madVR - high quality video renderer (GPU assisted)

[Razoola]

But are you having glitch-free playback with D3D9? You're using 120Hz, aren't you? Have you found a setting with D3D9 which works glitch-free?

With D3D nine its not totally glitch free, I only get 2 or 3 at the start of playback and thats it, no more incrementing for the rest of the movie. Yes Im at 120hz.

[dansrfe]

With D3D nine its not totally glitch free, I only get 2 or 3 at the start of playback and thats it, no more incrementing for the rest of the movie. Yes Im at 120hz.

Just wondering but I'm curious as to how you can output 120Hz through your graphics card :O I thought the max was 60Hz through HDMI then the TV doubles it or takes the right and left eye frames given in some sort of consecutive order then uses a multiplier to reach the maximum common denominator the TV supports for 3D.

[patul]

With D3D nine its not totally glitch free, I only get 2 or 3 at the start of playback and thats it, no more incrementing for the rest of the movie. Yes Im at 120hz.

Me too, I only get glitches and frame drops at the start of playback, which I can't complaint about though, since I'm running WinXP SP3 with lowly GF 9500 GT in my desktop computer (72Hz), and Vista Home SP2 with ATi Mobility Radeon HD 3400 Series (laptop - 60Hz). Tried most of the settings but I still get frame drops.

[Mr Alpha]

Oh boy, there are more problems with v0.59 than I anticipated! I take it you're using Windows 7? Which GPU? Can I have a log, please? It will probably not help (rarely does for crashes), but having a log is better than having nothing.Well, I have some good news. I don't know if it was the updated version of madVR or the new 11.4 GPU drivers that fixed it, but the DX9 crash is gone.

Here is the log for the DX11 crash. (http://mralpha.s3.amazonaws.com/madVR-log.rar)

[Andy o]

I didn't have a glitch problem before, and now I don't either with D3D11. I imagine there's no other benefit to using it, for people like me? Now that the buttons and window size issues have been addressed, D3D11 seems completely benign for me, neither positive nor negative.

[jmone]

* added option "treat 25p movies as 24p (requires Reclock)"

Out of interest does this also work with the MC16 Video Clock code or could JR make mods to support it?

[Darkimmortal]

Is it normal for madVR to cause the whole system (and itself; ~200ms rendering time spikes) to lag horribly while a video is playing in exclusive mode on an AMD graphics card? I've just got a new 6970 and this seems to be the case regardless of settings, whereas on my old Nvidia GTX 275 everything was perfect.

Neither CPU nor GPU is above 15% usage, so it's pretty weird that even the mouse cursor is lagging, which normally only occurs under extreme load.

Various info if it's relevant (http://imgkk.com/i/814h.png), also the log file (http://shit.weeaboo.com/madVR%20-%20log.zip).

Player is PotPlayer, configured as per this guide (http://imouto.my/configuring-potplayer-for-gpu-accelerated-video-playback-with-dxva-or-cuda-and-also-high-performance-software-decoding/). The same issue occurs in Media Player Classic HomeCinema. I've tried numerous versions of madVR, all of which used to work perfectly.

Is there some Catalyst driver setting I'm missing or something?

[HTPC-User]

Hello!

I would like to thank madshi for his great renderer! The picture quality is great, although I must admit, that for me the most important feature is the build in CMS, which hopefully will be developed further in the near future. Many thanks to yesgrey for his yCMS program (I hope the development isn`t stalled or cancelled!).

After a long time of testing a few madvr versions I recently tried to integrate madvr in a daily used system, without success. Hopefully this great community can give me some tips and clues.

Here we go:
System: Win7 without ServicePack, ATI 5770, Catalyst Version 10.12, MPC-HC 1.5.2.3052, ffdshow tryouts R3562, madvr 0.61, DirectX June2010 (although from 2011 as stated on MS website)

1.) Mouse pointer is visible all the time:
After loading a video, the mouse pointer doesen`t disappear. I have to move the mouse a little bit and after that the pointer disappears after a while. Laying the mouse on the ground to prevent some kind of vibrations doesn`t help either.
Am I the only one with this problem? Madshi, which further information do you need to solve that?

2.) Seekbar without function:
The 0.6x release is the first madvr release which shows me the madvr seekbar. I testet some early 0.2x, some 0.3x and some 0.4x releases, but I never saw the seekbar. If I move the mouse to the bottom of the screen the seekbar is popping up, but it has no function: I can click in the bar, I can try to grab and move the slider (which isn`t possible), the result is always the same: nothing happens. Is that by design or a bug or just me?

3.) qdvd workaround isn`t working properly:
I know that madvr currently has problems with DVD menus. I tried the qdvd workaround (copied the qdvd from XP into system32 and syswow64) with only limited success.
Are there some new tricks or workarounds to get DVD menus working with madvr? Madshi, any plans to solve that problem in the near future?

4.) DVD menu problem with EVR CP in the latest MPC-HC versions:
I know this seems to be a bit off topic, but I hope this could be perhaps a valuable hint for madshi to get DVD menus working: I tested the MPC-HC releases 1.5.2.3052, 1.5.2.3045 and 1.5.2.3016 and all of them show the same problem: If I enable the fullscreen exclusive mode then even with EVR CP the menu doesn`t work. Disable the exclusive mode and everything is fine.
Can anyone confirm this or is my system a little bit special? :)
Perhaps this helps madshi.

5.) forced subtitles:
What are the correct settings to get forced subtitles with madvr working? I tried many things and searched the web, but either the subs are shown all the time and I have to switch them off manually or they are not shown at all and I have to switch them on manually. I would like to see the forced subs automatically. Can anyone please give me a hint where to look or where to search? :confused:

Or are there other working solutions to show forced subs, for example ffdshow?

:thanks:
A HTPC-User

[Razoola]

Just wondering but I'm curious as to how you can output 120Hz through your graphics card :O I thought the max was 60Hz through HDMI then the TV doubles it or takes the right and left eye frames given in some sort of consecutive order then uses a multiplier to reach the maximum common denominator the TV supports for 3D.

This is not through HDMI but dual DVI to a nvidia certified 3D desktop monitor. I use HDMI to a secondary display (plasma) only.

[oddball]

The latest version is causing havok with the latest MPC-HC. I can't get proper windowed mode to work and left clicking the mouse won't pause the video. If I bring up the options windows I cannot select any of the options therein. It was working yesterday too. I am unable to launch MPC-HC on it's onw and access the options on the primary display anymore too. Gonna require a reboot it would seem. I noticed when I tried to go windowed mode yesterday on the first install of .61 with all the new options ticked that going to windowed mode would crash MPC-HC entirely.

[noee]

You might try a complete uninstall/reset settings/install of MPC-HC. I've had to do so recently due to some very strange behavior.

[sneaker_ger]

5.) forced subtitles:
What are the correct settings to get forced subtitles with madvr working? I tried many things and searched the web, but either the subs are shown all the time and I have to switch them off manually or they are not shown at all and I have to switch them on manually. I would like to see the forced subs automatically. Can anyone please give me a hint where to look or where to search? :confused:

This is not a job for the renderer.
LAV Filters (http://forum.doom9.org/showthread.php?p=1496581) have the function you need.

[oddball]

I just did that and it's all fine again.

Now to report problems.

I don't know if this is a problem with MadVR 0.61 or MPC-HC 3052. If I enable "Present several frames in advance" in either DX9 or DX11 mode and switch between fullscreen on my secondary display and windowed on my primary display MPC-HC hangs, glitches (Jerky playback on secondary display), presents a blank screen on the secondary display or after a couple of times will crash MPC-HC entirely. It does not matter if I tick or untick the sub options for this setting or not. If I untick 'Present several frames in advance' it does not crash as often but it still does it on occasion. It also does not give jerky playback or a blank screen when disabled.

Switching between fullscreen and windowed seems to bug out in any mode I have tried to some extent though.

[yesgrey]

Many thanks to yesgrey for his yCMS program (I hope the development isn`t stalled or cancelled!).
It's not. :)

[iSunrise]

Thanks for 0.61 madshi.

I´ve tried the D3D11 path again with your new option "overshoot max frame latency", but the already reported presentation glitches are still there, unfortunately. I can only explain this behaviour (since it seems to work for you) with different OSes and/or different DirectX runtimes on our sides. Windows 7 comes natively with WDDM 1.1 support, while with Vista you have to install the Windows Platform Update to even get full D3D11 support, which I don´t have installed. So I guess that this path should work just fine without glitches, if I would just install this Platform Update or Windows 7.

I will stay with the D3D9 path for now, as I only have to set the "use a seperate device for presentation" option. No additional tinkering at all needed. Since the D3D9 path is working flawlessly and is doing everything without any drops, delayed frames or glitches I will now use that for my everyday video playback. I have watched a lot of TV series episodes and movies (23.976fps or 24fps) or videos (25fps, 29.970fps, 30fps, 50fps, 60fps) lately without any hiccups, whatsoever. The old rendering path also works flawlessly, but the D3D9 path does faster windowed-fullscreen switching.

As far as I´m concerned, you can now add some new (requested) features for your renderer. I wouldn´t invest too much time into that D3D11 path, yet, but that´s up to you, of course.

[oddball]

Madshi I have a question. Do you have 2 displays or do you do all your dev work on a single display? Just curious because nearly ALL of the problems I have with your renderer are to do with 2 displays. I noticed that if I open the MPC-HC window on my secondary display in windowed mode. Play. Then go fullscreen and back I do not get nearly as many crashes or hangs as if when I have it windowed on one display and going fullscreen on the other. If you do not have a secondary display perhaps we can open up a donation Paypal account for you to get one as it would save a lot of headaches.

Ignore this if you do have a secondary display.

EDIT: Yup! That's why. I can confirm that if I enable present frames option with all sub options and in DX11 mode with windowed mode open on secondary display and switch between fullscreen and back to windowed it does not hang, black screen or crash. It's only if I start out with it windowed on one display and go fullscreen on the other.

[peckec]

madVR is really amazing, madshi, thank you very much!

0.61 and D3D11 works great on my HD5450 (CCC 11.3, Win7), no glitches or dropped frames.
8 presentation frames and all flushes disabled. Only those multiple display switches are a little bit annoying when switching to FSE.

But i don't understand why D3D11 path depends on so called "glitch hack"? I can't enable it without enabling "use a separate device for presentation" first.
On my machine 0.58 worked great without glitch hack as well.

It seems to me that madVR works much better with ATI/AMD cards.

[oddball]

I found another problem with DX11 mode. If I start a video fullscreen on the secondary display and then click on another open application window (Firefox for instance) on the primary display the video stops and I cannot start it again. All I can do is exit the player. If I do the same with DX9 it continues playing and lets me pause etc no problemo.

BTW 0 glitches or delayed frames with 16 preframes and no flushes. I get occasional dropped frames when the CPU is under load.

OK one more annoyance to add to the list. Whenever I have the mouse cursor below a certain point on the primary display the seekbar appears on the secondary display. THAT I could live without happening. It should only popup when I hover the mouse on the bottom half of the secondary display.

Madshi I hope you are not getting tired of my pointing all these problems out?

EDIT: I just reinstalled .58 and it does not have the same issue with the seekbar appearing when the mouse is below a certain point on the primary display. It only appears if I hover the mouse cursor on the secondary display. Bug in 0.61? I've disabled the seekbar for now.

EDIT2: Chapter skipping causes dropped frames. Not sure if that's normal as it's not something I use that often.

[madshi]

Does glitche expected to happen when I suscessively seek? I actually just relized that glitches you're talking about is not something you see in the picture but stat, so my "problem free" report before was wrong. I rarely have gliches when switch to and from FSE, and occationally have gliches when suscessively seek (MPC-HC jump forward multiple times). DX9/11 doesn't seem much different. Still trying to test more.
During start of playback and during seeking frame drops, delayed frames and glitches are acceptable to a certain degree. I will probably even stop counting them in a future version, in the first maybe 3 seconds after playback start and after every seek.

I was always wondering what really a glitch is, if presenting device runs 23.976 fps, video as well and the playback is not smooth. Then it should be either dropped or delayed frame, shouldn't it? Or is it a skipped frame?
A glitch is a delayed frame which happened due to a problem in either Direct3D itself or in the GPU drivers.

I reinstalled W7 x64 recently and I was never able to configure smooth playback since then (I used to be running MPC-HC (older) + EVR + Reclock, projector @ 24Hz to get smooth playback those days). With new setup (the same computer, Thinkpad W510, i7, nVidia Quadro FX 880M) only VLC was smooth (unless it encountered complex scene hehe), in MPC-HC I was having a lot jerkiness. Only recently discovered new madVR saved the day (I tried madVR at early stages and it was not working for me at that time) - ctrl+J showed me that I'm having glitches and with some tweaking (D3D9 exclusive for 23.976 content and D3D9 windowed for 25fps content) everything is perfect (or almost perfect) :-) Thanks, Madshi.
:)

Just a small question - I was using shaders, Complex Sharpen 2 from MPC-HC, it was really bringing picture to another level. Do you plan (at any future) to support custom shaders?
At some time in the future, maybe yes.

This is slightly OT, but does anyone know how to force my HD5450 to get into 1080p29?
PowerStrip can probably do that, but I doubt that your display will support it.

madVR versions 0.59+ are causing increased present times in fullscreen windowed mode, which appears to be resulting in 1-4 delayed frames per minute.

0.58 | Avg Present 6.57ms | Max Present (5s) 7.07ms | Delayed frames are rare
0.59+| Avg Present 6.97ms | Max Present (5s) 23.05ms | Frequent delayed frames
Is this 100% reproducible? The only thing that I've changed is that I'm polling the VSync scanline information from a separate Direct3D device instead of using the same device for everything. That has improved the situation for most people. It has also fixed the non-working refresh rate indicator you reported earlier. Other than that I haven't changed anything for windowed mode. So if there's any change at all, it should be for the better.

There may also be an issue on my ATI/Win7 box in fullscreen exclusive mode. With versions after 0.58, present queues are no longer stable at 14/14, and there may be a decrease in fullscreen exclusive smoothness on that ATI box as well (I deleted a previous post about this because I was unable easily re-confirm my observation), but I wonder if the above is related. I've yet to test 0.61 on the Win7 box.
No explanation for that, unless you use different tweak options when testing v0.58 vs. v0.60. All I've changed are adding some tweak options.

p.s. Still getting random madVR hangs on close, but I keep forgetting to use the debug build (I'm hesitant to leave it running and produce multi-GB logs...). Do you remember which revision I first reported this with? It was before you started calculating refresh rate in a new D3D instance, correct?
I don't remember the specific version, but a log from any (newer) version would be ok.

My earlier report about the DX11 presenter preventing dropped frames with my 720p60 file in FSE mode were actually not correct. Something else that changed between 58 and 60 fixed this and in 61 I am getting no dropped frames or glitches in FSE without use a seperate device for presentation checked, present several frames in advance checked, 4 frames, and all flush options set to don't flush. All this with a fullscreen resolution change to boot. From a performance standpoint, at least for me, you have absolutely nailed it. Fantastic work!
:)

Yep, that does fix the issue. Not sure why the two are overlapping like this, but at least it's a workaround. :thanks:
Don't know why they're overlapping, either! Really weird...

With D3D nine its not totally glitch free, I only get 2 or 3 at the start of playback and thats it, no more incrementing for the rest of the movie. Yes Im at 120hz.
Glitches at the start of playback do not really count for me.

I didn't have a glitch problem before, and now I don't either with D3D11. I imagine there's no other benefit to using it, for people like me? Now that the buttons and window size issues have been addressed, D3D11 seems completely benign for me, neither positive nor negative.
There's no need/reason to use D3D11 if you're happy with D3D9.

Out of interest does this also work with the MC16 Video Clock code or could JR make mods to support it?
All this option does is to make madVR switch to 24Hz/23.976Hz instead of 25/50Hz for PAL movies. What MC16 VideoClock makes out of this I don't know. Just give it a try. Maybe it already works.

Is it normal for madVR to cause the whole system (and itself; ~200ms rendering time spikes) to lag horribly while a video is playing in exclusive mode on an AMD graphics card?
No, it's not normal at all. I don't think anybody else has this problem. At least it wasn't reported yet, IIRC.

1.) Mouse pointer is visible all the time:
After loading a video, the mouse pointer doesen`t disappear. I have to move the mouse a little bit and after that the pointer disappears after a while. Laying the mouse on the ground to prevent some kind of vibrations doesn`t help either.
Am I the only one with this problem? Madshi, which further information do you need to solve that?
Can anybody else reproduce this? Might be a media player bug, not sure. I don't think madVR actually does anything to the mouse, I think the media player does that, but I'm not 100% sure right now.

2.) Seekbar without function:
The 0.6x release is the first madvr release which shows me the madvr seekbar. I testet some early 0.2x, some 0.3x and some 0.4x releases, but I never saw the seekbar. If I move the mouse to the bottom of the screen the seekbar is popping up, but it has no function: I can click in the bar, I can try to grab and move the slider (which isn`t possible), the result is always the same: nothing happens. Is that by design or a bug or just me?
You can't *drag* the seekbar, it's not designed for that. But you should be able to click on it to seek. Works for me, at least.

3.) qdvd workaround isn`t working properly:
I know that madvr currently has problems with DVD menus. I tried the qdvd workaround (copied the qdvd from XP into system32 and syswow64) with only limited success.
Are there some new tricks or workarounds to get DVD menus working with madvr? Madshi, any plans to solve that problem in the near future?
I have not investigated this any further recently and it's not on my near time schedule. Some time later maybe.

I´ve tried the D3D11 path again with your new option "overshoot max frame latency", but the already reported presentation glitches are still there, unfortunately. I can only explain this behaviour (since it seems to work for you) with different OSes and/or different DirectX runtimes on our sides. Windows 7 comes natively with WDDM 1.1 support, while with Vista you have to install the Windows Platform Update to even get full D3D11 support, which I don´t have installed. So I guess that this path should work just fine without glitches, if I would just install this Platform Update or Windows 7.

I will stay with the D3D9 path for now, as I only have to set the "use a seperate device for presentation" option. No additional tinkering at all needed. Since the D3D9 path is working flawlessly and is doing everything without any drops, delayed frames or glitches I will now use that for my everyday video playback.
As I've said times before, if you're happy with D3D9 there's no need to bother with D3D11.

Madshi I have a question. Do you have 2 displays or do you do all your dev work on a single display? Just curious because nearly ALL of the problems I have with your renderer are to do with 2 displays. I noticed that if I open the MPC-HC window on my secondary display in windowed mode. Play. Then go fullscreen and back I do not get nearly as many crashes or hangs as if when I have it windowed on one display and going fullscreen on the other.
I develop on a single display, but whenever I need to I can easily setup a dual monitor setup. I don't think others have as many crash problems as you have. Personally, switching monitors like you describe seems to work just fine for me. For now crashes are hard to debug for me, if they don't occur on my PC. That might change in some future version. But for now, there's not much I can do.

But i don't understand why D3D11 path depends on so called "glitch hack"? I can't enable it without enabling "use a separate device for presentation" first.
The glitch hack internally always was "use a separate device for presentation", I just didn't name it that way until recently. The only way for me to use D3D11 is to use a D3D11 device for presentation and a D3D9 device for rendering. So the D3D11 path does require me to "use a separate device for presentation". That's just the way it is.

On my machine 0.58 worked great without glitch hack as well.
Well, then don't bother with the new tweak options. If you don't have to use them, then don't. They're just hacks to workaround problems in Direct3D and/or the GPU drivers.

It seems to me that madVR works much better with ATI/AMD cards.
FWIW, I have the same impression, but it seems to apply mainly to situations where the display refresh rate is higher than the movie framerate. If you have a 1:1 match things seem to work just fine for NVidia users, as well.

I found another problem with DX11 mode. If I start a video fullscreen on the secondary display and then click on another open application window (Firefox for instance) on the primary display the video stops and I cannot start it again. All I can do is exit the player. If I do the same with DX9 it continues playing and lets me pause etc no problemo.
Can anybody else reproduce this?

Madshi I hope you are not getting tired of my pointing all these problems out?
Yes, it is quite tiring to get so many bug reports. I do want to get bug reports. But I still don't like them. It is especially frustrating if I can't reproduce a problem.

[madshi]

I plan to always "use a separate device for presentation" in the next build, and to remove the option from the settings dialog.

Any objections?

Speak up now or it will be too late soon.

[Hprd]

Well on v59b+ it doesn't make any difference to me, lol. And on v59 (which i'm using because it gives me perfect playback, despite the few issues that've been fixed since), it's the only way I can get FSE to work properly at all (with dx11 mode), lol. Maybe somehing you change will make it like v59 again then, so go ahead ;)

(this post is somewhat directed at ppl who get glitches on v60+, but never tried v59 ;))

[fastplayer]

I don't care because I don't notice a difference either way.
Choose whatever makes technically more "sense"! :)

[SamuriHL]

I'm good with that, madshi. I have all the glitch hacks turned off and am using the d3d9 path on all my machines. I'm not having any problems.

[sneaker_ger]

I'm good with that, madshi. I have all the glitch hacks turned off and am using the d3d9 path on all my machines. I'm not having any problems.

Madshi's plan is to force the glitch hack on, it's now called "use a separate device for presentation".

[SamuriHL]

It made no difference at all in my configuration. So, as I said, I'm good with that.

[Plutotype]

"use a separate device for presentation" - Im ok to have it ON too.

[cyberbeing]

Is this 100% reproducible? The only thing that I've changed is that I'm polling the VSync scanline information from a separate Direct3D device instead of using the same device for everything. That has improved the situation for most people. It has also fixed the non-working refresh rate indicator you reported earlier. Other than that I haven't changed anything for windowed mode. So if there's any change at all, it should be for the better.
It is 100% reproducible.
0.27-0.58 = Good
0.59-0.61 = Worse

I thought that new D3D instance for VSync change was for Fullscreen Exclusive mode only? You added it to Windowed mode in 0.59? Reverting that may be a good start, since I never had any issues with Windowed mode VSync prior to 0.59. Whatever you did from 0.58->0.59 is causing high present time spikes within every 5s polling period using Windowed mode. Those constant spikes are causing an increase in delayed frames.

No explanation for that, unless you use different tweak options when testing v0.58 vs. v0.60. All I've changed are adding some tweak options.

The unstable present queue is 100% reproducible in 0.59-0.60, even with the same tweak options (none) as 0.58.

0.58 = stable 14/14 present queue, 24p on 60Hz somewhat smooth
0.59-0.60 = fluctuating 13/14 present queue, 24p on 60Hz not very smooth

If it's noteworthy, for some reason the drop/repeat every dialog is appearing for non-mulitple refresh rates now. When did that change? Previously it only appeared for refresh rates which are a multiple of the video frame rate.

I plan to always "use a separate device for presentation" in the next build, and to remove the option from the settings dialog.
Enabling the glitch hack (use a separate device for presentation) and all other tweaks just make things worse on the ATI/Win7 box, but I'll defer to the majority. In a few days I'm going out of town (without guaranteed internet access :mad:) and won't be back until 3rd week of May. Obsessing over 24p on 60Hz playback on a secondary computer I rarely use for playback, is not really something I care to do anyway.

[oddball]

Thanks Madshi. To be honest my setup is a little esoteric in that I use CoreAVC as the decoder and ffdshow for video filters and audio decoding. It's possible that my issues are caused by the the path before it hits MadVR but I will need to test. I am also using the latest 270.61 Nvidia drivers if that's of any help. I've disabled the D3D hack since you decided it was not a very good thing to do and just set MPC-HC to change refresh rates according to the video being played in fullscreen mode. The seekbar issue surely can't be just me? I'd be glad to hear if others have the same issue (mouse cursor on one display causes seekbar to appear on the other playing the video). I can workaround the crashing by not having windowed mode start on the primary display.

[mr.duck]

Has everyone here run dxwebsetup.exe recently? Could that effect people's problems with glitches?

[yesgrey]

I plan to always "use a separate device for presentation" in the next build, and to remove the option from the settings dialog.

Any objections?
Yes. The same device always work the best for me.

With the same device I only get a presentation glitch at the beginning and then it's rock solid till the end, without any more glitches. It gives me always the most stable and smooth image.

Using a separate D3D9 device is the worse for me, lots of presentation glitches.

When using a separate D3D11 device is not that bad. For example, it starts without any glitches, but then, near the end, one glitch happens. I guess it might be related to presentation back buffers. With D3D9 device it always fills up the total number of back buffers, but with D3D11 it never fills up more than 6-7.

[madshi]

It is 100% reproducible.
0.27-0.58 = Good
0.59-0.61 = Worse

I thought that new D3D instance for VSync change was for Fullscreen Exclusive mode only? You added it to Windowed mode in 0.59? Reverting that may be a good start, since I never had any issues with Windowed mode VSync prior to 0.59.
The separate D3D instance for VSync was already in 0.57, so that can't be the cause of the problem. Actually I'm pretty sure that having a separate instance for VSync is all good with no possible negative side effects.

I've just rechecked the changes from 0.58 to 0.59 and there's one thing that caught my eye: In 0.58 I'm running the VSync thread with the highest possible priority. This priority is "out of this world" and should normally not be used. The value is 15, IIRC. The recommend use for high priority is 2. So in 0.59 I've lowered priority to 13. Thought that shouldn't change anything. But maybe that's what caused the difference for you, not sure. You can try this v0.61 special build where I've restore the original highest possible value of 15:

http://madshi.net/madVRvsync.zip

If it's noteworthy, for some reason the drop/repeat every dialog is appearing for non-mulitple refresh rates now. When did that change? Previously it only appeared for refresh rates which are a multiple of the video frame rate.
Nothing changed in that regard. 2.5x is a supported factor, has always been, IIRC.

just set MPC-HC to change refresh rates according to the video being played in fullscreen mode.
Maybe that's the cause of the instability you're experiencing? Try using madVR's internal refresh rate changer instead. Maybe that reduces the number of crashes?

Yes. The same device always work the best for me.

With the same device I only get a presentation glitch at the beginning and then it's rock solid till the end, without any more glitches. It gives me always the most stable and smooth image.

Using a separate D3D9 device is the worse for me, lots of presentation glitches.

When using a separate D3D11 device is not that bad. For example, it starts without any glitches, but then, near the end, one glitch happens. I guess it might be related to presentation back buffers. With D3D9 device it always fills up the total number of back buffers, but with D3D11 it never fills up more than 6-7.
Too bad... Ok, so I guess I'll have to keep the option for now...

[mrcorbo]

Since we seem to be back to reporting bugs...

I have observed the following behavior when there is an external resolution change:

When the DX11 presenter is activated FSE mode will not activate. Also, fullscreen windowed mode drops frames like crazy as does regular windowed mode.

With DX9 presenter, FSE activates and works great (as I reported earlier), but changing back to windowed mode, even by just using the right-click menu in MPC-HC, causes massive frame drops. Returning to FSE mode results in perfect playback again.

[pirlouy]

I jsut quote this post, for those who missed it. :D

@Guys, thanks for all the bug reports. I'll look into them sooner or later.

But what I'm *really* interested in is whether you still have glitches or not. The purpose of the D3D11 option and all the other new tweak options is to get rid of glitches. So instead of being flooded with millions of bug reports it would be nice to get some (more) feedback on whether glitches are still there or not.

Or let me say it in different words: In the next 3 days, nobody is allowed to post a bug report :devil:, unless you've first thouroughly tested for glitches and reported here whether or not you still have glitches and which options work best for you in terms of glitches - while ignoring bugs during the glitch testing.

[cyberbeing]

I've just rechecked the changes from 0.58 to 0.59 and there's one thing that caught my eye: In 0.58 I'm running the VSync thread with the highest possible priority. This priority is "out of this world" and should normally not be used. The value is 15, IIRC. The recommend use for high priority is 2. So in 0.59 I've lowered priority to 13. Thought that shouldn't change anything. But maybe that's what caused the difference for you, not sure. You can try this v0.61 special build where I've restore the original highest possible value of 15:

http://madshi.net/madVRvsync.zip


Problem solved in that build :)

It's now comparable to 0.58 and prior builds. I still get occasion max present (5s) spikes and delayed frames in those builds as well, but that's 'normal' and usually doesn't cause significant problems. The exception is the new Fullscreen Exclusive mode, where those present time spikes (above my refresh interval) are what make things unstable for me.

This makes me wonder if setting the VSync thread priority even higher, into the 'Real time' realm of 16-31 would solve all my delayed frame woes. Thread priority 24+ can cause system problems, but 16-21 may be safe enough, 22-23 who knows.

Can you make me please make me a couple of additional 'VSync > ALL' experimental builds? One with VSync priority set to 16 and another set to 21? Those two builds should be enough to test if higher priority is helpful or not. Even if you don't feel comfortable keeping it that way, I'd still like to test it. If it actually works, 16 (which would place it directly above all 'normal' threads at the lowest possible 'real time' level) would probably be enough.

[fps]

Tried various flush settings with the DX11 mode and some combinations didn't cause any glitches :).

settings: 4 pre rendered frames, DX11 mode, none of the 3 tweak options enabled, v0.61, Win7/32, Nvidia Geforce Go 7400, 1280x760@60Hz

-) don't flush/don't flush/don't flush/don't flush: 0 drops / 12 glitches after 8min, all queues full

-) flush/flush/flush/flush: 0 drops / 31 glitches after 2 hours, present times go up from about 0.2ms to 20ms, queues not full

-) don't flush/don't flush/flush & wait (sleep)/don't flush: 0 drops / 0 glitches, queues not full

-) don't flush/don't flush/flush/don't flush: 0 drops / 0 glitches, queues not full

On other thing, after the clock deviation is pretty much not changing any more, i get the following stats.
http://img31.imageshack.us/img31/3452/unbenanntoz.th.jpg (http://img31.imageshack.us/i/unbenanntoz.jpg/)
According to them I should get 1 drop every second, but I don't get any drops even after some minutes, how is this possible?

[madshi]

Problem solved in that build :)
@Hprd, maybe you want to try the 0.61 special build, too?

This makes me wonder if setting the VSync thread priority even higher, into the 'Real time' realm of 16-31 would solve all my delayed frame woes. Thread priority 24+ can cause system problems, but 16-21 may be safe enough, 22-23 who knows.
AFAIK 15 is the max I can set. The thread priorities are not absolute values but "difference" values to the process priority. If you want max priority you can try setting the media player process to "realtime" with the task manager. Doing so should make the VSync thread have priority 31.

On other thing, after the clock deviation is pretty much not changing any more, i get the following stats.

According to them I should get 1 drop every second, but I don't get any drops even after some minutes, how is this possible?
I would guess that the clock deviation measurement went wrong for some unknown reason.

[Andy o]

Could it be connected to the option for treating 25p as 24p and/or not using ReClock (or not using it properly)?

I've meaning to ask you about this madshi, but was waiting for this glitch testing to calm down like you asked. When not using ReClock, my stats look like this for a 23.976 movie (let the whole 1hr40min movie run):

23.97583Hz
-0.00103% deviation
one frame rept every 7.54 hr

The frame rept counter settles at that. Of course I can't verify it's right, but I assume it is.

When using ReClock with bitstreaming enabled (and media adaptation disabled), my clock deviation so close to zero as to be negligible (-0.00001), and I get 1 frame rept every 1.43hr or so, about what one should expect when one calculates the difference between the movie rate (24/1.001) and the measured refresh rate, without counting the system clock (which explains why it's close to zero). Does madVR adjust the system clock, and ReClock somehow is disabling this behavior (which would make this 1.43hr measurement accurate), or when ReClock is in the chain madVR can't measure reliably cause it doesn't know what ReClock is doing (supposedly nothing when bitstreaming), and the 7.54hr number still applies?

[jmone]

All this option does is to make madVR switch to 24Hz/23.976Hz instead of 25/50Hz for PAL movies. What MC16 VideoClock makes out of this I don't know. Just give it a try. Maybe it already works.

Nope - madVR seems to report more droped frames per second than exist! Does not look to good either!

[Andy o]

madshi, to add something I just realized to my post above: even taking ReClock completely off the chain, I get different results when I decode the audio (DTS core from a DTS-HD MA stream, using ffdshow) than when I bitstream. When I bitstream is as above when not using ReClock, while when I decode, I get similar timings than I did when I used ReClock+bitstreaming. I get around 1 frame repeat every 1.45hrs. The difference now is that the clock deviation settles at about -0.00171% when decoding.

[suanm]

After replacing madvr 0.61 version with madVRvsync.zip(2011.04.30 version),madvr workes so great,dropping frames dassent seem obvious,playback is smoother than 0.49,0.56 and 0.58 version. thank you,madshi

[Hprd]

Hmm well the special version didn't do anything, but i found something else out... I guess b4 i didn't test it quite enough on dx9 mode :o, as i have found out that using the "overshoot max frame latency" on dx9, dx11 still getting glitches no matter what fyi, (and nothing else more is needed, if i put the limit option on, the que goes down, so maybe i'd be more resiliant that way, if i had the thread option on as well, it would get some glitches once and a while, i'll need to test with that 13 minute clip again i think to find some conclusive evidence), i get playback that appears to be the same as dx11 mode on v59 with both options checked ;). This was with v61, or the special build (so i just put the regular v61 back in).

[Hprd]

Ok, after some testing, i can conclude that playback with the glitch hack on (or well what used to be call that), but dx11 off, plus ONLY overshoot max frame latency, i get the same results as v59, zero drops/delays or glitches on my 13 minute test video. Putting limit rendering times on actually i saw could lead to glitches. As on one video the present que got down to 0-4/10, then had a glitch b4 filling up a bit (back up to 2-4/10, or something like that), so this, plus run in a seperate thread are quite useless now (as having all three i noticed more glitches than with just the other two, and i'm certain it wasn't a coincidence. Plus having both the limit option and this on, but the other off, is no different as having all 3 off.)

Also, another idea for a future feature perhaps... Have a check to turn resolution switching completly on or off (for the most part i wouldn't use it, as on testing is probably when it'd be most effective, still there are some situations when at least i would benifit from it otherwise, not sure about other ppl though), as sometimes if i want to play back a video at 60hz for example, the best way to do it currently is to delete all the resolutions (as putting 60p in the filename just makes madvr have to due more work, and thus creating more glitches, maybe there could be a tweak with this instead though?), then retype them back in when i want the auto switching back (which isn't THAT big of a deal though, just sort of tedious, lol).

EDIT: Actually, due to something, i just found that if i put 60p in the filename, i don't get glitches anymore like on v59, so tweaking that option doesn't seem useful anymore, at least for me.

[Superb]

Bug (?):
using the d3d11 path, when double clicking out of an exclusive fullscreen mode to windowed mode, the mpc-hc window covers the entire screen (instead of getting back to the original size).
Happens on my secondary tv-monitor.
Doesn't happen on my primary monitor or when using d3d9 path.

--------------------------

I just upgraded from madVR 0.58 to 0.61.
Now, I have a TV running at 23.976Hz as a secondary monitor.
The only option I changed was setting flush & wait (sleep) on the third box (for exclusive mode), as it helped me avoid glitches in 0.58.
I can't stop frame drops every 3 minutes (as madVR reports). Why? The clocks are nearly the same (23.976 video).

[Superb]

Just happened when double clicking to go full screen:
http://k.min.us/jlkZI4.png

EDIT: Actually, I'm not sure if it was double clicking, pressing CTRL+E or pressing CTRL+R which caused it. :/

[oddball]

With experimental vsync build. Used my 60Hz monitor and disabled ReClock.

DX11 I seem to get glitches no matter what settings I use.

DX9 mode, 4 preframes, 3 options unticked. all options set to don't flush = glitches.
DX9 mode, 4 preframes, all options ticked. All set to don't flush. Very few dropped frames at start. 0 dropped frames. 0 delayed frames. Deviation hovers around 0.00100 most of the time.
DX9 mode, 4 preframes, all options ticked. Flush, Flush wait and sleep, don't flush, don't flush. Very few dropped frames at start (maybe 6-9) 0 dropped after. Deviation is lower at around 0.00050
DX9 mode, 4 preframes, only last sub option ticked, flush, flush wait and sleep, don't flush, don't flush. Dropped frames on occasion. I get better results if I tick at least the bottom two sub options.

ReClock gives slightly more initial dropped frames and 1 or 2 delayed frames at startup.

There appear to be no gains from increasing the number of preframes from default.

[Superb]

Bug: sometimes when right clicking the tray icon, another tray icon menu opens up (the clock's right click menu). I remember eMule had the same issue and one of the developers found the cause. Maybe google will point in the right direction.

EDIT:
after some searching in Google... maybe you catch WM_RBUTTONDOWN instead of the proper (http://www.codeguru.com/forum/showthread.php?t=359190) WM_RBUTTONUP? :)

[Superb]

DX9 mode, 4 preframes, all options ticked. All set to don't flush. Very few dropped frames at start. 0 dropped frames. 0 delayed frames. Deviation hovers around 0.00100 most of the time.
DX9 mode, 4 preframes, all options ticked. Flush, Flush wait and sleep, don't flush, don't flush. Very few dropped frames at start (maybe 6-9) 0 dropped after. Deviation is lower at around 0.00050Will try these two here and report back.

EDIT: Tried the first one. Playback went crazy after 8 minutes.
Frames jumping back and fourth. As if the frames were showing in the wrong order.
Maybe a thread synchronizing issue? Pausing and resuming fixed playback.
Pressing CTRL+J revealed no glitches at all, and no delayed frames. Dropped frames were crazy (>400), which is consistent w/ what i saw.
I'll try the exact same setup w/ the first option unticked ("run presentation in a separate thread") and report back.

EDIT2: after 35 mins of a smooth playback, I think that "run presentation in a separate thread" was probably the cause for the "jumpy" playback I experienced in the first EDIT. maybe it has a bug?
I had 12 dropped frames in 35 minutes (i pressed CTRL+R after seeking to 8:xx) . 0 delayed. 0 glitches.

[cyberbeing]

AFAIK 15 is the max I can set. The thread priorities are not absolute values but "difference" values to the process priority. If you want max priority you can try setting the media player process to "realtime" with the task manager. Doing so should make the VSync thread have priority 31.

I see what (http://msdn.microsoft.com/en-us/library/ms685100%28v=vs.85%29.aspx) you mean. (http://technet.microsoft.com/en-us/library/cc940378.aspx) The calling process would need to be in the 'Real Time' class, and force all threads into the 16-31 range (which is bad). The Idle, Below Normal, Normal, Above Normal, and High process priority classes do seem to max out at 15 for thread priority. Thread priority is the important value, while process priority is just for boosting or lowering the base thread priority (minimum 1 to max 15). Time-critical appears unique, that it never changes from 15, and has the same effect even when the process priority is set to Idle. In that case, scratch that idea, and just keep it at 15, at least it's better than 10 (there is no 13) which 0.59-0.61 was set to. By the looks of it, Reclock also uses a value of 15 for one of its active non-resampler threads.

[nevcairiel]

I finally got around to testing 0.61 a bit, and i found one case that bugged out on me.

With "use separate device for presentation" ON, and "run presentation in a separate thread" also ON, i get massive glitching (visible, too).

If i turn either one off, its perfect again. DX11 didn't seem to make a difference.

This was on a NVIDIA GTS450.

PS:
Both off is perfect, too.

[Superb]

I finally got around to testing 0.61 a bit, and i found one case that bugged out on me.

With "use separate device for presentation" ON, and "run presentation in a separate thread" also ON, i get massive glitching (visible, too).

If i turn either one off, its perfect again. DX11 didn't seem to make a difference.

This was on a NVIDIA GTS450.

PS:
Both off is perfect, too.Consistent w/ what I saw (see EDIT (http://forum.doom9.org/showthread.php?p=1497096#post1497096) here), only that it manifested as jumpy playback on my machine.