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

[JanWillem32]

Originally Posted by Liisachan

If you're implying VSFilter is not handling subpixels or it's not doing anti-aliasing, that is incorrect. The following images show, left to right: 1) Not anti-aliased; 2) Anti-aliased, pixel-accuracy; 3) Half-pixel accuracy; 4) Quarter-pixel accuracy; 5) VSFilter (I think this is 1/8-pixel accuracy).

My exact words were "really badly at". The pixel rendering kernel in the subtitle renderer merely does an extreme amount of supersampling on already badly truncated vertices. It can't hold a candle in regard to the initial sub-pixel accuracy of vertices, the configurabilty (think of ClearType, anti-aliasing levels and such) and the performance-to-quality ratio, compared to any of the other renderers I've seen.
The images you posted feature basic upright letters and the picture has a large contrast. Those are easy to render, and hides the 6-bit-at-best color rendering quite well. (The colors degrade even more if heavy grading of pre-multiplied alpha comes into play, and after the weird R'G'B'->Y'CbCr limited range conversion is used.)
The subtitle renderers are really not easy to edit. (Not only the vector renderers are odd.The bitmap-type renderers also do weird things.) The bad performance of the vector subtitle renderers is mostly due to sloppy programming and poor design choices. These two reasons are also the main cause of the awful rendering quality, which bothers me the most.
In the case of rendering SSA/ASS subtitles, there's also the factor of the color tags for the awkward R'G'B' format the standard calls for. For some reason a notation that only takes 256 levels of R'G'B' color and alpha was chosen, instead of a regular floating-point notation. This means that only few colors can be rendered at all, which is just a pity.

I've been trying to edit a few of the worst things where I could, but it's very hard. The double dependency of both VSFilter and the DirectX 9 allocator unit on the same subtitle renderers makes it a menace as well. I mostly edited the subtitle queue handler, SSE function implementations and editing the DirectX 9 allocator unit to clear it from bugs (or more commonly adding 'features' of the subtitle renderer) for the texture unit and color correction. It gained in quality and performance and I'm happy that that work paid off. It's just by far not good enough, though.

@MasterNobody: I'll give it a try. Thank you for reporting, though normally this should go on the bug tracker.

[cyberbeing]

Originally Posted by Liisachan

Similarly, things should be tested properly before imported from xy-VSFilter

I agree, especially since xy-VSFilter is based on VSFilter 2.39.

For this reason, the tentative plan was to merge anything MPC-HC needed from VSFilter 2.41 into xy-VSFilter rather than the other way around. I still haven't forgotten xhmikosr's request to get the xy-VSFilter dev on IRC to enter discussions with the MPC-HC team. Unfortunately this became delayed longer than expected. For a long time the project was on haitus, and now he's pre-occupied with completing xy-VSFilter's side of the new subtitle interface designed with madshi and nevcairiel as a replacement for the MPC-HC ISR.

Originally Posted by JanWillem32

My exact words were "really badly at". The pixel rendering kernel in the subtitle renderer merely does an extreme amount of supersampling on already badly truncated vertices. It can't hold a candle in regard to the initial sub-pixel accuracy of vertices, the configurabilty (think of ClearType, anti-aliasing levels and such) and the performance-to-quality ratio, compared to any of the other renderers I've seen.

I completely agree. Some fonts display quite horrible aliasing when rendered by VSFilter, which in many ways mimics the native rendering of its GDI32 backend. The good news is the xy-VSFilter dev has been tentatively planning as a long-term goal to completely rewrite the rendering code and the rest of VSFilter as fully floating-point along with implementation of high quality anti-aliased font rendering.

Originally Posted by Liisachan

I read somewhere that xy-VSFilter crashed just because you used a negative value between {\p1}...{\p0}, showing it's not very stable yet, though promising.

Do you remember where you read that, or rather how long ago? I'm not able to reproduce any crashes caused by negative {\p1}...{\p0} values, and I never remember receiving any bug reports about such. I'll need an example script, if there really is such a bug in the current versions.

[Liisachan]

Originally Posted by cyberbeing

Do you remember where you read that, or rather how long ago? I'm not able to reproduce any crashes caused by negative {\p1}...{\p0} values, and I never remember receiving any bug reports about such.

What I read was: "Fixed crash when \clip runs out of video frame"
http://code.google.com/p/xy-vsfilter/wiki/ReleaseNotes?tm=6
It was not {\p1}...{\p0}, but {\clip(...)}. Sorry!

@JanWillem32
@cyberbeing
I'm really happy and excited to know that there are people who are actually trying to improve VSFilter. And yes! I have to agree, in general, with what you guys are saying (though I don't really know the technical details).

Just in case there is some renewed interest: I posted these to MPC-HC bug tracker while ago:
*#2460 ASS Shadow alpha is wrong if subtitles are overlapped in a certain way
*#2461 \r in ASS can break "Position subtitle relative to the video frame"

Both are very old problems for me, but a few years ago, it was like, people were like "Reporting a bug of VSFilter? You're wasting your time. No one is working on it." I didn't know the latest buzz at all, and am happy to hear about xy-VSFilter.

[sneaker_ger]

I think #2461 does not apply to stand-alone (xy-)vsfilter anyway, since it only works on the input by the decoder. But that is not the only problem with MPC-HCs renderer and the positioning IIRC. For example positioned subtitles, boxes, etc. would also overlap into the black borders.

[MasterNobody]

Originally Posted by JanWillem32

@MasterNobody: I'll give it a try. Thank you for reporting, though normally this should go on the bug tracker.

To which one? You still use one at sf.net? Anyway thx.

P.S. btw. Link "Bugs can be reported here" from first post doesn't work.

[JanWillem32]

The support readme: https://github.com/mpc-hc/mpc-hc#readme
The support tracker is still the same as it was indeed. For your sample, it doesn't crash for me. It only rarely throws an exception that the crash reporter catches. The video keeps playing for me. It's a bit hard to debug, but I'll try some more.

@Liisachan:
For #2460; A layer property is probably stored wrong. I don't exactly know where to look for this code. I generally don't touch the code that handles these sorts of things.
For #2461; I re-wrote the DirectX 9 texture unit in the renderer fixes builds. It seems to work for the sample.

By the way, in the picture you posted, the sample for "no anti-aliasing" shows pixel blur (with a strong ClearType tinting of red on the left). If I render fonts without anti-aliasing, absolutely no color blending occurs. Also, the color is different than the other samples. Do you have any further source information about this picture?

[Liisachan]

@JanWillem32
The first picture is drawn as SSA by VirtualDub's old plugin, called Subtitler, with the Advanced Rasterizer disabled. The last picture is drawn normally (as text) by VSFilter. The other pictures are drawn by VSFilter too, but as a path, as {\p1}...{\p0}, {\p2}...{\p0}, and {\p3}...{\p0} respectively. I'm on Win XP, with USP10 1.0626.7600.20602. It's just showing VSFilter is anti-aliasing, and (my guess is) it's internally equivalent to {\p4}...{\p0}. I didn't mean anything else.

EDIT: Wait, captions like "no anti-aliasing" are drawn normally on an image editor. They are not part of demo, but simply captions.

[cyberbeing]

Originally Posted by Liisachan

*#2460 ASS Shadow alpha is wrong if subtitles are overlapped in a certain way

xy-VSFilter doesn't exhibit this issue.

By the looks of it, the bug was inadvertently fixed in xy-VSfilter's October 30th 2011 build, and has never resurfaced since.

[Pulstar]

I still can't figure out the 10bit option under Presentation. Using stats the output is still 8bit, even if it is forced, and no decoder can output 10bit through EVR(CP) at all. Is there a point to enabling it? 10bit sources are seldom seen, and madVR is overkill for my needs.

[JanWillem32]

The 10-bit output is a renderer feature, independent from video source. (The renderer can operate without video input as well, and output in 10-bit.)
Requirements for this feature:
-operating on Windows 7
-AMD Radeon/FireGL X1??? generation video card or newer, using a D-Sub analog (verified), HDMI (verified) or DP (no reports yet) connection, or one of the enabled Nvidia Quadro video cards since 2007 with a DP connection (no reports yet), I don't have information on Intel GPUs and 10-bit output capability yet
-D3D fullscreen exclusive mode enabled
-a display that will accept the 10-bit RGB signal (Note that the options for the Y'CbCr and 'limited range'/'studio range' RGB formats in the video card configuration panels force conversion of the pictures after the renderer, which degrade the image. Only full-range RGB formats are available for the renderer to output. Trying to output natively on a surface in a Y'CbCr format for display output has not been successful yet.)

[gilic]

Originally Posted by JanWillem32

-AMD Radeon/FireGL X1??? generation video card or newer, using a D-Sub analog (verified), HDMI (verified) or DP (no reports yet) connection

I can confirm that 10bit output over displayport works (for me) under statistics display says A2R10G10B10.

[Pulstar]

Ah yes, it works now, though the D3D option is a bit inconvenient. Cheers!

Edit - Actually it doesn't. Setting output to P010 in ffdshow just crashes the player, and using internal filter the output is NV12.

[ajp_anton]

What can I do to always automatically move a fullscreen video to the very top to give more space for subtitles in the bottom black bar (so they don't overlap with the movie so much)?

[Mikey2]

Can anyone please help with this problem I Have been having for a long time?

Basically, my "mpc-hc.ini" file very often gets corrupt. Whenever I save anything from the "options" menu, MPC freezes for 10-30 seconds. It took me a while to see that in Windows explorer it is rebuilding "mpc-hc.ini". (I can see the file-size go up from 0 to the current size of 254 KB.) If I do anything, stop anything, open another instance of MPC, etc while it is re-building this file, it stops right there, leaving me with an uncomplete ini file.

I'm a software engineer, so I can get pretty technical, but it seems to me that in order to fix other possible defects that MPC is rebuilding the entire ini file needlessly (often nothing changes at all...You can reproduce the problem by selecting "Options" then simply pushing "OK.")

Thanks in advance for any help. (And let me know if I can do anything to help!) This initially was just an annoyance, but it is getting real bad now...

Oh currently I"m on version 1.6.4.5881 (2196459) - however, as I mentioned, I have had this issue for as far as I can remember. (Although I think it is getting worse as new options are added in subsequent versions.) I'm also on Windows 7 x64, but I also noticed this on my 32-bit Windows 8 laptop.

thanks again,
MikeY

[Pulstar]

Originally Posted by ajp_anton

What can I do to always automatically move a fullscreen video to the very top to give more space for subtitles in the bottom black bar (so they don't overlap with the movie so much)?

Use the numpad keys to shift the picture in any direction you want. You can also untick the "Position subtitles relative to the video frame" option under Default Style.

[Tiduz]

edit: problem solved.

[ajp_anton]

Originally Posted by Pulstar

Use the numpad keys to shift the picture in any direction you want. You can also untick the "Position subtitles relative to the video frame" option under Default Style.

Well, that's not really automatic, and it will rarely align perfectly with the top, but I guess it's better than nothing (once I remap the keys to my laptop keyboard).

[pulbitz]

Originally Posted by pulbitz

It is deblocking problem.
Other decoders are the same condition.
Anyway I turn it off and caputre.

mpc-hc_dxva_decoder_does_not_deblocking.jpg:
other_decders_are_OK.jpg:


Edit:
ffdshow_libav_skip_deblocking_always.jpg:
I have compared between MPC-HC H.264 DXVA decoder and 'skip deblocking always' in ffdshow libavcodec.
They are same quality.

bump.
Not yet fix. (MPC-HC H.264 DXVA decoder doesn't support deblocking for Intel HD Graphics)

sample.flv
http://www.sendspace.com/file/xdrg1y

[JanWillem32]

Originally Posted by Pulstar

Edit - Actually it doesn't. Setting output to P010 in ffdshow just crashes the player, and using internal filter the output is NV12.

The EVR and VMR-9 mixers require support from the video card drivers to handle formats. The Y'CbCr formats supported by all reasonably modern video cards are: NV12 (8-bit 4:2:0), YUY2 (8-bit 4:2:2) and UYVY (8-bit 4:2:2). Nvidia also added YV12 and I420/IYUV (both 8-bit 4:2:0) support. It's not likely that newer drivers will support more formats anytime soon. Equipping the renderer with a custom mixer that supports additional formats is possible, but creating a custom mixer would require quite a bit of programming work.

[GrofLuigi]

Originally Posted by Mikey2

Can anyone please help with this problem I Have been having for a long time?

Basically, my "mpc-hc.ini" file very often gets corrupt. Whenever I save anything from the "options" menu, MPC freezes for 10-30 seconds. It took me a while to see that in Windows explorer it is rebuilding "mpc-hc.ini". (I can see the file-size go up from 0 to the current size of 254 KB.) If I do anything, stop anything, open another instance of MPC, etc while it is re-building this file, it stops right there, leaving me with an uncomplete ini file.

I'm a software engineer, so I can get pretty technical, but it seems to me that in order to fix other possible defects that MPC is rebuilding the entire ini file needlessly (often nothing changes at all...You can reproduce the problem by selecting "Options" then simply pushing "OK.")

Thanks in advance for any help. (And let me know if I can do anything to help!) This initially was just an annoyance, but it is getting real bad now...

Oh currently I"m on version 1.6.4.5881 (2196459) - however, as I mentioned, I have had this issue for as far as I can remember. (Although I think it is getting worse as new options are added in subsequent versions.) I'm also on Windows 7 x64, but I also noticed this on my 32-bit Windows 8 laptop.

thanks again,
MikeY

I have never seen anything like that (in the sense that MPC-HC is doing it), and I work on multiple computers/OS-s, often comparing the .ini-s and sorting the sections. If I change one option, only that option is written. There are some sections that MPC-HC likes to sort in its own way, and they aren't rewritten when I just visit the Options page(s).

I think in your case it might be caused by virus/antivirus or UAC File System Virtualization.

GL