Log in

View Full Version : ffdshow development #2


Pages : 1 2 3 [4] 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

virus
8th October 2004, 22:30
Don't get me wrong, but I still have some doubts. I'm really just presenting my opinion, and I have no real proof that things are as I think they can be.

If there's something that I've learned in the last 7 years using gcc, is that gcc almost never fails when doing basic optimization of some C code. To be more precise: I've never seen -O2 produce b0rked code on my programs with -O1 solving the problem. And when -O2 failed, it was because of my errors... you know, sometimes bugs are not exposed until the code gets really and deeply optimized, instructions reordered, redundant expressions eliminated, and so on. Again, I might be wrong, but I'm still not 100% sure that the error actually comes from gcc itself.

just my 2 cents

netchris
9th October 2004, 01:06
Since virus you have the experience please try to make your own builds with -O and the other options, and prove me right or wrong.
I would like to give you my build, so you can test it but it seems that attachements here are not working, and I dont have any free web space to upload it.
I didn't change anything in the code, I took what was given by the ffdshow and x264 team and played with it. It was trial and error.
The point is that in the current state even if there is something wrong with the code, there is a workaround to achieve a fast build without the white blocks bug.

celtic_druid
9th October 2004, 07:45
8th build up, with -O

virus
9th October 2004, 11:25
@celtic_druid:
build 8 works OK, thank you. I've crosschecked build 7 and 8 to see what was wrong. It turned out that the "broken" build 7 encode uses higher QPs on some scenes (example: 21-22 instead of 20). The picture looks OK, just with a slight detail drop. Don't ask me why this happens.

@netchris:
uh! oh! don't be so upset :)
I do appreciate that you found a workaround. And I'm not trying to "prove you wrong"... I'm just outlining a possible source of the problem, maybe unlikely, but still possible. And no, whatever the error is, you're not the culprit ;)


EDIT: btw, Milan has already added the GUI options I requested yesterday... what a speedster :)

netchris
9th October 2004, 13:09
@celtic_druid
Has the 8th build been removed? I dont see it.

@virus
Sorry I sounded like this, I was a little drunk yesterday so I was maybe a little harsh,but i didnt feal offended or upset.



I have a lot of respect for you virus as well as for celtic_druid and akupenguin (and all the other members of the community).
:thanks: for your time, efforts and explenations!

celtic_druid
9th October 2004, 13:38
There never was a link for me to remove.

I didn't bother posting a direct link to 8... kinda figured that people could just swap an 8 with one of the other links.

Kurtnoise
12th October 2004, 17:38
ffdshow_20041012 is out on SF (http://sourceforge.net/projects/ffdshow) !!! Available in 3 versions (Normal, SSE and SSE2).

Changes:
MS ADPCM support
two preferred vobsubs languages
configurable swscaler gaussian blur strength for vobsubs
support bmp and gif as bitmap overlay image
option to use WAVEFORMATEX instead of WAVEFORMATEXTENSIBLE on output if no custom channelmapping
bug fix: don't always recompute letterboxing size and don't always check for subtitles file
more blending modes
IDR interval and deblock parameters for x264
working on image overlay
vobsubs scaling
proper alignment in ICL+GCC libavcodec build
key shortcut for subtitles language cycling
possibilty to override vobsubs position
dwstring substr and c_str related fix
recognize cz for Czech in idx, don't show duplicate languages in vobsub config
don't set merit to "not used" if it previously had different value than one of defined in ffdshow
yet another fix in corrections
another subtitles correction fix
fix in capital letters correction
more keys working
keyboard shortcuts and remote messages for cycling presets
orthography correction from subrip
separate page for text subtitles options, more complete subtitles correction from Subrip
adding text subtitles correction from subrip
support for 8-bit samples on input
fixed another crash in vobsub
fixed crash in getNext with pictFull==true and pictHalf==true
enabled trellis quantization for H.263(+)



Notes:
This build should fix Windows XP explorer crashes.

Three editions are available: special for SSE and SSE2 capable CPUs and generic for all others.
SSE a SSE2 build were compiled using Intel C++ Compiler 8.1, generic build was compiled with Microsoft VC6.

Known bug: in SSE and SSE2 builds Avisynth image processing filter in ffdshow isn't working.

Subtitles correction may not work correctly in all cases (but should not crash).
These feature was ported from SubRip and if SubRip will give different resulf for a given subtitles, please send them to me.

Inc
13th October 2004, 10:59
First I do apologize in andvance if some questions now of mine have been already treated in here BUT the browsing/searching through this forum actually is a pain via t-online provider in germany as it seems.

These following questions do refer to the last ffdshow version BEFORE this actual release of 2004/10/12! Im at work on a MAC and just found this new release of 20041012 in here, so I cant test it right now.

1. A Makeavis generated "AVIS" AVI wont be decoded in an fvw decoding environment like Vdub or CCE, even if it has benn set in the vfw configuration of ffdshow (decode Avisynth AVIS). In an Dshow decoding environment its no problem.

2. Also according to vfw decoding environments: If in ffdshows vfw decode setup a decoding of for instance RAW YV12 decoding has been setup, Vdub still uses my XVID decoder in the informations Box of Vdub when decoding an avsiynth .avs with an outgoing YV12 colorspace (same thing in case of other outgoing colorspaces as FFdshow/vfw wont be used in Vdubor CCE).
If for instance capturing an AVI by using "FFDS" FourCC in ffdshow via VirtualVCR .... VirtualDub does decode that one via FFdshow/vfw! BUT no filtering within ffdshow's vfw setup will be performed. I have to do a Graphedit .grf workout via Directshowsource() to get filtering via ffdshow enbaled when opening an avs in vdub.

3. When trying to use Avisynth filtering within! ffdshows filter section, it doesnt work anymore. An access violation error is reported by avisynth WITHIN ffdshows decoding when doing the preview of that via ffdshow decoded AVI file.




I do love ffdshow! But to me it seems beside some very nice additions in new releases, there are some risks that things of the past wont work anymore as for instance Forcc "AVIS" decoding via Vdub or CCE never was a Problem when using the old FFvfw from autum of 2003 for instance.

Thanks a lot in advance!!
Inc.

Leak
13th October 2004, 12:00
Originally posted by incredible
3. When trying to use Avisynth filtering within! ffdshows filter section, it doesnt work anymore. An access violation error is reported by avisynth WITHIN ffdshows decoding when doing the preview of that via ffdshow decoded AVI file.

This seems to be a problem caused by compiling ffdshow with Intel's C compiler; it should work in the latest ffdshow version without SSE/SSE2 optimizations - see here (http://ffdshow.sourceforge.net/tikiwiki/tiki-read_article.php?articleId=12) and here (http://sourceforge.net/tracker/index.php?func=detail&aid=1040852&group_id=53761&atid=471489)...

np: Vladislav Delay - Anima E (Nohuume)

pogo stick
13th October 2004, 15:27
Is it just me or new ffdshow is not appearing in VDub's encoders list? :confused:

netchris
13th October 2004, 15:35
I have the same problem with the latest build, so I after i istalled it, I copied the ffdshow's files to a temp folder, then uninstalled it.
I installed an older build that doesnt have this problem, and copied the files from the temp folder to the ffdshow folder, to replace the old ones with the new.

stephanV
13th October 2004, 15:42
doesnt show up here either

celtic_druid
13th October 2004, 16:24
Go into your registry and change the vidc.ffds entry to progra~1 instead of "program files". Should sort it.

stephanV
13th October 2004, 16:46
yep, thats it... its weird though

Yong
14th October 2004, 06:15
Can i post the ffdshow audio decoder bug report in here?

yaz
14th October 2004, 08:26
i got the same problem(s!) as incredible.

no way of using 'makeavi' anymore :-((( loading the fake avi into mencoder makes a crash (windows say msvcrt.dll is crashing???) is there anyone succeeding in this business? pls, drop a workaround, if u're aware of any.

avisynthing in ffdshow still seems to crash (or i do in a wrong way?)

thx
y

Leak
14th October 2004, 09:38
Originally posted by yaz
avisynthing in ffdshow still seems to crash (or i do in a wrong way?)

Works for me - what exactly are you doing?

Also, did you make sure to install the unoptimized version? The SSE/SSE2 versions will still crash.

np: Vladislav Delay - Huone (Nohuume)

yaz
14th October 2004, 10:26
Originally posted by Leak
... The SSE/SSE2 versions will still crash ... well. i was playing with the sse ver. i give a try to the non-opt ver. thx for the answ.
thx
y

Inc
14th October 2004, 10:40
Originally posted by yaz
i got the same problem(s!) as incredible.

no way of using 'makeavi' anymore :-((( loading the fake avi into mencoder makes a crash (windows say msvcrt.dll is crashing???) is there anyone succeeding in this business? pls, drop a workaround, if u're aware of any.

avisynthing in ffdshow still seems to crash (or i do in a wrong way?)

thx
y

a) For using avisynth scripting "within" ffdshow it seems that you have to install the generic build - I hadnt time to install the newest for testing till now.

b) Mencoder worked without problems in the past when using ffdshows! Makeavis (I use it via commandline to just go in using an avs into the batch file). I just copied the "old" ffvfw.dll & mplayerlib.dll into the same folder as mencoder.exe and added the description in the codecs.conf file thats it. But as I do remember now ... wasnt it you who gave me that advice? So did something change also related to "this" in the very recent release of ffdshow??? :eek:

yaz
14th October 2004, 11:38
Originally posted by incredible
... wasnt it you who gave me that advice? So did something change also related to "this" in the very recent release of ffdshow??? :eek: yep, it was me ... but yesterday evening it all drove me mad. whatever i tried all i got was a crash in msvcrt.dll. and no reading, of course :-( i'll give it another go this evening. maybe, a 'clean-the-house-before' will help.
i'm just annoyed, becuse the log says 'makeavis support is back'. but for me 'back' means 'away' :-(

the bests
y

Audionut
14th October 2004, 12:01
The latest SEE2 version works fine for me.

Just had to run "regsvr32 c:\**\ffdshow.ax" from the command line.

It appears that it does not register itself.:confused:

http://sourceforge.net/tracker/index.php?func=detail&aid=1046034&group_id=53761&atid=471489

Yong
15th October 2004, 05:26
Here's my ffdshow audio decoder settings,
Format Decoder
MP2 libmad
MP3 mp3lib
AAC libfaad2
Vorbis themor
MSADPCM Libavcodec
IMA ADPCM Libavcodec
FLAC can't work.

Allowed sample format for sound processing:
16 bit integer

Support output sample format:
32 bit float
Floating to integer conversion:
Dithering(Selected), Noise shaping: heavy

When playing AAC ,MP2 ,WMA, and Vorbis with only the 24 bit output sample format selected, the audio will become full of distrosion.
AAC only have problem with 24 bit out.
While MP3 only can work with 16/32 bit float output, when select 24 bit the sound will lowered and full of distrosion, select 32 bit integer out will crash.
Vorbis only can play with the 32bit integer /float ouput but have some distrosion, select 16 bit out will crash.
playing MP2 with 16 bit out will crash.
IMA / MS ADPCM crash with 32 bit integer out.
WMA is work fine but can't properly decode some WMV audio.
Only the 32 bit float is safe to use.

ffdshow video decoder bug:
Playing mpg with ffdshow now can work, but still can't play some video-stream-only mpg file.

Sorry for my english...:stupid:

Mug Funky
15th October 2004, 14:34
w00t! the vobsub functionality is really good.

i'm pushing my machine to it's limit, and it seemed vobsubbing through ffdshow and disabling Dvobsub was just enough to decode in real time.

how's the development of the loading of embedded subs? i'm experimenting with muxing the .idx into a .mkv file, but nothing except media player classic can find these subs, and it renders them too slow.

i'm playing back xvid @ 50fps, 512x384... just to see if i can. i was so sick of field-blends and what to do with them that i just up and smartbobbed the whole thing and encoded it without touching the blends.. . this presents the problem of decoding speed, of course, which is why i'd like ffdshow to do as much of the processing as it can.

i suppose for now i can just leave the .idx and .sub in the same directory, but i find whole files are tidier.

oh, also, yellow subs seem to come out grey...?

APF_Gandalf
16th October 2004, 22:48
Originally posted by Audionut
The latest SEE2 version works fine for me.

Just had to run "regsvr32 c:\**\ffdshow.ax" from the command line.

It appears that it does not register itself.:confused:

http://sourceforge.net/tracker/index.php?func=detail&aid=1046034&group_id=53761&atid=471489

it seems to me that it's the "old" folder name bug that is back. if you install it in a folder without spaces, it works: "c:\ffdshow" works, "c:\program files\ffdshow" fails even if I try to register it myself using "c:\program files\ffdshow\ffdshow.ax" or "c:\progra~1\ffdshow\ffdshow.ax"
I submitted a way to avoid this bug long ago when it first appeared. it worked, but it seems that it was forgotten in the latest official builds.

arman68
16th October 2004, 23:58
Originally posted by Mug Funky
w00t! the vobsub functionality is really good.

i'm pushing my machine to it's limit, and it seemed vobsubbing through ffdshow and disabling Dvobsub was just enough to decode in real time.

Yes, it is coming along very nicely. I did some tests on a P4 2.2Ghz, and VSFilter adds an extra 30% cpu usage, as opposed to VobSub in ffdshow. And that's using swgaussian blur, which is very nice. For the first time I can now use the post-processing options I wanted :D

Seems that the new keyboard shortcut for subtitle cycling are broken, and broke the toggle as well though.

kittychan
17th October 2004, 08:48
yooooo

I am quite happy ffshow can replace the vsfilter.dll and decrease the cpu load ^_- , but I can't play subtitle embedded in .ogm ot .mkv...
I did tick the "accecpt embedded sutitles" box, but no subs appears, so I still have to use vsfilter for .ogm (and .mkv)

Did I do something wrong?

++

kittychan

EDIT : OK my wrong....
I have to set the VRM7 for the video direct show in MPC option...
sorry for having ask ^^

++

2nd EDIT....
re-sorry, when set the VRM7 for the video direct show in MPC option, sub are played by MPC but still not by ffdshow...
I still can not make ffdshow play ogm sub.

++

Mug Funky
17th October 2004, 13:35
there's a little tooltip that says "very incomplete and experimental. Send me samples which don't work".

i'm guessing it'll only do textsub? not sure, but i'm certain it'll be completed soon.

in the meantime you'll have to live with cluttered directories full of .idx and .sub files :)

The Shemeta
17th October 2004, 14:51
hi,

i upgraded to the latest ffdshow-20041012 & now i don't have the vfw when i open a movie in VirtualDubMod.

anyone with the same problem?

the good thing is that if fixed the explorer.exe crash.

esby
17th October 2004, 16:36
@The Shemeta:

you could READ this thread a bit...

Now take this answer, go back 4-5 replies ago, and READ!

esby

therealjoeblow
18th October 2004, 03:03
The .idx file of a VobSub set has numerous user-definable settings that can be changed to suit, including custom colors:

custom colors: OFF, tridx: 1000, colors: D2D2D2, 000000, 000000, 000000

Would it be possible to add support for overriding this setting when ffdshow reads the file (colors would do fine for now, and maybe in the future, all of the special settings when you have time)?

If it's a big deal to add this functionality, then don't worry about it, but I just thought the ffdshow configuration interface could have a couple of more tabs where a user could enter their preferred color (and other) settings, and then ffdshow would use those if they were entered, instead of the ones contained in the .idx file.

Many thanks,

therealjoeblow
18th October 2004, 16:51
One more note on the vobsub colors:

There appears to be a small glitch in basic interpretation and display of colors. The following line shows correctly with directvobsub as white text outlined in black for one particular set of subs:

custom colors: ON, tridx: 1000, colors: 000000, 000000, 000000, D2D2D2

However, with ffdshow, it shows the text in all black with black outline, impossible to read. If I change the line as follows, then the subs appear correctly with white text outlined in black:

custom colors: ON, tridx: 1000, colors: D2D2D2, 000000, 000000, 000000

I'm not sure exactly which one is correct (directvobsub or ffdshow), but since vobsub was around much longer, I'm assuming it's interpretation would govern, having been burned to thousands of cd's around the world already.

LigH
18th October 2004, 17:33
If that was about the order of the masking (transparency) bits, it would have been at least understandable:

Either "tridx: 3210" (bit-position order) or "tridx: 0123" (ascending order) would look possible to me; I'd prefer "ascending" order.

But the colors... shall always be connected in ascending order, shouldn't they?

colors: (col. for pal.-index 0) (col. for pal.-index 1) (col. for pal.-index 2) (col. for pal.-index 3)

I would be very confused if it wasn't so! :scared:

So the second line shall be nonsense, because color 0 is transparent, so it shouldn't matter which color it got - the fact that it is white, obviously shows a shift or mirror in the palette index interpretation.

Yong
19th October 2004, 06:28
ffdshow version Oct 12 2004 08:06:04 SSE2
Bug report:

My computer spec,
Pentium 4 2.4GHz
512 MD DDR 266
Windows XP SP1
DirectX 9.0c
ATI RADEON 9600XT

Software:
Media Player Classic 6.4.8.2
Matroska container,
Using Haali Matroska Splitter Sep 8 2004 14.06.02

ffdshow configuration:
All video decoder using libavcodec,
except MPEG1 and MPEG in AVI using libmpeg2.
Only the postprocessing filter is used.
Other setting use default setting.

ffdshow postprocessng configuration:
[* = Selected option]
*process whole image
*only right half
*Postprocessing

*Custom
--*Dering, Luminance
--*Deblock H, Luminance
--*Deblock V, Luminanve
Processing Strength: 512

Processing method:
--*mplayer
--*Accurate deblocking
Level fix
--*luminance
--*Full luma range

--*Nic's Nic's First
--* X threshold: 20
--* Y threshold: 40

I can start playing video normally sometimes when using setting above, But i drag new video to the player, it crashed.
If the video can start normally, it will crashed if i close the video.
Sometimes can't start playing at all, crashed.
If i unselect the "Dering, Deblock H, Deblock V(All Luminance)" option, crashed won't happen again.
This three settings crashed with "only right half" option selected.

select "Dering , Deblock H, Deblock V(all Chroma)" don't have this problem.

BoNz1
19th October 2004, 06:32
Encoding video with x264 with ffdshow video codec in virtualdubmod crashes here. Anyone else with the same problem?

LigH
19th October 2004, 08:14
No crash with x264; but some strange over-softening occured (I wonder if the loop filter was the reason, or enabling all subdivisions); I miss a "reset" to default options (except for re-installing).

yaz
19th October 2004, 10:41
Originally posted by LigH
No crash with x264; but some strange over-softening occured (I wonder if the loop filter was the reason, or enabling all subdivisions); I miss a "reset" to default options (except for re-installing). the same here. with celtic's compile it 'cuts like hell' for me.
the quality is amazing (for me) but, yes, the overall look is somehow softer than i like. (yes, i'm one of that crisp-freaks:-)
the bests
y

akupenguin
19th October 2004, 20:39
Originally posted by LigH
some strange over-softening occured (I wonder if the loop filter was the reason, or enabling all subdivisions)
Do you have normal ffdshow postprocessing on, in addition to the H.264 loopfilter?

LigH
20th October 2004, 08:30
Of course! ;) - I tried both; It looks like after a scene change, the first frames are quite crisp, then suddenly the following frames become flat. I'm not sure if this still occurs with the current version shipped with the latest ffdshow, it occured with the previous version and celtic_druid's test version 8. I'll try to check it more soon (sorry, don't have much free time)...

celtic_druid
20th October 2004, 09:37
As far as I can tell there haven't been any changes to src/codecs/x264 since I compiled those versions, although there have been a couple of changes to x264.

hellfred
20th October 2004, 13:03
VDubMod and ffdshows x264 works for me,too.
Win98
PIII
VDubMod 1.5.10.1 build2439
ffdshow 20041012 with cletic_druids ff_x264.dll #8
2pass with default settings. quant 12

Hellfred

celtic_druid
20th October 2004, 13:25
In case anyone is interested:
http://celticdruid.no-ip.com/test/ff_x2648-AthlonXP.7z
http://celticdruid.no-ip.com/test/ff_x2648-Pentium4.7z

Might give a bit of a speed boost over the previous pentiumpro build.

yaz
20th October 2004, 14:50
Originally posted by celtic_druid
In case anyone is interested ... many many thx ! just for sure; did u recompile the previous source or is it a new cvs-snapshot ?
this or that, it always occurs to me that would we ever be able to make any return for u for your contribution ? if u got a clue, just call me :-)))
thx again
y

celtic_druid
20th October 2004, 15:12
Well I have done several cvs updates since the last compile, however the src/codecs/x264 files have not changed so yeah it is basically a recompile.

yaz
21st October 2004, 08:28
Originally posted by yaz
... if u got a clue, just call me :-))) ... thx again
y Microsoft Internet Explorer Detected - Downloaded Blocked would it be that ? :-(((
y

Yong
21st October 2004, 08:29
I can't download...
When i click the link above,
after 2 swcond i was see "Microsoft Internet Explorer Detected - Downloaded Blocked"...
can someone plz tell me why...:confused:

EDIT: plz ignore my post, it worked:D :D :D

LigH
22nd October 2004, 22:46
I'd guess, the MSIE detected a META-REFRESH or script initiated download, and the MSIE blocked a not-user-initiated download (for security reasons).

Your message looks like the webserver detected an access using MSIE, and the webserver blocked the download - what I would not expect at all... :D

celtic_druid
23rd October 2004, 07:28
Originally posted by LigH
Your message looks like the webserver detected an access using MSIE, and the webserver blocked the download - what I would not expect at all... :D

That is exactly the case, however I turned off the IE block days ago.

Yong
23rd October 2004, 11:13
Big Thanks celtic_druid for the ff_x264 dll and the mencoder!!:D
But still encounter an error when encoding video using the h264 codec... :eek:
My comp, spec. is P4 2.4GHz.
I've read the mencoder+x264 thread, i'll try the new build [mplayer20041022p4].
It's worked. :D

@LigH:
I'm using shared computer(internet cafe), so i don't know the computer admin's using what kind of secutity setting...
I'm using firefox(R) at my own computer!;)

New bug report here:
ffdshow config:
format: MPEG1/2
codec: libavcodec

ffdshow will crash when playing mpeg 1/2 video with the visualization "Motion vector" is on or selected while playing.

Rumbah
24th October 2004, 14:22
The ffdshow SSE2 build crashes Premiere Pro 1.5 with ffv1 (I have a P4). With the non SSE2 bild, everything works fine.

celtic_druid
26th October 2004, 10:29
updated x264: filesize (bits) in a 32 bit int will overflow after 250MB, screwin.
core/common.h
encoder/encoder.c

Someone want to test this (http://celticdruid.no-ip.com/test/ff_x2649.7z)
, not sure if it effects the problem with -o3 or not, so I compiled this new dll with -o3.

Doubt it fixes the -o3 so I also put up two new dll's with -o.

Links for the updated dll's with -o in case anyone was wondering:
http://celticdruid.no-ip.com/test/ff_x264-AthlonXP.7z
http://celticdruid.no-ip.com/test/ff_x264-Pentium4.7z