Media Player Classic Home Cinema (MPC-HC) - DXVA!

[tetsuo55]

I posted a 50mb sample of BBC HD which brings about the issue as well :) see my post aboveAwesome, we can finally try to reproduce

[Ger]

also, i have to untick MPEG PS/TS/PVA source filter as otherwise it either goes green or is very very jerky. you have this issue as well?
To be honest, I haven't used the internal TS splitter in a long time, since there are so many issues with it. I'm just keeping it unchecked, using Haali's splitter until someone starts working on the internal splitter.

[Momber]

I posted a 50mb sample of BBC HD which brings about the issue as well :) see my post above
Thank you very much. That saves me a lot of time upping a sample through my modem (!) connection.

[Inspector.Gadget]

Hi all,

I can't get the VC1 DXVA decoder working on my laptop. Am I correct in thinking that my 9650M GT doesn't support the correct DXVA mode(s) to be compatible with the internal DXVA decoder? Thanks.

http://img23.imageshack.us/img23/9475/dxvachecker.png (http://img23.imageshack.us/i/dxvachecker.png/)

[tetsuo55]

VC1 VLD is not listed, unless its the random number one

[Inspector.Gadget]

No, just MoComp. Tetsuo55, are there any DXVA VC-1 decoders that can work without VLD? I've tried the various VC-1 Tweakers for Arcsoft and Cyberlink decoding and been unable to get them to work on my Vista x64 system. Thanks.

[tetsuo55]

PowerDVD, arcsoft, Windvd, Windows, they all support assisted decoding.

[Inspector.Gadget]

Is there a way to check if the assisted decoding is being used? I look through the output pin info from the WMVideo Decoder DMO and don't see any reference to DXVA (using EVR custom). Thanks.

[tetsuo55]

Is there a way to check if the assisted decoding is being used? I look through the output pin info from the WMVideo Decoder DMO and don't see any reference to DXVA (using EVR custom). Thanks. Lower CPU usage is the only way to see it.

[Inspector.Gadget]

Ah. So DXVA is subject to Heisenbergian indeterminacy. Thanks :D

[tetsuo55]

@Everyone,

We still need someone to write a guide on how to compile MPC-HC, in both 32bit and 64bit flavors.

A good place to start would probable be ffdshow's guide:
http://ffdshow-tryout.sourceforge.net/wiki/devel:building

[Inspector.Gadget]

I found something else interesting in MPC-HC, though I'd hesitate to call it a bug. When playing an H.264/AAC/MP4 file with corrupt headers, MPC-HC instead loads a few corrupted seconds of one of the VOB files I ripped from "Running Man" in an entirely different directory.

The catch? I deleted the Running Man directory from my HDD weeks ago. :eek:

I think DXVA decoding (MPEG-2 and H.264) might be the common culprit here: is it possible that part of the MPEG-2 stream is just kicking around my video memory? That doesn't seem like a satisfactory explanation, though, because I also get some audio...

[Klaus_1250]

Just installed Cat 9.6 on a HD3850AGP (XP32, VMR9 Renderless) and DXVA seems to be borked. Files play, but every so many frames, an old frame is displayed. Cyberlink plays them fine.

[73ChargerFan]

The trick is to skip the Catalyst drivers ending in an even number.

[Octo-puss]

The trick is to skip the Catalyst drivers ending in an even number.

I assume you also need to play the movies 47 seconds after midnight while humping a pillow?

[tetsuo55]

Just installed Cat 9.6 on a HD3850AGP (XP32, VMR9 Renderless) and DXVA seems to be borked. Files play, but every so many frames, an old frame is displayed. Cyberlink plays them fine.Could be related to a known regression in build 1141, what happens when you use an older build?

[tetsuo55]

Cat9.6 is out.

some people are reporting broken DXVA playback (black screen)
I don't have any trustworthy confirmation yet though.

[tetsuo55]

We are almost ready to release a new stable build.

the question now is are there any remaining regressions between the last stable 908 and the current SVN 1161?

We know about:
* H264 decoding issue, casimir is working on it(it's not easy to fix)
* Refresh rate detection issues, beliyaal is working on it(but the current code is almost perfect, frameworks are severely limiting our options here)
* Bilinear renderering, beliyaal is working on it (technically not a bug, it has simply been broken for years, he is trying to get some more speed out of it though)

[Kurtnoise]

TS splitter is broken w/ a lot of TS files (mainly AVC+AAC as streams I'd say)...

[mark0077]

Blu-Ray Problem
Yeah I know its probably not a regression but not being able to seek in m2ts files without the risk of getting a crash, or not being able to use another decoder without horrible stuttering within m2ts will cause alot of grief for people trying to play blu-rays..... For a new version of "homecinema" I honestly think this needs to be fixed before its released to the wider public.

DVD Quality Problem
I still think defaulting to having de-interlacing on for mpeg2 decoder is still.... well not a good idea. Users of mpc-hc decoder "out of the box" will still get half the resolution they should be getting because of 99% of badly marked DVDs out there. (marked as coming from interlaced sources, and actually coming from progressive sources). IMO the wider public "out of the box" would be better off having full resolution, non bob de-interlaced mpeg2 / dvd content, instead of defaulting to having de-interlacing enabled. We can't trust the DVD flags and with the majority of dvd's being progressive sources, off should be the default. First thing I do after installing mpc-hc is set de-interlacing to Force Weave to make my dvd's look half reasonable. We really should think of whats going to be best for the majority of users IMO. (A software solution to detecting interlaced vs progressive still hasn't appeared so defaulting to off should be best, unless the user chooses to output the flag to their gfx card etc), but default should be off (force weave).

[Mario Bros]

These two were not in the latest stable release:
*MPCHC freezing issues
http://sourceforge.net/tracker/?func=detail&aid=2800414&group_id=170561&atid=854651
http://sourceforge.net/tracker/?func=detail&aid=2799089&group_id=170561&atid=854651

[littleD]

Catalyst 9.6 with dxva works good. No problems since last drivers.

[tetsuo55]

TS splitter is broken w/ a lot of TS files (mainly AVC+AAC as streams I'd say)...But wasn't that already broken in 908?Blu-Ray Problem
Yeah I know its probably not a regression but not being able to seek in m2ts files without the risk of getting a crash, or not being able to use another decoder without horrible stuttering within m2ts will cause alot of grief for people trying to play blu-rays..... For a new version of "homecinema" I honestly think this needs to be fixed before its released to the wider public.

DVD Quality Problem
I still think defaulting to having de-interlacing on for mpeg2 decoder is still.... well not a good idea. Users of mpc-hc decoder "out of the box" will still get half the resolution they should be getting because of 99% of badly marked DVDs out there. (marked as coming from interlaced sources, and actually coming from progressive sources). IMO the wider public "out of the box" would be better off having full resolution, non bob de-interlaced mpeg2 / dvd content, instead of defaulting to having de-interlacing enabled. We can't trust the DVD flags and with the majority of dvd's being progressive sources, off should be the default. First thing I do after installing mpc-hc is set de-interlacing to Force Weave to make my dvd's look half reasonable. We really should think of whats going to be best for the majority of users IMO. (A software solution to detecting interlaced vs progressive still hasn't appeared so defaulting to off should be best, unless the user chooses to output the flag to their gfx card etc), but default should be off (force weave).m2ts problem is not a regression as far as i know, fixing it is planned for a later stable release. Ill ask casimir about the dvd thing, if it's really easy to fix he might do it.These two were not in the latest stable release:
*MPCHC freezing issues
http://sourceforge.net/tracker/?func=detail&aid=2800414&group_id=170561&atid=854651
http://sourceforge.net/tracker/?func=detail&aid=2799089&group_id=170561&atid=854651Yeah your right, i don't think these will get fixed for this release though (MPC is already correctly presents everything correctly to windows)Catalyst 9.6 with dxva works good. No problems since last drivers.which os are you using?

[mark0077]

But wasn't that already broken in 908?m2ts problem is not a regression as far as i know, fixing it is planned for a later stable release. Ill ask casimir about the dvd thing, if it's really easy to fix he might do it.Yeah your right, i don't think these will get fixed for this release though (MPC is already correctly presents everything correctly to windows)

OK Cheers. If I had control I would simply set the default de-interlacing setting from Auto to Weave. I imagine it would take two minutes to change. That will give most users watching most DVD's, the full quality they deserve, but at the cost of having bad interlaced content. We have to go with the majority I think, which is forcing weave for the benefit of the mostly progressive sourced DVD content OR if it wasn't difficult for a future version, to detect gfx card and enable outputting the interlaced flag so the gfx card can do de-interlacing. (might be tricky to decide this though). Forcing weave is much better than the current default of auto though.

[Kurtnoise]

But wasn't that already broken in 908?
nope...as far I remember, all worked fine before Beliyaal merges.

[tetsuo55]

I don't see any critical TS bugs on the tracker.
Can you point me to the exact bug reports you are talking about??

All i'm really seeing is incorrect splitter output (showing extra non existing tracks)

[Ger]

Just to make sure there are no misunderstandings. Kurtnoise's statement may be true for his samples, but there are certainly many issues (both bugs and missing features) that have been present in the TS splitter "forever". Audio tracks not recognized. Subtitles not recognized. Language codes for both are ignored. Picture freeze when skipping or starting playback. Slow skipping etc.

I''ll try to look through the bug tracker and add new ones (with samples) eventually, but I don't have time yet, and my issues are not regressions (except for the 1141 regression that Casimir is working on) so they probably won't be prioritized anyway.

[tetsuo55]

I am writing an extensive article on the built in splitter, which will be used to point into the direct we wish to go with it, and will allow aspiring developers to help out with clearly defined goals.

(Same will happen for subtitle renderer and playlist)

[73ChargerFan]

Just to make sure there are no misunderstandings. Kurtnoise's statement may be true for his samples, but there are certainly many issues (both bugs and missing features) that have been present in the TS splitter "forever". Audio tracks not recognized. Subtitles not recognized. Language codes for both are ignored. Picture freeze when skipping or starting playback. Slow skipping etc.
+1
I clear the settings when I upgrade or test for bugs, and then the next mkv or ts hd file I watch, I have to switch back to haali splitter because internal isn't acting as I expect (slow, locks up, lag on seek, no seek, no start.)

Shouldn't block the next stable, though.

Next version, splitters!

After that, interface!

[Mercury_22]

We are almost ready to release a new stable build.

the question now is are there any remaining regressions between the last stable 908 and the current SVN 1161?

We know about:
* H264 decoding issue, casimir is working on it(it's not easy to fix)
* Refresh rate detection issues, beliyaal is working on it(but the current code is almost perfect, frameworks are severely limiting our options here)
* Bilinear renderering, beliyaal is working on it (technically not a bug, it has simply been broken for years, he is trying to get some more speed out of it though)

Here is one ! somewhere between SVN 1043 and SVN 1050 EVR CP got broken resulting in a very choppy play (with some files) !

Bug report: http://sourceforge.net/tracker/?func=detail&aid=2802538&group_id=170561&atid=854651

Samples here (http://sharebee.com/e3a81240) and here (http://sharebee.com/4b4c33ab)

EDIT: I found what's causing the problem: it's the "Enable Frame Time Correction"
So this looks like part of the "Refresh rate detection issues" ?!:confused:

[ADude]

We are almost ready to release a new stable build.

the question now is are there any remaining regressions between the last stable 908 and the current SVN 1161?

Beliyaal said he would answer the questions about what the Renderer Settings actually do (in a technical sense), for example what "Vsync Off" actually turns off.
It has been several weeks since he said that, and no answer has occurred.
Until he has done so, we still don't know what his new sync code does, so we cannot know if there are any regressions.

[tetsuo55]

Beliyaal said he would answer the questions about what the Renderer Settings actually do (in a technical sense), for example what "Vsync Off" actually turns off.
It has been several weeks since he said that, and no answer has occurred.
Until he has done so, we still don't know what his new sync code does, so we cannot know if there are any regressions.Like i said earlier, disabling everything matches the old behaviour.

So all you have to do is compare to 908 with everything disabled in 1161
Lock back buffer is now called flush gpu after present

Beliyaal is really busy with private life, so the technical explanation will probably take some more time

[Momber]

Just installed Cat 9.6 on a HD3850AGP (XP32, VMR9 Renderless) and DXVA seems to be borked. Files play, but every so many frames, an old frame is displayed. Cyberlink plays them fine.
That sounds pretty much like the thing that has been bugging me, and others, for months.
In all fairness, it IS better under Vista than under XP. In that sense, installing this horrible interim OS on my HTPC has paid off.
For instance, I just finished watching Sin City (BR remuxed to mkv) and did not spot a single flickering back and forth of frames, using the built-in H.264 decoder and EVR CP.
Under XP, such a pleasant experience was unthinkable.

Also, it would appear that the problem is related to PS 2.0 (down-)scaling because I never get it with 720p x264 files I watch with DXVA (my pj is 720p), only with 1080i/p...
And only with H.264 AVC, never with VC-1.

Before publishing a new release labelled "stable" I would suggest looking into the inability to play back BBC HD caps though, with internal filters.

[mark0077]

I don't see any critical TS bugs on the tracker.
Can you point me to the exact bug reports you are talking about??

All i'm really seeing is incorrect splitter output (showing extra non existing tracks)

http://sourceforge.net/tracker/?func=detail&aid=2793531&group_id=170561&atid=854651

This bug hasn't been fixed. It basically means I cannot seek in any m2ts vc-1 content without the risk of crashing mpc-hc, or causing repeated frames for infinity (usually the number of frames set to be buffered, repeated over and over, eg 5 frames repeated forever after seeking). Another bug related to this (close because duplicate of this), also means external vc-1 decoder can't be used or massive amounts of dropped frames are seen. Huge bug for me anyways.

[kumi]

Added Bug #2807358: Subtitle renderer induces tearing.

Originally posted (http://forum.doom9.org/showthread.php?p=1272429#post1272429) in the old beliyaal thread, and still having this issue.

[ADude]

Like i said earlier, disabling everything matches the old behaviour.

How does the phrase "disabling everything" correlate to the new Renderer Settings ?

Clicking "Vsync Off" does not disable the other settings.

As I previously commented:

I tried the new sync display. With "Vsync" unchecked in the Renderer settings, other settings will change the graph. There is no documentation on the graph, but perhaps it is just showing what the code would do if it is turned on ? I am assuming that unchecking "Vsync" in the Renderer settings turns off all of MPC-HC's Vsync code and allows the graphic card drivers to do the sync in whatever manner they have been set to do - is that correct ? There are different graph results with the 4 different combinations of "Vsync" and "Alternate Vsync" checked, for example.

The current situation is equivalent to the Renderer Settings being:

- Something
- Another thing
- A setting
- Something related to the other ones in some way

When the options and settings are "play" "pause" " open file " "exit" and so forth, they are obvious or correspond to the same words used elsewhere.

But when the settings correspond to tiny different branches in code having to do with subtle video playback differences, then the documentation has to be equally specific and technical.

When that is done, then - and only then - could anyone actually test the settings to see if they actually do what they are supposed to do.

[tetsuo55]

I understand your point.
And you are right that the technical documentation on each setting has to be updated asap (but probably won't be before the next stable release).

I would rather see more options be removed and default behaviour work for everyone.

what i mean is, if you "uncheck" every "new" option (new compared to 908, introduced with the renderer changes) the renderer should act almost the same as 908.

One option seems to have been removed, "lock back buffer", this is not the case, the option is now named "Flush GPU after present"

---

the question we have at the moment is not "what do these settings do", but rather:
With all settings at default(908) and all settings are default(1161), which file has problems playing correctly in 1161 that played correctly in 908.

[Kurtnoise]

I don't see any critical TS bugs on the tracker.
Can you point me to the exact bug reports you are talking about??

#8369 (http://forum.doom9.org/showthread.php?p=1296567#post1296567)
#8125 (http://forum.doom9.org/showthread.php?p=1292262#post1292262)
#7703 (http://forum.doom9.org/showthread.php?p=1285347#post1285347)
#6734 (http://forum.doom9.org/showthread.php?p=1257649#post1257649)

more or less, what Ger & 73ChargerFan said...

[tetsuo55]

#8369 (http://forum.doom9.org/showthread.php?p=1296567#post1296567)
#8125 (http://forum.doom9.org/showthread.php?p=1292262#post1292262)
#7703 (http://forum.doom9.org/showthread.php?p=1285347#post1285347)
#6734 (http://forum.doom9.org/showthread.php?p=1257649#post1257649)

more or less, what Ger & 73ChargerFan said...I don't think these will get fixed before the next stable. Some of those are not on the tracker, so someone will have to add them.

The splitter is basically made up of 2 parts, with different evolutions:

PS: raw mpeg1, raw mpeg2, vob, evo.
TS: raw mpeg1, raw mpeg2, TS, M2TS(AVCHD)

Each new name adds extensions to the previous one.

So to get 100% compatibility we have to be spec compliant (which means we need the itu documentation) for all these revisions, but theoretically simply having the full spec for EVO and M2TS should be enough to support all the older versions.

EDIT: i wonder how legal/licence compatible it is to use the information in libavformat to improve the internal splitters

[mark0077]

tetsuo, any idea on the progress with the m2ts problem I linked to you earlier in this thread. Beliyaal apparantly implemented the "Frame Time Correction" setting to get around this problem temporarily but that seems like months ago now. The same timestamp problem is apparantly what is causing my vc-1 in m2ts seeking crashes also. Means I can't seek in any of my blu-ray collection.

http://sourceforge.net/tracker/?func=detail&aid=2793531&group_id=170561&atid=854651

[tetsuo55]

Yeah i'm not 100% sure what it all means, but timestamps should be part of the stream, and the splitter should read them.

The option beliyaal added basically lets you get away with not reading the timestamps (or something to that extent, could be the other way around).

Still that is not the solution to the problem, what we need is accurate support.

for now i think our best bet is to see if we can use parts of the libavformat source to update the internal splitter, to see if that would even help the m2ts files could be tested on ffplay.

[Casimir666]

New website is online : http://mpc-hc.sourceforge.net/

Congratulation to Martin for his great work on new design :)

[fastplayer]

New website is online : http://mpc-hc.sourceforge.net/

Congratulation to Martin for his great work on new design :)
+1 :)
Way better than the old FrontPage hack :D

[ADude]

I understand your point.
And you are right that the technical documentation on each setting has to be updated asap (but probably won't be before the next stable release).

:eek: :eek: :eek:

So you are going to release it - after months of no new stable release - before the documentation that tells the user what the settings mean ?

And I still do not think you understand my point - that we cannot tell if there is a regression until we know exactly what the settings do.

I would rather see more options be removed and default behaviour work for everyone.

what i mean is, if you "uncheck" every "new" option (new compared to 908, introduced with the renderer changes) the renderer should act almost the same as 908.

I have yet to see any feedback from Beliyaal that confirms that "unchecking" ever Renderer Setting is equivalent to 908.

Also, your use of "almost" indicates another possible place for regressions to pop up...

My experience with big hardware and software companies is that you save a lot of time by producing all possible documentation before the release, otherwise you have the same questions asked hundreds of times.

[tetsuo55]

New website is online : http://mpc-hc.sourceforge.net/

Congratulation to Martin for his great work on new design :)I agree with GTPVHD that the new site looks a bit cheezy, still its 10x better than the old site.

[tetsuo55]

:eek: :eek: :eek:

So you are going to release it - after months of no new stable release - before the documentation that tells the user what the settings mean ?The importance of documentation should not be underestimated, but many features are undocumented. It is not a blocking issue.

And I still do not think you understand my point - that we cannot tell if there is a regression until we know exactly what the settings do.Sure you can, with default settings, does the same file that worked on 908 still work on the latest SVN? (And if you want to compare apples with apples you have to make sure all new options are disabled)


I have yet to see any feedback from Beliyaal that confirms that "unchecking" ever Renderer Setting is equivalent to 908. He explained it spread over several posts(tiny bits and pieces of information), i also talked to him outside of the thread.

Also, your use of "almost" indicates another possible place for regressions to pop up...Several functions where improved, and no longer act 100% the same as 908.

My experience with big hardware and software companies is that you save a lot of time by producing all possible documentation before the release, otherwise you have the same questions asked hundreds of times.Thats the biggest difference between big companies that make money (one way or another) and open source volunteer projects that take time away from wife and kids.

[Snowknight26]

The new site has horrible grammar. :\

[neoufo51]

The new website is better looking, but you guys HAVE to get rid of the photo of the family watching the laptop. It makes it look so cheesy. You're supposed to make this player look like an advanced tool for video purists, not a Disney-branded piece of family education software.

[littleD]

@Adude
Have u seen a that (http://forum.doom9.org/showthread.php?t=123537&page=379) beliaal post? Some of descriptions dont tell much, but u have to play with options by yourself anyway as player with beliaal's changes can behave on many system differently.
And i would change default settings. I would turn on only vsynch on XP (off when using aero vista/7), flushes leave as is and the rest stuff off as it should be tuned only when problems occured.

[khagaroth]

Ditto about the family photo, cheesy as hell, and that package photo and the huge download button on the download page just feel really bad as well (have seen too many scam/crapware sites with similar images). Apart from that, it's a nice design.