New ffdshow build (?)

[clsid]

resize crashes are been fixed in latest revision

[haruhiko_yamagata]

resize crashes are been fixed in latest revision
Not only comminting the patch, he added dialog for setting and fixed my bug. Thank you, milan!

[foxyshadis]

So basically ffdshow needs to be re-engineered to do multithreading all the way through, presumably by pre-buffering several frames of audio or video data (which can decide how to best parallelize that part), decoupling the filtering engine from the decoding engine and feeding into a worker thread per core like vm9 says, then having a pool at the renderer that ensures frame order and timing are correct. Shouldn't be too hard... for server engineers. ;) I wish I knew more about implementing threading, I'd love to work on it. I'm curious how much of that structure is already done.

Until then, having portions of it multithreaded is still better than nothing, though.

[issa]

Rev. 2545 build,

Compiler ICC 9.0.30 (msvcrt 8 included) + GCC 4.1.1
CPU extension: SSE or SSE2
Download: RapidShare (http://rapidshare.de/files/19652727/ffdshow-2545-icc-sse-20060505.exe.html) or MegaUpload (http://www.megaupload.com/?d=5537E1BA)

How many people still want to try pure GCC 4.1.1 build?

[videomixer9]

So basically ffdshow needs to be re-engineered to do multithreading all the way through, presumably by pre-buffering several frames of audio or video data (which can decide how to best parallelize that part), decoupling the filtering engine from the decoding engine and feeding into a worker thread per core like vm9 says, then having a pool at the renderer that ensures frame order and timing are correct. Shouldn't be too hard... for server engineers. ;) I wish I knew more about implementing threading, I'd love to work on it. I'm curious how much of that structure is already done.

Until then, having portions of it multithreaded is still better than nothing, though.

ffmpeg will add MPEG4 and other multithreaded for sure. Then ffdshow can just set an amount of threads to be used for decoding and libavcodec will do effective video decoding multithreading. ffdshow's filtering is another case, but most filters aren't that cpu heavy anyways, and for some the original devs will prolly also do something about it.

[foxyshadis]

I'm beginning to think people are intentionally agitating videomixer.... You could just read the last page.

http://forum.doom9.org/showthread.php?p=822612#post822612

[MatMaul]

Same problem (crash of MPC after seeking) with rev 2544.
I stay on rev 2539.

[videomixer9]

there is this nifty insurgent thing on the CCCP website, maybe try that and post the results of the log somewhere. If that problem is persistent than it may have to do with something else on your system that conflicts with MPC and ffdshow ...

http://www.cccp-project.net/downloadi.php

you can post result on nopaste.info or similar for better overview.

ffdshow rev. 2546

requires: SSE
compilers: GCC 4.1.1 + MSVC 8
specials: high accuracy tremor, high accuracy libmad, runtime cpu detection
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060509-rev2546.exe)

changes: edge emulation for deBand - works only when processing full frame

[MatMaul]

I don't think it's a filter problem because when I render the file with mpc the filter chain seems to be normal :
AVI MPC source -> ffdshow MPEG4 decoder -> video renderer

log file of CCCP (http://www.nopaste.info/index.php)

[MatMaul]

The uninstaller doesn't remove audxlib.def and audxlib.dll.

[asasadad_1]

lpcm decoding crashes in ffdshow rev. 2546(by videomixer9)

[videomixer9]

hehe is there any directshow filter available you didn't install? :O Well if the filters used are really only those. There are some evil things in that list but they usually don't interfere with avi playback. Btw. are those crashes really produced by video rendering? Try to disable sound output and seek around, dunno which audio filters you use but they may also be the culprit.

did lpcm still work in 2545, if yes then I wonder how it got broken ... need to check that but don't have any lpcm ... guess I try search the forum for some sample.

Ah dammit forget to add the delete command in the uninstall section, thx MatMaul. reuploaded the version with hopefully fixed uninstalling.

[MatMaul]

if it's an audio problem, ffdshow is also the problem because it decode the audio of those files with ffdshow :p
I have removed the audio of a file and same problem......
I think it's a problem of the couple ffdshow mt patch-MPC...

EDIT : same crash with last rev of mpc (rev609 compiled by CelticDruid)

[asasadad_1]

hehe is there any directshow filter available you didn't install? :O Well if the filters used are really only those. There are some evil things in that list but they usually don't interfere with avi playback. Btw. are those crashes really produced by video rendering? Try to disable sound output and seek around, dunno which audio filters you use but they may also be the culprit.

did lpcm still work in 2545, if yes then I wonder how it got broken ... need to check that but don't have any lpcm ... guess I try search the forum for some sample.

Ah dammit forget to add the delete command in the uninstall section, thx MatMaul
here is a sample:http://www.mplayerhq.hu/MPlayer/samples/MPEG-VOB/LPCM/lpcm.vob

[LoRd_MuldeR]

here is a sample:http://www.mplayerhq.hu/MPlayer/samples/MPEG-VOB/LPCM/lpcm.vob

I can reproduce the problem with that sample. If I enaable LPCM support in ffdshow I get only noise. Otherwise it's fine...

[videomixer9]

hm I tried an example from here:
http://forum.doom9.org/showthread.php?p=603581#post603581

and it works fine. o_O that mplayer sample I only get noise too.

[MatMaul]

anyone have a pentium M dothan and can try to reproduce my bug ?

[asasadad_1]

hm I tried an example from here:
http://forum.doom9.org/showthread.php?p=603581#post603581

and it works fine. o_O that mplayer sample I only get noise too.
that mplayer sample is ok via ffdshow-rv2526 by you:)

[videomixer9]

lpcm doesn't work with issa build either here, though it either constantly peeps or you hear nothing, there were several changes since that revision on the file decoding lpcm. milan might have broke it unintentionally. From looking a bit around ... could it be 64bit CPU users don't have this problem? are you guys using 64bit CPUs, I'm not. Just an idea I got from the code changes ...

Other than that, it first appears as a problem in rev2544 for me ... I didn't do 42 and 43 builds. Inbetween there was some quicktime audio stuff added that changed some audio decoding stuff, I guess I'll try and revert those files affect in that patch and try if the error still occurs.

May also be related to floating point code generation ...

edit: the sample is 24bit lpcm? older builds says so, newest one only detects standard lpcm and may thus decode it as 16bit resulting in the noise. I hardcoded the 24bit part and then it works ... but that's not a real solution hehe.

[sleepking]

Same problem (crash of MPC after seeking) with rev 2544.
I stay on rev 2539.



The same problem,and ffdshow audio decoder-->codercs-->uncomperessed "all supported" don't work on my computer.

[issa]

Rev. 2546 build,

Compiler: ICC 9.0.30 (msvcrt 8 included) + GCC 4.1.1
CPU Extension: SSE or SSE2
Download: RapidShare (http://rapidshare.de/files/19739749/ffdshow-2546-icc-sse-20060506.exe.html) or MegaUpload (http://www.megaupload.com/?d=FZCJ1D0O)

Temporary fix LPCM problem by assume 16bit PCM when fail to detect the stream type.

[max-holz]

Rev. 2546 build,

Compiler: ICC 9.0.30 (msvcrt 8 included) + GCC 4.1.1
CPU Extension: SSE or SSE2
Download: RapidShare (http://rapidshare.de/files/19739749/ffdshow-2546-icc-sse-20060506.exe.html) or MegaUpload (http://www.megaupload.com/?d=FZCJ1D0O)

Temporary fix LPCM problem by assume 16bit PCM when fail to detect the stream type.

Where is link for SSE2 build?

[videomixer9]

the binaries automatically select the available SIMDs and use the one according to your CPUs capabilities. Code is generated for both SSE and SSE2 with ICL and embedded into the same binary. So the link to SSE and SSE2 version is the same ...

[MatMaul]

@ videomixer9 :I see on your website :
applied patches : runtime cpu detection, advanced gcc optimizations makefile patch
what are these patch ?
Can you publish them please ?
I actually try to make my own build and I'm interested by these patchs.
thanks !

[Adub]

Videomixer, are you going to make some builds with sse2 optimizations? That would be great man!

[videomixer9]

just use the build from issa, I won't do SSE2 builds.

[issa]

Rev 2546 Build version 2,

Compiler: ICC 9.0.30 (msvcrt 8 included) + GCC 4.1.1
CPU Extension: SSE or SSE2
Download: RapidShare (http://rapidshare.de/files/19816562/ffdshow-2546-icc-sse-20060507.exe.html) or MegaUpload (http://www.megaupload.com/?d=M8CGQY9T)

Temporary try to fix LPCM problem by using the old detecting code when new code fail to detect the stream. I think milan still working on the new detecting code.

[haruhiko_yamagata]

I apologize for lacking consideration for ffmpeg project/libavcodec.
Please forget the prior version of document and words about Superman trailer.
I am really sorry for it.

To 2546 updated document.
http://sourceforge.net/tracker/index.php?func=detail&aid=1472926&group_id=53761&atid=471491

[MatMaul]

@ videomixer9 :I see on your website :
applied patches : runtime cpu detection, advanced gcc optimizations makefile patch
what are these patch ?
Can you publish them please ?
I actually try to make my own build and I'm interested by these patchs.
thanks !

Up, thanks !

[_xxl]

ffdshow-20060508-2546.exe Intel 9.0.30
http://rapidshare.de/files/19931630/ffdshow-20060508-2546.exe.html

[Liisachan]

ffdshow-rev2546-SSE2.exe (http://ffdshow.faireal.net/mirror/ffdshow/ffdshow-rev2546-SSE2.exe)

3826010 Bytes
2006-05-08 13:14:20 UTC

[celtic_druid]

Not 100% rev2546. I added the multichannel vorbis patch from ffmpeg so 5.1 vorbis should sound fine now when decoded by libavcodec. Tested ok here.

[Liisachan]

Wow, I'll test it now.

EDIT
YES! That problem has been fixed in this new build.
In case someone needs a sample to test: xvid+6ch_vorbis.ogm (http://ffdshow.faireal.net/tmp/xvid+6ch_vorbis.ogm)

I was wondering why you made yet another one when there are already many builds lately, but now I know the reason. You're aggressively creative, which i'm really appreciating.

[videomixer9]

Hm, I also applied the patch and updated my latest build to use it. I usually don't check ffmpeg mailing lists so I missed it ...

[asasadad_1]

some lpcm decoding seems didn't work in ffdshow-rev2546-SSE2(by celtic_druid ).
here (http://www.mplayerhq.hu/MPlayer/samples/MPEG-VOB/LPCM/Fever.vob) is a sample.
ffdshow-svn2523-20060420(by clsid) is ok.

[issa]

some lpcm decoding seems didn't work in ffdshow-rev2546-SSE2(by celtic_druid ).
here (http://www.mplayerhq.hu/MPlayer/samples/MPEG-VOB/LPCM/Fever.vob) is a sample.
ffdshow-svn2523-20060420(by clsid) is ok.

I think the lpcm detect code in svn had not finished yet, if you need lpcm decoding, try my build on http://forum.doom9.org/showthread.php?p=824440#post824440.

[_xxl]

http://rapidshare.de/files/20008336/ffdshow-20060509-2546.exe.html
http://rapidshare.de/files/20034853/gcc.jpg.html
http://rapidshare.de/files/20034897/icl.jpg.html

[Lemonzest]

using videomixer9's builds, latest one and wma2 has stopped working, it is recognised but just muted/no sound have to disable libavcodec in the options and let windows decode it. i made a new test file and its the same.

[videomixer9]

works fine for me with a sample I got off mplayer ftp as I don't have such trash on my hd anymore :p there are some aggresive gcc optimizations enabled that may break stuff randomly, I usually test them only for aac/ac3/dts sound though as those are the usual things that break with them.

[Lemonzest]

i just did a quick encode with wma2 audio, and have to disable "WMA 8/9" Audio in the codec part of the audio decoder, for it to play, windows explorer says the audio is "Windows Media Audio 2" and in the info panel of the audio decoder it says "libavcodec wmav2" just no sound :(

http://lemonzest.tastyspleen.net/Temp.avi

[Kostarum Rex Persia]

Can someone tell me why build 2546 doesn't support fourCC IV50( Indeo 5.11 codec).

I can't play Indeo 5.11 files at all, can someone make a patch for decoding this stuff?

[foxyshadis]

I've never seen IV50 or IV40/41 in ffdshow, you've always had to go to ligos's website to get the codecs. As the decoders aren't present in lavc at all, it would take a lot more than a quick patch to get it in.

http://www.free-codecs.com/download/Intel_Codec_Installer.htm

[Lemonzest]

videomixer9, just got your new build and no idea what you did but wma v2 now plays fine :D Thanks.

[3dsnar]

I have posted this request through sourceforge,
but I would like to know what other users think about such approach
(I think this is the least confusing and possibly the most stable apprach)

-------------------------------------------------------------
Currently Aud-X MP3 5.1 is available as one of the mp3 decoding libraries.

My request is:

1) Please, add new format: Aud-X MP3 5.1 to the format list.
The user will be able to chose audxlib or disable the format.

---- If Aud-X MP3 5.1 is selected:
audxlib.dll would be used ONLY for Aud-X streams.
For regular MP3 streams other mp3 decoder would be used (selected with the MP3 format drop list)

---- if not selected, audxlib.dll would not be used.

2) Please add description to "info & debug"
about the mp3 5.1 stream (if it is such).

-------------------------------------------------------------

Please share your thoughts within this forum,
but also here:
http://sourceforge.net/tracker/index...61&atid=471492

[videomixer9]

ffdshow rev. 2546 ICL9.1

minimum requirement: SSE
compilers: GCC 4.1.1 + ICL 9.1.022
specials: high accuracy tremor, high accuracy libmad, runtime cpu detection, libavcodec 6ch vorbis fix
download: here (http://bittekeinspam.googlepages.com/ffdshow-20060511-rev2546-icl91.exe)

I thought it was time to try and get things on ICL maybe, test it and find out if it works better for you. It uses autovectorization and auto-parallelization (automatic optimization for dualcore). Raw decoding performance shouldn't be very much better as libavcodec is still gcc compiled, but some filters built into ffdshow.ax should work better. Compile time for this is very long so I don't know if I do any built with ICL 9.1 from now on. I prolly release another built with some other settings later.

[celtic_druid]

Parts of libavcodec can be built with ICL9 to. Yeah, compiling ffdshow with ICL does take ages.

[videomixer9]

libavcodec ICL 9.1 fails on missing links for some things that I'm trying to figure out to get it compiled with it, interesting is the auto-parallelization part. Seems to have something to do with dsputil ...

[LoRd_MuldeR]

Are ICL builds okay for AMD CPU's too or is it recommended to use GCC builds on my AthlonXP ???
Will do some testing myself too...

[videomixer9]

From my expierence upto now ICL isn't really usefull for main ffdshow part. I had better results making up performance by using more optimization for libavcodec on gcc. Also mostly totally stupid stuff is autovectorized or parallelized, e.g. dolby decoder or mixer.

Would be nice to have timecodec results, for me the icl build is not a bit better for just decoding, must be cause all needed operations are optimized anyways, and the rest of the code not being that bad. Considering ICL needs almost a half hour and MSVC 2005 only like 2-3 mins ... not even talking about stuff like kerneldeint which throws out of memory after a while here ...

And yes better not use it on AMD the libs aren't patched and I think it still is nazi against AMD, but rather than applying a patch I think I'd stick to MSVC compiler as it performs well enough that way. If it'd patch any libs I'd also only patch them now to check for AuthenticAMD so Intel people may cry, got sick of that company.

[LoRd_MuldeR]

I've just noticed that there's a big difference between RGB and YUV output. With YUV output (ffdshow default setting) it looks "mealy". If I uncheck YUV modes and force RGB output, the colors look *much* better. So it's a great improvement for image quality. Any explanation why it's that way?

Sample: YUV (http://mulder.dummwiedeutsch.de/etc/mpc_yuv.png) <-> RGB (http://mulder.dummwiedeutsch.de/etc/mpc_rgb.png)