madVR - high quality video renderer (GPU assisted)

[blackjack12]

Originally Posted by madshi

Thanks. I do wonder why you upload a text file unzipped, but anyway. Can you please try these 2 test builds:

http://madshi.net/madVR837test.rar

Does either of them fix the problem (if so, which)? If neither of them fixes the problem, please double check whether the problem also occurs with v0.83.1, which you can download here:

http://www.videohelp.com/tools/madVR/old-versions#download

Here are the results from testing all 4 versions

TEST 1 Trial - madVR837test1:
Windowed Mode: Bad – but seems to take more jumps before problem
FSE Mode: Good

TEST 2 Trial - madVR837test2:
Windowed Mode: Worse
FSE Mode: Good

TEST 0.83.1:
Windowed Mode: Bad
FSE Mode: Good

TEST 0.82.5:
Windowed Mode: Good
FSE Mode: Good

[mzso]

Originally Posted by oddball

Systray icon

Same thing. How do you imagine it being exclusive when something else is on screen than madVR? (BTW the systray brings up a context menu too)

Originally Posted by Moony349

For FSE in Potplayer:

General - Skin - Control - Method = Disable

This works for me. Any method besides disable (D3D9 or Image Overlay) will not allow MadVR to go FSE for me.

Not my experience. the d3d9n adds the capabilit to show the potplayer skin in FSE.

[Blight]

Madshi:
To cover all the bases I would like the option to:
1. Rotate 90, 180, 270
2. Flip vertical & Horizontal

And yes, monitor orientation doesn't have anything to do with this feature request as I intend it to be a user-function.
You should make the user-initiated rotation value relative to the monitor orientation if you support that feature too.


Prinz:
Sadly, I don't know what's causing this error, in the next build, you wont see any error as ZP's code now catches it.
I'm not sure though what side-effects it may have (hopefully none).

[omarank]

I installed madVR v0.83.7 on an old laptop, which has mobile Intel 4 series express graphics. I wasn’t expecting it to handle madVR but to my surprise it handled pretty well. I just had to play with the scaling algorithms and I could play Blu Ray videos without any dropped/ delayed frames. I could set any of the chroma algorithms other than Lanczos and Spline, but I could only set Nearest Neighbour or Bilinear for luma. I did my testing using Mitchell Netravalli for chroma and Bilinear for luma.

I found two issues however:
1. When I play any interlaced video, I see a “deinterlacing failed” message.
2. If I select Overlay mode, I see a jerky video (like having back and forth motion). Even though the video is jerky, queues remain stable and dropped/ delayed frames or glitches counters don’t go up.

There are two observations:
1. If I use yCMS, I can’t play videos without any dropped/ delayed frames even if I select all the “trade quality for performance” options with the 3dlut bit depth set as 6 bit and use Bilinear for both chroma and luma scaling.
2. I see one or two presentation glitches every 15-20 minutes in some videos.

I don't consider the above observations as issues, simply due to the fact that the hardware used is rather old.

I just wanted to confirm if deinterlacing and overlay mode are not supported for old Intel integrated graphics.

[mbordas]

Can anyone verify a problem with the latest LAV filters (0.51.3-89-g8c3b2ec) with dvd decoding? Certain dvds seem to display a black screen for the menu (eg Michael Clayton, Syriana, Princess Mononoke) and the only way to fix it is to disable automatic source detection in madVR. Many other dvds play just fine.

Unfortunately, disabling source detection isn't a good solution (for me) cause it leads to other problems - the display refresh rate is for some reason set to 23.9999 instead of 23.976 even though this isn't in madVRs list of display modes..?

I'm posting here rather than in the LAV forum since the fix seems related to madVR, but given that the dvd decoding is a brand new feature in LAV the problem may actually be there.

[Pat357]

Originally Posted by mbordas

Unfortunately, disabling source detection isn't a good solution (for me) cause it leads to other problems - the display refresh rate is for some reason set to 23.9999 instead of 23.976 even though this isn't in madVRs list of display modes..?

You don't have 24Hz in madVRs list of display modes..?
How do you play real movie-content then ??

[mbordas]

Well, with nvidia drivers, I've always had to create a custom resolution. You're right that it says 24Hz but it's really 23.976. So I guess I don't know, where does the "true" 24 hz refresh rate come from? (and I don't think I have any content that is true 24 Hz anyway).

I'm getting weirdly different results from different dvds. Some exhibit the behavior I described, some don't, some just hang. So I'm just really confused, and curious if anyone else is seeing these problems.

[6233638]

Slight bug with the stats when the clock stabilises:



I assume this happens when it reaches 1 frame repeat every 10 days, but I've only ever noticed once it has already happened.

I don't think this is anything new, I only started noticing it after making some tweaks to the system timers.

Originally Posted by mbordas

Well, with nvidia drivers, I've always had to create a custom resolution. You're right that it says 24Hz but it's really 23.976. So I guess I don't know, where does the "true" 24 hz refresh rate come from? (and I don't think I have any content that is true 24 Hz anyway).

I'm getting weirdly different results from different dvds. Some exhibit the behavior I described, some don't, some just hang. So I'm just really confused, and curious if anyone else is seeing these problems.

If you want 24/1.001, you should specify 1080p23, rather than 1080p24. (1080p24 = 24.000Hz)

I recommend using 1080p24 and ReClock however.

[cyberbeing]

In the past madshi has said 1#J days was likely a math overflow, which basically means you're exceeded the maximum duration which madVR is capable of measuring. This has been a known issue since madshi first added these readings to the OSD. It's common to see 1#J days when using Reclock, though still possible if less common without Reclock, by just using an extremely fine tuned refresh rate.

[nevcairiel]

Just feel happy when you see it there.

[6233638]

Originally Posted by cyberbeing

In the past madshi has said 1#J days was likely a math overflow, which basically means you're exceeded the maximum duration which madVR is capable of measuring. This has been a known issue since madshi first added these readings to the OSD. It's common to see 1#J days when using Reclock, though still possible if less common without Reclock, by just using an extremely fine tuned refresh rate.

With ReClock I was previously getting times in the 3-5 hours range, but after making tweaks to the system timers, that now stabilises at days. (seems to get more stable the longer it's active)

What I did was:

1. Enable HPET by turning it on in the BIOS and running bcdedit /set useplatformclock true in a command prompt (Windows 7) then rebooting.
2. Set Timer Resolution to 0.5ms with TimerResolution

Originally Posted by nevcairiel

Just feel happy when you see it there.

I'm definitely not complaining, just hadn't seen any discussion of it before.

[madshi]

Originally Posted by blackjack12

Here are the results from testing all 4 versions

TEST 1 Trial - madVR837test1:
Windowed Mode: Bad – but seems to take more jumps before problem
FSE Mode: Good

TEST 2 Trial - madVR837test2:
Windowed Mode: Worse
FSE Mode: Good

TEST 0.83.1:
Windowed Mode: Bad
FSE Mode: Good

TEST 0.82.5:
Windowed Mode: Good
FSE Mode: Good

Weird. Ok, I guess the only way to get to the bottom of this is create some test builds which are somewhere between 0.82.5 and 0.83.1 to find out which exact change cause the problems. I'll do that later today, or maybe during the weekend.

Originally Posted by Blight

Madshi:
To cover all the bases I would like the option to:
1. Rotate 90, 180, 270
2. Flip vertical & Horizontal

Ok, will put that on my to do list. But it might take a while until I get to that.

Originally Posted by omarank

I installed madVR v0.83.7 on an old laptop, which has mobile Intel 4 series express graphics. I wasn’t expecting it to handle madVR but to my surprise it handled pretty well. I just had to play with the scaling algorithms and I could play Blu Ray videos without any dropped/ delayed frames. I could set any of the chroma algorithms other than Lanczos and Spline, but I could only set Nearest Neighbour or Bilinear for luma. I did my testing using Mitchell Netravalli for chroma and Bilinear for luma.

I found two issues however:
1. When I play any interlaced video, I see a “deinterlacing failed” message.
2. If I select Overlay mode, I see a jerky video (like having back and forth motion). Even though the video is jerky, queues remain stable and dropped/ delayed frames or glitches counters don’t go up.

Not sure why deinterlacing fails. I think it should work. Or are you still using XP? In that case maybe you don't have .NET 3.0 (or higher) installed? That's required for DXVA2 deinterlacing. Or maybe the Intel drivers don't support DXVA2 deinterlacing in XP? Don't know. If you're on a newer OS then in theory I think it should work. It would be very hard for me to fix. Does deinterlacing appear to work in EVR?

Can't say much about Overlay mode. Well, I'm presenting frames in a different thread, maybe that confuses the Intel driver. But reportedly it works fine on newer Intel GPUs, so it's probably a driver bug. Nothing I can do about it...

Originally Posted by mbordas

Can anyone verify a problem with the latest LAV filters (0.51.3-89-g8c3b2ec) with dvd decoding? Certain dvds seem to display a black screen for the menu (eg Michael Clayton, Syriana, Princess Mononoke) and the only way to fix it is to disable automatic source detection in madVR.

What "source detection" do you mean exactly? FWIW, madVR does not officially support DVD playback yet. If this problem is reproducable and only occurs with LAV, then maybe you could ask nevcairiel to look into it? Not sure if he's interested in fixing that, though. Of course it could also be a bug in madVR, can't really say...

[madshi]

madVR v0.84.0 released

http://madshi.net/madVR.zip

* added media player screenshot functionality (IBasicVideo::GetCurrentImage)
* added "Jinc" image upscaling algorithm
* added anti-ringing filter for downscaling
* added "user interface\keyboard shortcuts" settings page
* added option to apply keys "only if media player has keyboard focus"
* added options to adjust the keys for all existing keyboard shortcuts
* added shortcut option for directly selecting specific source primaries
* added shortcut option for directly selecting specific deint content types
* added shortcut option for en/disabling automatic fullscreen exclusive mode
* added shortcut option for disabling exclusive mode for 10 seconds
* added shortcut option for en/disabling the display mode switcher
* added shortcut option for en/disabling dithering
* added shortcut option for changing the display bitdepth
* added various shortcut options for selecting scaling algorithms
* added IMadVRExclusiveModeCallback interface for exclusive mode notifications
* renamed "luma up/downscaling" to "image up/downscaling"
* renamed "use 10bit luma buffer ..." to "use 10bit image buffer ..."
* scaling algorithm parameters (softness, sharpness, taps) are remembered now
* scaling algorithm red/green graphs (settings dialog) now cover anti-ringing
* modified shader storage to decrease madVR.ax file size
* fixed: KMPlayer + madVR decoders + "delay playback start" was always paused
* fixed: settings dialog had problems with yCMS data when using Overlay mode
* fixed: cosmetical issue in fullscreen windowed mode when changing video files

Some notes:

(1) Screenshots are done after deinterlacing and color conversion, but before scaling and aspect ratio correction. Screenshots are always converted to 8bit fullrange RGB (0-255), using error diffusion, which is a higher quality dithering algorithm compared to what madVR does during playback. Screenshots only work if the media player supports it. MPC-HC does out of the box, haven't tested other media players, but they will likely work, too.

(2) The "Jinc" (3 or 4 taps) image upscaling algorithm is totally new. It's somewhat similar to Lanczos, but Jinc is slower and has slightly higher quality. You can think of Lanczos as a rectangular resampling filter. In comparison Jinc is a circular resampling filter. Jinc is slightly softer than Lanczos, but it has very low aliasing artifacts (lowest of all algorithms) and ringing is less strong compared to Lanczos. Furthermore the madVR anti-ringing filter works especially well together with Jinc, so with the anti-ringing filter Jinc is pretty much ringing-free without any negative side effects that I could see. In certain situations Lanczos might look better (because it's slightly sharper than Jinc), but overall I believe Jinc beats Lanczos in quality, thanks to a more natural look with less ringing and aliasing artifacts. Jinc comes at a quite big performance cost, though. At low scaling factors, Jinc is comparable in performance to Lanczos8. However, with bigger scaling factors, Jinc3 can be twice as slow as Lanczos8 and Jinc4 twice as slow as Jinc3. Ouch...

[ryrynz]

Holy crap it's Christmas, and I was going to head to bed.. This improved 8 bit dithering done with screenshots, how come this isn't an option in real time? Is it that slow?
I thought MadVR took "no shortcuts, highest quality has priority over anything else"

[madshi]

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.

[SamuriHL]

Jinc eh? Guess I'm gonna have to get my head out of the Android world for a bit and do some testing. Thanks!

Sent from my Xoom using Tapatalk 2

[ryrynz]

The Keyboard shortcuts are awesome, I was hoping for an auto populate when the keys were pressed instead of typing them in but that's okay. Status messages on the hotkey activations as well! It's everything I asked for last year. Thanks!
Jinc is very similar to Lanczos as you said. I think I'd generally prefer Jinc but as far as ringing goes they're both very close to each other at least as far as dark line animated content anyway..
Is there any ETA on the more advanced anti-ringing filter?

I think though what I love most of all about this build is this

Originally Posted by madshi

* fixed: cosmetical issue in fullscreen windowed mode when changing video files

Ive noticed that on animated content Jinc with 4 taps has ringing closer to dark lines on animated content which doesn't really look better or worse than 3 taps as far as ringing is concerned but with anti-ringing enabled
4 taps due to having it's ringing closer to the dark lines has more of it's ringing wiped out than 3 taps. As a result 4 taps looks better than 3 taps when anti-ringing is enabled, it murders my HD3000 however, so no go.

Madshi what's your thoughts on having a upscaling or downscaling value listed in the OSD? Or perhaps a combined output res along with upscaling/downscaling value depending on what's being done.

[madshi]

Originally Posted by ryrynz

Jinc is very similar to Lanczos as you said. I think I'd generally prefer Jinc but as far as ringing goes they're both very close to each other at least as far as dark line animated content anyway..
Is there any ETA on the more advanced anti-ringing filter?

The Jinc anti-ringing algorithm is already more advanced. You still see problems with dark lines with Jinc anti-ringing?

[Moony349]

Madshi,

Would it be theoretically advantageous/possible to allow more taps for Jinc upscaling and/or chroma upscaling with Jinc?

Also: MadVR is amazing! Forgot to mention that one.

[THEAST]

Originally Posted by madshi

Weird. Ok, I guess the only way to get to the bottom of this is create some test builds which are somewhere between 0.82.5 and 0.83.1 to find out which exact change cause the problems. I'll do that later today, or maybe during the weekend.

Actually I had the exact same issue as blackjack12 with the builds between v0.82.5 and v0.83.7 both on my dekstop (AMD HD 7950) and laptop (Nvidia 520M) but I just tested build v0.84 on my desktop and the issue is gone. Though I was on XP at that time and now I'm on 7.