madVR - high quality video renderer (GPU assisted)

[huhn]

Ah yes, powerstate. Probably makes sense to fix the powerstate to a specific value (max?) while benchmarking rendering times...

at least this is impossible with my card when dxva decoding is used it goes 500/~800 and stays there even with jinc 8 ar. if i force the high powerstate the player will crash with dxva decodering and i'm pretty sure the 6770 isn't the only card with that "awesome" powerstate design. so we have to exclude dxva decoding and maybe other dxva parts like deintlacing if this is crippled other cards too.

[nevcairiel]

at least this is impossible with my card when dxva decoding is used it goes 500/~800 and stays there even with jinc 8 ar. if i force the high powerstate the player will crash with dxva decodering and i'm pretty sure the 6770 isn't the only card with that "awesome" powerstate design. so we have to exclude dxva decoding and maybe other dxva parts like deintlacing if this is crippled other cards too.

A long standing issue with AMD cards, as soon as you use DXVA it forces itself into the "video" power state, and anything you do to change that - it won't like.

My assumption is that it also controls the clock of the decoder, and it crashes if its used at a too high clock, but thats just guessing.
Might be worth to test if just using DXVA deint also has that effect, or if its only decoding.

[Stereodude]

Question (hopefully an easy one instead of all the bug reports)... If I'm playing a VFR mkv that has 30000/1001fps and 24000/1000fps content should I see the "movie frame interval" in the OSD update during playback as it the frame rate changes? Also, the OSD says "smooth motion is off (settings)". The file starts with 30000/1000fps content so that might be normal, but should I see if turn on when the framerate in the mkv drops to 24000/1000fps? madVR is set to enable smooth motion frame rate conversion (only if there would be motion judder without it...) and when I play back 24000/1000fps cfr files the OSD says it is on.

[noee]

madshi:
Not bad news, but, fwiw, I spoke too soon on the perf note before. I was set at Spline64/AR when I tested .86 and .87(1)(b) for deint. All is good, but just to pass on that .86 and .871b both only allow me to go max Spline64/AR image with 480i material. If I go to Jinc3 on either version, I get drops, so at least perf, in this case, appears to be about the same from .86->.871b for 480i->1080 with deint.

My assumption is that it also controls the clock of the decoder, and it crashes if its used at a too high clock, but thats just guessing.

Good guess, it's been an issue as the DPM and VDPAU stuff has been implemented in later linux kernels/drivers. Quite quirky depending on asic and they don't seem to really know the fix or have time for it because it's probably different per asic. Hell, in the last dri pull, they disabled default DPM on Turks, Cayman, Bonaire?. because of this quirkiness. That said, I run the DPM (Turks) but I don't use the UVD and it works great over there.

[cyberbeing]

If I'm playing a VFR mkv...

madshi has yet to add any VFR detection code to madVR. It may happen someday, but he said it was rather complex to get right, and he wouldn't want to implement such a thing unless he could make it behave reliably.

[huhn]

Question (hopefully an easy one instead of all the bug reports)... If I'm playing a VFR mkv that has 30000/1001fps and 24000/1000fps content should I see the "movie frame interval" in the OSD update during playback as it the frame rate changes? Also, the OSD says "smooth motion is off (settings)". The file starts with 30000/1000fps content so that might be normal, but should I see if turn on when the framerate in the mkv drops to 24000/1000fps? madVR is set to enable smooth motion frame rate conversion (only if there would be motion judder without it...) and when I play back 24000/1000fps cfr files the OSD says it is on.

i played at lot with vfr file i suggest to use 60 hz and force smoothmotion.

vfr detection is not very good in madvr and splitter (most likely thanks to missing information's in the file it self)

[Stereodude]

madshi has yet to add any VFR detection code to madVR. It may happen someday, but he said it was rather complex to get right, and he wouldn't want to implement such a thing unless he could make it behave reliably.Where does the timecode information that's muxed into the .mkv go? madVR can't access that?

[huhn]

intel hd deint:

86.11 left first 2 and right 87.1b only screen...
http://abload.de/img/86.11deinthighqueueezuy7.pnghttp://abload.de/img/86.11deintlowqueuem8usw.png http://abload.de/img/87.1bdeinthighqueuenrpvq.png

long story short 87.1 plus intel = issue

render deint times are pretty unstable so don't take then to serious the gpu usage is very stable.

86.11 high queue got max power state at ~82%
86.11 low queue got max power state at ~79%
87.1b queue doesn't matter max powerstate ~92%

the problems start with upload already. the pc i used here got 1333 mhz ram (so i set it up wrong in the bios... good to know)

My assumption is that it also controls the clock of the decoder, and it crashes if its used at a too high clock, but thats just guessing.
Might be worth to test if just using DXVA deint also has that effect, or if its only decoding. my card only cares about decoding deint works at max powerstate. but i think is possible that amd messed this up tooon some cards too we are talking about amd here.

[huhn]

Where does the timecode information that's muxed into the .mkv go? madVR can't access that?

the time code info's reach madvr but smoothmotion doesn't engage on file with 29p or 23p files

[sneaker_ger]

The b test build fixes both the settings delay and the restart/next file problem for me. (Didn't test test build a.)

There has also been an mpc-hc commit about something similar, it seems:
https://github.com/mpc-hc/mpc-hc/commit/1143d90a28d3ffffc101f8412a501f04e3eb59f3

[sajara]

Madshi, just to report my experience:

OS: Win 8.1
GPU: AMD 6570M (rebranded 5730M)
CPU: i7 2630QM
Drivers: 13.12

video used: NTSC 720x480i Mpeg2 upscaled to 1366x768

86.11:
average stats
rendering ~15.00ms
present ~0.16ms
max stats (5s)
rendering ~19.70ms
present ~0.35ms

GPU usage: ~26%

87.1b:
average stats
rendering ~5.63ms
present ~0.18ms
max stats (5s)
rendering ~6.96ms
present ~0.29ms

GPU usage: ~25%



OpenCL: no problems - gpu:~53%
NNEDI3: no problems, just gpu too weak to be usable even with 16 neurons
Debanding: no problems - gpu: ~31%

For me, no problems or bugs so far.
Thank you for your great work.

[Thunderbolt8]

testing the ayaka MSSL12.ts clip for performance (mpeg2, 1440x1080 upscaled to 1080p (jinc 3 taps + ar); 29,97 fps), 0.871b is a little faster for me than 0.8611

0.8611:
without deinterlacing: 17.20ms
with deinterlacing: 17.50ms + deint: 3.30ms


0.871b:
without deinterlacing: 16.00ms
with deinterlacing: 16.90ms + deint: 2.95ms
just to compare, same clip on my intel HD 4600:

0.8611:
without deinterlacing: 50ms
with deinterlacing: 35ms + deint: 24ms

0.871b
without deinterlacing: 51ms
with deinterlacing: 60ms + deint: 10ms


edit: well I forgot to disable deinterlacing and stuff in the intel HD settings before I did this comparison. after doing so, rendering times for 0.871b for deint are ~3.30ms

[cyberbeing]

Where does the timecode information that's muxed into the .mkv go? madVR can't access that?

It already does, but getting an accurate frames-per-second value out of timecodes requires averaging over possibly hundreds to thousands of frames, depending on how much jitter exists. VFR means framerate changes can occur much more frequently than this. This then means that madVR would need to implement a large lookahead, create a threshold + averaging method which can accurately detect VFR segments containing jitter, and ultimately have an average fps of each segment before it is time to display it.

Look at Reclock's frame rate detection on VFR videos for example. It takes quite a few seconds for it to detect the VFR framerate changes during playback, and smaller VFR segments go completely unnoticed since they do not have enough frames to accurately average a fps value. Such a delayed detection implementation would not be up to madVR's quality standards, not to mention that smooth motion likely wouldn't take kindly to turning on and off frequently.

[clsid]

madshi, can you add a command-line parameter to madHcCtrl.exe that directly opens the settings window and also exits the process when closing the settings window (if the process wasn't already running)?

[cyberbeing]

madshi, can you add a command-line parameter to madHcCtrl.exe that directly opens the settings window and also exits the process when closing the settings window (if the process wasn't already running)?

@madshi

Speaking of which, there is a new issue in 0.87.1b with offline setting editing which now causes madHcCtrl.exe to continue running in the background even after the settings window is closed.



@clsid

You can already do this with the following, but it has the 5 second delay bug:

madHcCtrl.exe editLocalSettings


The version without the 5 second delay is the following, but unfortunally madHcCtrl.exe doesn't automatically close with this version:

madHcCtrl.exe editLocalSettingsDontWait

[madshi]

Madshi, just to report my experience:

OS: Win 8.1
GPU: AMD 6570M (rebranded 5730M)
CPU: i7 2630QM
Drivers: 13.12

video used: NTSC 720x480i Mpeg2 upscaled to 1366x768

86.11:
average stats
rendering ~15.00ms
present ~0.16ms
max stats (5s)
rendering ~19.70ms
present ~0.35ms

GPU usage: ~26%

87.1b:
average stats
rendering ~5.63ms
present ~0.18ms
max stats (5s)
rendering ~6.96ms
present ~0.29ms

GPU usage: ~25%



OpenCL: no problems - gpu:~53%
NNEDI3: no problems, just gpu too weak to be usable even with 16 neurons
Debanding: no problems - gpu: ~31%

For me, no problems or bugs so far.
Thank you for your great work.
Looks good, thanks for the feedback.

madshi, can you add a command-line parameter to madHcCtrl.exe that directly opens the settings window and also exits the process when closing the settings window (if the process wasn't already running)?
The version without the 5 second delay is the following, but unfortunally madHcCtrl.exe doesn't automatically close with this version:

madHcCtrl.exe editLocalSettingsDontWait
^

Noticed the bug with not closing myself. Should be fixed in v0.87.2. The correct parameter to use should be "editLocalSettingsDontWait".

[Stereodude]

i played at lot with vfr file i suggest to use 60 hz and force smoothmotion.You're suggesting I set my display to 60Hz, not setting the mkv file to 60FPS right?

[madshi]

madVR v0.87.2 released

http://madshi.net/madVR.zip

* fixed: D3D9 device wasn't freed properly, resulting in stability problems
* fixed: using shared NV12 surfaces resulted in DXVA performance drop
* fixed: opening settings without connected decoder took 5 seconds
* fixed: chroma Jinc keyboard shortcut activated NNEDI3
* fixed: chroma NNEDI3 keyboard shortcut was not listed in the settings
I think the performance problems with deinterlacing should finally be fixed now. So I hope this build will make v0.87 equal to (or better than) v0.86.11 in terms of stability and performance, as long as NVidia users don't touch OpenCL. Please test and let me know. If all is well now with this release, I can move on to trying to fix OpenCL with newer NVidia GPUs and drivers...

[DragonQ]

Nice work, the performance regression when deinterlacing is fixed for me. Comparable stats to 0.86.11. It looks like I can activate AR for chroma now too, although the GPU is dangerously close to 100% when using this for the hardest scenario (1080i at just under 1080p resolution). :)

The tray icon doesn't appear any more though, even after clicking the "Show tray icon" button. :/

[truexfan81]

What is this split queue I'm seeing in those pics? I don't have one of those.

[huhn]

You're suggesting I set my display to 60Hz, not setting the mkv file to 60FPS right?
use that what works best for you.

[huhn]

What is this split queue I'm seeing in those pics? I don't have one of those.

because you are an amd user right?

[madshi]

Nice work, the performance regression when deinterlacing is fixed for me. Comparable stats to 0.86.11. :)
Yey! Thanks for being persistent with your feedback. Your settings file and comments finally helped me reproduce the problem.

What is this split queue I'm seeing in those pics? I don't have one of those.
DXVA2 decoding and deinterlacing outputs NV12 surfaces. Unfortunately pixel shaders can't use them directly. So madVR has to convert the NV12 surfaces somehow to make them pixel shader compatible. There are 3 different ways madVR can use to do that. Solution (1) is copyback (downloading the NV12 data to CPU RAM, then re-uploading it to GPU in a different format). Solution (2) is conversion/processing via OpenCL, introduced in v0.87.0. And solution (3) is a copy operation on the GPU, which can be lossless with AMD GPUs, but produces a slightly blurred chroma channel with NVidia and Intel GPUs. There's a trade-quality-for-performance option to switch between (1) and (3). And there's a new option in the "rendering -> general settings" to enable/disable (2).

You get the "split" queue in the OSD when "copyback" is used. It shows how many frames were already converted to a pixel shader compatible format by "splitting" the NV12 surfaces via copyback.

[DragonQ]

Yey! Thanks for being persistent with your feedback. Your settings file and comments finally helped me reproduce the problem.
Don't leave me hangin', I'm gonna need an explanation of what the problem was. I may not be as proficient as yourself but I am a programmer too. :p

[madshi]

The changelog already explains it: The slowdown was caused by asking D3D9 to allocate *shared* surfaces for DXVA processing. Sharing is necessary to make OpenCL perform well (with AMD and Intel GPUs, NVidia on the other hand doesn't like sharing at all). But it seems that using shared surfaces slows down DXVA decoding/processing a lot under specific circumstances. Which doesn't make much sense to me. But that's the way it seems to be. I'm not sure if this only affects Intel or not. Could be an Intel only problem.

[ThurstonX]

Sorry for the noob question, but I read about the number of frames presented in advance being increased to 8 (from 4?) for exclusive mode. I'm looking at the "exclusive mode settings" tab and "how many video frames shall be presented in advance" is set to "4". Saw that on 0.87.1; this is with 0.87.2. GPU queue size is set to "8". Am I looking in the wrong place? Do I need to reset the settings? Normally I just unzip the new version, overwriting without question.

I didn't do nearly the extensive testing as others, but for my usual use on a Radeon HD 6670 (and less used 5670), I didn't notice any performance hit. My default settings have deband on (default setting), Smooth motion FRC on (only if there would be motion judder without it...), and Jinc 3 + AR from chroma and image upscaling. Pretty sure "trade quality for performance" is default (3 checked).

nnedi @ 32 neurons was OK if I used lesser scaling (I left deband and Smooth motion on). OpenCL error diffusion proved a little much, though I didn't try disabling debanding and Smooth motion, or going really low with upscaling. I'll likely stick with what I normally use, unless I find time to do some serious testing.

I used a 23p SD (probably recorded from DVR using S-Vid to Hauppauge; lo-fi style ;-) file and a 1280x544 23p DVD encode. I have some copy free MPEGs from Comcast (so, interlaced), but haven't checked those yet.

core i5, 8 GB RAM, always software decoding. 1920x1080 60 Hz LG LED TV.

[truexfan81]

because you are an amd user right?

nope nvidia gtx650

[za222]

Nice! v0.87.2 also has fixed the green/white/black screen issues with deinterlacing for me!
Force Film / Force Video / Auto now all work perfectly, again!

[TheShadowRunner]

madshi, I'm uncertain if it's still expected at this point with 0.87.2 but I'm still getting a blackscreen with all videos with barebone settings (using nvidia 332.21).

[Stereodude]

nope nvidia gtx650You're probably using CUVID not DXVA2 for decoding then which is why you don't see it.

[truexfan81]

You're probably using CUVID not DXVA2 for decoding then which is why you don't see it.

nope, i have LAV using dxva2cb (copy back)

[nevcairiel]

You would only see it with native DXVA2. Any other decode modes look just like software decoding to the renderer, which is part of their beauty.

[madshi]

Sorry for the noob question, but I read about the number of frames presented in advance being increased to 8 (from 4?) for exclusive mode. I'm looking at the "exclusive mode settings" tab and "how many video frames shall be presented in advance" is set to "4". Saw that on 0.87.1; this is with 0.87.2. GPU queue size is set to "8". Am I looking in the wrong place?
I've just changed the *default* settings. Those are the settings that are used when you start madVR the first time. Changing the default settings only affects new madVR users. Or it affects you if you manually reset madVR to default settings.

Nice! v0.87.2 also has fixed the green/white/black screen issues with deinterlacing for me!
Force Film / Force Video / Auto now all work perfectly, again!
Good to hear!

madshi, I'm uncertain if it's still expected at this point with 0.87.2 but I'm still getting a blackscreen with all videos with barebone settings (using nvidia 332.21).
That is certainly not expected. My best guess would be that you still have one of the OpenCL options active. Make sure that you:

(1) ... have all options in "image doubling" turned off.
(2) ... have chroma upsampling set to anything but NNEDI3.
(3) ... have the "trade quality..." option "use random dithering..." *enabled*/checked.

[James Freeman]

Madshi, Thank you very much for 87.2.

A quick question,
When I switch to Film Mode (29.97 interlaced video), detected ivtc cadence is 3:2 (24fps), isn't smooth motion should automatically disable itself with 72Hz refresh rate?
Its currently does not (maybe an old bug).

Can you please elaborate about the debanding mechanism and how it does its magic (in simple terms)?
I also see the Custom debanding left from the beta builds which is great (including the arrow controls).

Somebody needs to write a full comprehensive manual for MadVR with all its "hidden" functions.

[djfred93]

MadVR don't load with the latest version, it is stuck at opening the file and it crash on the setting window. The debug don't work either.

Log (http://www.mediafire.com/view/vmvdkk1ww0fiss4/madVR%200.87.2%20log.txt)

My system : Intel Core i7@920, Ati Radeon HD 5770, Windows 8.1, MPC-HC 1.7.2

[DragonQ]

Dunno if this helps your nVidia investigations but with my GTS 250 using 331.82 drivers, I only get the blue and red chroma channels (no luma or green) when using NNEDI3 to double luma resolution. If I set chroma upscaling to NNEDI3 I only get green and luma (no red or blue).

[TheShadowRunner]

That is certainly not expected. My best guess would be that you still have one of the OpenCL options active. Make sure that you:

(1) ... have all options in "image doubling" turned off.
(2) ... have chroma upsampling set to anything but NNEDI3.
(3) ... have the "trade quality..." option "use random dithering..." *enabled*/checked.

I'm sorry, but yes every setting is set as above :S
I made a log: http://videoff7.free.fr/madVR_log_black0872.zip
Thanks for your support.

[madshi]

When I switch to Film Mode (29.97 interlaced video), detected ivtc cadence is 3:2 (24fps), isn't smooth motion should automatically disable itself with 72Hz refresh rate?
Its currently does not (maybe an old bug).
This is probably an old bug. If you want it fixed, please enter it into the bug tracker (if it's not already there, I think I've seen something like that there already).

Can you please elaborate about the debanding mechanism and how it does its magic (in simple terms)?
It replaces every pixel with an average of 4 surrounding pixels of random distance - but only if the surrounding pixels are near to the target pixel.

MadVR don't load with the latest version, it is stuck at opening the file and it crash on the setting window. The debug don't work either.
You probably just copied madVR.ax over? You need to replace all files, especially mvrSettings.dll.

Dunno if this helps your nVidia investigations but with my GTS 250 using 331.82 drivers, I only get the blue and red chroma channels (no luma or green) when using NNEDI3 to double luma resolution. If I set chroma upscaling to NNEDI3 I only get green and luma (no red or blue).
I'm going to reproduce this myself.

[Nachbar]

Thank you for fixing the previous problems I had everything that I used works like it used to before.

Maybe add in parenthesis next to dxva2 and nnedi3 in the scaling algorithms section that its for intel/amd ? Unless its a bug and its supposed to work for nvidia. Same with the image doubling section.

[Soukyuu]

Hmm... what would be some good material to compare chroma upscaling with? I just tried switching on a couple of anime episodes and there is barely any difference, even NN looks quite acceptable most of the time. What do I have to look out for? Image upscaling on the other hand is quite obvious. Basically, I'm trying to figure out why I'd even want NNEDI for chroma upscaling.

0.87.2 works fine here btw, minus the openCL, so at least no regressions.

[madshi]

I'm sorry, but yes every setting is set as above :S
I made a log: http://videoff7.free.fr/madVR_log_black0872.zip
Thanks for your support.
Ah yes. You've disabled "prerender several frames in advance" in the "exclusive mode settings". That for some reason doesn't work with v0.87.2. I'll look into that for v0.87.3. For now just re-activate the "prerender" option and you should be good to go.

[madshi]

Maybe add in parenthesis next to dxva2 and nnedi3 in the scaling algorithms section that its for intel/amd ? Unless its a bug and its supposed to work for nvidia. Same with the image doubling section.
It is supposed to work for NVidia, sooner or later.

Hmm... what would be some good material to compare chroma upscaling with? I just tried switching on a couple of anime episodes and there is barely any difference, even NN looks quite acceptable most of the time. What do I have to look out for?
Best thing to check this with is content where there's lots of black and red. E.g. red fonts on black background. Or red numbers on an old alarm clock. Or some of the Battlestar Galactica scenes in the Cylon mothership.

Chroma upscaling is often not very important. I don't really recommend using NNEDI3 for chroma upscaling - unless your GPU doesn't have anything else to do, maybe.

[TheShadowRunner]

Ah yes. You've disabled "prerender several frames in advance" in the "exclusive mode settings". That for some reason doesn't work with v0.87.2. I'll look into that for v0.87.3. For now just re-activate the "prerender" option and you should be good to go.
I just tried and that's not it either.
It has to do with the openCL stuff again as 0.87.2 behaves exactly like 0.87 here, black screen while audio is playing then always player crash on exit..

[madshi]

I just tried and that's not it either.
It has to do with the openCL stuff again as 0.87.2 behaves exactly like 0.87 here, black screen while audio is playing then always player crash on exit..
Well, I can reproduce the problem if I uncheck the "prerender multiple frames" option off on my PC. That's definitely the reason why you got a black screen. Maybe there is an additional reason. But I can't see it in the log because the unchecked "prerender multiple frames" option already did all the damage. If you can create another log with the option checked, I can have another look. Maybe I can see what's going on.

OpenCL should not even be initialized at all, if you have those 3 options set like I said earlier. So I don't think this has anything to do with OpenCL.

[djfred93]

You probably just copied madVR.ax over? You need to replace all files, especially mvrSettings.dll.

I delete all the file except setting.bin and it's still the same issue.

[TheShadowRunner]

Well, I can reproduce the problem if I uncheck the "prerender multiple frames" option off on my PC. That's definitely the reason why you got a black screen. Maybe there is an additional reason. But I can't see it in the log because the unchecked "prerender multiple frames" option already did all the damage. If you can create another log with the option checked, I can have another look. Maybe I can see what's going on.

OpenCL should not even be initialized at all, if you have those 3 options set like I said earlier. So I don't think this has anything to do with OpenCL.
Oh my bad, sure, here it is with "present multiple frames" enabled: madV_logTSR.zip (http://videoff7.free.fr/madV_logTSR.zip)

[Stereodude]

I'm a little confused... I have these profile auto select rules, but it always picks the progressive profile even when the source is definitely interlaced (madVR reports deinterlacing is on and shows a time for it).

if (srcInterlaced = true) "interlaced"
else if (srcInterlaced = false) "progressive"

I'm using .87.2 if that matters.

Edit: False alarm. I didn't realize it doesn't re-evaluate the source (like when playing a DVD going from the previews to the menu to the movie), nor did I realize that the madVR settings window doesn't stay in sync with madVR as it's running.

[noee]

try this:

if (srcInterlaced = true) "interlaced" else "progressive"

[iSunrise]

I'm a little confused... I have these profile auto select rules, but it always picks the progressive profile even when the source is definitely interlaced (madVR reports deinterlacing is on and shows a time for it).

if (srcInterlaced = true) "interlaced"
else if (srcInterlaced = false) "progressive"

I'm using .87.2 if that matters.
You need to use "==" and not "=", since this is a boolean check.

You can also use if (srcInterlaced) "interlaced" or (!srcInterlaced) to make it even simpler.

[Stereodude]

You need to use "==" and not "=", since this is a boolean check.No you don't. link (http://forum.doom9.org/showpost.php?p=1271417&postcount=3)

However, I've edited my post as I believe I understand why I wasn't seeing the correct information.