ffdshow development #2

[iron2000]

Is the postprocessing faulty again?
I always use the same video to test the codec.

The video and audio is unsynced when postprocessing is on.
It played well on the 27-May-2005 build though.

Info on video from GSpot:
http://img226.echo.cx/img226/1285/vidprop9sx.png


Hmm..

I tried playing it with Koepi Xvid latest beta codec.
Plays fine without postprocessing but unsynced with post processing.

Is it a file problem?

[Tuesday]

is anyone else having trouble with the www.aziendeassociate.it site?

Every file i get off it seems to be b0rked :S

Can anyone provide a mirror please ?

[Sharktooth]

Yup, i mantain the site and it's constantly overloaded...:(
Yesterday 540 concurrent connections was reached...

[iron2000]

Heres a mirror for ffdshow:
http://m17n.cool.ne.jp/freeware/mpc/
theres links for other mirrors in there too.

[Revgen]

I go to bob0r's site (http://x264.nl) if I can't get it from Sharktooth's server.

[Rash]

If I'm writing a DVD (data) with one or more AVI files whilst ffdshow is installed NERO either locks up or crashes.
Does that happen with any ffdshow version or just this 11th June one? Because I have never had this problem. But I didn't test with the latest ffdshow release yet.

[Yong]

the jun 11 2005 version of ffdshow seems have problem with YUY2 colospace with "resize & aspect" filter.
ive tried play XviD(XviD dshow decoder) and MPEG 1 video(ms mpeg 1 decoder),
then i force ffdshow decode raw video by using YUY2 colorspace,
only the luma method:simple/warped resize dun have problem.
heres the pic output by ffdshow:
http://img117.echo.cx/img117/4044/yuy24wf.png

[wata]

i have a avi with a corrupt frame 0
all version after ffdshow-20050406.exe crash the player with this fault

Faulting application mplayerc.exe, version 6.4.8.4, faulting module libavcodec.dll, version 0.0.0.0, fault address 0x001fe3ef.

[vortex_hl]

@celtic_druid
icl9.0 released.
will you compile ffdshow with icl9.0? what are you (and milan) thinking about it?

[celtic_druid]

http://www.intel.com/software/products/compilers/cwin/whatsnew.htm
Not sure that there would be much of a difference other than for dual core CPU's.

[LigH]

The new "Cinderella" MPEG4-AVC-HP trailer from Ateme (http://forum.doom9.org/showthread.php?p=672441#post672441) does not play in ffdshow or Mplayer. It seems that they used features which are not yet supported in libavcodec:

MPlayer dev-CVS-050609-00:54-3.4.4 (C) 2000-2005 MPlayer Team
CPU: Advanced Micro Devices Duron Spitfire (Family: 6, Stepping: 1)
Detected cache-line size is 64 bytes
CPUflags: MMX: 1 MMX2: 1 3DNow: 1 3DNow2: 1 SSE: 0 SSE2: 0
Compiled for x86 CPU with extensions: MMX MMX2 3DNow 3DNowEx


85 audio & 199 video codecs
Playing cinderella-ateme-3000k-hp.mp4.
ISO: Unknown File Type Major Brand: avc1
QuickTime/MOV file format detected.
--------------
Warning! pts=1 length=0
MOV track #0: 1 chunks, 0 samples
Generic track - not completely understood! (id: 0)
--------------
Warning! pts=1 length=0
MOV track #1: 1 chunks, 0 samples
Generic track - not completely understood! (id: 1)
--------------
Warning! pts=1484398397 length=0
MOV track #2: 72 chunks, 3559 samples
MOV: AVC decoder configuration record atom (49)!
MOV: Found unknown movie atom btrt (20)!
Image size: 1280 x 720 (24 bpp)
Display size: 1280 x 720
Fourcc: avc1 Codec: 'AVC Coding'
--------------
MOV: longest streams: A: #-1 (0 samples) V: #2 (3559 samples)
VIDEO: [avc1] 1280x720 24bpp 23.976 fps 0.0 kbps ( 0.0 kbyte/s)
==========================================================================
Opening video decoder: [ffmpeg] FFmpeg's libavcodec codec family
Selected video codec: [ffh264] vfm:ffmpeg (FFmpeg H.264)
==========================================================================
Audio: no sound
Starting playback...
[h264 @ 00A3A468]custom scaling matrix not implemented
[h264 @ 00A3A468]non existing SPS referenced
Error while decoding frame!

P.S.: I still didn't find out for sure where to report this "issue" (not quite a bug, right?). Mplayer and FFmpeg are related to the libavcodec library, but who/where are the "core developers"?

[Cyberace]

didn't find out for sure where to report this "issue" (not quite a bug, right?)... ...who/where are the "core developers"?FFmpeg bug-reports:
http://ffmpeg.sourceforge.net/ffmpeg-bugreport.html

MPlayer bug-reports:
http://www.mplayerhq.hu/DOCS/HTML/en/bugreports.html
http://bugzilla.mplayerhq.hu

(I believe the "core developers" are the same as for x264 codec (http://www.videolan.org/x264.html) and also FFmpeg lead-developers (http://sourceforge.net/project/memberlist.php?group_id=16082))

[Sharktooth]

SPS is not mandatory for AVC decoders, so ffmpeg should simply ignore it (or add support for SPS), so this is a bug.
For what concerns the custom matrices they are in the HP specs, so they should be implemented to be compliant. This is a missing feature but since the decoder is not compliant to the standard it can be considered a bug.

[LigH]

I was just unsure: Both ffmpeg and Mplayer are based on libavcodec, but there is no special libavcodec buglist.

I sent a mail to the ffmpeg developer mailing list, that seemed most suitable to me.

[Inventive Software]

libavcodec is part of the FFMPEG project. So if you want to bitch about libavcodec, talk to the guys at FFMPEG.

[Rash]

http://www.intel.com/software/products/compilers/cwin/whatsnew.htm
Not sure that there would be much of a difference other than for dual core CPU's.
Well, it supports the 64-bits instructions, doesn't it? But I don't think ffdshow's case is as simple as compiling it in 64-bits. :cool:

[multiblitz]

Well, does this mean that compiling ffdshow simply with Intel V.9 compiler gives us a multi-threaded ffdshow ? To good to be true !!!

[LigH]

I don't think so. Multiple threads must be handled by one "base" thread, must be synchronized. This requires code in the source. A compiler might optimize multi-threading for optimal performance on real multi-processor or multi-core systems, but the thread forking and control must be present in the source already.

Multi-Threading routines will still work even with only one CPU, anyway - but only time-sliced.

[madman1980]

What's up with the SPP Deblocking in Postprocessing anyway? I thought that when i upgraded to A64 3800+ I could run it, but still as slow as ever. No one can use this thing. :angry:

[Shinobu]

SPP Deblocking is ok 720*576 on my athlon 2000+ (set up 2 (H+V luma), more than 2 is ways to blurry) and mpc take about 85% cpu with it

++

[madman1980]

Ok, that worked better. All the six boxes were marked by default. Still a little slow though, especially with soft treshold.

[bob0r]

New ffdshow build needed :eek:
lossless playback! http://forum.doom9.org/showthread.php?t=95962
:thanks: :sly:

[celtic_druid]

Don't think that libavcodec in ffdshow has been updated yet.

[Egh]

Bug or feature?

(finally able to post in the forum after initial 5 days:)

So -- is this bug or feature?
In ffdshow one can enable visualisation of motion vectors and quantizers. If libavcodec is used, everything is ok. If xvid is used, only quantizers will be shown.

[Egh]

2 celtic_druid:

also, could you please explain what presicely "Fix luminance level" option does?

[bob0r]

Don't think that libavcodec in ffdshow has been updated yet.

Check again please, sorry i was a bit hasty :o

[celtic_druid]

http://mirror05.x264.nl/celtic_druid/force.php?file=./ffdshow-20050619.exe

[LigH]

Yes! You guys are tight at the heartbeat of development!

[Liisachan]

When I first downloaded it from the above URL, the filesize was 2480344, but it installed fine. I downloaded the same file several times by several methods (to make sure I'll mirror the correct file), and I think the original filesize is 2480343. I have no idea why the filesize was different when I first downloaded it...

[celtic_druid]

2,480,343 bytes.

[Liisachan]

k, thanks.

[moonman]

I have mirrored it here:
http://svartling.hopto.org/index.php?q=node/view/1039

[bob0r]

Good work.
The lossless encodes i did all work!

Mirrored also, ofcourse :cool:

[FredThompson]

@celtic_druid,

How about creating a shortcut to the localized help file during installation?

[LigH]

And since x264 now supports custom scale/quant matrices, I hope that the ffmpeg team will update their routines as well; MP4 files created by x264 with "--cqm jvt", for example, play well with the Ateme beta decoder filters.
__

I just found a strange bug in the FourCC registration on my system!

Although I registered ffdshow's VfW interface to decode FFV1, applications like VDub(Mod) tell me they cannot find a decompressor for 'ffds' or 'FFV1'. Reason:

[HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Drivers32]
vidc.ffds = ":ff_vfw.dll"

Please note the leading ':' colon !

I'm not sure if this was a mistake which was introduced in earlier versions. So I uninstalled and re-installed ffdshow (2005-06-19), and watched the registry -- the colon did not appear again. Who knows...

So if someone else wonders why he cannot decode a video encoded by ffdshow's own VfW codecs, he might want to look at this registry entry.

[Sharktooth]

sorry for the delay, i've been busy but finally i mirrored ffdshow too.

[Egh]

AVIsynth routine bugs (?)

If avisynth script is used inside ffdshow while playing video, then unchecking box "add ffdshow as video source" is followed by an instant playback stop and ffdshow+mpc shutdown. Quite inconvinient, could a fool-proof protection be implemented here?

Also, if "Avisynth" box in ffdshow is enabled (despite box "add ffdshow as video source" being checked!), and mpc is restarted by user, then _randomly_ ffdshow refuses to start (the moment mpc window appears after double-clicking on a video file it closes again). If this happens, the workaround is to go directly to ffdshow's configuration dialog and uncheck Avisynth box. Then you may double-click the same file, it starts nicely. And the very same avisynth script works again after being enabled (so it's certainly not a particular script problem). I'd thought that happens cause during initiation of videoplayback ffdshow doesn't always provide usable videosource to avisynth but tries to use it (prooly leads to internal exception of some sort, but no messages are shown). Maybe caused by increased CPU load too (several filters enabled).

Also, requesting info on filters priveleges in ffdshow. For example, if i have aWarpsharp, bicubic resizer and avisynth script simultaneously enabled, is resizer the last one to be executed ?

[vinetu]

Egh,
Right now ,there is some incompatibility between Avisynth and winXP Service Pack 2 ,thanks to m$...
- if something is wrong with the *.AVS script the host application will just disappear instead of showing the error message .

Try to put in ffdshow some "wrong proof" commands like this:

code:

ShowSMPTE(25)

and see what happens ...

[Shirokuu]

I encountered a bug in the june 19th FFDSHOW compile. Playing a particular MKV (XVID/VORBIS) with bsplay or zoom player works perfectly. But as soon as i browse the playing video to any spot on the timeline the movie starts to stutter. Audio and video run out of sync every few sync. The strange thing is that OSD shows that bitrate is dropping in a slow steady pace from 2000 down to null. While playing the same MKV from the start gives a nice almost steady bitrate of around 1900~2000.

I don't know for sure if the problem is related to FFDSHOW, since I use the latest Haali splitter and the OGG directshow filter to decode the audio.

EDIT: OOPS, problem fixed. Using FFDSHOW to decode the Vorbis audio stream as well fixed the problem. I will leave the post here if people encounter the same problem.

[Rash]

Hey guys. I just noticed the "2/2 quadro" preset on the Mixer (audio filter) doesn't relay the LFE channel to the front R, L anymore. And it used to. Does anyone know why? I used to like it. :(

No more LFE relay to the .0 presets » http://img147.echo.cx/my.php?image=ffdshowaudio6yq.png

ffdshow release ffdshow-20050619.exe

[pankov]

I'm having problems with the latest build (20050619).
The playback of DTS wav files using libdts as a decoder is horrible on my NF3 onboard ALC850 soundcard. The sound is VEEEERY loud and corrupt althogh I can guess what's been played so it do decode it correctly. So this makes me think that the problem is in ffdshow itself not libdts. When I use the 20050303 version everything is perfect.
can someone confirm this or this is only with my setup?

[pankov]

I found what was causing the problem:
32 bit floating point output
I knew that I have heard the same horrible sound before and after changing different settings I found it
:)
The problem was solved when I disabled it.

I suppose that it was intentional to change the priority of the different output sample formats and make the 32bit float the default one in the new version. But I don't think it's a good idea because as the note bellow it sais there can be driver incompatabilities.

[Defiler]

I've never read through the ffdshow source code, but I'm wondering how hard it would be to split the decoder functions and postprocessing functions into two separate modules.
At this point, it's extremely difficult for testers to help pinpoint the cause of a particular playback problem, because ffdshow is so feature-rich.
If there were a little checkbox called "Connect to post-processing module?", you could disable all of those code paths and easily verify that changes to one of the post-processing features hadn't broken something else.

Just a thought.

[ExtraEye]

but if you just disable all the features wouldn't it work just as well? I mean, what's the point of seperating it?

[Defiler]

You'd think so, but take a look at how many posts there are in this thread along the lines of "New version improves feature X, but now I can't play files encoded with codec Y."

Sometimes life is simple enough that your code can look like:
if(blue_widget_checked) { tweakMagicKnob(z); }

Other times, for various reasons, you might be stuck doing:
z = 0
tweakMagicKnob(z);
In theory, both should do "nothing".. but if a bug has just crept into "tweakMagicKnob", any process that passes through it may be affected.

There's a reason that all the video card drivers in the world don't come in one big file called "video_card_drivers.exe". Heh.

[DeeGee]

Was someone from the x.264 team doing custom quantizer support for h.264 in libavcodec? I'd like to use ffdshow to watch all my encodings...

[LigH]

Look into the libavcodec CVS (http://www1.mplayerhq.hu/cgi-bin/cvsweb.cgi/ffmpeg/libavcodec/?cvsroot=FFMpeg&sortby=date) for yourself if someone committed an according patch recently: As far as I can see while writing this reply - not yet...

[foxyshadis]

I encountered a bug in the june 19th FFDSHOW compile. Playing a particular MKV (XVID/VORBIS) with bsplay or zoom player works perfectly. But as soon as i browse the playing video to any spot on the timeline the movie starts to stutter. Audio and video run out of sync every few sync. The strange thing is that OSD shows that bitrate is dropping in a slow steady pace from 2000 down to null. While playing the same MKV from the start gives a nice almost steady bitrate of around 1900~2000.

I don't know for sure if the problem is related to FFDSHOW, since I use the latest Haali splitter and the OGG directshow filter to decode the audio.

EDIT: OOPS, problem fixed. Using FFDSHOW to decode the Vorbis audio stream as well fixed the problem. I will leave the post here if people encounter the same problem.
I had this problem with the radlight vorbis codec in the 6/11 and 6/19 release too, even when I didn't seek, always at the same points. (Drove me insane when I couldn't get it to use ffdshow or any other vorbis decoder without uninstalling.) It occured only in mkvs (several versions of haali splitter), ogms played fine. Apparently some incompatibility creeped in.

[nanoflower]

I've seen something like that (stuttering picture) recently with the recent celtic_druid build. What I see is when I have two copies of MPC open watching two different videos they both start to stutter using ffdshow. It seemed that the problem was with the libavcodec. If I changed the decoder for Xvid videos to be Xvid instead of libavcodec the problem went away when I restarted the clips.

[namchik]

Does any1 know when Milan is going to release his new builds?
his latest build of March, 3 was good and fast... but old =(