madVR - high quality video renderer (GPU assisted)

[namaiki]

You can't use ffdshow to view the subtitles. SSA and other formats work with ffdshow/MadVR. If you use EVR/VMR you also don't need DirectVobSub. But for VobSubs on MadVR you need DirectVobSub.

Madshi, is that anything that can be easily fixed?

Here is a screenshot with madVR displaying subtitles rendered with FFDShow Video decoder (even though they look worse than the subtitles that VSFilter (http://img251.imageshack.us/img251/6136/vsfilter.png) displays, you can still see something): http://img801.imageshack.us/img801/50/ffdshow.png

They work fine. I don't know what you are doing wrong...

[nevcairiel]

Please don't tell me I can't use WMP/WMC with ffdshow+madVR ?

WMP/WMC can only use their own built-in renderer, which is EVR.
Oh, and additionally, madVR does not work with DXVA

[SundaY82]

Oh, and additionally, madVR does not work with DXVA
It does if you use nvidia gfx and CoreAVC, ie CUDA ;)

[namaiki]

It does if you use nvidia gfx and CoreAVC, ie CUDA ;)

CoreAVC's CUDA thing uses the GPU to decode video, but it's not DXVA.

[SundaY82]

CoreAVC's CUDA thing uses the GPU to decode video, but it's not DXVA.
Ok thats true. Same Same but different ;)

[bur]

Here is a screenshot with madVR displaying subtitles rendered with FFDShow Video decoderAnd those subtitles are in VobSub format? If so, that's strange. ASS, SRT, TXT, all work with ffdshow/MadVR, just VobSub doesn't for me. I recreated the path in GraphEdit and MadVR only shows the VobSubs if DirectVobSub is involved. Maybe it's special circumstances on my system that causes this.

[namaiki]

And those subtitles are in VobSub format? If so, that's strange. ASS, SRT, TXT, all work with ffdshow/MadVR, just VobSub doesn't for me. I recreated the path in GraphEdit and MadVR only shows the VobSubs if DirectVobSub is involved. Maybe it's special circumstances on my system that causes this.

The track is:

Codec : VobSub
Codec/Info : The same subtitle format used on DVDs
Title : Vobsub full subtitles
Language : English


:/ Well MediaInfo doesn't give out more information, but it is indeed from DVDs.

Using VSFilter/DirectVobSub, at the very worst, you can rest assured that you are seeing better looking* VobSubs.

[bur]

Using VSFilter/DirectVobSub, at the very worst, you can rest assured that you are seeing better looking* VobSubs.True. Thanks for your help.

[SantinoSan]

I am using BD rips and the native PGS subs, ffdshow does not look very good, but I'm not sweating it as madVR is slated to render subs eventually. I'm just happy I am able to display the subs now.

[pie1394]

CoreAVC's CUDA thing uses the GPU to decode video, but it's not DXVA.

CUDA still relies on VP2 (or VP3/VP4 ...) to perform MPEG-2 / VC-1 / H.264 decoding. It will not be efficient for SP to perform bitstream decoding. Only the BSP (Bit-stream Processor inside VPx) can handle it at very high bit rate per second.

The output image format of VPx engine is NV12. The CUDA API has an example to perform NV12 --> RGBA32 conversion before it is composited with any other OSD image(s) / subtitle(s).

[szabi]

I tried this madvr but i got an annoying problem.

http://www.abload.de/img/madvr3dlut_ffdshowffmedy5r.png

Watch the middle-left part of it.

I am using mpc-hc v1.4.2499 with the following settings:
madvr v0.31 with yCMSv1.8 (3dlut activated, all other settings default) + ffdshow_rev3507_20100707_clsid_icl10 (ffmeg-mt is used)

This problem is also existed if i change the ffdshow setting to libavcodec.
Even more ffdshow changed to mpc-hc(ffmeg) or divxh264 decoder this problem still appear.

The problem is gone when renderer changed to haali or evr or wmr9.

Do you have any idea why i have this?

bye
szabi

[nevcairiel]

Did you try turning off the 3dlut?

[szabi]

You are right, the problem is gone when i switch off 3dlut.
Anyway the picture should have been better with 3dlut. Am I right?

[yesgrey]

You are right, the problem is gone when i switch off 3dlut.
Can you upload a sample where the problem happens and your yCMS input file so I can take a look?

[szabi]

Yes, this is the video (http://www.2shared.com/file/pBahLvsz/Bleach_-_0011.html) and this is the entire directory i am using for madvr (http://www.2shared.com/file/DvlutK1e/MadVR.html).
Thnx for it. :)

bye
szabi

[sepheas]

mh I've a question. For best results, no processing before sending to madvr ?

I use LSFmod as an avisynth script in ffdshow and I want to keep this.

[leeperry]

is it me or switching refresh rates when mVR is already running seems to make mVR(in windowed mode) freeze up as a black window for a few secs? It used to pick up changes much faster for me in older versions.

For best results, no processing before sending to madvr ?

I use LSFmod as an avisynth script in ffdshow
LSF only sharpens the luma and looks wunderbar through mVR, thanks to its very soft chroma.

Oh, and BTW the exclusive mode looks über-smooth too, but I'm not learning anything to anyone...same ole' same ole' :D

[XPC]

Hello Madshi,

I have some updates on the issues with MPC-HC DVB-t:


1- When switching channels, at some point the mpc-hc hangs after rendering audio of the new channel for few seconds while the last frame of the previous channel frozen. I get this in two different machines, the first one only when switching between different types of stream (MPEG-2 to h.264 and then switching back to MPEG-2) while the other (an i7 machine with GTs260 card on W7 64 bits) happens even just switching only between MPEG-2 channels (usually at the second or third switch). Could you please take a look to the logs so we can diagnose what is happening?
XPC
I noticed that this happens when switching between channels with different AR. When switching to or from a 1440x1080 (16:9) H.264 channel the application always freezes. If I use a decoder that doesn't read the AR from the stream, this doesn't happen.


2- AR and resolution: when the connection between the Demux and the decoder is established, the information on AR and resolution is taken initially from the container to allow the connection. AR dinamically changes for each program and resolution could also be changed when switching channels. This information is taken from the contents by the decoder (some decoders better than others...). Madvr currently takes the initial information and does not allow any changes in AR dinamically like EVR-CP does. Do you think that this could be addressed in the future?
XPC
This seems to be related with the previous one. We have recently fixed the MCP-HC MPEG-2 internal decoder to read AR from the stream and we found that after this change, when using this decoder with madVR, the application crashes.


3- When opendevice, from time to time I get no image. Taking a look into the log files, I saw that this happens when, for any reason, two instances of Madvr are opened at the same time. I know that you have this in your to-do list, but I think the problem is more on the mpc-hc side and fixing MadvR just hides the real problem. This is something that was addressed sometime ago in mpc-hc but from time to time fails. I don't think you should do anything about this, but I believe this could happen not only for DVB-t but also when playing any files or DVDs with MPC-HC, so you just need to be aware.

I believe I managed to fix this one. When building the filter graph, MPC-HC tryes recursively to connect several filters. If one does not connect, it is removed from the filter graph and the application tryes with the next filter in the list. The problem was that it was only removed from the filter graph but not from memory so if this happened with madVR, at some point (and only in some cases) we could have two instances of the renderer at the same time.

Regards
XPC

[memmerson]

Hi Madshi, just wondering, is the 3DLUT file stored and read entirely from memory during playback or is it referenced from disk at intervals?

[yesgrey]

Hi Madshi, just wondering, is the 3DLUT file stored and read entirely from memory during playback or is it referenced from disk at intervals?
Stored and read entirely from video memory.

[yesgrey]

Yes, this is the video and this is the entire directory i am using for madvr.
It worked fine for me, so it shouldn't be any problem with the 3DLUT file creation... Are you using YV12 output on ffdshow?

[leeperry]

PotPlayer has an A/B repeat mode: in windowed mode it loops perfectly, but in FSE mode it always hiccups at the beginning of the sequence.

I've tried to capture it in a log w/ a 29.97fps movie in 89.910Hz(using Reclock+ffdshow): http://www.mediafire.com/?pg7jb5qx5e6d2uq

[janos666]

That is exactly the point. I'm doing *zero* work on CMS. [...] I did *not* want to do the CMS work myself (via shader math).

Did you consider to implement Little CMS 2.0 (http://www.littlecms.com/)? It's ready and available for free!
In some cases, a good ICC profile (created from numerous multidimensional and iterative test patch measurements; processed with hard work) with a good ICC based CMS solution can be theoretically better than the current yCMS 3DLUT solution.
It doesn't increase the loading times as much as the reading of the 3DLUT files. It can produce better quality on LCD and plasma screens (the current yCMS solution can't really adapt to fluctuating white balance and non-additive behaviors while there is a complete solution for this problem with the existing ICC based CMS softwares - yes, the other solution is to create 3DLUTs with an ICC based CMS software but you lose the other benefits then). It is more flexible: You need only one system-wide ICC profile and you can select the desired output characteristics easily.
As much as I know, the MPC-HC team used this software in their latest SVN builds. The only problem with the MPC-HC solution that it works with EVR which is clearly inferior in many aspects. So, it simply doesn't worth it.

How much work would it take to implement this CMS solution? (Or do you resist to use other's codes, even they are free to use? - The 3DLUTs are external works too. :))

[szabi]

It worked fine for me, so it shouldn't be any problem with the 3DLUT file creation... Are you using YV12 output on ffdshow?

I am using these settings:

http://www.abload.de/img/ffdshw_outputjxc3.gif

http://www.abload.de/img/ffdshw_convertlbmw.gif

Do you think the problem is from this settings?
Anyway, are there any decoder which is better than ffdshow?

bye
szabi

[yesgrey]

I am using these settings...

Do you think the problem is from this settings?
Anyway, are there any decoder which is better than ffdshow?
Your settings are fine. I've downloaded and installed the ffdshow version you posted but still can't reproduce the problem, so you should be fine with ffdshow too, because if it was any bug on it I would get the same results.

Which OS, CPU, and graphics card are you using?

[szabi]

Win7x64, i7-860, and gainward hd4850.
I will install new catalyst, perhaps it can solve this issue.

bye
szabi

[yesgrey]

I'm also on W7 x64, but I have a GF8600GT. How have you setup your graphics card drivers? I know there are several setting for the output. Are you outputting YCbCr, RGB Full or RGB limited?

[szabi]

I do not find any kind of settings in control panel.
Are you sure these settings should be there?

[yesgrey]

Are you using HDMI or DVI?

[szabi]

I am using DVI.

[Razoola]

I am getting tearing in fullscreen window mode, I'm not sure with which version this has started but its a type of tearing I have seen before and is in a static location about 6 pixels from the bottom of the screen, I have tried all the flush settings but nothing helps. This seems to happen on the secondary monitor. I have aero disabled and have tried other renders to confirm those are working ok.

[janos666]

I have aero disabled

I am just wondering but does the Aero status matter when you run a D3D fullscreen renderer?
I can remember that old Vista builds had to disable Aero when you launched a Dx or OpenGL game but the new WDDM 1.1 allows us to keep it in the background. So...?

[yesgrey]

I am using DVI.
I don't know if those options are also available when using DVI... Sorry, but I can't help you more than this, because I don't have an ATI card...

I am getting tearing in fullscreen window mode, I'm not sure with which version this has started but its a type of tearing I have seen before and is in a static location about 6 pixels from the bottom of the screen, I have tried all the flush settings but nothing helps. This seems to happen on the secondary monitor. I have aero disabled and have tried other renders to confirm those are working ok.
This always happened to me with Aero disabled, that's why I only switched to W7 after the creation of the exclusive mode.

[Razoola]

This always happened to me with Aero disabled, that's why I only switched to W7 after the creation of the exclusive mode.

Its the entire opposite for me, this only used to happen with aero enabled. I remember with the ERV custom renderer and MPC-HC I needed to use the alt vsync method to fix it. This situation only happened on the secondary monitor btw.

Exclusive mode does fix the issue, the problem is I get dropped frames which don't happen in windowed mode. It seems exclusive mode interfers with reclock somehow, It seems to drop frames when it does not need to.

Its possible this issue has started with the new 260.63 nvidia drivers, I'm not sure.

[dansrfe]

Dear madshi,

Can you please add a "manual colorspace output" option, a "display half-res on initial start when source res > screen res option", and integrated subtitle rendering (I know you're working on this). Please respond whenever you get a chance. I really, really, really am impressed by your renderer and want to continue to use it with the options that I have listed. I'm sure these three would be helpful to other people as well even if they don't actively post in this thread. If not in this thread there are hundreds of people checking your threads as guests everyday :P

[namaiki]

a "display half-res on initial start when source res > screen res option"

If you are using MPC-HC: View-> Options-> Tick "Auto-zoom" and set the option to "Auto Fit."

[iaTa]

I was previously using the subtitle function of ffdshow with madVR but for some reason I can no longer get them working. I'm using the latest madVR and always keep up to date with the latest MPC-HC and ffdshow builds.

Has anyone else experienced issues getting the subs working?

[robpdotcom]

ffdshow stop displaying subtitles for me a few builds back. Not sure exactly which build it began with, but I'm currently using rev. 3533 with no problems.

[STaRGaZeR]

Yup, it was fixed in r3576

[iaTa]

Sods law, had just upgraded to r3576 but didn't test before I posted.

Thanks guys.

[janos666]

It's interesting. I softmodded my 8600M GT to a Quadro FX 1700 (to make it better for AutoCAD Civil3D 2011) and I have significantly lower rendering times in madVR 0.31 now. I had to use bilinear for every resizers but I can use BiCubic for luma resizing now.
(But may be it's only a coincidence. I used a different driver version -> a quadro set instead of geforce sets - but they are interchangeable with simple inf mods. It is the 259.57-Quadro-win7 by the way.)

[ajp_anton]

How does madVR's speed compare to other renderers when using the same scaling algorithms?
Just wondering if madVR is still worth it on slow GPUs.

[dansrfe]

madVR is definitely worth it ;)

[namaiki]

How does madVR's speed compare to other renderers when using the same scaling algorithms?

There isn't another video renderer that does the same scaling algorithms as madVR, unless you are talking about Bilinear or Bicubic, however the quality does not compare.

[Qaq]

How does madVR's speed compare to other renderers when using the same scaling algorithms?
Just wondering if madVR is still worth it on slow GPUs.
It's not slow GPUs but slow ... OS. With my ATI 5450 madVR works slow under XP and pretty smooth under win7 with extreme quality settings. 5450 hasn't fast GPU, you know :)

[Peuj]

Hi,

I've an issue with madVR, after the switch from one video to another the image has been lost (black screen with sound). I was on "Exclusive mode".
I use the latest PotPlayer version.

Here's the madVR log: http://www.megaupload.com/?d=4YC128DU

Thanks

GPU: nvidia 8600M GT with driver 257.15
OS: Vista
Aero: on
1920x1080 24Hz (TV)
Dual monitor PC/TV

[madshi]

when playing 25fps@100.000Hz or 24fps@96.000Hz, I usually get a rough estimate of one duped frame every....+5 hours:

You said that the duped frame estimate was currently wrong, but if it's not that off...
The estimate is wrong because my calculation used the clock deviation with the wrong sign. Since your clock deviation is very small, it isn't much of a problem in your case. The bug will be fixed in the next build, of course.

maybe James would agree to add a debug option to disable resampling when the frame rate/refresh rate match exactly, that could be fun http://forum.slysoft.com/images/smilies/agreed.gif

or maybe we could trigger the "BypassResampler" reg entry manually throught the VBS script, not sure.
I had PMed James via SlySoft PM, asking him for doing some Reclock <-> madVR communication, but he didn't seem to be very interested. At least he stopped replying after my 2nd PM.

Do you (madshi) plan to write an Audio renderer as well? :cool:
No.

Madshi, I know madVR doesn't do deinterlacing, but can it work with the card outputting 30i? I get freezes with it. Got an Nvidia GTX460 (258.96 drivers) on Win 7-64. I can't remember this being a problem with the ATI 5770, but you know, I don't remember having tested it particularly either.
madVR currently expects video frames to be delivered progressively. So you should deinterlace before sending the data to madVR (e.g. by using ffdshow raw video processor). Maybe things will change in a future version, but for now that's the way it is.

1. fps video is taken from the start of the grath, but it does not take into account that in the processing of video, he can change.
I actively use svp (Smooth Video Pack) - and he was in the process of video processing increases fps by constructing an intermediate frame (ffdshow avisynth + mvtools2.dll)
eventually to the renderer submitting more than in the source file frame rate, composition rate in old version is set 59.940 despite the fact that the video from ffdshow already comes with display fps (60.015)
perhaps because of this periodical are jerks images? unclear deemed clock deviation in 0.31- for the initial data are obtained are incorrect?

Parameters of the last file processed video:
Frame size: 1280x534 pixels
Initial frequency of video frames: 23.976 fps

Define parameters of the monitor
Width: 1360 pixels
Height: 768 pixels
Refresh rate: 60.015 Hz

Data on improving the smoothness of video
Factor increasing smooth: 5:2
The resulting video frames rate: 59.94 fps
After ConvertFps: 60.015 fps
Hmmmm... This sounds like a complicated situation. Some comments:

(1) If you modify the frame rate by adding intermediate frames with ffdshow/avisynth, the media type frame rate information *should* be updated to reflect the modified frame rate. If ffdshow/avisynth don't do that, I consider that a bug in ffdshow/avisynth. You may want to report that to the ffdshow tryouts thread.

(2) Currently the movie frame rate shown in the madVR OSD does not have any meaning, other than to inform you what the media type information block says. In a future version madVR will do automatic refresh rate switching, and for that the media type information will be used. So it *is* important that this information is correct. So again, see (1), you may want to report this to the ffdshow devs. A future madVR version may be able to detect that the frame rate information in the media type is incorrect and adjust itself properly, but that will only work after a few seconds of playback, so it will not be ideal. Ideally, the media type information should be correct.

(3) What purpose does "ConvertFps" have? Is that just another name for "ffdshow avisynth + mvtools2.dll"? Or is it something different?

2. because the video card has ati1950 render much larger number of staff - have to change their algorithms upscaling depending on the playback file. uncomfortable.
is it possible to set this selection algorithm based on presets, depending on the size of the input images?
Well, or an automatic reduction algorithm in case of constant drop frame
I don't really like to talk about what I've planned for future madVR versions. But I can tell you that a future madVR version will take care of that problem...

Can you add an option to manually select BT.601 or BT.709 in madVR since I resize to 1080p with A/R consideration of the original video in ffdshow and output YV12 from it. i want to make sure everything is shown in the correct coefficients and not always in BT.709 with my setup.
At the minimum I think my main problem would be solved if there was an option to select BT.601 or BT.709 in madVR manually and preferably with some kind of hotkey association facility.

[...]

Also I had another idea. Is there a way you can add the feature for madVR to somehow connect to ffdshow to find out the input video's resolution coming into ffdshow? That would solve the problem of resizing to 1080p in ffdshow's AviSynth tab and consequently madVR determining the wrong colorimetry to use.
These things are already on my to do list.

Also like in EVR-CP can you add the feature in madVR to display what option has been selected for a split second in the upper right corner and for movie related stuff such as chapters, duration, etc. to be displayed in the upper left corner.
As I said earlier, the current MPC HC OSD messages are not displayed when using madVR. That should be fixed sooner or later. Let's first wait for that fix. I can add further messages afterwards. If I add messages now, they might be duplicates of what MPC HC already shows by itself (once the OSD starts working).

Can you please add a "manual colorspace output" option
You are repeating yourself, for the 3rd time now. Please post every wish only once. I don't need reminding, nor repeated postings of the same wish.

a "display half-res on initial start when source res > screen res option"
You've already asked about this in the past. My answer is still the same: I think this should be done by the media player and not by the renderer. Ask the devs of your favorite media player to offer such an option. Haali implemented this feature by actually lying to the media player. I don't like that approach.

The closest I can get to 23.976 is 24hz and 59hz for 29.970
Why is that?

I also have a problem to report with exclusive mode though I am not sure if there is anything that can be done to solve this. I have the nvidia 3D glasses and on my computer panel I use a 120hz refresh rate. When using this refresh rate in fullscreen exclusive mode it causes the nvidia 3D software to kick into 3D mode. There is nowhere in the driver setting to disable this on set programs and the only solution is to disable 3D vision I have found.
I'm not sure if I can do anything about that. But let's wait for the automatic refresh rate switching functionality. Maybe it will fix that. Or not...

Regarding the subtitle problem I had. If anyone else ever encounters the same:

VobSub subtitles only work with MadVR if you use DirectVobSub/VSFilter. You can't use ffdshow to view the subtitles. SSA and other formats work with ffdshow/MadVR. If you use EVR/VMR you also don't need DirectVobSub. But for VobSubs on MadVR you need DirectVobSub.

Madshi, is that anything that can be easily fixed?
Nope, not easily, not right now.

Idea / Feature suggestion : Display homogenization
For example, we measure 12 points on the screen, interpolate a map (like a 3D surface model from survey points) and apply some compensations. (EIZO have been doing it internally by the controller for their high-end displays.)
I am not talking about full colorimetric compensations here, but luminance differences only.
Or... but it would be very hard to do: We make a highres digital photo of the display and do the map from it. (It could allow some colorimetric compensations as well. But I think that would be superfluous overkill...)

What do you think? How does it sounds: hard or easy, good or useless idea?
I've already thought about that myself. But it's one of those fancy extra features which only few people would really use, so it's WAY down my priority list. I've sorted it under "maybe sometime, maybe never".

I tried this madvr but i got an annoying problem.

Watch the middle-left part of it.

I am using mpc-hc v1.4.2499 with the following settings:
madvr v0.31 with yCMSv1.8 (3dlut activated, all other settings default) + ffdshow_rev3507_20100707_clsid_icl10 (ffmeg-mt is used)
Hmmmmm... This looks like a problem older yCMS/cr3dlut versions had. Is it possible that you installed yCMSv1.8, but that the 3dlut files were created with an older yCMS/cr3dlut version? Simply updating the yCMS version does not automatically update the 3dlut files. You need to delete the old files to force madVR to recreate them with the latest yCMS version. Does that help?

is it me or switching refresh rates when mVR is already running seems to make mVR(in windowed mode) freeze up as a black window for a few secs? It used to pick up changes much faster for me in older versions.
You're not really supposed to switch refresh rates behind madVR's back. But anyway, "For a few seconds" is not what I would expect. Can you create a log for me? As short as possible, please. Thanks.

Did you consider to implement Little CMS 2.0 (http://www.littlecms.com/)? It's ready and available for free!
In some cases, a good ICC profile (created from numerous multidimensional and iterative test patch measurements; processed with hard work) with a good ICC based CMS solution can be theoretically better than the current yCMS 3DLUT solution.
As far as I can see, Little CMS 2.0 is a standard Windows software and has nothing to do with video rendering. Which means that it's pretty much useless to me. I'd have to rewrite practically every line of code to make it run on the GPU.

I do not plan to do any CMS kind of stuff myself. That is yCMS' job. What you could do is go to the yCMS thread and ask yesgrey there to add an ICC -> 3dlut conversion utility.

I am getting tearing in fullscreen window mode, I'm not sure with which version this has started but its a type of tearing I have seen before and is in a static location about 6 pixels from the bottom of the screen, I have tried all the flush settings but nothing helps. This seems to happen on the secondary monitor. I have aero disabled and have tried other renders to confirm those are working ok.
Can you go back version by version to find out with which version the problem has started?

Exclusive mode does fix the issue, the problem is I get dropped frames which don't happen in windowed mode. It seems exclusive mode interfers with reclock somehow, It seems to drop frames when it does not need to.
I'd like to look into this problem. Can you create a log for me? Also I'd need a detail description of your setup (software + hardware + GPU graphics mode etc + Reclock config etc).

Have you disabled Reclock's VSync correction? Have you tried cleaning up the Reclock video clock timing database?

It's interesting. I softmodded my 8600M GT to a Quadro FX 1700 (to make it better for AutoCAD Civil3D 2011) and I have significantly lower rendering times in madVR 0.31 now. I had to use bilinear for every resizers but I can use BiCubic for luma resizing now.
(But may be it's only a coincidence. I used a different driver version -> a quadro set instead of geforce sets - but they are interchangeable with simple inf mods. It is the 259.57-Quadro-win7 by the way.)
That really is interesting, but the 8600M GT is very old. I'd be more interesting to check whether such a mod also helps for newer GPUs.

How does madVR's speed compare to other renderers when using the same scaling algorithms?
Just wondering if madVR is still worth it on slow GPUs.
madVR scaling consumes more GPU power than other renderers, but the quality is higher.

[madshi]

1- When switching channels, at some point the mpc-hc hangs after rendering audio of the new channel for few seconds while the last frame of the previous channel frozen. I get this in two different machines, the first one only when switching between different types of stream (MPEG-2 to h.264 and then switching back to MPEG-2) while the other (an i7 machine with GTs260 card on W7 64 bits) happens even just switching only between MPEG-2 channels (usually at the second or third switch). Could you please take a look to the logs so we can diagnose what is happening?
I've checked out the logs and from what I can see, madVR accepts the switching just fine, but the decoder doesn't start sending frames, for whatever reason. I don't know why...

2- AR and resolution: when the connection between the Demux and the decoder is established, the information on AR and resolution is taken initially from the container to allow the connection. AR dinamically changes for each program and resolution could also be changed when switching channels. This information is taken from the contents by the decoder (some decoders better than others...). Madvr currently takes the initial information and does not allow any changes in AR dinamically like EVR-CP does. Do you think that this could be addressed in the future?
This is supposed to work, and it does work with a few samples I have here, where AR changes after playing the first video frame.

I noticed that this happens when switching between channels with different AR. When switching to or from a 1440x1080 (16:9) H.264 channel the application always freezes. If I use a decoder that doesn't read the AR from the stream, this doesn't happen.

This seems to be related with the previous one. We have recently fixed the MCP-HC MPEG-2 internal decoder to read AR from the stream and we found that after this change, when using this decoder with madVR, the application crashes.
Is there an easy way for me to reproduce these problems on my own PC?

[madshi]

PotPlayer has an A/B repeat mode: in windowed mode it loops perfectly, but in FSE mode it always hiccups at the beginning of the sequence.

I've tried to capture it in a log w/ a 29.97fps movie in 89.910Hz(using Reclock+ffdshow): http://www.mediafire.com/?pg7jb5qx5e6d2uq
Hmmmmm... How many loops are in each log? I've a bit confused by the logs. In the FSE log there's a "flush" every 2.5 seconds. In the windowed log there's a flush every 2.0 seconds. So the loop was about 2-2.5 seconds? Did you use different loop intervals for FSE and windowed testing (2.5 seconds vs. 2.0 seconds)? One thing that irritates me is that in the windowed mode log it takes about 400ms after every flush, before the decoder starts sending new frames again. But for FSE mode there's no such delay. Another thing irritating me is that in the FSE log you seem to switch between windowed and exclusive mode at least twice, but not for every loop.

Can you describe the "hiccup"? How does it look like? Does madVR list dropped frames when the hiccup occurs?

[madshi]

I've an issue with madVR, after the switch from one video to another the image has been lost (black screen with sound). I was on "Exclusive mode".
I use the latest PotPlayer version.
Thanks, the renderer thread seems to be stuck somewhere. I'll try to find out what happened. Can you reproduce the problem, or was it a one-time problem?