appear in the sample, because I didnt mux subs in it this time. when deactivating directvobsub, the shaking is gone.
would be nice if this interaction could be improved.
What happens if you use the MPC-HC internal subtitle renderer instead of DirectVobSub? It is a bit problematic if DirectVobSub draws subtitles on the not-yet-deinterlaced video. The same problem should appear with EVR. There's nothing I can do about it.

Your change did indeed help a little. My CPU load is almost back to normal. It was ~6-7% before and now I'm at ~2.5-3.5%.
Nice! Maybe I can improve it further.

I am curious as to how MadVR commands the use of DXVA deinterlacing. What mode does it choose, seeing as different GPU's need different flags.
madVR always uses the deinterlacing mode which is listed first. That is supposed to be the highest quality mode, according to DXVA2 rules.

I get no video picture when using LAV video, audio and splitters when using my MadVR settings or when restoring the default settings.

Also, clicking "Ok" on the MadVR properties window doesn't close the window and you have to close the entire player to get rid of the MadVR properties window.

This is on 32 bit Windows 7 with MPC-HC 1.5.3.3816 and latest stable Nvidia drivers.
I can't reproduce this problem on my PC. Can anybody else?

I have a rather interesting problem with Dscaler IVTC mod. For some reason it reports to madvr that the fps is film speed and deinterlacing is turned off. I don't know if the problem is with Dscaler or MadVR, perhaps i will try the non ivtc mod version since its ivtc function never worked for me anyways.
Not sure what problem you're seeing there. The DScaler IVTC mod is supposed to do IVTC, that's what it's for. madVR knows that and so treats the DScaler IVTC mod output as 24p. That's intentional madVR behaviour and I see nothing wrong with it.

Madshi, could you please add a force deinterlace option with the Ctrl+Alt+Shift D shortcut in case the source doesn't report the interlace status correctly?
Not sure what you mean? If you press Ctrl+Alt+Shift+D twice, you can force deinterlacing on. Isn't that what you want?

Thanks madshi, deinterlacing is now working at 50p for my material instead of at 25p.
Great!

The other thing I reported where 29i material with a "24p" tag in the file name would switch to 50p or 60p has also been fixed, but then deinterlacing gets turned off. The way I used to do it with CUVID was the decoder outputting 30p and then decimating with ffdshow processor/avisynth. I suppose it's not possible to decimate like this with madVR's deinterlacing, right?
Oh well, so you want 24p display mode, but you still want madVR to deinterlace? I hope you can see that this is not intuitive to madVR. You could try "24i" or "48i" instead of "24p", but then you need to turn off the option "don't deinterlace 48i" in the madVR settings.

is it possible that it doesnt work so well when when I also use sharpening with ffdshow?
I'm not sure if it's a good idea to process video in ffdshow/aviSynth, if you want to use DXVA2 deinterlacing, because usually you should deinterlace *first* and process afterwards. Sharpening should ideally be performed after deinterlacing + upscaling.

Also deinterlacing gives rise to noticeable jerkiness and aliasing.
More details, please? Do you see dropped or delayed frames? Or presentation glitches? Which GPU? Which OS? Which movie resolution? Which output resolution? Which display refresh rate? Which decoder? Are you just deinterlacing, or up/downscaling as well?

madshi, one other thing, would it be possible to add the option to force 29i/30i whenever I add a "29i" or "30i" to the file name, even when there's a "1080p60" in madVR options? Right now I have both 1080i29 and 1080p59, and it always switches to the latter.
It makes sense to switch to 1080i59/60. But I don't see why it makes sense to activate deinterlacing in that case?

How does madVR deal with the refresh rate reported by the source filter or splitter and ffdshow's video decoder and then if deinterlacing is activated within ffdshow?

For example:

I play a VOB file hard-telecined/3:2 pulldown at 29.970 reported by the source filter (navigator) and ffdshow, but then I enable deinterlacing within ffdshow. Now pure deinterlacement would keep the same fps of 29.970 and I realize that but what happens if I use AviSynth internally within ffdshow and inverse telecine it using tfm().tdecimate()? How does madVR deal with that? Does ffdshow report 23.976 as the new fps? Does madVR still use the rate reported by the source filter?
As far as I know, ffdshow does *not* update the fps information when changing frame rates. This is IMHO a bug in ffdshow and there's nothing I can do about it. I can't somehow guess that ffdshow is performing IVTC or that it's adding additional frames or whatever.

madVR 0.79's GPU RAM IN USE number is either inaccurate on WinXP SP3 or something is wrong with the queues. The number madVR shows is up to 3x that GPU-Z or NVIDIA Inspector shows.
GPU memory usage, and estimates based on it seem inaccurate. By default, it sets the queues to: 20/16/16 and reports memory usage of 709/1280MB.
Turning that up to 32/24/24 reports 989/1280MB, when Nvidia Inspector and MSI Afterburner both report a maximum usage of 555MB.[/list]
Well, madVR calculates the number of MB it allocates, but this includes both GPU RAM and PCIe RAM. I'm not sure which textures/surfaces Direct3D allocates in which RAM. I suppose GPU-Z and NVidia Inspector etc report only GPU RAM usage, while madVR counts all the RAM it uses in either GPU RAM or PCIe RAM. I'm not sure if I can improve that. Maybe I can figure out which resources are allocated where and take that into account. Not sure...

@cyberbeing, does deinterlacing work for you now? I think it didn't work for you in v0.78 due to GPU RAM problems? Which queue sizes are working for you now?

I have noticed two things:[list=1] When setting the queue sizes to their maximum, they now stay almost maxed out:31–32/32
23–24/24
22–24/24
15–16/16
Previously they were close to, but not reaching the maximum with the old default values, so to fill that many extra queues, it seems like performance must have gone up quite a bit.
Maybe it has, not sure. So any improvements or negative side effects by increasing the queue sizes?

I maxed out all the queues and I don't see any problem. GPU memory usage is 50MB below the max with all queues filled to the max with very slight fluctuations.
Good. Any improvements by maxing out the queues? Or any negative side effects?

Thanks! I will however not till Saturday, I am away from home.
No worries, I'm not in a hurry.

[LigH]

madVR currently does not allow you to specify OSD size or font. I might add such features into a future version. But for now my priority is making madVR feature complete first. Cosmetical things like changing OSD looks is pretty low on my priority list right now. You'll have to wait a while until I get to these things, I'm sorry. I hope you'll agree that things like adding post processing shader support, DVD support, DXVA deinterlacing etc are much more important things compared to changing OSD font sizes... :)

No panic, madshi. First solve the serious issues. :cool:

But when you get there some day ... I was looking for a way to call the configuration dialog while no video was played. You might add a matching batch file to your package to allow this, I guess it would be a call of RunDLL?

[madshi]

But when you get there some day ... I was looking for a way to call the configuration dialog while no video was played. You might add a matching batch file to your package to allow this, I guess it would be a call of RunDLL?
This is on my to do list. Currently the settings dialog only works if there's a madVR video instance running. Changing that is possible, but not a 5 minute job.

[Andy o]

Oh well, so you want 24p display mode, but you still want madVR to deinterlace? I hope you can see that this is not intuitive to madVR. You could try "24i" or "48i" instead of "24p", but then you need to turn off the option "don't deinterlace 48i" in the madVR settings.

The way it worked for me with CUVID was that the decoder did IVTC to 30p (no decimating), then I would decimate with ffdshow/avisynth. When I played a 29i file then, it would be IVTCd and decimated to 24p, which is why I wanted to force madVR to switch 29i files to 24p manually by tweaking the file name.

I think avisynth/ffdshow decimating wouldn't work with madVR deinterlacing though, the way it does with CUVID, right? Cause the deinterlacing is being done after ffdshow's processor.

It makes sense to switch to 1080i59/60. But I don't see why it makes sense to activate deinterlacing in that case?
What does madVR do with interlaced content when deinterlacing is not active? Looks pretty bad with interlacing lines all over. The way I'm figuring it works when deinterlacing is on with telecine content, is it reconstructs the frames into 60p and a 3-2 pulldown pattern. Then the video card re-interlaces it into 1080i and my TV again reconstructs the frames, this time to 24p into 72 Hz.

Pioneer TVs and probably other TVs/processors do this automatically, but only with up to 1080i sources. If it were possible to pass the original interlaced video it would probably be ideal, but I don't think it is, and besides with lower than 1080i content there would be problems with scaling, cause I always keep my desktop to 1080x1920.

I understand this probably affects only people with Pioneer TVs and then very few of us, if not only me. I thought it would be an easy fix, if not, don't worry about it.

[nx6]

It was not removed. If you don't get the message then probably you have exclusive mode disabled accidently? Or maybe it stopped working for you? You can check via Ctrl+J. It tells you whether madVR is in windowed or exclusive mode.
How does one disable Exclusive Mode? :confused:

I've updated to 0.79, "uninstalling" the old version first. Also reset to default settings, no change.

Is there an actual difference in rendering quality on Windowed vs Exclusive mode, or it just that Exclusive allows better system performance?

[nevcairiel]

What does madVR do with interlaced content when deinterlacing is not active?

It does nothing. Interlaced content is typically already weaved into one frame, and madVR just shows you that. Of course, you'll see the interlaced artifacts/combing.

[madshi]

The way it worked for me with CUVID was that the decoder did IVTC to 30p (no decimating), then I would decimate with ffdshow/avisynth. When I played a 29i file then, it would be IVTCd and decimated to 24p, which is why I wanted to force madVR to switch 29i files to 24p manually by tweaking the file name.

I think avisynth/ffdshow decimating wouldn't work with madVR deinterlacing though, the way it does with CUVID, right? Cause the deinterlacing is being done after ffdshow's processor.
You're right, that wouldn't work. So it probably makes no sense to use either "24p" or "48i" tags, unless the content is soft-telecined. In the long run I may implement decimating, but it will be difficult to do on the GPU. I'll probably have to use CUDA or OpenCL, and unfortunately ATI doesn't offer Direct3D9 <-> OpenCL interopability, so I'm not sure if it will be possible at all. There may be a different solution to remove the 3:2 judder, without decimating. But let's talk about that later...

The first step was to implement DXVA2 deinterlacing. To be honest, I consider is mostly useful only for native video type content. It's also moderately useful for movie content. But movie handling is definitely not ideal yet. But I gotta take one step at a time. At least I'm at the same level as EVR now, in terms of deinterlacing capability. Further improvements are planned for future versions (but probably not too soon).

Pioneer TVs and probably other TVs/processors do this automatically, but only with up to 1080i sources. If it were possible to pass the original interlaced video it would probably be ideal, but I don't think it is, and besides with lower than 1080i content there would be problems with scaling, cause I always keep my desktop to 1080x1920.

I understand this probably affects only people with Pioneer TVs and then very few of us, if not only me. I thought it would be an easy fix, if not, don't worry about it.
It's not a question of easy or not. Changing this would indeed be easy. The question for me is whether it makes generally sense to deinterlace 1080i60 content if the GPU outputs 1080i60, anyway. Maybe it does make sense, maybe not. I'm not 100% sure. Any more opinions on that?

How does one disable Exclusive Mode? :confused:

I've updated to 0.79, "uninstalling" the old version first. Also reset to default settings, no change.
There's an option in the madVR settings to disable exclusive mode. It's rather unlikely that you enabled that option by accident, but not impossible. Anyway, resetting to default settings should make sure exclusive mode is enabled. So the problem must be something else. So does Ctrl+J definitely confirm that you're in windowed mode and not in exclusive mode, even when going fullscreen?

Is there an actual difference in rendering quality on Windowed vs Exclusive mode, or it just that Exclusive allows better system performance?
Exclusive mode usually performs better and pretty much guarantees that you don't get any tearing. In windowed mode you may get tearing and the danger of getting frame drops is usually higher. However, if windowed mode works perfectly for you, with no tearing, no frame drops and no performance problems, then there's no reason to use exclusive mode. Anyway, it might still make sense to figure out why it's not working for you (in case it really isn't).

[thuan]

That's a known problem for some users. Unfortunately it doesn't occur on my PC, so I don't know right now what to do about it. I've implemented a little change in v0.79, though, which *may* help (but I rather doubt it). So you can retest this with v0.79.
It's good to be doubtful, but with 0.79 that problem goes away :D.

[ranpha]

Sample. (http://www.filesonic.com/file/20150847)

madVR won't deinterlace this file, while with EVR custom presenter the deinterlacing will happen.

Filter graph: LAV Splitter/Haali + LAV Video Decoder (software mode).

[gendouhydeist]

Ok I've set my Display to 60Hz but why madVR display as 59.93Hz and 60Hz Composition rate instead of 60Hz Display even in exclusive mode
http://i.imgur.com/5FuRW.png

[SamuelMaki]

I have maybe a problem with the presentation queue... Now with 0.78 version presentation queue is 0-3/4, when 0.77 reports 3-4/4 all the time... I do not have any presentation qlitches or dropped frames thought... So, what is the problem, or is there any problem with ctrl+j reporting? I am using radeon hd 5650...

Playback is perfect with 0.79:thanks:
Queues stay with maximum/maximum -1...

[nx6]

So does Ctrl+J definitely confirm that you're in windowed mode and not in exclusive mode, even when going fullscreen?
Yup, goes from "windowed mode" to "fullscreen windowed mode" now.

[madshi]

It's good to be doubtful, but with 0.79 that problem goes away :D.
Yey! Good news... :)

Sample. (http://www.filesonic.com/file/20150847)

madVR won't deinterlace this file, while with EVR custom presenter the deinterlacing will happen.

Filter graph: LAV Splitter/Haali + LAV Video Decoder (software mode).
Good sample, thanks. It doesn't have any frame rate information, which confuses madVR's "deinterlacing on/off" decision. Will be fixed in the next build.

Ok I've set my Display to 60Hz but why madVR display as 59.93Hz and 60Hz Composition rate instead of 60Hz Display even in exclusive mode
madVR's display information is usually very reliable. So it seems that your GPU really does output 59.93Hz instead of 60.00Hz. That is actually a good thing because by far most consumer content (broadcasts, Blu-Rays, DVDs) is 60/1.001 = 59.940fps.

Playback is perfect with 0.79:thanks:
Queues stay with maximum/maximum -1...
Great! Looks like v0.79 seems to be a nice improvement all around.

Yup, goes from "windowed mode" to "fullscreen windowed mode" now.
Hmmmm... Please create a log and uploaded it (zipped). Please try to keep it small. Which means: Start your media player, go to fullscreen mode, wait 10 seconds, close your media player. Done. Thx.

[Gser]

Not sure what problem you're seeing there. The DScaler IVTC mod is supposed to do IVTC, that's what it's for. madVR knows that and so treats the DScaler IVTC mod output as 24p. That's intentional madVR behaviour and I see nothing wrong with it.


It shows film speed for all material - even natively interlaced 30fps material. If I decode the file with your internal filters or FFDShow MadVR shows 30fps. For v 0.78 it still deinterlaced it but for 0.79 it does not. Forcing deinterlacing seems to work for now. Don't know it MadVR is rendering at film speed or just claims to.

[madshi]

It shows film speed for all material - even natively interlaced 30fps material. If I decode the file with your internal filters or FFDShow MadVR shows 30fps. For v 0.78 it still deinterlaced it but for 0.79 it does not. Forcing deinterlacing seems to work for now. Don't know it MadVR is rendering at film speed or just claims to.
The DScaler IVTC Mod is meant to be used mainly for movie content. If you use it for that purpose, the movie should already fully be deinterlaced and decimated when it reached madVR, resulting in 24p. And that is what madVR expects from the DScaler IVTC Mod. Consequently, the deinterlacer gets deactivated, because why deinterlacing progressive 24p content? For natively interlaced material the DScaler IVTC Mod is not the choice. It might accidently treat some scenes as being movie content, resulting is dropped fields, artifacts and non-smooth motion.

[Gser]

The DScaler IVTC Mod is meant to be used mainly for movie content. If you use it for that purpose, the movie should already fully be deinterlaced and decimated when it reached madVR, resulting in 24p. And that is what madVR expects from the DScaler IVTC Mod. Consequently, the deinterlacer gets deactivated, because why deinterlacing progressive 24p content? For natively interlaced material the DScaler IVTC Mod is not the choice. It might accidently treat some scenes as being movie content, resulting is dropped fields, artifacts and non-smooth motion.

I also tried the non IVTC version. Same thing. I also have IVTC set as off. The IVTC part never worked for me anyways.

[madshi]

madVR checks whether the registry value "HKCU\Software\DScaler5\MpegVideo Filter\Inverse Telecine" exists. If it does, it considers DScaler to be the IVTC Mod. If you want to go back to the non-modded version, maybe deleting that registry value will help?

But why are you using DScaler in the first place, if the IVTC Mod doesn't work for you? It's very much outdated, based on rather old and buggy decoder sources. I'd suggest to use LAV Video Decoder or the internal madVR decoder instead. Both are better MPEG2 decoders, IMHO, they just don't do IVTC.

[pie1394]

It's not a question of easy or not. Changing this would indeed be easy. The question for me is whether it makes generally sense to deinterlace 1080i60 content if the GPU outputs 1080i60, anyway. Maybe it does make sense, maybe not. I'm not 100% sure. Any more opinions on that?



As long as the deinterlacing process does not touch the original decoded field's content, it won't affect the visual quality. Just a waste of GPU processing power to generate the deinterlaced field (will not be output). If original video field contents are touched, my eyes can often observe the lower visual quality on interlaced CRT / PDP device.

If madVR does not send it to do deinterlacing process, in theory, the 25 / 30 fps frame presentation speed should be ok for 1080i50 / 1080i60 contents during playback. But my very old test with madVR showed jaggy moving objects on nVidia GPU (both G80/GT100). Maybe it work if one interlaced frame is presented 2 times?

For the odd-field-first video contents w/o deinterlace, it is no doubt that frame presentation method must be still the one with odd-field-first since it has smaller Presentation Timestamp than the combined even-field parts output by MPEG video decoder.

So this is not a problem on my HTPC(Win7x64 , GTX260+) / Hitachi 42" PDP P42A01A system to do the frame-rate-doubling deinterlacing + original cotent's interlaced mode output. The video output mode is always set 1920x1080i60 to play 29.97/30 double-frame-rated deinterlaced MPEG-2 contents.

Yet I did a comparsion of following 2 playback methods, and found the method 1 shows that human skin colors are somehow more vivid and stereo.

[1] LAVVideo CUDA MPEG2 (+deinterlace) + madVR0.79
[2] LAVVvideo MPEG2 + madVR0.79 (+ DXVA2 deinterlace)

Yet I cannot tell the difference via both methods on Ion + Samsung 203B LCD monitor (6-bit TN + FRC).

Thus I suspect that nVidia's driver did more high-precisioned video image post-processing on GTX260+ just like what madVR does. Otherwise there is no way that same madVR video scaling method can produce such difference.

[iSeries]

According to the log file, the splitter outputs really bad timestamps. Often multiple frames get exactly the same timestamp. That's a splitter problem.

Hi,

I was reluctant to say it was a splitter problem as I am using LAV as are many others here. I don't understand 1. why playing the m2ts directly plays smoothly and 2. reverting back to MadVR 075 (the last version I have) playing the index.bdmv or mpls, playback is as smooth as it should be?. Something must have changed between version 075 and 078 / 079?

[leeperry]

It works perfectly fine with my HD 6850 and HD 4770 on XPSP3. Haven't tested NVidia on XPSP3 yet, but in theory it should work, too. LAVVideo does not deinterlace itself (unless you're using CUVID). Normally the renderer does the deinterlacing. CUVID is the exception.

So did you use CUVID? Or did you use software decoding?
I had not tried yet when I posted this, but I just did and indeed as you said today I do need deinterlacing to be performed prior to running my Avisynth script, so either way I would have no use for deinterlacing being performed by the VR :o

CUVID does an amazing job on 29.97 true interlaced DVD's, but it's a major failure on 1080i IME...each scene change making nasty artifacts.

Anyway, as I feared DXVA2 doesn't seem to exist in XP (http://www.google.com/search?hl=en&q=dxva2+xp) so as much as ATi might convert them to DXVA1, all I get is a black screen on my 8800GS and not even mVR's OSD will appear. If I mark this 29.97fps DVD as progressive, then I get the usual combing.

I might soon stop the XPSP3 testing, because I want to benefit from the HPET for audio purposes :cool:

[Gser]

madVR checks whether the registry value "HKCU\Software\DScaler5\MpegVideo Filter\Inverse Telecine" exists. If it does, it considers DScaler to be the IVTC Mod. If you want to go back to the non-modded version, maybe deleting that registry value will help?

But why are you using DScaler in the first place, if the IVTC Mod doesn't work for you? It's very much outdated, based on rather old and buggy decoder sources. I'd suggest to use LAV Video Decoder or the internal madVR decoder instead. Both are better MPEG2 decoders, IMHO, they just don't do IVTC.

Sweet thanks, deleting the registry value worked. I've noticed that using Dscalers reference idct results in slightly more sharper and detailed picture while the ones in ffmpeg result in a softer picture. Don't know why the mods IVTC feature never works for me. I get lots of combing with it and miss-matched fields.

[cyberbeing]

Well, madVR calculates the number of MB it allocates, but this includes both GPU RAM and PCIe RAM. I'm not sure which textures/surfaces Direct3D allocates in which RAM. I suppose GPU-Z and NVidia Inspector etc report only GPU RAM usage, while madVR counts all the RAM it uses in either GPU RAM or PCIe RAM. I'm not sure if I can improve that. Maybe I can figure out which resources are allocated where and take that into account. Not sure...
You've mentioned this before about PCIe RAM. What exactly is PCIe RAM and where is it being allocated, as it certainly doesn't seem to be going into System RAM or GPU RAM?

For example, with Queue maxed out on 1440x1080i30 16:9:

madVR reports 1188MB/512MB GPU RAM in Use.
Actual GPU RAM use is only 392MB.
MPC-HC is only using 115MB.

Assuming madVR is reporting GPU + PCIe RAM, is there any reason that combined number should be kept below GPU RAM capacity? Is it just some sort of theoretical maximum size that madVR would use in cases of massive upscaling/downscaling?

This also has me curious if madVR in any way confirms that 16bit/32bit FP textures are being stored properly by the GPU? Seeing as some other people has much higher GPU RAM use numbers in GPU-Z (side-effect of Aero?), hopefully the low numbers don't mean the GPU is actually holding 8bit RGB textures in the queues?

@cyberbeing, does deinterlacing work for you now? I think it didn't work for you in v0.78 due to GPU RAM problems? Which queue sizes are working for you now?
It doesn't crash now which is a success, but I still need to set all scaling to Bilinear and disable the 3DLUT for it to be usable without delayed/dropped frames on 1440x1080i30 16:9 video.

Could you add an option under Trade Quality for Performance for something like 'Use Bilinear and Shaders Only when Deinterlacing or 60fps+ framerates' to temporarily set these options when deinterlacing is active and/or just high framerate content in general? That's the only practical way I could see to keep deinterlacing enabled in madVR. What this comes back to is my GPU can barely playback 60fps content with madVR even with progressive video.

Another thing I note is performance is much much slower (3-6ms deint vs 0.15-0.25ms deint) if Inverse Telecine enabled in the NVCPL. Since madVR doesn't seem to support decimation like VMR (?), I guess everybody should disable that setting? No change in quality as far as I can tell using madVR, just slower. I'll test a DVD later which VMR9 decimates correctly with the NVIDIA PureVideo MPEG2 decoder (even changing between 24p and 30p on hybrid content) with that 'Inverse Telecine' setting enabled to see if anything changes in madVR.

As for queue sizes for deinterlacing
Absolute minimum: 8CPU/6GPU (below this results in massive dropped frames)
Absolute maximum: 32CPU/24GPU
Requirement 1: Inverse Telecine is disabled in NVCPL
Requirement 2: CPU Queue is at least +2 greater than GPU queue

[madshi]

As long as the deinterlacing process does not touch the original decoded field's content, it won't affect the visual quality. Just a waste of GPU processing power to generate the deinterlaced field (will not be output). If original video field contents are touched, my eyes can often observe the lower visual quality on interlaced CRT / PDP device.
I don't know whether DXVA deinterlacing touches the original fields or not. Haven't tested that. Might depend on ATI/NVidia.

If madVR does not send it to do deinterlacing process, in theory, the 25 / 30 fps frame presentation speed should be ok for 1080i50 / 1080i60 contents during playback. But my very old test with madVR showed jaggy moving objects on nVidia GPU (both G80/GT100). Maybe it work if one interlaced frame is presented 2 times?
In D3D9 exclusive mode with "prepresented frames" active, madVR will already show each weaved frame twice, provided that the timestamps are clean. If the timestamps jitter a lot, nasty things like 2:2:2:3:1:2:2:2 can happen.

For the odd-field-first video contents w/o deinterlace, it is no doubt that frame presentation method must be still the one with odd-field-first since it has smaller Presentation Timestamp than the combined even-field parts output by MPEG video decoder.
If you turn deinterlacing off, madVR does *nothing*. If the fields are swapped then madVR does not correct that.

So this is not a problem on my HTPC(Win7x64 , GTX260+) / Hitachi 42" PDP P42A01A system to do the frame-rate-doubling deinterlacing + original cotent's interlaced mode output. The video output mode is always set 1920x1080i60 to play 29.97/30 double-frame-rated deinterlaced MPEG-2 contents.
You just said above that activating deinterlacing would be a waste of GPU processing power. But now you're saying in your own setup you have deinterlacing active even when using 1080i60 output? That's kinda confusing.

Yet I did a comparsion of following 2 playback methods, and found the method 1 shows that human skin colors are somehow more vivid and stereo.

[1] LAVVideo CUDA MPEG2 (+deinterlace) + madVR0.79
[2] LAVVvideo MPEG2 + madVR0.79 (+ DXVA2 deinterlace)
Can you make comparison screenshots, please?

I was reluctant to say it was a splitter problem as I am using LAV as are many others here. I don't understand 1. why playing the m2ts directly plays smoothly and 2. reverting back to MadVR 075 (the last version I have) playing the index.bdmv or mpls, playback is as smooth as it should be?. Something must have changed between version 075 and 078 / 079?
The splitter (or decoder??) is definitely producing bad timestamps. It is possible, though, that some older madVR versions handled the problem better than the latest builds. I recently did a change in this area, which may have made things worse for you. I'd like to debug that. Can you please create a log with v0.79? I know you already created one with v0.78, but it didn't contain enough information. I've added more logging to v0.79, so that should help.

Anyway, I still recommend trying to figure out where the bad timestamps are coming from. It would be better to have that problem fixed, too, instead of relying on madVR to work around it somehow.

Anyway, as I feared DXVA2 doesn't seem to exist in XP (http://www.google.com/search?hl=en&q=dxva2+xp).
That's not true. That website you're quoting is either wrong or outdated. EVR + DXVA2 gets retrofitted into XP by installing .NET 3.0 or 4.0. On my PC (with an ATI GPU) DXVA2 deinterlacing works exactly the same as in win7, while that link you posted says "The DXVA 2 API requires Windows Vista or later". Clearly the link is wrong.

so as much as ATi might convert them to DXVA1, all I get is a black screen on my 8800GS and not even mVR's OSD will appear.
Do you get DXVA2 deinterlacing with EVR? Have you installed .NET 3.0?

Anybody using NVidia on XP? Does madVR's DXVA2 deinterlacing work for you?

[nevcairiel]

Another thing I note is performance is much much slower (3-6ms deint vs 0.15-0.25ms deint) if Inverse Telecine enabled in the NVCPL. Since madVR doesn't seem to support decimation like VMR (?), I guess everybody should disable that setting? No change in quality as far as I can tell using madVR, just slower.)


Disabling Inverse Telecine in the NVIDIA Control Panel can result in artifacting on Telecined discs, if anything it might do full deinterlacing instead of just weaving the fields back together properly, which could result in an image degradation.
As always however, trust your eyes over anything else. :)

[dansrfe]

I see dropped frames when upscaling more than approximately 1280 x 720. No delayed frames or presentation glitches.
GPU: ATI Mobility Radeon HD 4570
OS: Windows 7 x64
Movie Resolution: 704 x 480 (16:9) DVD
Output Resolution: 853 x 480 (default 16:9) or upscaled to 1920 x 1080
Display Refresh Rate: 59.940Hz
Decoder: ffdshow decoder

I'm deinterlacing from 29.970 interlaced (3:2 pulldown, but I guess that doesn't make a difference).

Scaling: Spline @ 3 taps on all. Scaling in linear light off.

Basically after upscaling to a certain resolution the queues progressively start to fill up less and less until it's at 0-2 like 400 horizontal pixels before I can hit fullscreen 1920 horizontal resolution. I'm guessing I have to get a faster GPU to fix this or is there hope in me waiting for more speed improvements in DXVA2 implementation of madVR?

Is there an equivalent of CUDA in ATI that you can utlize to attain GPU acceleration in ATI cards?

Also I have noticed that madVR is still not accurately detecting if the input video needs to be deinterlaced or not. I just played a 25fps PAL VOB and it showed that deinterlacing was on. I'm guessing that this has to do with incorrect flag information in the VOB.

Also, the gpu ram usage section in the OSD seems to be a bit off or something else is happeneing with my GPU drivers. When I fill up all the queues on a 1080p file or something large it shows 1068/512. My card is a 1GB card however I'm used to seeing it say 512MB everywhere (e.g. in the driver control panel, windows display information etc). What does this mean?

[nevcairiel]

That's not true. That website you're quoting is either wrong or outdated. EVR + DXVA2 gets retrofitted into XP by installing .NET 3.0 or 4.0. On my PC (with an ATI GPU) DXVA2 deinterlacing works exactly the same as in win7, while that link you posted says "The DXVA 2 API requires Windows Vista or later". Clearly the link is wrong.

The DXVA2 DDI API requires Windows Vista and a WDM display driver, however you're not calling the DDI directly, you call the Direct3D9 interfaces.

http://msdn.microsoft.com/en-us/library/windows/desktop/cc307941%28v=vs.85%29.aspx

"If the graphics drivers uses the older Windows XP Display Driver Model (XPDM), DXVA 2 API calls are converted to DXVA 1 DDI calls. "

So, while XP doesn't support DXVA2 natively, if you install the .net framework 3 or 4 (not sure which version), there will be an emulation layer in place that can convert DXVA2 API calls to DXVA1 hardware calls.
Note that this also applys to very old drivers on Vista/7, which didn't implement WDM yet. However those drivers won't really be around anymore. ;)

I think when you query for a DXVA2 device, there should be some information flag somewhere that tells you if its a emulated device based on DXVA1.

PS:
Using DXVA2 *decoding* on XP is another matter, and alot harder (close to impossible), however video processing is possible.


Is there an equivalent of CUDA in ATI that you can utlize to attain GPU acceleration in ATI cards?

ATI doesn't have a real equivalent. They have OpenCL based APIs to decode video, however they failed to properly link that to D3D, so its not really possible to couple that with a Direct3D based rendering chain.