madVR - high quality video renderer (GPU assisted)

[Thunderbolt8]

about chroma upscaling, is this always involved when watching something, also when no up- or downscaling is done at all? just wondering why its name hasnt been changed to something like "picture upscaling" or similar as well.

[noee]

fyi, HD6570, Win7 Ult, playback on second monitor, SD=>1080

Jinc 3tap/no AR, GPU ~36%
Jinc 3tap/AR, GPU ~40%

Jinc 4tap/no AR, GPU ~51%
Jinc 4tap/AR, GPU ~57%

0 dropped, 0 glitches

One thing I noticed, not sure if it's related to recent madVR changes or not, but the OSD is now showing the monitor (LG HDTV) refresh rate at 24.00051Hz, dead on steady. Before the last few builds, it would never set below 24.00083Hz. This is HDMI with the HD6570=>Onkyo=>LG HDTV. Seems strange that the madVR "timer" changes would affect this.

Gonna try a low-end Llano laptop (6520G) next, see what it will do.

[madshi]

about chroma upscaling, is this always involved when watching something, also when no up- or downscaling is done at all? just wondering why its name hasnt been changed to something like "picture upscaling" or similar as well.
It's always involved, at least for normal movie sources. However, differences between different chroma upsampling algorithms are often only visible in specific scenes. Anyway. The reason why it wasn't renamed is because it has the correct name right now. Chroma upscaling only upscales the Chroma channel.

fyi, HD6570, Win7 Ult, playback on second monitor, SD=>1080

Jinc 3tap/no AR, GPU ~36%
Jinc 3tap/AR, GPU ~40%

Jinc 4tap/no AR, GPU ~51%
Jinc 4tap/AR, GPU ~57%

0 dropped, 0 glitches
Interesting. Seems I underestimated modern GPUs. I wouldn't have expected the 6570 to handle Jinc 4 that easily. Good news, I guess...

[HitomiKun]

madVR v0.84.0 now switches to 100 Hz for no valid reason on a 23.976 fps video... Recently I just use 120 Hz for 24 fps movies, but now it switches to 100 Hz instead (which I set for some European movies that run at 25 fps). I tried every sample I had, various movies / anime, same problem everywhere, always switching to 100 Hz, doesn't matter if it's 23.976, 24 or 25 fps.
Here's a log:

madVR - log.7z (https://dl.dropbox.com/u/48947610/madVR%20-%20log.7z)
^I've been running at 120 Hz from the start, but it switches to 100 Hz for no reason. Am I the only one getting problems here with a 120 Hz monitor? xD

Edit: MadVR mode change settings (https://dl.dropbox.com/u/48947610/MadVR%20settings.png) <- not sure if this helps, but I posted it anyways.

[madshi]

The log claims you've only one mode in the madVR settings: 1080p100. That's why madVR is always switching to that. Please double check. If your settings do have multiple modes in there, please send me your "settings.bin" file. Or if you don't have such a file, export "HKEY_CURRENT_USER\Software\madshi\madVR\Settings" and send me that.

[HitomiKun]

The log claims you've only one mode in the madVR settings: 1080p100. That's why madVR is always switching to that. Please double check. If your settings do have multiple modes in there, please send me your "settings.bin" file. Or if you don't have such a file, export "HKEY_CURRENT_USER\Software\madshi\madVR\Settings" and send me that.

So that means I gotta add 1080p120 Hz too? I thought MadVR calculates if it needs to change the mode to match the fps or not.

Ok. I added 1080p120 and everything works now. Sorry for the false alarm. ^^'

[Hypernova]

I can easily do 704x480 to 1920x1080 which uses 71-74% of my HD4000 with no dropped frames in windowed or exclusive mode.
My settings are as follows:

CPU queue size: 4
GPU queue size: 4

-Windowed mode-
Backbuffers: 8
GPU flushing set to flush & flush & wait, don't flush, don't flush.

-Exclusive mode-
Frames in advance: 3
GPU flushing set to flush, don't flush, don't flush, flush and wait.

Hope that helps a little but with you're dealing with a 0.92MP to 3.8MP upscale so whilst it's around 2x on x and y axis individually it's 4x the total detail.
Your GPU is dealing with 3x as many pixels than mine at the start and almost twice as many with the final result.
My upscale is 0.33MP to 2.07MP, the work required for your upscale could be around twice much as mine and therein likely lies your issue.

I see. That make sense. When you said 2x I was thinking 2x in one dimension so 720p->1440p is a "2x." When I switch to watching on the projector (1080p) then everything is great. It's kind of a pain to switch back and forth but giving up on Jinc altogether is a shame for me too. So I'm in a bind. Thank you for the answer though.

[madshi]

So that means I gotta add 1080p120 Hz too? I thought MadVR calculates if it needs to change the mode to match the fps or not.

Ok. I added 1080p120 and everything works now. Sorry for the false alarm. ^^'
madVR only changes to one of those modes you listed. It tries to find the best matching mode of those you listed. But if you only list one mode, madVR has no choice but to always switch to that.

[ajp_anton]

I would like to keep the size of the OSD as small as possible/reasonable for now.Just assign a keyboard shortcut to the scaling algorithm "toggle" action and then pressing it once will show you which algorithm is active without changing it. I'd really like to keep the OSD as small as possible for now.Another subsection for "user interface"? =)

[namaiki]

I can easily do 704x480 to 1920x1080 which uses 71-74% of my HD4000 with no dropped frames in windowed or exclusive mode.


Aaah! 704x480 ~24fps -> 1920x1080 eats up about 95% of my GeForce 330M. 720p -> 1080p reports as 99% usage, but mostly no dropped frames reported in OSD and looks smooth. I can overclock core from 475-> 620 to get it down to about 80% but Jinc is mostly usable.

edit: Oh, didn't notice that the HD4000 beats my card in 3DMark Vantage...

[nlnl]

What am I supposed to see in those logs? I'm not sure. I do see some frame drops in the logs, but it's always after something happened and the frame drops stop very quickly afterwards, going back to fluid playback. At least according to what I could see from a quick look at the logs.
madshi

Sorry, did not tell you.
When I do: Start, Seek, Switch Film -> Video -> Film ... (23->59->23), video stops or sometimes MPC-HC freezes.
Sometimes I can exit player but the process MPC is in memory.
Sometimes i can not and see black screen.

Edit:
Another log
https://dl.dropbox.com/u/43376972/madVR%20-%20log.7z

[Hypernova]

I just played around with overclocking HD 4000. Max overclock I can do is 1500 MHz and that brings the rendering time down to ~43ms. If I use tread quality options I can bring it down to ~40ms which will make the video smooth. But I don't want to do that.

So closed...

[Blight]

Madshi:
I encountered several issues:

FSE gets used even if FSE is disabled in the MadVR options (both FSE and Overlay are unchecked).
To duplicate, all you need to do is:
1. Make sure FSE is disabled in MadVR options.
2. Make sure "Adv. Options / Playback / Video - Video Covers entire display area" is enabled in Zoom Player.
3. Press Alt+Enter.

Secondly,
Possibly the same as nev's report... If I try running MadVR under the Delphi debugger, all I see is a black screen, seeking doesn't do anything and closing the player causes a MadVR crash:
https://dl.dropbox.com/u/104094858/madvr_debug_crash.rar

Lastly,
You mentioned that IQualProp interface will be active in the new version and I can't get it with QueryInterface, are you sure it's implemented?

P.S.
The "," char being used for a decimal point is country-specific, some countries (Germany I think) uses "," for a decimal point.
Windows actually has a native decimal point character regional setting somewhere.

[cyberbeing]

I don't think it has anything to do with the anti-ringing filter. Try it without, you'll see similar results. Jinc simply looks different because it's a circular resampler compared to rectangular Spline or Lanczos. Jinc does look softer in some situations, but on the positive side it has noticeably less aliasing than even Lanczos. I already mentioned in my v0.84 announcement post that Lanczos still may look better in some situations due to being sharper. If you test the algorithms with different material, you'll see the positive sides of Jinc, too, I'm sure. E.g. have a look at these pictures:

Lanczos 4 - anti-ringing (http://madshi.net/Lanczos4ar.png) | Jinc 4 - anti-ringing (http://madshi.net/Jinc4ar.png)

I think this test picture (enlarged 400%) shows quite nicely the characteristics of Lanczos and Jinc: Lanczos is able to pull out some details that Jinc is smoothing over. So some fonts look more detailed in the Lanczos upscale. However, Lanczos shows a lot of aliasing (although less than Spline) and some fonts look rather "digital". On the other hand, Jinc is a bit softer overall (much softer in some places, not much/any softer in other places), but the image looks more natural and organic/analog overall. So which is better? Lanczos or Jinc? I think it's a matter of taste. Personally, I think I prefer the look of Jinc.

Judging by that image comparison, Jinc4 has much more residual bright ringing around dark edges compared to Lanczos4. I can see this in my own tests that Jinc4 has more residual dark ringing than Lanczos4 as well. If that's nothing you can improve upon in the anti-ringing filter, you should probably update the graph representation. Judging by how Jinc4 anti-ringing seems to behave, maybe something like this (http://imageshack.us/a/img21/7734/jinc4antiringing.png) would be a more accurate representation.

The main positives I see of Jinc is lack of aliasing in a decently sharp algorithm. Residual ringing, general softness of fine detail, and slowness being the trade-offs. It's an interesting algorithm, maybe useful for doing extreme upscaling, but for general case I don't see it being to my tastes.

[wOxxOm]

MadVR's hotkeys don't work here at all when FFDShow video decoder is used in filter chain with its internal hotkey control is enabled, even though the hotkeys used are quite different, like Ctrl-Shift-Alt-D, which is intercepted by ffdshow as its Ctrl-Alt-D.

[omarank]

Error diffusion can't be done with GPU pixel shaders because error diffusion processes one pixel at a time, using the result of the previous pixel calculation for the next pixel. GPU pixel shaders more or less work on all pixels at the same time, which is the opposite of what error diffusion needs. *Maybe* it might be possible to do error diffusion using OpenCL/CUDA, but it will be difficult to do and likely not perform very well. For screenshots error diffusion is easy because madVR is doing the screenshot processing via CPU instead of GPU.

Is it not possible to use CPU for dithering and dither via error diffusion, as dithering is applied at the last step of processing? I am just wondering. I am very much satisfied with the existing method of dithering, though.

Then I don't know why it doesn't work with madVR. Might be a bug in madVR, but since it seems to work with newer Intel GPUs and all NVidia and AMD/ATI GPUs I don't really know how to fix this, without having access to hardware for testing purposes...

Is there anything I can do here which can help you diagnose the problem?

[Plutotype]

fyi, HD6570, Win7 Ult, playback on second monitor, SD=>1080

Jinc 3tap/no AR, GPU ~36%
Jinc 3tap/AR, GPU ~40%

Jinc 4tap/no AR, GPU ~51%
Jinc 4tap/AR, GPU ~57%

0 dropped, 0 glitches

One thing I noticed, not sure if it's related to recent madVR changes or not, but the OSD is now showing the monitor (LG HDTV) refresh rate at 24.00051Hz, dead on steady. Before the last few builds, it would never set below 24.00083Hz. This is HDMI with the HD6570=>Onkyo=>LG HDTV. Seems strange that the madVR "timer" changes would affect this.

Gonna try a low-end Llano laptop (6520G) next, see what it will do.

Thanks for this, my 5670 should be fine with Jinc 4 taps then.
EDIT: With Jinc@4taps, my HD5670 does not handle 1280x720 to 1920x1080 without dropping frames. GPU load peaks up to 50-60%, which leads to drop outs in madvr queues.

[madshi]

FSE gets used even if FSE is disabled in the MadVR options (both FSE and Overlay are unchecked).
To duplicate, all you need to do is:
1. Make sure FSE is disabled in MadVR options.
2. Make sure "Adv. Options / Playback / Video - Video Covers entire display area" is enabled in Zoom Player.
3. Press Alt+Enter.
This should be fixed in the next build.

Possibly the same as nev's report... If I try running MadVR under the Delphi debugger, all I see is a black screen, seeking doesn't do anything and closing the player causes a MadVR crash
Yes, looks similar to nev's report. Should hopefully be fixed in the next build, too.

You mentioned that IQualProp interface will be active in the new version and I can't get it with QueryInterface, are you sure it's implemented?
No, I meant the "IMadVRExclusiveModeCallback" interface is now implemented. The one you asked for which calls you whenever exclusive mode is entered/left. IQualProp is not implemented yet.

The "," char being used for a decimal point is country-specific, some countries (Germany I think) uses "," for a decimal point.
Windows actually has a native decimal point character regional setting somewhere.
Yes, I know. Here in Germany "," is the usual decimal separator. I've told Delphi to always use ".", though. And for some reason Delphi reverted my "." change whenever madVR used Overlay mode, in v0.83. I still don't understand how that happened. But the problem is fixed in v0.84.

Judging by that image comparison, Jinc4 has much more residual bright ringing around dark edges compared to Lanczos4. I can see this in my own tests that Jinc4 has more residual dark ringing than Lanczos4 as well.
Now that you mention that, I actually see it in the test image. Didn't notice it before, to be honest. I did notice, though, that the ringing at the left side of the parking meter was quite visible with Lanczos, but mostly gone with Jinc. I'll check if I can improve the Jinc anti-ringing to fix that bright ringing near dark edges.

MadVR's hotkeys don't work here at all when FFDShow video decoder is used in filter chain with its internal hotkey control is enabled, even though the hotkeys used are quite different, like Ctrl-Shift-Alt-D, which is intercepted by ffdshow as its Ctrl-Alt-D.
If ffdshow eats the keys before madVR gets access to them, there's not much I can do about it, unfortunately. It seems that ffdshow doesn't check for "Shift", and so eats keys regardless of whether "Shift" is pressed or not. You'll have to complain to the ffdshow devs about this.

Is it not possible to use CPU for dithering and dither via error diffusion, as dithering is applied at the last step of processing? I am just wondering.
Nope, not possible. Dithering is the last step of processing, and that is done on the GPU, not the CPU.

Is there anything I can do here which can help you diagnose the problem?
I don't know. Maybe it would be possible to diagnose and fix it somehow. But to be honest, it would probably cost me many hours, maybe days, and I've so many other things to do. I don't think I'm going to work on that, unless someone insists on sending me an old laptop with this hardware for testing.

[robpdotcom]

Also, could we get the disable/enable FSE mode as a toggle?

But it's already there!

Right now, you must assign one key to enable, and another to disable. I meant just have one shortcut to switch back and forth. Unless that is already there, and I'm missing it.

[madshi]

Why do you need a toggle? Aren't separate actions just as fine?

[madshi]

madVR v0.84.1 released

http://madshi.net/madVR.zip

* added Jinc 8-tap option
* added support for Jinc chroma upscaling
* added shortcut options for Nearest Neighbor scaling
* fixed: exclusive mode was used in ZoomPlayer even if it was disabled
* fixed: settings dialog showed blank page when starting with key shortcut page
* fixed: crash when creating & freeing madVR instance without playing a video

[madshi]

madshi

Sorry, did not tell you.
When I do: Start, Seek, Switch Film -> Video -> Film ... (23->59->23), video stops or sometimes MPC-HC freezes.
Sometimes I can exit player but the process MPC is in memory.
Sometimes i can not and see black screen.

Edit:
Another log
https://dl.dropbox.com/u/43376972/madVR%20-%20log.7z
The problem is that you just send a log and then say "sometimes this happens" and "sometimes that happens". But I've no idea which of the things that sometimes happen your log is supposed to show and at which runtime in the log. Please understand that these log files contains tens of thousands of lines of text and I can't just read every line to try to find the place where things went wrong etc. If you upload a log please let me know what you did exactly to create that log, which problem occurred exactly, what you did after the problem occurred etc. Only that way *maybe* I have a chance to find something in the log file.

If the media player freezes, please press Ctrl+Alt+Shift+Break to create a freeze report.

[robpdotcom]

Why do you need a toggle? Aren't separate actions just as fine?

I just thought one key made more sense, since you can't have it enabled and disabled at the same time.

Anyway, no big deal. I don't want to be a pain.

[Superb]

Freeze report when using "Save Thumbnails" in MPC-HC: http://www.sendspace.com/file/ol66yq

madVR changed settings:
Disabled decoding. Using LAV Video.
Enabled windowed overlay.
Set both display devices to "Digital Monitor / TV".
All other options are default.
MPC-HC 1.6.5.6069 (96c9d77) x86.
madVR v0.84.1.

[ryrynz]

Madshi, 0.84.1 doesn't hold the number of taps when I set anti-ringing on Jinc. I set Jinc to 8 taps leave the options, enable anti-ringing and it defaults to 4 taps.

Hard to see any real difference with Jinc 8 taps without zooming in greatly, would be overkill me thinks.

[Blight]

Madshi:
v0.84.1 no longer shows the black screen and crash when running under a debugger. Thanks.

[madshi]

Freeze report when using "Save Thumbnails" in MPC-HC: http://www.sendspace.com/file/ol66yq

madVR changed settings:
Disabled decoding. Using LAV Video.
Enabled windowed overlay.
Set both display devices to "Digital Monitor / TV".
All other options are default.
MPC-HC 1.6.5.6069 (96c9d77) x86.
madVR v0.84.1.
On a quick check I don't see anything wrong in that freeze report. Doesn't seem to me that madVR is frozen. Maybe MPC-HC itself is frozen? Don't know for sure. Can you reproduce the problem? If so, how exactly?

Madshi, 0.84.1 doesn't hold the number of taps when I set anti-ringing on Jinc. I set Jinc to 8 taps leave the options, enable anti-ringing and it defaults to 4 taps.
Will be fixed in the next build.

Hard to see any real difference with Jinc 8 taps without zooming in greatly, would be overkill me thinks.
Agreed.

[madshi]

v0.84.1 no longer shows the black screen and crash when running under a debugger.
Glad to hear that. Hope the FSE problem is fixed, too?

[nlnl]

The problem is that you just send a log and then say "sometimes this happens" and "sometimes that happens". But I've no idea which of the things that sometimes happen your log is supposed to show and at which runtime in the log. Please understand that these log files contains tens of thousands of lines of text and I can't just read every line to try to find the place where things went wrong etc. If you upload a log please let me know what you did exactly to create that log, which problem occurred exactly, what you did after the problem occurred etc. Only that way *maybe* I have a chance to find something in the log file.

If the media player freezes, please press Ctrl+Alt+Shift+Break to create a freeze report.
OK. I get it.
1) Start playing file, then seek, then switch Film-Video-Film (23-59-23), video stops, press ALT-X, exit MPC, kill MPC process.
Log attached.
2) Start playing te same file, then seek, then switch Film-Video-Film (23-59-23), black screen, press Ctrl+Alt+Shift+Break, press ALT-X, exit MPC, kill MPC process.
Freez report attached.

Thanks for patience :)
https://dl.dropbox.com/u/43376972/madvr%20freez%20and%20log.7z

Edit:
Sorry, I did the following:
Start playing file, then seek, then switch Film-Video-Film (23-59-23), black screen, press Ctrl+Alt+Shift+Break, press ALT-X, exit MPC, kill MPC process.
Freez and log report attached (log and freeze reports for the same playback)

[mzso]

Nope, not possible. Dithering is the last step of processing, and that is done on the GPU, not the CPU.

Isn't it just infeasible because of the need to pass the data back and forth between the two processors? Because that might just change next year. Just wondering....

[jokerb47]

* fixed: KMPlayer + madVR decoders + "delay playback start" was always paused
:thanks:madshi. Now it's fine with KMPlayer.
I have a question about internal decoder. Is there a way to use it with J.River? It doesn't allow to choose madVR as decoder.

[leeperry]

I don't get any picture with 0.84.1, everything is fine with 0.84 :o

I'm feeding 0-255 RGB32 and I've got "force full range input" enabled. I can seek and PotP works fine, but all I get is a black screen with the latest versions of Pot, LAV Splitter, Reclock and ffdshow and a 8800GS on XPSP3 with the old rendering path. 0.84 works perfectly.

Lemme know if you want a log of 0.84 & 0.84.1 in order to compare them? :thanks:

[Gser]

With 0.84.1 Jinc 8 drops frames like crazy no matter if its luma, chroma, anti-ringing or linear light. Win7 x64, ati 5870.

[mzso]

Is it possible that I'm getting GPU memory leaks with Potplayer+madVR? Because looking at the graphs it seems like most of the memory isn't freed when I stop playback only after a I close the player.

[nevcairiel]

With 0.84.1 Jinc 8 drops frames like crazy no matter if its luma, chroma, anti-ringing or linear light. Win7 x64, ati 5870.

Jinc8 was not available before. :p
madshi even said that Jinc8 will be *very* demanding, and he was not sure if any GPU can properly run it in real-time.

[iSunrise]

Bug report for 0.84.1

Trying to assign a keyboard shortcut on any action like Ctrl+Alt+Shift+# results in an error dialog:

http://www.abload.de/img/errorzqp83.png

[SamKook]

A 320x240 video in FSE 1366x768(I only have a 720p TV) only uses about 50% of the GPU with Jinc 8 taps on my 560Ti so there still plenty of room for crazy settings. It gives around 16.1 ms of rendering time.

[pie1394]

Jinc scaler is indeed quite GPU power demanding.

With the following configuration, it takes 32ms to render a video frame for 1280x720 --> 1920x1080 upscaling by madVR.

If Luma upscaling is changed to Lanczos3 w/o Anti-Ring, it just takes 17ms or less.



OS: Win7 x64 SP1
CPU: E8400@3.6 GHz
RAM: 8GB
GPU: MSI Lightning GTX260+ 1792MB (216 cuda core)

[madvr 0.84]
Chroma: Lanczos3 + Anti-Ring
Luma: Jinc3 + Anti-Ring
Other options: Q8, FSE
Gamma correction: off

[noee]

Fwiw, low-end Llano A6-3400M/6520G (stock speeds) SD=>1600x900 (laptop monitor), FSE, jrMC18/LAV/VideoClock ON:

Jinc3 w/AR: GPU ~49%
Jinc3 no AR: GPU ~44%

Jinc4 w/AR: GPU ~97% (drops frame about every 5 secs)
Jinc4 no AR: GPU ~95% (drops frame about every 10 secs)

Edit:

SD=>1080p upscale results

Jinc3 w/AR: GPU ~83%
Jinc3 no AR: GPU ~81%

Jinc4: GPU 100% (drops frames like crazy, not watchable)

[6233638]

Oh well, I remember. The point is: I'm afraid of making the settings dialog too complex. It's already frightening to new users as it is right now. Maybe some time in the future...It seems to me that while madVR is still under active development, it might be beneficial to allow control over the anti-ringing, bicubic parameters etc. While I'm sure it's complicated, and you have done a good job tuning the filters, it might not hurt to have a few other sets of eyes look over it, and see if it can be improved in any way.

Perhaps the solution is to have a "basic" and "advanced" UI.

[SamuriHL]

Jinc8 was not available before. :p
madshi even said that Jinc8 will be *very* demanding, and he was not sure if any GPU can properly run it in real-time.

We gotta try it though. ;) I mean, if the 3770k and the 680 can't handle it, well, then it's not a very useful option. LOL! :D

[vomanci]

Tested Jinc in image upscaling 720x480 -> 1920x1080 with 8800gt OC 700/1800 (from 600/1800) using FSE (using chroma and image downscaling Lanczos3, other setting the default ones) :

Jinc3 w/AR: GPU 50% rendering 19.75 ms
Jinc4 w/AR: GPU 85% rendering 29.50 ms
Jinc8 w/AR: GPU: 100% rendering 105ms (not playable)

Lanczos3 w/AR: GPU: 13% rendering 6ms
Lanczos4 w/AR: GPU: 20% rendering 6.65ms
Lanczos8 w/AR: GPU: 23% rendering 9.90ms

While Jinc is perfectly usable, i can't see any difference (in my tests). Not like i'm zooming while playing a movie, i tested in real usage. I remained at Lanczos3 mostly because i underclock my videocard to 300/450 and i still have low gpu usage and rendering times.

Good work with this renderer, realy nice.

[vivan]

Since 0.84.1 4:4:4 input is broken (green screen).

[Superb]

On a quick check I don't see anything wrong in that freeze report. Doesn't seem to me that madVR is frozen. Maybe MPC-HC itself is frozen? Don't know for sure. Can you reproduce the problem? If so, how exactly?The freeze happens every time (when using the "Save Thumbnails" functionality). I might go over MPC-HC's code to find the point of freeze (if it's there).
Does "Save Thumbnails" work on your machine?

[Budtz]

Tested Jinc in image upscaling 720x480 -> 1920x1080 with 8800gt OC 700/1800 (from 600/1800) using FSE (using chroma and image downscaling Lanczos3, other setting the default ones) :

Jinc3 w/AR: GPU 50% rendering 19.75 ms
Jinc4 w/AR: GPU 85% rendering 29.50 ms
Jinc8 w/AR: GPU: 100% rendering 105ms (not playable)

Lanczos3 w/AR: GPU: 13% rendering 6ms
Lanczos4 w/AR: GPU: 20% rendering 6.65ms
Lanczos8 w/AR: GPU: 23% rendering 9.90ms

While Jinc is perfectly usable, i can't see any difference (in my tests). Not like i'm zooming while playing a movie, i tested in real usage. I remained at Lanczos3 mostly because i underclock my videocard to 300/450 and i still have low gpu usage and rendering times.

Good work with this renderer, realy nice.

Yes descussions regarding this often revolve around zooming in. I cant much see the point in differences that cant be seen when just normally watching a movie. Is there any real-world difference between Jinc and the others? I am only waching 720p or 1080P content on a 40 inch tv from 2011. Would it even be possible for the human eye to see a differnce on such a setup?

[madshi]

OK. I get it.
1) Start playing file, then seek, then switch Film-Video-Film (23-59-23), video stops, press ALT-X, exit MPC, kill MPC process.
Log attached.
2) Start playing te same file, then seek, then switch Film-Video-Film (23-59-23), black screen, press Ctrl+Alt+Shift+Break, press ALT-X, exit MPC, kill MPC process.
Freez report attached.

Thanks for patience :)
https://dl.dropbox.com/u/43376972/madvr%20freez%20and%20log.7z

Edit:
Sorry, I did the following:
Start playing file, then seek, then switch Film-Video-Film (23-59-23), black screen, press Ctrl+Alt+Shift+Break, press ALT-X, exit MPC, kill MPC process.
Freez and log report attached (log and freeze reports for the same playback)
Hmmmm... I've looked at the freeze report and the log and it seems that the cause of the problem is that asking the GPU VSync position stops working for some reason. I'm not sure why. Which GPU, which drivers and which OS are you using? Have you tried updating to the latest drivers?

Isn't it just infeasible because of the need to pass the data back and forth between the two processors? Because that might just change next year. Just wondering....
Yes, that's the main reason. But why would that change next year?

I have a question about internal decoder. Is there a way to use it with J.River? It doesn't allow to choose madVR as decoder.
I'm not sure. It'd be better to ask the J.River guys about this.

I don't get any picture with 0.84.1, everything is fine with 0.84 :o
Since 0.84.1 4:4:4 input is broken (green screen).
Ouch. Just wrote "=" instead of "==" at one place. Sometimes I hate C++. Such stupid bugs could never happen in Delphi.

With 0.84.1 Jinc 8 drops frames like crazy no matter if its luma, chroma, anti-ringing or linear light. Win7 x64, ati 5870.
I kinda expected that. Simple solution: Don't use Jinc 8! I've added it on request, but I don't recommend using it.

Is it possible that I'm getting GPU memory leaks with Potplayer+madVR? Because looking at the graphs it seems like most of the memory isn't freed when I stop playback only after a I close the player.
I think PotPlayer keeps madVR running even when you stop playback, because PotPlayer draws some of its GUI through madVR sometimes (depending on PotPlayer configuration). If you think there might be a leak, the first thing would be to double check with other media players. If the same problem occurs with e.g. MPC-HC then it might be a bug in madVR. Otherwise it might either be intentional behaviour of PotPlayer, or it might be a bug in PotPlayer.

Trying to assign a keyboard shortcut on any action like Ctrl+Alt+Shift+# results in an error dialog
Oh well. I'm currently supporting key shortcuts only for common keys like chars, numbers, or most of the special keys like arrows, tab, enter, home etc. The problem with keys like "#" is that the same key could behave different with a different language setting or a different keyboard layout. So for now I'm sorry to say, but you gotta stick to the keys that are supported. I'll probably add support for all keys in a future version.

It seems to me that while madVR is still under active development, it might be beneficial to allow control over the anti-ringing, bicubic parameters etc. While I'm sure it's complicated, and you have done a good job tuning the filters, it might not hurt to have a few other sets of eyes look over it, and see if it can be improved in any way.

Perhaps the solution is to have a "basic" and "advanced" UI.
Any other wishes which would consume multiple days of my time? ;) No thanks. I'd really like to implement all the missing features first before doing stuff like basic/advanced UI switches or other fancy stuff like that. There will be a time for fine tuning, cosmetical stuff and user interface polishing, but it is not now.

The freeze happens every time (when using the "Save Thumbnails" functionality). I might go over MPC-HC's code to find the point of freeze (if it's there).
Does "Save Thumbnails" work on your machine?
Oh well, just tried that for the first time, and yes, I can reproduce the freeze here. However, it seems that madVR continues to run just fine. Well, playback stops, but try pressing Ctrl+J to show/hide the debug OSD, it still works and the debug OSD is updated regularly. That indicates that the madVR private threads still run just fine. The freeze seems to be somewhere inside of MPC-HC. I think it's likely a bug in MPC-HC, but I can't say for sure. Maybe if you upload a freeze report from this situation, created with an MPC-HC debug build (with PDB files), I might be able to comment on where MPC-HC is stuck and why.

[madshi]

Yes descussions regarding this often revolve around zooming in. I cant much see the point in differences that cant be seen when just normally watching a movie. Is there any real-world difference between Jinc and the others? I am only waching 720p or 1080P content on a 40 inch tv from 2011. Would it even be possible for the human eye to see a differnce on such a setup?
It all depends on multiple things:

(1) Scaling factor: The bigger the scaling factor the bigger the visible difference between various scaling algorithms. Upscaling from 720p to 1080p is not a very high scaling factor, so the differences will be much smaller compared to DVD -> 1080p upscaling.

(2) Viewing angle. A 40 inch TV isn't very big. It might take a front projection setup to see differences with real-world material when upscaling 720p to 1080p. Or at least you'd have to sit very near to that 40 inch TV.

(3) Some people are much more sensitive to specific artifacts than others. E.g. some people see DLP rainbows very easily, others not at all. Some people are terribly bothered by ringing artifacts, others are not. Some people are terribly bothered by aliasing artifacts, others are not. Etc etc.

[Blight]

madshi:
Yes, FSE being used while unchecked in the madvr options is no longer an issue.

[madshi]

madVR v0.84.2 released

http://madshi.net/madVR.zip

* fixed: 4:4:4/RGB videos -> green or black screen (introduced in v0.84.1)
* fixed: Jinc8 was not always properly remembered

[nlnl]

Hmmmm... I've looked at the freeze report and the log and it seems that the cause of the problem is that asking the GPU VSync position stops working for some reason. I'm not sure why. Which GPU, which drivers and which OS are you using? Have you tried updating to the latest drivers?


ATI 5750, 12.8, W7. Yes, I have updated.
Thanks!