ffdshow tryouts project: Discussion & Development

[Jay Bee]

I see, learned something new. Anyway, even without multithreading libavcodec vc-1 is still very impressive and should be good enough for playback of HD files without dxva on most modern CPUs. The tests I did were with 2.7 GHz X2.

[Inventive Software]

Yep, but it's not complete until interlacing's added, and for that we gotta wait for ffmpeg to implement it. :)

[Jay Bee]

I'm not hoping for that anymore, even AVC decoders still have problems with interlacing and that's a format that has been around for ages now, since European HDTV started. Interlaced VC-1 is nonexistent in comparison.

[Kurosu]

VC-1optimized code has been around for 6 months but only got into the CVS around a month ago. From this post (http://article.gmane.org/gmane.comp.video.ffmpeg.devel/53133) you can see improvements. From those values, only one function accounting for 18% of the decoding time is left to optimize in the case of this sample (YMMV).

[krisq]

Hi guys, I've search the thread and didn't found anything about my problem so here goes. I've installed the latest build by clsid (rev1751 20070105). It was a jump from pretty old version for me, which worked flawlessly BTW :)
After the update, Adobe After Effects started to act funny on me, it can't initiliaze correctly. It gives me an error at startup ('Bad digit in number'). I've seen such errors before in other apps and it all goes to differences between "," and "." in regional settings of OS. I'm using polish Windows XP SP2 and here default separator for numbers is "," which apparently is causing problems. When I uninstall ffdshow or change to "." in regional settings, everything is back to normal, AFX starts without problem.
I hope you can help me with this because I don't want to revert to old version of ffdshow nor change separator "," to "." :(

BTW I'm 100% sure it's because of ffdshow, it's started happening the same day and after uninstall everything's back to normal.

[fastplayer]

Then blacklist the Adobe app:
DirectShow control --> Don't use ffdshow in: --> <enter EXE filename>

[haruhiko_yamagata]

@krisq
Could you pin-point the revision which the problem starts? Many old builds are available for download.
I don't have Adobe After Effects nor Polish version of Windows. I'm trying to recall the changes related to locale but nothing come to me.

[krisq]

Then blacklist the Adobe app:
DirectShow control --> Don't use ffdshow in: --> <enter EXE filename>

Yeah, that was the first thing i did :)
Unfortunately it didn't help, error is still there :(

@krisq
Could you pin-point the revision which the problem starts? Many old builds are available for download.
I don't have Adobe After Effects nor Polish version of Windows. I'm trying to recall the changes related to locale but nothing come to me.

I'll try.

[Delerue]

Can anyone confirm a bug with high frame rate videos? With this sample (http://www.zshare.net/download/6304318cfc0301/) I got three hard stutters with MPC (clsid january build) and FFDShow _xxl 1757, although the CPU usage has an average of 18%. The problem seems to be with the libavcodec version, because with CoreAVC or MPC-HC internal decoder there's no problem (except for the fact that MPC seems to limit the FPS to the actual refresh rate, even with vsync off). Also, VLC and SMPlayer can run just fine.

[krisq]

@krisq
Could you pin-point the revision which the problem starts? Many old builds are available for download.
I don't have Adobe After Effects nor Polish version of Windows. I'm trying to recall the changes related to locale but nothing come to me.

Ok, apparently something has changed between clsid's rev1431 and rev1487. 1431 works flawlessly, 1487 throws out an error in AFX.

[Yong]

i think the encoder part was bugged when testing with gcc-4.2 pre-releases(svn) quite a while ago, not sure if anything was done about it
so which ffdshow build have gcc3 or 4.0/4.1 compiled libavcodec? i would like to test it to see wether it is gcc or libavcodec bugged :p
i got gcc 3.4.6, 4.0.3 and 4.2.2,
but none of them work for me as they will throw an weird error message when using them :sly:

[fastplayer]

Can anyone confirm a bug with high frame rate videos?
I get no stutters at all with MPC+ffdshow. The file just plays back slow with <40% CPU load. :confused:
VLC on the other side uses 95-100% and plays the file faster than MPC with some micro-stutters. Probably because of the CPU load hitting 100% from time to time...
I'm using clsid's latest MPC and ffdshow versions.

[clsid]

Ok, apparently something has changed between clsid's rev1431 and rev1487. 1431 works flawlessly, 1487 throws out an error in AFX.
Could you also test 1437 (http://downloads.sourceforge.net/ffdshow-tryout/ffdshow_rev1437_20070823_clsid.exe?modtime=1187883251&big_mirror=0)?
Then I suspect that the bug was introduced in revision 1434. Since that revision there also is a weird compatibility problem with a game called "Silent Hunter 3". In that case blacklisting also doesn't help.

Does the ',' versus '.' problem apply to the decimal symbol or the the digit grouping symbol in the regional options? Or both?

[haruhiko_yamagata]

Can anyone confirm a bug with high frame rate videos? With this sample (http://www.zshare.net/download/6304318cfc0301/) I got three hard stutters with MPC (clsid january build) and FFDShow _xxl 1757, although the CPU usage has an average of 18%. The problem seems to be with the libavcodec version, because with CoreAVC or MPC-HC internal decoder there's no problem (except for the fact that MPC seems to limit the FPS to the actual refresh rate, even with vsync off). Also, VLC and SMPlayer can run just fine.It's smooth for me. Perhaps it's your video driver. Frame rate that is higher than refresh rate may not be supported.

[Thunderbolt8]

thanks for that black border brightness fix! that brightness was suddenly there although I was pretty sure this wasnt the case in former days and already wondered why this was the case.

[haruhiko_yamagata]

Could you also test 1437 (http://downloads.sourceforge.net/ffdshow-tryout/ffdshow_rev1437_20070823_clsid.exe?modtime=1187883251&big_mirror=0)?
Then I suspect that the bug was introduced in revision 1434. Since that revision there also is a weird compatibility problem with a game called "Silent Hunter 3". In that case blacklisting also doesn't help.

Does the ',' versus '.' problem apply to the decimal symbol or the the digit grouping symbol in the regional options? Or both?
It must be 1434. I have committed the fix.

@krisq
Please test and let us know when next Nightly build is uploaded.

[clsid]

A new build is online.

[haruhiko_yamagata]

A new build is online.
:thanks:

[krisq]

It must be 1434. I have committed the fix.

@krisq
Please test and let us know when next Nightly build is uploaded.

A new build is online.

Thank you so much, guys!!
rev1763 is working perfectly :):thanks:

[cc979]

so which ffdshow build have gcc3 or 4.0/4.1 compiled libavcodec? i would like to test it to see wether it is gcc or libavcodec bugged :p
i got gcc 3.4.6, 4.0.3 and 4.2.2,
but none of them work for me as they will throw an weird error message when using them :sly:

gcc-4.0.4 is what i use mostly, which encoder seems to cause problem ?

[cc979]

ffdshow-rev1763.7z
md5: E5358DEF6DFB6D775074C9D24C2759E9
http://www.speedyshare.com/600685999.html

gcc-4.0.4 and gcc-4.2.1

[LoRd_MuldeR]

so which ffdshow build have gcc3 or 4.0/4.1 compiled libavcodec? i would like to test it to see wether it is gcc or libavcodec bugged :p
i got gcc 3.4.6, 4.0.3 and 4.2.2,
but none of them work for me as they will throw an weird error message when using them :sly:

The "official" builds by clsid have gcc 4.2.2 compiled libavcodec.dll (according to the version info window)

[Delerue]

I get no stutters at all with MPC+ffdshow. The file just plays back slow with <40% CPU load. :confused:
VLC on the other side uses 95-100% and plays the file faster than MPC with some micro-stutters. Probably because of the CPU load hitting 100% from time to time...
I'm using clsid's latest MPC and ffdshow versions.

It's smooth for me. Perhaps it's your video driver. Frame rate that is higher than refresh rate may not be supported.

Hmmm... I finally discovered. The problem appears when you set the render to 'Overlay' or 'VMR7 Windowed' (worst on that). 'VMR9 Windowed' or any 'Renderless' doesn't have any stutter.

Haruhiko, the FPS limit is gone with 'Overlay' render. And all the other decoders play fine, without stutter. Do you really think it's a problem with Catalyst 7.11?

Thanks

[Shakey_Jake33]

Slightly off topic (and I have asked this in other topics too, but I figure some people in this topic may know).
Does ffdshow get it's AC3 Encode Mode (to send 5.1 AAC over SPDIF as AC3) code from ffmpeg? I've been assuming so, and have been trying to work out how to use it in ffmpeg!/mplayer

[LigH]

AC3 encoding is one part; S/P-DIF output is another. You seem to refer to the latter?

[haruhiko_yamagata]

Hmmm... I finally discovered. The problem appears when you set the render to 'Overlay' or 'VMR7 Windowed' (worst on that). 'VMR9 Windowed' or any 'Renderless' doesn't have any stutter.

Haruhiko, the FPS limit is gone with 'Overlay' render. And all the other decoders play fine, without stutter. Do you really think it's a problem with Catalyst 7.11?

ThanksSorry, it's VMR7 then. CoreAVC have the same behavior. It's not a bug of ffdshow.

[haruhiko_yamagata]

Does ffdshow get it's AC3 Encode Mode (to send 5.1 AAC over SPDIF as AC3) code from ffmpeg?Yes, it does.

[Yong]

ffdshow-rev1763.7z
md5: E5358DEF6DFB6D775074C9D24C2759E9
http://www.speedyshare.com/600685999.html

gcc-4.0.4 and gcc-4.2.1
hmm:cool:
gcc 4.0.4 crash immediately after encoding start, with mpeg2/4

gcc 4.2.1 show "assertion failed!" with mpeg1/2

i was try to trascode some mpeg4 file to mpeg1/2/4 with ffdshow for testing.

just tried clsid build, same error i got with 4.2.1, mpeg 4 will crash if disable qpel.

[clsid]

There also are some encoding bugs when compiling ffdshow.ax with Visual Studio, so it is not just GCC's fault. Of course, those could be different bugs than the ones you are having.

[Yong]

There also are some encoding bugs when compiling ffdshow.ax with Visual Studio, so it is not just GCC's fault. Of course, those could be different bugs than the ones you are having.
hmm but cc979 compiled ffdshow are pure gcc build.

i post the video encoding settings some other infos here incase u guys wan to fix it :p
system spec:
amd x2 6000+
windows vista
2gb ram.
virtualdub 1.7.7 build 28312/release.
ffdshow rev 1763 msvc 2008 x86 unicode build.

here are the encoding setting,
fast recompress mode,
Encoder: mpeg2, one pass, 2201kbps
max i frame interval: 300, min: 30
bframes enabled, max consecutive bframe: 3, dynamic, closed gop,
number of threads: 2

macroblock decision mode: rate distortion, always try no motion, RD optimization for CBP,
ME/subpixel/macroblock/prepass compare function: SAD with chroma.
EPZS/prepass diamon size: adaptive with 3,
ME pre-pass: always.

Quant type: mpeg with trellis, MPEG2 DC precision: 11bits

Luminance/Darkness/Border masking enabled with default value, normalize adaptive quant.

i dont use it for real encoding, just for testing,
will try to narrow down the problem :)

[cc979]

i've just tested my build of ffdshow-1763 (gcc-4.2.1) with windows xp, encoding with virtualdub using mpeg2 or mpeg4 is fine, maybe its a vista/virtualdub bug

[Delerue]

i've just tested my build of ffdshow-1763 (gcc-4.2.1) with windows xp, encoding with virtualdub using mpeg2 or mpeg4 is fine, maybe its a vista/virtualdub bug

I confirm; 1763 and XP too. The only problem (my problem?) is with x264 multi-thread encode. I can't use the other core. :(

[Yong]

hmm, the "assertion failed!" message is gone after change the encoding thread to 1 :p multithread lavc encoding will crash.
but x264 multithreaded encoding work for me lol.

[draggoon01]

i'm trying to use ffdshow for subtitles (srt files) but when i go full screen the text is blurry/jaggie. how can i fix this?

[Yong]

resize ur video to higher resolution, or use VMR7/9/EVR or haali video renderer.

[leeperry]

just wanted to point out that all issues with ffdshow & MPC HC + EVR on XP seem to be fixed.

I've watched several movies w/o any freezes using this package, with updated DLL's of EVR from Vista SP1 :
http://www.badongo.com/fr/file/7196279

Keep up the good work :D

[Atak_Snajpera]

Everything above ffdshow_rev1724_20071225_clsid does not decode correctly this ts


http://x264.nl/h.264.samples/force.php?file=./luxe.hd.ts

[wiak]

just a stupid question will there be a auto update future?, hehe now that ffdshow gets new builds nearly every day, i want to be up to date to keep my upcoming AVC+HE-AAC with LATM dvb-t setup stable :thanks:

[Delerue]

Everything above ffdshow_rev1724_20071225_clsid does not decode correctly this ts

http://x264.nl/h.264.samples/force.php?file=./luxe.hd.ts

No problem here. XP SP2 + build 1763 + MPC clsid january version + VMR7 Windowed. The only 'problem' is the high CPU demand, hehehe. ;)

[Bathrone]

Hi folks :) Searches havent found answers so can I please ask:

1. Does ffdshow hook into the DXVA acceleration on Vista with suitable GPUs?

2. When the tryouts FAQ says that xvid and h.264 encoding is multi-threaded, does it actually mean multithreaded decoding or can ffdshow somehow be used for encoding?

[Delerue]

1. Does ffdshow hook into the DXVA acceleration on Vista with suitable GPUs?

Try the MPC Home Cinema at www.codecs.com.

2. When the tryouts FAQ says that xvid and h.264 encoding is multi-threaded, does it actually mean multithreaded decoding or can ffdshow somehow be used for encoding?

FFDShow can encode and decode xVid/x264 in multi-thread. But the x264 decode has some limitations; can't work in multi-thread with recently x264 encodes.

[ACrowley]

Try the MPC Home Cinema at www.codecs.com.


FFDShow can encode and decode xVid/x264 in multi-thread. But the x264 decode has some limitations; can't work in multi-thread with recently x264 encodes.

With latest CABAC Multithreading Patch it works for x264 too , so far its CABAC encoded. Latest XXL Builds are already patched with it.
Its working perfect multithreaded on all my x264 Files

~60% CPU Load for 1080p x264 12Mbps and ~85% for BluRay H264

[Leak]

can ffdshow somehow be used for encoding?
Well, ffdshow comes with both an "ffdshow video encoder" DirectShow video compressor filter and an old-fashioned "ffdshow video codec" for VfW, so - yes? :)

np: Kid606 - I Miss You (Resilience)

[clsid]

With latest CABAC Multithreading Patch it works for x264 too , so far its CABAC encoded. Latest XXL Builds are already patched with it.
Its working perfect multithreaded on all my x264 Files

~60% CPU Load for 1080p x264 12Mbps and ~85% for BluRay H264
All recent builds have it, since the patch was committed to SVN.

[clsid]

Everything above ffdshow_rev1724_20071225_clsid does not decode correctly this ts

http://x264.nl/h.264.samples/force.php?file=./luxe.hd.ts
Does it work if you change the number of decoding threads in ffdshow options to 1?

[fastplayer]

just a stupid question will there be a auto update future?
I'd look at auto-update from another perspective:
Who's gonna provide and pay for the bandwidth?

[Inventive Software]

Sourceforge?

[clsid]

I would prefer to continue putting our efforts into improving existing ffdshow functionality, instead of adding non-essential stuff like auto-update.

[Delerue]

With latest CABAC Multithreading Patch it works for x264 too , so far its CABAC encoded. Latest XXL Builds are already patched with it.
Its working perfect multithreaded on all my x264 Files

~60% CPU Load for 1080p x264 12Mbps and ~85% for BluRay H264

Well, I told that because of this (http://forum.doom9.org/showthread.php?p=1085619#post1085619). What you think?

[leeperry]

I would prefer to continue putting our efforts into improving existing ffdshow functionality, instead of adding non-essential stuff like auto-update.

how about some AR buttons for easy access so I don't need to use the virtual keyboard on my PCHC to reach 1.85 ?

just 4 buttons : 1.33 / 1.78 / 1.85 / 2.35 :p