madVR - high quality video renderer (GPU assisted) [Archive] - Page 114

Ger

21st February 2011, 06:48

Well, it sounds like motion compensated deinterlacing. I'm just not 100% convinced, because that would almost be too good to be true.

Found some info in this article (http://ixbtlabs.com/articles2/video/avivo.html) from 2005-12-30:

The best de-interlacing methods, yielding the highest quality, are complex. There are no simple ways to do this job to eliminate artifacts in all cases. Old de-interlacing methods, simple and non-adaptive to motion speed changes, are becoming a thing of the past, providing not the best quality. Modern video chips use complex de-interlacing, so called adaptive methods. Adaptive de-interlacing methods selectively cut pixels, based on information about motion in a frame.

Before AVIVO, ATI products already used adaptive methods. But it's R520, RV530, RV515 video chips supporting AVIVO, where the company introduced the advanced method - vector adaptive de-interlacing. This improved algorithm resembles the new method in NVIDIA PureVideo - spatial-temporal. In brief, it works in the following way: the algorithm selects data to build a progressive frame from either the raw field data, when motion is detected to be low; in case of fragments with fast motion, it uses video data that is interpolated along several vectors, which provides the greatest amount of details to the progressive frame. Vector adaptive de-interlacing removes the comb effect very well even in traditionally difficult scenes (for example, diagonal lines at a small angle, or flying colors ) without much quality losses and blurring.

madshi

21st February 2011, 09:23

I have been testing with the same clip yes, and it's nothing that should be so processor-intensive anyway, it's a video of someone playing a SNES game. With Haali there's little to no increase of CPU usage, with madvr on default except for disabling exclusive mode my fan is spinning 700rpm faster and one of my CPU's cores is maxing out (According to Speedfan anyway, not the most reliable program I know, but there's definitely quite a big different in power used regardless)
In the "processes" tab of the task manager, which CPU consumption do you get for Haali, madVR and VMR/EVR?

Have you modified the flush settings in the madVR settings dialog? Avoid "flush & wait (loop)", if you want to keep CPU consumption low.

I already tested that awhile back, forcing full 3D Clocks makes no difference with the 3DLUT problem. Not to mention, forcing 3D Clocks isn't even needed, since they stay at full in such situations anyhow.
Too bad. But maybe it would fix the pause problem?

Found some info in this article (http://ixbtlabs.com/articles2/video/avivo.html) from 2005-12-30:
Thanks. It's not as clearly described as I would like, though.

Luv

21st February 2011, 10:05

I'm loosing it,I can feel it.
I found this:

HD 1080i Test Pattern - detect Vector Adaptive Deinterlacing + others
http://www.avsforum.com/avs-vb/showthread.php?t=1157287

If I check Vector Adapt (V.A) in the CCC + pulldown,the test tells me that I'm using it in wmp9 and TMT5 .187 (Automatic deinterlacing) .
ZP is only using Weave (According to the test).
Ok,such results were expected (No HA with madVR,etc)
BUT,in this case,I have obvious combing if I use 1080i material with wmp9 and TMT while absolutely none in ZP with madVR.
Moreover,choosing V.A or automatic deinterlacing in the CCC does not change the fact that I have absolutely no combing in ZP with madVR.Hence my "extraordinary results" expression.

PS:
about my rendering times,Madshi,it doesn't seem to impact anything.Very relieved.

madshi

21st February 2011, 11:48

BUT,in this case,I have obvious combing if I use 1080i material with wmp9 and TMT while absolutely none in ZP with madVR.
Moreover,choosing V.A or automatic deinterlacing in the CCC does not change the fact that I have absolutely no combing in ZP with madVR.Hence my "extraordinary results" expression.
Hmmmm... If you run that cheese slices test pattern with ZP + madVR, what image quality do you get? No combing, you say. But is it really VA quality?

I'll be really glad if you look into it.
Here is a sample you can use for testing (make two copies of the file in the same folder)
http://www.mediafire.com/?zmn2nyyrzat
but in reality it should happen with other files too.

About the splitter - I'm mainly using Haali's one(v1.10.262.12) and the error is more rare with it but with MPC's one (v1.4.2667.0) it's happening every time.
In the following archive
http://www.mediafire.com/file/5s14932rm5si6ja/madVR%20-%20log%20-%20access%20violation%20%28files%20to%20reproduce%29.rar
you'll find both .ax files for the splitters, my ZP settings (both a .reg file and and an archive of the MediaAutoGraph folder located in ZP's install folder, which contains the configuration files for media playback - source filters/splitters/decoders) and zplayer.exe (it's the latest beta build).
To get the same configuration as me (as close as possible) you'll have to replace your zplayer.exe and MediaAutoGraph folder. You'll also have to import the .reg file (you can rename HKCU\Software\VirtuaMedia to VirtualMedia_orig so you don't lose your current settings ... and the same with the MediaAutoGraph folder)
After this, start the first sample, go to fullscreen (double click on the video ... or Alt+Enter) and then pres "]" or "PgDn" (which should load the next file) and then you should, hopefully, get the access violation.
I've done as you suggested - but I don't get any crashes here, with your ZP exe, your reg file and your MatroskaSplitter.ax. Maybe it is related to the h264 decoder you're using? On my PC ZP chose to use ffdshow for decoding. Which decoder are you using? Try different decoders...

(Edit: I'm testing with my latest sources, of course. There's also a (very small) chance that the bug is fixed with some other changes I've done. So you should retest with the next madVR version, just to be safe.)

namaiki

21st February 2011, 12:46

pie1394

21st February 2011, 17:40

Not really. I know all the deinterlacing algorithms and which is better than which. I'm just not sure if ATI/NVidia do real motion compensated deinterlacing or not.

Well I don't know that either. But I remembered both the slow and quick motioned deinterlaced video is still very stable and sharp when I viewed it generated from 8800GTS(G80) before.

Unforturnately my current TV does not need it at all for 1080i video, and there is no way to see the difference on Hitachi ALiS PDP panel.

Just curious about motion-compensated method. Do you think it will perform better for very complex scenes like explosion?

It is also believed a HW Motion Estimation engine is required to do this algorithm efficiently. Even the advanced vector processor cannot do that quickly if the Stream Processor clock is not very high or parallel processing units are not enough.

The Spatial-temporal method for Edge-based deinterlace has the fixed computation power and memory bandwidth requirements depending on the actual implemented mode. Thus I think this is what current GPU / Video processor is capable. Maybe I am wrong...

What I did 4 years ago was just something of improved 2D Edge-based DCDi. With pure 2D Edge-based DCDi, it is even jaggy on some common color edge patterns ... Just looks like how low-end TV performs.

3D Edge-based method should improve that. But it needs the HW to do median number selection among at least 5 numbers. Not quite easy to vectorize the calculation just like median of 3 numbers.

My final choice was VT(Vertial Temproal) / 2D Edge-based adaptive per pixel to eliminate the jaggy edges w/o consuming too much computation power. It does create stable static and moving images althrough this is not as sharp as what nVidia G80 can produce.

dawar49

21st February 2011, 18:42

Disabling what? Power management?

As I just mentioned, the 3DLUT massive frame-drop problem isn't the result of the GPU downclocking (it's triggered by various rare out-of-gamut scenes alone). I've even tried forcing an overclock and it makes no difference. The only solution I've found was tossing a more powerful GPU (I tested a GTX 470) at the problem, which had a similar rendering time spike, but was powerful enough to muscle though the 3DLUT problem sections.

what's your memory usage with and without a 3dlut? also, in gpu-z is you gpu load constant (within 1% or 2%) or does it fluctuate? i had a similar out of gamut issue and found it related to insufficient gpu ram. if madvr was able to load everything in the gpu ram, the gpu usage would stay fixed and everything worked fine. if there was insufficient ram, gpu usage would fluctuate heavily and i'd get dropped frames in areas with significant out of gamut colors.

madshi

22nd February 2011, 15:28

madshi, could you please edit the first post:
"- no subtitle support yet"

I know it means that it doesn't support MPC-HC's internal subtitle renderer, but for a few people that I've recommended madVR to, they have initially said that they simply can't use it because they need subtitles, so I've had to explain that they could still use DirectVobSub/VSFilter/FFDShow subtitle renderer.
Will do.

Just curious about motion-compensated method. Do you think it will perform better for very complex scenes like explosion?
It is my understanding that a well designed motion-compensated deinterlacer should only use motion-compensation for image areas where motion estimation seems to be spot on. In other image areas the deinterlacer should fall back to the next best solution. If implemented this way, motion compensation should only ever improve the results, but never make them worse.

Of course motion estimation for complex scenes is difficult. And very probably the performance of motion-compensated deinterlacing will differ a lot, depending on how well motion estimation is implemented in any given deinterlacer.

My current guess is that ATI and NVidia do true motion-compensated deinterlacing. But I'm not sure how good/reliable their motion estimation is, and I'm also not sure how good their "fallback" implementation is.

What I did 4 years ago was just something of improved 2D Edge-based DCDi. With pure 2D Edge-based DCDi, it is even jaggy on some common color edge patterns ...
Did you work on a software or hardware implementation? Which one? (If you're allowed to say.)

3D Edge-based method should improve that.
The 3rd dimension being time (making use of multiple previous and future fields)? Or something else?

My final choice was VT(Vertial Temproal) / 2D Edge-based adaptive per pixel to eliminate the jaggy edges w/o consuming too much computation power. It does create stable static and moving images althrough this is not as sharp as what nVidia G80 can produce.
I'm wondering if DCDi filters are necessary at all, if you use a good resampling algorithm from the get go. E.g. ICBI automatically takes care of smoothing diagonal lines etc. But then you could probably say that ICBI has DCDi type algorithms built in.

madshi

22nd February 2011, 15:48

madVR v0.37 released

http://madshi.net/madVR.zip

* IBasicVideo::SetDestinationPosition not needed, anymore (for J.River MC)
* exclusive -> windowed switch should be a bit faster in Vista+ now
* added options for disabling desktop composition
* added workaround for MPC-HC MPEG2 decoder crash with VOB files (decoder bug)
* settings dialog now remembers the page you've been on
* fixed small bug in video levels output math
I've invested some time to improve the exclusive <-> windowed mode switch in Vista and Windows 7 (it was already working fine in XP before). There are 3 problems known to me right now:

(a) switching takes too much time
(b) an old video frame is sometimes shown for a short moment
(c) mouse cursor is reset to middle of screen

Problem (a) should generally be slightly improved in v0.37 now. But I've found a key solution to make it *much* faster (as fast as XP): By disabling desktop composition. Furthermore I've also found that disabling desktop composition also takes care of problem (b). As a result I've added new options to the madVR settings dialog to define if and when desktop composition is disabled. There are 5 options:

(1) don't disable desktop composition
This is similar to what older madVR versions did. Problem (a) should be improved slightly, but probably not by much. Problem (b) will still occur.

(2) disable desktop composition during exclusive -> windowed mode switch
This will disable + reenable desktop composition whenever madVR switches from exclusive mode to windowed mode. Not sure about problem (a), might eventually be worse for some people, not sure. Problem (b) should be fixed.

(3) disable desktop composition while madVR is in exclusive mode
Not sure about problem (a), might be slightly better or worse. Problem (b) should be fixed.

(4) disable desktop composition while media player is in fullscreen mode
This should fully fix problems (a) and (b). The disadvantage is that context menus and option windows in fullscreen mode will look uglier because Aero is disabled.

(5) always disable desktop composition
This should also fully fix problems (a) and (b). But Aero will be disabled even when running madVR in a small window.

I'm not sure about problem (c). Some of the options above might help with (c) or not. Personally, I didn't experience (c) lately at all.

Please try the options above and give me feedback on which option works best for you. I would like to get rid of most of these options again. My personal favorite is option (4). I hope keeping options (1) and (4) will satisfy everybody. So for anyone who prefers options (2), (3) or (5), please let me know why you prefer them. Thanks.

adam777

22nd February 2011, 16:13

Hi madshi,
I've used options (1) and (4) and here are my findings.
Using option (1)
exclusive -> windowed - (a), (b) - screen shows an old video frame, goes to black and then switch.
windowed -> exclusive - (a) - screen goes to black and then switch.

Using option (4)
exclusive -> windowed - (b) - old frame still shown, plus right click menu not always visible (entries show when you hover over where they should exist, but not the whole menu). switch itself much better.
windowed -> exclusive - almost perfect, no problems.

Adam.

fastplayer

22nd February 2011, 16:22

When you open the settings dialog, the window stays on top of all other windows which is kinda annoying.
Steps to reproduce:
1. Play video with MPC.
2. Go to Play --> Filters --> madVR Renderer --> Edit Settings.
3. Open another app or Explorer window.
4. madVR's settings dialog stays on top of all other currently opened windows.

Edit:
System setup: MPC-HC rev2940, madVR 0.37, Win7 32-bit (Aero enabled).

noee

22nd February 2011, 16:29

Playback with jRiver's MC (beta):

(2) Disastrous results, playback extremely choppy, constant on and off (it appears) of Aero until machine hard locks. I've never had this happen before on Win7, quite interesting.
(3) stable, I see neither (b) or (c)
(4) same results as (3)
(5) As you state, however, Aero never came back, even after I stopped the video, until I stopped MC.

madshi

22nd February 2011, 16:37

Using option (4)
exclusive -> windowed - (b) - old frame still shown, plus right click menu not always visible (entries show when you hover over where they should exist, but not the whole menu).
That's very disappointing... :( Is that fully reproducible? Which GPU and OS are you on?

When you open the settings dialog, the window stays on top of all other windows which is kinda annoying.
This is as intended. I didn't have it like that originally, but then sometimes it opened behind the media player window and people got confused. You're supposed to edit the settings and then close the box again, not to keep it open for hours... :p

(2) Disastrous results, playback extremely choppy, constant on and off (it appears) of Aero until machine hard locks. I've never had this happen before on Win7, quite interesting.
(3) stable, I see neither (b) or (c)
(4) same results as (3)
(5) As you state, however, Aero never came back, even after I stopped the video, until I stopped MC.
So which one do you prefer?

Option (5) only reenables Aero when the madVR instance is destroyed. When it is destroyed depends on MC.

fastplayer

22nd February 2011, 16:39

This is as intended. I didn't have it like that originally, but then sometimes it opened behind the media player window and people got confused. You're supposed to edit the settings and then close the box again, not to keep it open for hours... :p
If it's by design, then I can live with it. :)

noee

22nd February 2011, 16:45

adam777

22nd February 2011, 17:08

That's very disappointing... :( Is that fully reproducible? Which GPU and OS are you on?

Windows 7 x64, Mobility Radeon 4570 512MB.
Problem isn't always reproducible, only happens from time to time (without any logic I could realize).

Here are three screenshots
First one (http://img602.imageshack.us/i/exclusivewindowed.png/) shows a double menu, which is something I can always see when switching from exclusive to windowed, along with the old frame.

Second one (http://img828.imageshack.us/i/exclusivewindowed2.png/) shows what seems like a normal menu, however, when I took it the actual display was of the disappearing menu described above.

Third screenshot (http://img441.imageshack.us/i/exclusivewindowed3.png/) shows a phenomena similar to the first one (with madVR setting window this time), and I think I can conclude that in both cases the menus shown are the correct, current one and the one I opened previously.

Hope this would be of assistance.

madshi

22nd February 2011, 17:12

I prefer option 5. However, is option 4 disabling aero in both full screen modes (windowed and excl)? Maybe it's just my old card, but it seems like with option 4, there is a lot of flashing as the modes switch.
With option 4 Aero is disabled if the media player window covers the full screen (including the taskbar) *and* if the madVR rendering window touches at least 3 edges of the screen.

That means: If you go into fullscreen mode (taskbar not visible, anymore), Aero should be disabled at once. Opening context menus or option windows should not reenable Aero, as long as the media player stays in fullscreen.

You can see if Aero is enabled/disabled in the madVR OSD: If the OSD lists the "composition rate", Aero is enabled. Otherwise it's disabled. FWIW, sometimes it takes a few seconds for Aero to reenable.

madshi

22nd February 2011, 17:16

Windows 7 x64, Mobility Radeon 4570 512MB.
Problem isn't always reproducible, only happens from time to time (without any logic I could realize).

Here are three screenshots
First one (http://img602.imageshack.us/i/exclusivewindowed.png/) shows a double menu, which is something I can always see when switching from exclusive to windowed, along with the old frame.

Second one (http://img828.imageshack.us/i/exclusivewindowed2.png/) shows what seems like a normal menu, however, when I took it the actual display was of the disappearing menu described above.

Third screenshot (http://img441.imageshack.us/i/exclusivewindowed3.png/) shows a phenomena similar to the first one (with madVR setting window this time), and I think I can conclude that in both cases the menus shown are the correct, current one and the one I opened previously.
Hmmmm... How did older madVR version behave? Did they show the same problems? Is the new madVR version at least an improvement? Or is it actually worse for you?

Please also try option (5) "always". Does that get rid of all problems? When using option (4), please enable the madVR OSD (Ctrl+J) and check whether/when the "composition rate" is shown. Thanks.

nlnl

22nd February 2011, 17:19

My video card is ATI 5450 + W7.
Madvr reports that my refresh rate is 23.978 but not exactly 23.976. And EVR CP (MPC HC) reports the same.

Does anybody have 23.976 exactly?
What card do you use?

I upgraded the card from 5450 to 5750 (gigabyte passive) and installed 11.2 driver.
And now Madvr (.36) reports that my refresh rate is 23.976 exactly :):):) !

Madshi thanks for .37 :thanks:

noee

22nd February 2011, 17:52

That means: If you go into fullscreen mode (taskbar not visible, anymore), Aero should be disabled at once. Opening context menus or option windows should not reenable Aero, as long as the media player stays in fullscreen....

Gotcha, yes, this appears to working correctly. It is working properly in dual-monitor playback as well.

adam777

22nd February 2011, 19:39

Hmmmm... How did older madVR version behave? Did they show the same problems? Is the new madVR version at least an improvement? Or is it actually worse for you?

Please also try option (5) "always". Does that get rid of all problems? When using option (4), please enable the madVR OSD (Ctrl+J) and check whether/when the "composition rate" is shown. Thanks.

Installed 0.36 to test.
Exclusive -> windowed does not seem to suffer from the disappearing menu syndrome (DMS - :)), nor do I remember ever seeing it before 0.37, old frame displayed + black screen for a brief moment.
Windowed -> exclusive - black screen for a moment.

I also tried the "always" option with 0.37.
Still seeing DMS from time to time and an old frame, but overall seems that transitions are a bit better comparing to "full screen" option (though it's probably placebo...).
Aero is disabled as expected (no "composition rate" in OSD).

Anyway, it really isn't a big issue for me and 0.37 bring obvious improvement comparing to 0.36 in the transition speed (though I was perfectly fine with 0.36 too, to be honest), it really is a "nice to have" but not a deal-breaker feature for me.
:thanks:

mark0077

22nd February 2011, 19:40

Hi Madshi, Using option 4 I'm pretty happy. My only issues so far with it are

1) Aero / Desktop composition seems to take about 10 seconds to come back, sometimes upto a minute, sometimes longer..., even with mpc-hc paused (ie 0 cpu usage).... Is madVR immediately reenabling aero or is it itself delaying the reenabling of aero?
2) When I right click on the fullscreen exclusive window, the menu appears first time (excellent improvement for me - cheers), but I often like to right click again to get out of long nested menus from decoders, but right clicking again makes madVR switch back to exclusive, then back to windowed. I'm sure with a tiny tweak here or there you can get rid of this unnecessary switch back to exclusive. You can reproduce by right clicking when in fullscreen, then rightclicking somewhere again.... Not a big deal, just a tweak ;)
3) With desktop composition disabled I get very bad tearing. Is madVR supposed to never allow tearing, even with desktop composition off? I'm using Windows 7 64bit, Nvidia GTX 295. If you need me to run any tests to help get rid of tearing in this config with desktop composition disabled thats no problem..

Thanks for 0.37 :)

dansrfe

22nd February 2011, 19:51

I'm on Windows 7 x64 with an ATi Radeon HD Mobility 4570 1GB RAM and I have the option to disable desktop composition unchecked. Going from windowed to exclusive is the smoothest thing I've ever seen. It's pretty much the same as going from Windowed to Fullscreen windowed. I'm very happy with 0.37 and it shows quite clear improvements from 0.36 in terms of performance and stability. Thanks madshi and keep up the good work! :)

nlnl

22nd February 2011, 22:04

0.37 W7 32 ATI5750 11.2
(2) disable desktop composition during exclusive -> windowed mode switch
(3) disable desktop composition while madVR is in exclusive mode
(4) disable desktop composition while media player is in fullscreen mode
(5) always disable desktop composition

(2) ghost images (menu ?)
(3)-(5) tearing and clicking audio

The best option is old and gold (1) - don't disable desktop composition :)

madshi

22nd February 2011, 22:18

Gotcha, yes, this appears to working correctly. It is working properly in dual-monitor playback as well.
But you said with option (4) there would be a lot of flashing going on. Where does that flashing come from, if Aero stays enabled while you are in fullscreen mode?

Installed 0.36 to test.
Exclusive -> windowed does not seem to suffer from the disappearing menu syndrome (DMS - :)), nor do I remember ever seeing it before 0.37, old frame displayed + black screen for a brief moment.
Windowed -> exclusive - black screen for a moment.
v0.37 should behave the same way as v0.36 if you keep Aero enabled all the time. Can you see a difference in switching speed? Is v0.37 quicker if you keep Aero enabled than v0.36? Thanks.

I also tried the "always" option with 0.37.
Still seeing DMS from time to time and an old frame, but overall seems that transitions are a bit better comparing to "full screen" option (though it's probably placebo...).
Aero is disabled as expected (no "composition rate" in OSD).

Anyway, it really isn't a big issue for me and 0.37 bring obvious improvement comparing to 0.36 in the transition speed (though I was perfectly fine with 0.36 too, to be honest), it really is a "nice to have" but not a deal-breaker feature for me.
:thanks:
So which option do you plan to use?

Hi Madshi, Using option 4 I'm pretty happy. My only issues so far with it are

1) Aero / Desktop composition seems to take about 10 seconds to come back, sometimes upto a minute, sometimes longer..., even with mpc-hc paused (ie 0 cpu usage).... Is madVR immediately reenabling aero or is it itself delaying the reenabling of aero?
madVR tries to enable Aero without delay. The delay comes from Aero itself. But maybe I can speed that up a bit by adding some flushes somewhere. I'll put that on my to do list.

2) When I right click on the fullscreen exclusive window, the menu appears first time (excellent improvement for me - cheers), but I often like to right click again to get out of long nested menus from decoders, but right clicking again makes madVR switch back to exclusive, then back to windowed.
Have you enabled or disabled the 3 second delay?

You can reproduce by right clicking when in fullscreen, then rightclicking somewhere again....
I can't, unfortunately, at least as long as I keep the 3 second delay activated.

3) With desktop composition disabled I get very bad tearing. Is madVR supposed to never allow tearing, even with desktop composition off? I'm using Windows 7 64bit, Nvidia GTX 295.
That tearing should be gone in fullscreen exclusive mode, though, correct? Theoretically madVR is not supposed to tear in windowed mode, either. I don't know why it tears for you. Might be a driver bug.

You could try option (3). That way you should never have tearing. However, that might slow down the exclusive <-> windowed switch a bit.

I'm on Windows 7 x64 with an ATi Radeon HD Mobility 4570 1GB RAM and I have the option to disable desktop composition unchecked. Going from windowed to exclusive is the smoothest thing I've ever seen. It's pretty much the same as going from Windowed to Fullscreen windowed. I'm very happy with 0.37 and it shows quite clear improvements from 0.36 in terms of performance and stability. Thanks madshi and keep up the good work! :)
Happy to hear that it works well for you with Aero turned on all the time. It doesn't with my NVidia 9400 IGP. But I guess it's different for everyone.

0.37 W7 32 ATI5750 11.2
(2) disable desktop composition during exclusive -> windowed mode switch
(3) disable desktop composition while madVR is in exclusive mode
(4) disable desktop composition while media player is in fullscreen mode
(5) always disable desktop composition

(2) ghost images (menu ?)
(3)-(5) tearing and clicking audio

The best option is old and gold (1) - don't disable desktop composition :)
You should get no tearing in exclusive mode, do you?

Is option (1) faster than with older versions for you?

madshi

22nd February 2011, 22:22

madVR v0.38 released

http://madshi.net/madVR.zip

* changed mouse cursor management (for J.River MC)
* added support for NV12 (4:2:0 8bit), P010 (4:2:0 10bit), P016 (4:2:0 16bit)
Of course support for high bitdepth input maintains full bitdepth throughout the whole processing chain.

Does anyone know a decoder or source filter which is able to output P010 or P016?

I'm asking because although I think madVR's P010/P016 support should work just fine, I haven't been able to fully test it, due to missing decoders / source filters which output these format.

nevcairiel

22nd February 2011, 22:34

Works with NV12 just fine, thanks! :)

leeperry

22nd February 2011, 22:46

* added support for NV12 (4:2:0 8bit)
sweet, 2 releases in a day..thanks very mucho! any chance we could input YV12 for 709 and NV12 for 601 or something? a simple option would allow this kind of trick for SD content.

noee

22nd February 2011, 22:46

But you said with option (4) there would be a lot of flashing going on. Where does that flashing come from, if Aero stays enabled while you are in fullscreen mode?

Sorry, I wasn't totally clear. I've been able to watch it closely now that I'm back in dual-mon mode. madVR is disabling Aero as expected, when the player is in fullscreen mode. The flashing, however, is coming from the switch from excl to windowed whenever I run the mouse up and MC's player bar is showed (which is obviously expected). The switch to windowed causes both screens to flash and the video gets a little choppy while the MC player bar is showing. Also, on the screen that is *not* showing the video, I get a screen view from sometime in the recent past (apparently a manifestation of issue (a)). Transition back to exclusive is essentially seamless and Aero stays disabled during this whole process.

madshi

22nd February 2011, 22:52

Works with NV12 just fine, thanks! :)
:)

any chance we could input YV12 for 709 and NV12 for 601 or something? a simple option would allow this kind of trick for SD content.
No, sorry, I don't like implementing weird hacks. There'll be a proper solution soon enough.

Sorry, I wasn't totally clear. I've been able to watch it closely now that I'm back in dual-mon mode. madVR is disabling Aero as expected, when the player is in fullscreen mode. The flashing, however, is coming from the switch from excl to windowed whenever I run the mouse up and MC's player bar is showed (which is obviously expected). The switch to windowed causes both screens to flash and the video gets a little choppy while the MC player bar is showing. Also, on the screen that is *not* showing the video, I get a screen view from sometime in the recent past (apparently a manifestation of issue (a)). Transition back to exclusive is essentially seamless and Aero stays disabled during this whole process.
Ok, thanks for the detailed description. Too bad. Not sure if the fault is with Aero or with the GPU drivers. Maybe both.

madshi

22nd February 2011, 22:53

madVR v0.39 released

http://madshi.net/madVR.zip

* fixed: 16:9 DVDs were shown as 4:3 (introduced in v0.37)

cca

22nd February 2011, 22:56

madshi

22nd February 2011, 23:00

Wanted to try madVR in my PC but exclusive mode was not working not matter what I tried. I did figured it out though...
madVR somehow triggers Crossfire to be used! If enable the options to show the Crossfire logo on the AMD drivers, when madVR goes in exclusive mode the logo suddenly appears. If I disable the Catalyst AI or Crossfire altogether, madVR works fine but obviously messing with options every time I want to watch a video is not user friendly. The only permanent solution I can see for now is to never use exclusive mode.
Hmmmm... That's weird. I'm no expert with Crossfire. Isn't there a way to make it not activate for madVR while still keeping is activated for games? madVR is not doing anything Crossfire specific. I can only guess that ATI is always enabling it if any software goes into exclusive mode. Maybe someone else with Crossfire experience knows a workaround that doesn't require constant option messing?

BTW, what exactly do you mean with "not working"? Did it just not activate? Or did it crash? Or did it result in weird images?

cca

22nd February 2011, 23:05

cyberbeing

22nd February 2011, 23:09

Changing the "Upload Frames in Render Thread" setting while a video is playing crashes madVR 0.39 on WinXP.

madshi

22nd February 2011, 23:17

Unfortunately AMD does not provide an application list to put an exception in. Crossfire tampers with the frame processing, attempting to process one frame in GPU 1 and the next in GPU 2, doing alternate frame processing. There is also another mode, not quite sure how that works. This does not happen with MPC's renderers, even if I use the D3D fullscreen mode.
Ok.

What exactly do you mean with "not working"? Does exclusive mode just not activate? Or does it crash? Or does it result in weird looking video playback?

Changing the "Upload Frames in Render Thread" setting while a video is playing crashes madVR 0.39 on WinXP.
That's already on my to do list, but not high on my priority. For now, after changing "Upload Frames in Render Thread", you should always restart the media player.

noee

22nd February 2011, 23:19

Ok, thanks for the detailed description. Too bad. Not sure if the fault is with Aero or with the GPU drivers. Maybe both.

Ok. Frankly, I don't really care because the only reason I use Aero is because SilverLIght "needs" it to work correctly with NetFlix, so most of the time, this machine will have Aero turned off. In addition, this is a really old video card, and while it works amazingly well in this setup, I will be upgrading soon (6850).

Other than that, a great experience with MC so far, great build.

cca

22nd February 2011, 23:22

Oh, it switches to fullscreen alright, but the video seems to play the frames out of order, producing a weird stuttering effect making it unwatchable. Since you use 3 buffers, imagine that instead of 1 2 3, it plays 2 1 3 or something like that. I can't explain much better.

nlnl

22nd February 2011, 23:31

You should get no tearing in exclusive mode, do you?

Is option (1) faster than with older versions for you?

No tearing in exclusive mode and (1) is a little bit faster (may be autosuggestion :))

adam777

22nd February 2011, 23:46

v0.37 should behave the same way as v0.36 if you keep Aero enabled all the time. Can you see a difference in switching speed? Is v0.37 quicker if you keep Aero enabled than v0.36? Thanks.

0.39 without "disable desktop composition" and 0.36 seems to be about the same - black screen on switch, old frame etc.
0.39 with "disable desktop composition" behaves as explained before - disappearing menu etc.

Anyway, as I've said before, it really isn't a big issue, whatever alternative I choose.
I'll probably try both (0.39 with and without) for a while, and see...
Thanks again.

Thunderbolt8

23rd February 2011, 00:52

dont know much about NV12, does it provide better quality than YV12?

leeperry

23rd February 2011, 02:08

I'm having a problem w/ all builds >0.36 when switching between FSE and FSW(XPSP3/CoreAVC CUDA, both in 1024*768 and 1280*720), basically PotPlayer's transport bar is not allowed to show up properly: http://thumbnails8.imagebam.com/12072/be0032120713303.jpg (http://www.imagebam.com/image/be0032120713303)

and if I hover over the buttons, I get this: http://thumbnails4.imagebam.com/12072/dbc8ae120713306.jpg (http://www.imagebam.com/image/dbc8ae120713306)

There's a zombie frame that refuses to leave and masks PotPlayer's GUI. It's never happened w/ <0.37, and I haven't changed any setting. It's a fairly random problem, but easy to reproduce.

PS: if I move from FSE/W to windowed while being paused, it often does the same w/ the windows tasktray/desktop(I use a dark gray Vista theme). I've double-checked, no such problem w/ 0.36.

Anima123

23rd February 2011, 03:27

madshi,

The seekbar sometimes still won't disappear after switched to exclusive mode. I thought this was an old issue.

Regards,

Toe Tag

23rd February 2011, 05:36

I post this here because where else but madvr do people get 0-255 source. Wanted to confirm, its good to use the full 0-255 levels with MadVR. My issue is that not all display devices, including many projectors, don't have DVI inputs. IIRC every JVC PJ for the last 4 years from the RS1 on up 1. has no DVI input 2. has a setting for enhanced 0-255 input (though where this 0-255 into the PJ HDMI input originates in the first place is unclear to me).

Now, nVidia cards by default put out 0-255 on the DVI port, and will only put 16-235 on the HDMI port. nVidia cards are willing to mix HD audio into the DVI port, but when you use a DVI-HDMI adapter on the DVI port, the nVidia card detects that, and forces you back to 16-235. And, nVidia doesn't offer you software control of this.

ATI (AMD) is similar. It does offer a control panel option to request 0-255 output, but it actually doesn't do anything at all on the HDMI port. Its been this way for a year and they seem uninterested in fixing it.

Many projectors, including my RS2, accept only HDMI, and no DVI. And they let you choose standard 16-235 or enhanced 0-255 input levels. (This implies that HDMI cables are capable of carrying enhanced signal levels). But it seems there is no way to get neither nVidia nor ATI graphics cards to push 0-255 over the HDMI cable.

Well, there may be one way. Use a DVI-HDMI adapter on a 5000-series-and-up ATI card's DVI output, and set the control panel for 0-255. ATI does seem to detect the adapter, but doesn't seem to force you back to 16-235 like nVidia does. Assuming you have set the ATI control panel to 0-255. Having negotiated the above land mines, ATI still does not jam HD audio (TrueHD etc.) into the DVI output, so you set up a second monitor, and run a cable from the HDMI output of your ATI card, into the HDMI input of your AVR. You lose the graphical overlay of the AVR, and the sound may be out of sync, but you can probably correct the sync.

Have I summarized the situation correctly? And, it will never change from the above "state of the art"? For people with e.g. JVC projectors that don't have DVI input, but do want the full 0-255 madVR goodness? Could displayport help?