Log in

View Full Version : New ffdshow build (?)


Pages : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 [31] 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72

Egh
6th June 2006, 03:25
I would like to watch some videos at a faster speed.

MPC can change video (and audio) playback rate. So I don't think changing video/audio rate is needed inside ffdshow.

Liisachan
6th June 2006, 11:45
videomixer9's ffdshow-20060604-rev2546.exe:

If "Queue output samples" is unchecked, you'll get a static picture (video frame) when you pause the video (which is normal).

If "Queue output samples" is ticked, you'll get nothing (blackness) when the video is paused, which is not too nice.

Anyone can reproduce that?
I'm on Windows 2000, MPC rev611 (aka the newest svn), VMR9 Renderless, RGB32.

videomixer9
6th June 2006, 12:55
Well for me it doesn't reproduce, maybe only happens on Win2k or using other specific stuff. Is it reproducable using Overlay or VMR7 modes? dunno what's with the kids and VMR9 obsession anyways especially with overlay resize quality being at the level of lanczos4 in many modern cards :devil:

Btw. Gabest's splitter seem to have problems with this patch unlike Haali Media Splitter, so it may be even recommended to use Haali Splitter, also for avi. Gabest's AVI Splitter e.g. just stalls frame displaying. when seeking. The matroska splitter only just falls out of audio sync when seeking sometimes but is fine, same for ogm.

Gabest avi splitter with RGB32 and VMR9 renderless produces a small flickering here. Might be it reproduces the picture that time, and maybe that fails for you and the screen goes black. Haali as AVI Splitter and this doesn't happen.

Liisachan
6th June 2006, 13:48
Not reproducable with VMR9 Windowed nor Overlay. It's apparently related with the MPC's internal sub renderer. Don't ask me about VMR7 as it's not officially supported on Win2k. I don't use VMR9 Renderless for resizing quality, but for softsubs. The reason why I use it is unrelated.

videomixer9
6th June 2006, 14:00
I actually tried with a subbed file too, so you do use the internal filters, all of them produce the flicker for me? Though it might indeed be related to the sub renderer.

Zarxrax
6th June 2006, 16:24
MPC can change video (and audio) playback rate. So I don't think changing video/audio rate is needed inside ffdshow.

If you don't mind my asking, where would I change this in MPC? I have never seen any such options.

videomixer9
6th June 2006, 16:47
Menu->Play->Increase Rate or Decrease Rate or just Ctrl+UP or Ctrl+Down

twist3d
6th June 2006, 16:51
videomixer9, your latest ffdshow build loads over gig-sized (1024mb) .mkv files for 30-60 seconds before they start to render (at least for me ;)) I first thought that this was a problem with latest mpc or haali's media splitter until I installed latest ffdshow from x264.nl (which starts rendering almost instantly).

videomixer9
6th June 2006, 17:14
Queue output samples on or off? It loads instantly here, however it also use automatic array prefetching code, I don't see a reason for this delaying thing this long though. x264 build doesn't have the same patches, especially not the same patchlevel on haruhikos patches ...

Are you using any filter like Reclock, or do you use other resampling? What kind of CPU and OS and output renderer?

Zarxrax
6th June 2006, 20:22
Menu->Play->Increase Rate or Decrease Rate or just Ctrl+UP or Ctrl+Down

Oh, that. Well, it only seems to give you two different speeds... one of which is about twice as fast, and the other being about half as fast. Both of these modes seem to work poorly and are pretty much unwatchable for me. So it would be much better to have a fully controllable speed from ffdshow.

twist3d
6th June 2006, 21:42
Queue output samples on or off?
no effect on long loading times with this option on or off
Are you using any filter like Reclock, or do you use other resampling? What kind of CPU and OS and output renderer?
no other filters (except if you count haali's one), no resampling or postprocessing of any kind enabled. amd64 3200+, nvidia 6600gt agp, winxp pro sp2 and i've enabled "use overlay mixer" on ffdshow (if that is what you mean with output renderer :D)

videomixer9
6th June 2006, 21:48
Output renderer as does your player use VMR7, Overlay Mixer or VMR9, the checkbox in ffdshow doesn't matter much really. It might though give errors if you're not use overlay mixer if the checkbox is checked and not in intermediate state.

I don't see why it would have this load delay. No clue, so as long as there is noone else with this problem it's again the I cannot reproduce it so I cannot fix it problem. Also this really does happen with any player?

Liisachan
7th June 2006, 00:53
fyi, I have some huge (>10GB) AVI files and they load instantly with or without that Queue thingy.

videomixer9
7th June 2006, 01:14
I wouldn't see any reason why the filesize should matter anyways, it just reads in a certain amount anyways at once and not everything. So also the queueing shouldn't matter really, it works on a small part of the file. So for every file it's basically the same, even if it's 1000 GB it shouldn't matter :p But wasn't 2 GB a limit for avi files or was that just due to old splitters and old shitty filesystem?

I bet on a random other error.

Liisachan
7th June 2006, 03:49
But wasn't 2 GB a limit for avi files...? Yes, it was, and it isn't (OpenDML). Incidentally, there's also the 4 GB limit on FAT32 (hence on Win98).

TheShadowRunner
7th June 2006, 04:29
hey all, sorry to interfere in your tech talks, but what do you consider to be the latest stable build?
See you,

TSR

haruhiko_yamagata
7th June 2006, 12:24
videomixer9's ffdshow-20060604-rev2546.exe:

If "Queue output samples" is unchecked, you'll get a static picture (video frame) when you pause the video (which is normal).

If "Queue output samples" is ticked, you'll get nothing (blackness) when the video is paused, which is not too nice.

Anyone can reproduce that?
I'm on Windows 2000, MPC rev611 (aka the newest svn), VMR9 Renderless, RGB32.
Hello.
I can't reproduce so far. I'm trying MPC/VMR9 renderless, RGB32, files with subtitle. If it is convenient to you, and if the files are not too large, please send me the files.

Liisachan
7th June 2006, 12:51
I can reproduce the probelm with any random file.
GRAY.avi (http://ffdshow.faireal.net/tmp/GRAY.avi) 75KB
Queue disabled (http://ffdshow.faireal.net/tmp/gray_no_q.jpg) (you can puase normally)
Queue enabled (http://ffdshow.faireal.net/tmp/gray_q.jpg) (you can NOT puase normally)

Another weird thing. libavcodec in ffdshow is weird with the above sample. If libavcodec is used, MPC reports it as like 10 ~ 16 fps. If xvid in ffdshow is used, or ffdshow is disabled and standalone XviD is used, MPC reports the correct fps, 23.976.
That's not the recent pb, as I can repro it with milan's 20051115. Usually libavcodec works ok with me, but not for the above thing. Can anyone confirm this or is it just me?

videomixer9
7th June 2006, 13:01
Might be a way to save CPU power and memory bandwidth, obviously there are no changes in the picture.

haruhiko_yamagata
7th June 2006, 13:31
I can reproduce the probelm with any random file.
GRAY.avi (http://ffdshow.faireal.net/tmp/GRAY.avi) 75KB
Queue disabled (http://ffdshow.faireal.net/tmp/gray_no_q.jpg) (you can puase normally)
Queue enabled (http://ffdshow.faireal.net/tmp/gray_q.jpg) (you can NOT puase normally)

Another weird thing. libavcodec in ffdshow is weird with the above sample. If libavcodec is used, MPC reports it as like 10 ~ 16 fps. If xvid in ffdshow is used, or ffdshow is disabled and standalone XviD is used, MPC reports the correct fps, 23.976.
That's not the recent pb, as I can repro it with milan's 20051115. Usually libavcodec works ok with me, but not for the above thing. Can anyone confirm this or is it just me?
Thank you. I'll test furthrer.
As for Gray.avi, I have frame rate reported 16 fps too.

wyrd
7th June 2006, 13:52
@Liisachan
I can reproduce the probelm.
but not with or without that Queue output sumple in ffdshow.

MPC VMR9(VMR MIxer mode on) + Queue output check on = OK
MPC VMR9(VMR MIxer mode off) + Queue output check on = NG(blackness)
MPC VMR9(VMR MIxer mode on) + Queue output check off = OK
MPC VMR9(VMR MIxer mode off) + Queue output check off = OK

ok.png (http://tirnanog.fate.jp/tmp/snap/ok.png)
ng.png (http://tirnanog.fate.jp/tmp/snap/ng.png)

my pc env: Pen4(singlecore),RADEON,XPsp2,latest DX9,
at celtic_druid's MPC611,videomixer9's ffdshow

(I don't know framerate probrem.. sorry,)

videomixer9
7th June 2006, 14:09
Yeah, this way the problem reproduces here too.

Liisachan
7th June 2006, 14:58
ok, we narrowed down the problem: in VMR 9 Renderless:
Queue ticked + Mixer mode unchecked + RGB out = problematic.

- If Queue is not ticked, there's no pb.
- If Mixer mode is ticked, there's no pb.
- If ffdshow's output color space is not rgb, there's no pb either.

hopefully haruhiko can find what is wrong now :)

haruhiko_yamagata
7th June 2006, 14:58
Thank you, wyrd. I still can't reproduce the problem, but it would be a nice hint.

haruhiko_yamagata
8th June 2006, 11:37
Thank you. I confirmed the problem. Perhaps someting around GetBuffer.

Livesms
8th June 2006, 12:02
Can anybody tell which ffdshow build has correct postprocessing by default with H264.

Problem is described here
http://forum.doom9.org/showthread.php?p=819640#post819640

and my quest:

http://forum.doom9.org/showthread.php?p=837917#post837917

Px
8th June 2006, 13:38
Oh, that. Well, it only seems to give you two different speeds... one of which is about twice as fast, and the other being about half as fast. Both of these modes seem to work poorly and are pretty much unwatchable for me. So it would be much better to have a fully controllable speed from ffdshow.
Use BSPlayer, it can control speed with 10% step

DeathTheSheep
8th June 2006, 19:16
Due to the incredibly nice picture of Haruhi present within it, I hereby declare my pride in using videomixer9's ffdshow build. Upon seeing this scintillating picture, I'm sure you'll all come around to my way of thinking :D

Zarxrax
8th June 2006, 21:06
Use BSPlayer, it can control speed with 10% step
Thanks, that was exactly what I needed.

Liisachan
11th June 2006, 13:26
@DeathTheSheep
I guess watching that too much is the reason videomixer9 gets suddenly arrogant from time to time :D
I just realized the author of the patch was Haruhiko. lol Is this just coincidence??

SeeMoreDigital
12th June 2006, 14:00
Does anybody know whether anybody is working on FFDshow's DVD decoding/parsing: -

http://img60.imageshack.us/img60/3534/ffdshow2vz.png


Cheers

videomixer9
12th June 2006, 17:04
works perfectly for me together with mpc, I'm using libmpeg2 though and my build should automatically select libmpeg2 too when you check mpeg1/2 in the installer. Also works in WMP11 fine, only problem is if the resolution is changing e.g. if menu and video differ in that, it sometimes hangs up then, seems to be more a renderer prob than ffdshow though.

btw. if the lame greek guy who stole my website for his codec pack builds reads this please remove the google analytics code, seeing stats of your page in my overview is annoying :p I have nothing against others using my beautiful xhtml code, but I have sth. against keeping the analyticscode :p

SeeMoreDigital
12th June 2006, 17:51
When playing DVD's with both MediaPlayer Classic and WMP11 using "only" FFDshow's filters/parsers I see a pumping/surging effect of the video stream...

EDIT: Strike that... libmpeg2 does indeed appear to work okay!


Cheers

videomixer9
12th June 2006, 17:54
well it doesn't use any deinterlacing and other processing many dvd decoders apply etc. by default.

ckjnigel
12th June 2006, 21:33
At least that's how it was on my father's PC with shared video RAM.
There had been vertical green bar artifacts in some XviD files using VideoMixer9's latest; the DirectX update fixed that.
My 83-year-old father and the 82-year-old girlfriend both especially appreciate the dialogue enhancing "crystality" filter.
Thanks to all involved!

MatMaul
13th June 2006, 17:38
I have made a very little patch to modify the behaviour when you use a ratio in the resize properties page : before the patch, ffdshow always apply this formule (aspect ratio = a1/a2) : y=a2*x/a1;x=x
it's good when you use a video with video ar>screen ar (an example (screen 16/10) : 576*240 => 576*360) because it's an upsize
but whith a video with video ar<screen ar (4/3 video) ffdshow does that : 720*576 => 720*450, I think it's not good because it's a downsize => lose of video information
I thing the good resize is : 720*576 => 924*576 (upsize)

My patch modifies ffdshow to do an upsize in all the situtations.

http://sourceforge.net/tracker/index.php?func=detail&aid=1505488&group_id=53761&atid=471491

SeeMoreDigital
13th June 2006, 19:00
Hi Matt,

I've performed extensive tests with FFDshow and as far as I'm able to determine the aspect ratio output performs exactly as it should, with MPEG-1, MPEG-2 and MPEG-4 sources containing AR signalling...

Unless you are referring to something else!

MatMaul
13th June 2006, 19:15
sorry if I am not clear...
no problem with aspect ratio embedded in mkv for example, this patch is only for people (like me) who use the option "specify aspect ratio" in "resize & aspect" (I use it to resize 4/3 source to 16/10).
the "problem" is, if video ar<screen ar (specified in "specify aspect ratio"), ffdshow downsize the image : I think it's better to upsize (=> the behaviour is good when video ar>screen ar)

haruhiko_yamagata
13th June 2006, 22:52
Hi, MatMaul.
It works for me.
Plainly it is better to upsize than to downsize.

MatMaul
13th June 2006, 23:18
cool !

NULUSIOS
13th June 2006, 23:35
Well I would say that it is better to DOWNsize than UPsize.

You see when you downsize you create less data from a larger data sample, where in upsize you (algorithmicaly) "guess" extra data from a smaller sample. I hope you get my point.

Sharktooth
14th June 2006, 00:11
i would say it's better to NOT resize at all...

Farhad
14th June 2006, 00:31
Hi,
I am having lot of compilation error issues with ffdshow
source downloaded from SourceForge.net while using
Microsoft VC6++/PlatformSDK while trying to generate
xvid.ax.
Is there an easy way around this problem? Can I use GCC for this--
if so then which version of GCC and where can I get
corresponding make file?

By the way, I was able to use GCC for building libavcodec.dll and libpostproc.dll.

Pointer to ffdshow source code buildable by GCC would also
be appreciated.

Thanks,

Farhad


Yeah ffdshow gcc 4.0.2, ffdshow.ax msvc 7.1, ff_libfaad2.dll msvc 8.0 is what i used (ffdshow-20051102.exe), who knows what msvc 8.0 does with it, thats why i put it there :)

Anyways, only one file left to fix for both gcc and msvc 8.0, ffdshow.ax

Edit:
ffdshow.ax with gcc 4.0.2 was "fixed" by using "make SSE2=no;"
ffdshow.ax with msvc 8.0 almost works, it compiles, only a registering bug left.

haruhiko_yamagata
14th June 2006, 11:01
Visual Studio 6 support was dropped (http://ffdshow.sourceforge.net/tikiwiki/tiki-view_articles.php).

To make ffdshow by GCC, first read this (http://ffdshow.sourceforge.net/tikiwiki/tiki-index.php?page=From+sources), though it is bit old.

Get GCC from
http://gda.utp.edu.co/~ceniza/GCC-4.0.3/
http://gda.utp.edu.co/~ceniza/GCC-4.1.0/
http://gda.utp.edu.co/~ceniza/GCC-4.1.1/
I recomend GCC-4.0.3 for those firstly try to compile ffdshow, others may recomend newer version though.
I don't know if this (http://sourceforge.net/tracker/index.php?func=detail&aid=1469519&group_id=53761&atid=471489) is fixed.

Get nasm. (http://sourceforge.net/project/showfiles.php?group_id=6208)

Get NSIS. (http://nsis.sourceforge.net/Main_Page)

DirectX SDK and Platform SDK is also needed.

Set environment variables properly.
For example, my environment variables:
CC=gcc
CPLUS_INCLUDE_PATH='C:\mingw\include;(DX9SDK)\Include;(PlatformSDK)\Include'
C_INCLUDE_PATH='C:\mingw\include;(DX9SDK)\Include;(PlatformSDK)\Include'
LIBRARY_PATH='C:\MingW\lib;(DX9SDK)\Lib\x86'

MSYS is recomended. You can complile without MSYS, but "make clean" is not easy without it.

Btw on my environments, MSVC2005 always crushes when it terminates after compiling ffdshow. I have tested two PC, both of them are the same. Does anybody experiencing the crush? Is it related to Japanese version?

draggoon01
15th June 2006, 11:19
when i play the hd trailers from apple (rename to avi), i get data excection prevention errors. any way to fix this?

videomixer9
15th June 2006, 11:30
so which idiot got you to rename them to avi, which is 100% incompatible with mov whatsoever?

mp4 is based on mov container roughly, so it's basically compatible, so renaming to mp4 usually works without problems, but avi isn't at all compatible.

There is no need to rename them at all, in Zoomplayer and MPC you just need to set quicktime playback to DirectShow instead of Quicktime.

Liisachan
17th June 2006, 20:26
A trivial problem.

Rounding of the frame timing to REFERENCE_TIME is not always strict, in OST | Frame timestamps.
Examples in 24000/1001 fps.
Frame 1 (0-based) is reported as 417083-831467, 0.083 416 667 rounded properly.
Frame 29 is reported as 12095416-12512499, off by 1rt, should be 1.251 250 000 exactly
etc etc
666.... sometimes gets 667, sometimes 666, I don't like this kind of looseness.

haruhiko_yamagata
18th June 2006, 02:23
A trivial problem.

Rounding of the frame timing to REFERENCE_TIME is not always strict, in OST | Frame timestamps.
Examples in 24000/1001 fps.
Frame 1 (0-based) is reported as 417083-831467, 0.083 416 667 rounded properly.
Frame 29 is reported as 12095416-12512499, off by 1rt, should be 1.251 250 000 exactly
etc etc
666.... sometimes gets 667, sometimes 666, I don't like this kind of looseness.

Hello. I don't fully understand what the numbers mean, but it seems to be too advanced by 1.2ms at frame 29.
Setting time stamps is usually parser's responsibility, ffdshow may change it though.
Which parser filter do you use?
Does it differ when you use different parser?

Liisachan
18th June 2006, 07:07
uh, no, it's not a bug report. at least i didn't mean that.
I see this 'problem' everywhere and I'm sure this is nothing new for an excellent coder like you, but some coders naively think that in CFR the current time should be duration-per-frame * frame_number, which is correct only mathematically.

a typical bad example

#define SEC_TO_RT(x) (x)*((float)(1000*1000*10))
char buf[ 1000 ];
int iFrameNumber = 30;
float one_frame = 1001.0f / 24000.0f; // seconds
float time_stamp = one_frame * iFrameNumber; // seconds
sprintf( buf,
"(Real) one_frame = 417083.333333333333\r\n"
"(Float) one_frame = %.12f\r\n"
"(Real) time_stamp = 12512500\r\n"
"(Float) time_stamp = %.12f\r\n",
SEC_TO_RT(one_frame), SEC_TO_RT(time_stamp)
);
MessageBoxA( NULL, buf, "Unit: 100ns", MB_OK );

/* Result:

Unit: 100ns
(Real) one_frame = 417083.333333333333
(Float) one_frame = 417083.315551280980
(Real) time_stamp = 12512500
(Float) time_stamp = 12512499.466538429000
*/

Farhad
19th June 2006, 18:28
Thanks Haruhiko for your advice. I found out that the recent
revisions of ffdshow downloaded from SourceForge.net builds
well with VS2005. I have done that and got ffdshow.ax and
a bunch of dlls (e.g., libavcodec.dll. mplayer.dll, etc.). For
installaling the plugin and the codec dlls into my Windows XP,
I registered ffdshow.ax by using command:
regsvr32.exe ffdshow.ax and then placed ffdhow/bin into
windows path to make all the dlls visible to Windows.
Then I tried to playback a Xvid MPEG-4 video
file in .avi format using Windows Media Player 10 (WMP10)
on my XP desktop PC. Unfortunately, only the audio plays
and video is just garbage on WMP10 display/screen!
Just before the audio playback, an error message appears
on WMP10 toolbar saying:
"Connecting....Error downloading codec".
Apparently WMP10 is unaware of the presence of
libavcodecs.....? I further noticed that WMP10 behaves the
same whether I register or deregister ffdshow.ax into XP
registry. This makes me guess that I am perhaps making some
error in the installation procedure of ffdshow.ax and the dlls
generated from the ffdhow build.
Or Could this be a problem of WMP10 on XP itself??
I am confused and would appreciate any help / feedback in
this regard.
Thanks,

Farhad


Visual Studio 6 support was dropped (http://ffdshow.sourceforge.net/tikiwiki/tiki-view_articles.php).

To make ffdshow by GCC, first read this (http://ffdshow.sourceforge.net/tikiwiki/tiki-index.php?page=From+sources), though it is bit old.

Get GCC from
http://gda.utp.edu.co/~ceniza/GCC-4.0.3/
http://gda.utp.edu.co/~ceniza/GCC-4.1.0/
http://gda.utp.edu.co/~ceniza/GCC-4.1.1/
I recomend GCC-4.0.3 for those firstly try to compile ffdshow, others may recomend newer version though.
I don't know if this (http://sourceforge.net/tracker/index.php?func=detail&aid=1469519&group_id=53761&atid=471489) is fixed.

Get nasm. (http://sourceforge.net/project/showfiles.php?group_id=6208)

Get NSIS. (http://nsis.sourceforge.net/Main_Page)

DirectX SDK and Platform SDK is also needed.

Set environment variables properly.
For example, my environment variables:
CC=gcc
CPLUS_INCLUDE_PATH='C:\mingw\include;(DX9SDK)\Include;(PlatformSDK)\Include'
C_INCLUDE_PATH='C:\mingw\include;(DX9SDK)\Include;(PlatformSDK)\Include'
LIBRARY_PATH='C:\MingW\lib;(DX9SDK)\Lib\x86'

MSYS is recomended. You can complile without MSYS, but "make clean" is not easy without it.

Btw on my environments, MSVC2005 always crushes when it terminates after compiling ffdshow. I have tested two PC, both of them are the same. Does anybody experiencing the crush? Is it related to Japanese version?