New ffdshow build (?)

[LoRd_MuldeR]

if I'm right ffdshow just load Microsoft decoder to do it anyway, so there's no point in enabling it.

Oh really ???

[videomixer9]

prolly, the libav decoder for WMV3 is in vc9.c and this doesn't come with ffdshow's version of libav, however it's based on the vc-1 reference decoder afaik and not recommended to be used anyways, hence you're better of using ms filter directly, comes with windows anyways.

[LoRd_MuldeR]

prolly, the libav decoder for WMV3 is in vc9.c and this doesn't come with ffdshow's version of libav, however it's based on the vc-1 reference decoder afaik and not recommended to be used anyways, hence you're better of using ms filter directly, comes with windows anyways.

okay :mad:

[haruhiko_yamagata]

Originally Posted by foxyshadis
Oh right, I need to report a bug in the mt builds. Seeking is screwy - it sometimes works, and sometimes just quits decoding video after the seeked-to frame. Maybe it has to do with landing on the wrong type of frame. It won't even restart at the next I-frame, I waited to make sure. Normal builds always seek just fine.

:thanks: I tested seek bar of MPC and Windows Media Player 10. It worked.
Because Zoom Player crashes without showing any frame,
I think I must test more applications.
Please tell me what application do you use.

I'm trying to debug, though it is hard ^^;

[Egh]

Milan actually started responding to the entries on the ffdshow bugtracker, so I guess some old bugs are going to be corrected soon :approved:

Important info for gcc builders from Milan Cutka:

Date: 2006-04-25 08:54
Sender: milan_cutkaProject Admin
Logged In: YES
user_id=547197

I'm afraid the problems with GCC 4.1.0 aren't caused by ffdshow source code but by GCC 4.1 bug (or at least by build for MinGW). I'm almost certain there is nothing I could do to fix it. I compared assembly output of GCC 4.0 and 4.1 and those generated by GCC 4.1 are missing underscores in mangled symbol names (such as _ZN11CBaseFilter15QueryVendorInfoEPPw) in the parts of code where those symbols should be defined. So for now only GCC 4.0.3 is supported - if you'd find a problem compiling ffdshow using this compiler, report it to me. I promise I'll respond quickly.
I also hope GCC 4.1 gets fixed.

[Liisachan]

Who uses 5.1 vorbis anyway? Not me, but Yong. He/She uses it and that's why he/she could report that. But what's more important is, now its clear that vorbis decoder in LAVC is buggy, and it's possible that the decoding quality for normal 2ch vorbis is not perfect. which matches my vague impression that CoreVorbis is the safest/best DS Vorbis decoder too.

[Egh]

Not me, but Yong. He/She uses it and that's why he/she could report that. But what's more important is, now its clear that vorbis decoder in LAVC is buggy, and it's possible that the decoding quality for normal 2ch vorbis is not perfect. which matches my vague impression that CoreVorbis is the safest/best DS Vorbis decoder too.

AFAIK the 6ch vorbis per se is buggy. I mean that most people do not recommend to use it at all at this moment. Some even promised to correct that and do proper 6ch vorbis... A year ago already :P

[Liisachan]

oh, I see. you mean that fishy org?
um, but then again, BeLight has an option to directly convert AC3 5.1 to Vorbis 5.1, so I guess there are a few ppl who want to do that. I would use AC3 as it is if I needed 5.1 tho. Or maybe AAC? But I don't own 5.1 sound system, so...

[foxyshadis]

5.1 vorbis only isn't recommended because it has no channel coupling logic, so even vorbis backers point people to aac for that right now. There's no quality bugs (with the reference decoder), just that it's a lot larger than it has to be for any given quality.

haruhiko, I'll give wmp a try, but I've been using zoom player and mpc to test with. Also, haali's splitter. I've only done testing with mpeg4 (asp/avc) and mpeg1 so far. But, I can't get it to happen now, after shutting windows down for the night, so I guess it was something else! Weird.

[Shirokuu]

Risking going slightly off topic with the next question, but i have for months - maybe years already - an issue with using FFDSHOW combined with the Haali splitter and TMPEGenc. For some reason it happens with any FFDSHOW build since about a year ago so i guess it's not FFDSHOW related perse. Maybe somebody happens to know a workaround?

When feeding TMPEGEnc a matroska file with multiple audio tracks and one or more subtitles tracks, the subs don't get displayed. FFDSHOW feeds TMPEGenc with video and audio and Haali takes care of parsing the subtitles embedded in the MKV. With regular playback through the core player, BSP or MPC the subs DO appear.

A workaround would be to use MKVExtract to extract the embedded to a SRT file, and using FFDSHOW itself to parse the subtitles from the SRT. Of course TMPEG displays the subtitles this way, but TMPEG becomes instable when converting the MKV to for instance a DVD compliant MPEG2 stream. And on top of that two Haali splitters icons appear in the system tray. Strangely on normal playback using any player of your choice the regular amount of icons appears, namely one.

Both on my regular PC with the old 2004 stable FFDSHOW build, Haali 1.5 and TMPEG 2.0 and on my laptop with the latest FFDSHOW, Haali 1.6 and TMPEG Xpress 3.0 this behaviour occurs.

[celtic_druid]

TMPGEnc's dshow input is somewhat buggy and I believe mkv is officially not supported. So if it works at all, I guess you should be happy.

[Shirokuu]

MKV is indeed not officially supported so i have no real reason to complain of course. But I hoped to get MKV's working by using the VFAPI dshow input ability offered by TMPGEnc. I guess i have to take the long route: splitting a MKV into seperate streams with MKVExtract and feeding TMPGEnc the audio, video and subtitle streams manually through FFDSHOW.

[issa]

SVN Rev. 2531 icc 9 with gcc 4.1.1 build,

unicode build: ffdshow-20060426-icc-rev2531-unicode.exe (http://rapidshare.de/files/19025631/ffdshow-20060426-icc-rev2531-unicode.exe.html)

[foxyshadis]

Shirokuu, you can probably just extract the subtitles and use vsfilter to put them on the movie as hardsubs (at least, I'm pretty sure vfapi allows using vdub plugins on the source, if not, there's always avisynth). No need to separate it all and then try to line it back up again. If you want softsubs, you need them external to convert to vobsubs anyway. (Assuming dvd, not mpeg-1.)

[videomixer9]

SVN Rev. 2531 icc 9 with gcc 4.1.1 build,

unicode build: ffdshow-20060426-icc-rev2531-unicode.exe (http://rapidshare.de/files/19025631/ffdshow-20060426-icc-rev2531-unicode.exe.html)

i did any build in unicode so far, i think it's not even worth mentioning anymore as it is naturally to use unicode if available. Besides that it seems that ICL still does a better job on libmpeg2 e.g. but only marginal sadly. The .ax file seems to be no different with any compiler still hm.

[Liisachan]

5.1 vorbis only isn't recommended because it has no channel coupling logic, I can see your point very well. The audio part of the above sample is like 500Kbps, but the original AC3 is 384kbps, this is totally meaningless except for a testing purpose.
What you mean by "no channel coupling logic" is, it's not taking advantage of channel-correlations, such as in "mid-side coding"? (I'm not sure about those technical terms, but I mean, "does it compress each chan independently when chans are sharing a lot of similar data?")


@Shirokuu
If I have time I'll do more investigating, but I reckon... what you are trying to do is essentially like this?
If so, it's not TMpgEnc, but VSFilter doesn't like it. Maybe. Not 100% sure because this is my 1st time to try 2 connect graph this way.

http://ffdshow.faireal.net/tmp/vs2avimux.png

[Shirokuu]

Liisachan, that IS exactly what I was trying to do. Thanks for trying to reproduce the problem. You would reckon it would make no difference in a real player or in TMPGEnc, but it does. Is there a way around VSFilter that I could try?

[Liisachan]

This is not ffdshow-related, but here's what I would do.

First do this:
mkvinfo in.mkv

check fps, and the track number of sub you want.
...
| + Default duration: 41.708ms (23.976 fps for a video track)
...
| + Track number: 5
...
| + Codec ID: S_TEXT/UTF8
| + Language: eng

now, do this:
mkvextract tracks in.mkv -c WINDOWS-1252 5:WinLatin.srt

-c is for charset. if omitted, the result is UTF-8 (in that case you might need to re-save the .srt in UTF-16 or something)
5 is the track number

Make my.avs which looks like....

LoadPlugin("path\to\VSFilter.dll")
DirectShowSource( "in.mkv", fps=23.976 )
## Or, + AssumeFPS
# DirectShowSource( "in.mkv", fps=23.976 ).AssumeFPS( 24000, 1001, false )
## If upside down do
# FlipVertical()
TextSub( "WinLatin.srt" )

Open my.avs by MPC to check if it's ok. If it's ok, load my.avs to TmpgEnc.
The above won't take more than 1 minute or so.

EDIT: If it's dual-audio, maybe you need to demux the audio to handle it separately. AssumeFPS is technically needed because ((float)23.976) is not equal to = 23976/1000 nor = 24000/1001 in binary. If it's 25.000fps. not needed at all.

[foxyshadis]

I can see your point very well. The audio part of the above sample is like 500Kbps, but the original AC3 is 384kbps, this is totally meaningless except for a testing purpose.
What you mean by "no channel coupling logic" is, it's not taking advantage of channel-correlations, such as in "mid-side coding"? (I'm not sure about those technical terms, but I mean, "does it compress each chan independently when chans are sharing a lot of similar data?")
Yep, that's exactly the problem. 5.1 mixing is more complex than stereo coupling, I understand, but same idea. It's been on the todo list, but I guess no one's been available to do it in a while (most of the development outside aoyumi's tweaks seems to concentrate on speex and theora now).

[videomixer9]

ffdshow rev. 2532

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor
download: here (http://rapidshare.de/files/19057679/ffdshow-rev2532.exe.html) or here (http://www.megaupload.com/de/?d=E6EGBBNQ) or here (http://www.filepoint.de/download/2841-dl-ffdshow-rev2532_exe) (no nagging)(2.93 MB)

[namchik]

Wow! It seems that latest SSE2 build by celtic_druid doesn't crash when libmpeg2 used :cool:

[videomixer9]

ffdshow rev. 2533

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor
download: here (http://rapidshare.de/files/19084161/ffdshow-rev2533.exe.html) or here (http://www.megaupload.com/de/?d=IZD51S29)

changes: VMR9 color controls

[issa]

SVN 2533 icl 9 with gcc 4.1.1 unicode build,

URL: RapidShare (http://rapidshare.de/files/19102971/ffdshow-rev.2533-20060428-icc.exe.html) or MegaUpload (http://www.megaupload.com/?d=0VVWDA4C) [no sse & sse2 require]
URL: RapidShare (http://rapidshare.de/files/19105309/ffdshow-rev.2533-20060428-icc-intel.exe.html) or MegaUpload (http://www.megaupload.com/?d=QSPWCRAA) [sse2 intel cpu require]

[namchik]

issa
sse2 intel cpu require
is it for intel cpus only? I'm asking coz i have AMD 64...
and does it crash when libmpeg2 is used for decodeing mpeg2?

[issa]

issa

is it for intel cpus only? I'm asking coz i have AMD 64...
and does it crash when libmpeg2 is used for decodeing mpeg2?

Use the [no sse & sse2 require] version, it will use sse & sse2 when you have it.
The [no sse & sse2 require] compiled with CFLAGS "-QaxKW",
the [sse2 intel cpu reqire] compiled with CFLAGS "-QxN" & "-msse2 -mfpmath=sse".

[JarrettH]

SVN Rev. 2531 icc 9 with gcc 4.1.1 build,

unicode build: ffdshow-20060426-icc-rev2531-unicode.exe

thanks...just the stuff i'm looking for:cool:

where can i see the changes between these various builds?

[Liisachan]

Log (http://ffdshow.faireal.net/bakaupdates.php), more (http://svn.sourceforge.net/viewcvs.cgi/ffdshow?rev=2534&sortby=date&view=rev)

[videomixer9]

ffdshow rev. 2534

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, more experimental gcc switches
download: here (http://www.filepoint.de/download/2870-dl-ffdshow-rev2534_exe) or here (http://rapidshare.de/files/19110555/ffdshow-rev2534.exe.html) or here (http://www.megaupload.com/de/?d=MKML8A61)

[Yama4050242]

@issa
your build can not decode the x264 clip i made long time ago, sample here
http://rapidshare.de/files/19117988/sample.rar.html

[videomixer9]

ffdshow rev. 2535 (unicode as always)

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, more experimental gcc switches
download: here (http://www.filepoint.de/download/2883-dl-ffdshow-20060428-rev2535_exe) or here (http://rapidshare.de/files/19125295/ffdshow-20060428-rev2535.exe.html) or here (http://www.megaupload.com/de/?d=H9P10Z1M)

changes: vertical and horizontal image flipping

[videomixer9]

ffdshow rev. 2536 (unicode as always)

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, even more experimental gcc switches
download: here (http://www.filepoint.de/download/2895-dl-ffdshow-20060428-rev2536_exe) or here (http://rapidshare.de/files/19132438/ffdshow-20060428-rev2536.exe.html) or here (http://www.megaupload.com/de/?d=GH3BR3ES) or here (http://bittekeinspam.googlepages.com/ffdshow-20060428-rev2536.exe) (2.90 MB)

changes: DeBand filter

[videomixer9]

ffdshow rev. 2537 x64 tryout build
download here (http://www.filepoint.de/download/2938-dl-ffdshow64-20060428-rev2537_exe)

beware ... SLOWWWW! not milans fault, only mine :O

[LoRd_MuldeR]

ffdshow rev. 2536 (unicode as always)

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, even more experimental gcc switches
download: here (http://www.filepoint.de/download/2895-dl-ffdshow-20060428-rev2536_exe) or here (http://rapidshare.de/files/19132438/ffdshow-20060428-rev2536.exe.html) or here (http://www.megaupload.com/de/?d=GH3BR3ES) (2.90 MB)

changes: DeBand filter

Hey, why you don't make a small Homepage for your builds? Or at least a pure realese therad where the link is always in the first post (like Shartooth's x264 builds)? Would be much nicer, since this way I could permanently link to that thread...

BTW: What does "DeBand" do ???

[videomixer9]

i don't have a real example of banding on my hand, however you can add nice ghost shadows to people on fine videos with it :p It removes vertical bands that some video sources have, it seems to be an error that occurs mostly for analog input sources through interferences or broken cables afaik.

as for a homepage try this one, remembered I could "abuse" google for this:
http://bittekeinspam.googlepages.com/

[Rash]

Please no spam!:thanks: :D

[bob0r]

I could fully automated ffdshow builds, just like x264, as it also uses svn.
But, i think its better you update ffdshow every few days or revisions numbers, even when Milan farts, its added to the SVN.

Just a though, its a free world, but i am not hosting each revision of ffdshow :cool:

[LoRd_MuldeR]

as for a homepage try this one, remembered I could "abuse" google for this:
http://bittekeinspam.googlepages.com/

Looks good :) I'll link to that one!

[Egh]

I could fully automated ffdshow builds, just like x264, as it also uses svn.
But, i think its better you update ffdshow every few days or revisions numbers, even when Milan farts, its added to the SVN.

Just a though, its a free world, but i am not hosting each revision of ffdshow :cool:

But it would be still nice if you did weekly builds :) More than a week passed since your last one... :P

[foxyshadis]

Personally, I'm only updating MT builds, so that alone is why I appreciate bob0r's contribution. ;)

[videomixer9]

ffdshow rev. 2538 (unicode as always)

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060429-rev2538.exe)

changes: minor msvc bugfixes by milan, I use this to try some more funny tweaks, this time some for caching.

[videomixer9]

ffdshow rev. 2538 MULTITHREADED VERSION

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060429-mt-rev2538.exe)

the patch is note exactly made for this revision so it's a unknown to me if it works.

[LoRd_MuldeR]

ffdshow rev. 2538 MULTITHREADED VERSION

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060429-mt-rev2538.exe)

the patch is note exactly made for this revision so it's a unknown to me if it works.

Seems to works here without problems. No Multi-CPU though...

[videomixer9]

means you got no Dual-Core processor? well okay i don't have one either and the build works for me too, just wonder if it really does it's work on dualcores. Patching app threw some errors but seemed to have applied most things except for the outdated patching of .nsis2 file.

One reason I don't like external patches is that they tend to not fit anymore with even some trivial code changes here and there and need extra updates. I seriously hope milan does something about this soon.

[foxyshadis]

Well, I don't have time to in-depth test it now, but it works fine during normal viewing, no crashing and burning or anything.

[haruhiko_yamagata]

I seriously hope milan does something about this soon.

Thank you very much. But I think I have to fix at least the major bug with Zoom Player(maybe concern to other applications) before accepted.

[MatMaul]

ffdshow rev. 2538 (unicode as always)

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060429-rev2538.exe)

changes: minor msvc bugfixes by milan, I use this to try some more funny tweaks, this time some for caching.

Can you also activate high accuracy for libmad please ?
You can find the diff patch for activate accuracy mode for libmad here (http://kurosu.free.fr/ffdshow-update.tar.bz2) (made by kurosu (http://kurosu.free.fr/ffdshow)).

[videomixer9]

indeed, seems like this patch vanished after I had it applied and fiddled around a bit. Doing a new build now with 2539 revision and libmad patched too. The multithreading seems to give no harm in performance on single cpu so I'll keep it, if anyone thinks this isn't the case inform me please. As for a bug with ZoomPlayer ... dunno about that but it seems to work fine here as far as I can see, if anyone gets errors with this patch I'll also do an patchless version.

ffdshow rev. 2539

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, high accuracy libmad, multithreading by haruhiko_yamagata
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060430-mt-rev2539.exe)

Also I did a real test with a multithreaded and a non-multithreaded version on my Athlon XP 3000+ with a regular MPEG4 video.

ffdshow MT
User: 87s, kernel: 0s, total: 87s, real: 95s, fps: 387.2, dfps: 356.4
ffdshow non-MT
User: 90s, kernel: 0s, total: 90s, real: 96s, fps: 374.5, dfps: 350.0

it seems that it also gain a bit of performance with single CPU systems. Repeated this a few times and it always ended up similar.

[MatMaul]

coool thanks

[MatMaul]

I have a problem with your multithreaded buids : when I play a video (I don't know if it appears with all my videos, I have test with only 2 videos) and I navigate on the video in MPC, after 5-6 seeking I obtain an error :

http://www.mezimages.com/image/matmaul/miniature/mini_error.PNG (http://www.mezimages.com/agrandir_membre.php?na=matmaul&fi=/error.PNG)

I obtain this error with your 2 last mt buids (fdshow-20060430-mt-rev2539.exe and ffdshow-20060429-mt-rev2538.exe) but I don't have any problems with your last non-mt build (ffdshow-20060429-rev2538.exe)

[videomixer9]

your cpu type, os and mpc version and evtl. haali version would be interesting. So far I didn't hear from anyone else of problems with the mt patch with single core, so I'm kinda curious and maybe it also has something in common with those ZP crashes haruhiko reported before. I'll build another version without the mt patch. Also interesting would be if the problem also occures with bobors mt builds.

ffdshow rev. 2539

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, high accuracy libmad
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060430-rev2539.exe)