ffdshow development #2

[Liisachan]

"Waiting for GCC 4.0 (http://ffdshow.sourceforge.net/tikiwiki/tiki-read_article.php?articleId=16&PHPSESSID=0a56077b3918dfedd0daff1c43d56009)" is related?


I'm waiting for GCC 4.0 to be released to make it one of compilers supported by ffdshow build process again.

[namchik]

+-- GCC 4.0 branch created ------+
| (Feb 25 2005) \
| v
GCC 4.1 Stage 1 (ends April 25 2005) GCC 4.0.0 release (April 20 2005)
| \
| v
GCC 4.1 Stage 2 (ends July 8 2005) GCC 4.0.1 release
|
v
GCC 4.1 Stage 3
|
:
v

http://gcc.gnu.org/develop.html#future
So GCC 4.0 wasn't released on April 20 ? =(

[Musafir_86]

-Does anyone know how well the support is it for Ogg Vorbis decoding in ffdshow? Is it better than CoreVorbis or any decoder based on libvorbis 1.1.1 (or at least same - but with lower CPU usage)? :confused:

-But..., are there any new DS decoder based on libvorbis 1.1.1? :stupid:

Thanks.

-Musafir_86.

[celtic_druid]

Last time I tried ffdshow compiled with gcc 4.0 it just crashed.

Radlight's filters I believe include a vorbis decoder.

By vorbis decoding in ffdshow do you mean libavcodec or tremor?

[Musafir_86]

-Of course I mean libavcodec; isn't Tremor dated back to year 2003 then? Also, it's based on fixed-point method, right?

-Musafir_86.

[Liisachan]

RadLight's filters and VSFilter don't mix well, and so not suitable for softsubbed OGMs, but RadLight's filters even can decode Ogg Theora via DS nicely.
Yet another Ogg DS decoder is by illiminable's
http://www.illiminable.com/ogg/

[madman1980]

Ffdshow really covers all my needs, but one really annoying thing is MPEG2 aspect ratio. It is almost always off when I play files in MPC. SVCD is not played with the right aspect ration, neither is PAL or NTSC vobs (dvd spec). I have to adjust it manually.

My solution has so far been to install Windvd in addition to ffdshow. But it is kind of a waste, since Ffdshow should be able to detect aspect ratio too, if it was just given some priority.

Any ideas if this will be fixed?

[eb]

use VLC player , then you are independent of M$ system, very important when playing back multiaudio files, MPC plays them all at the same time

[madman1980]

Sure, VLC is good. But i prefer MPC by far. FFdshow has made everything work except this, so I hope the developers can learn something from the WinDVD MPEG2 decoder with regards to defaulting to the correct aspec ratio.

[celtic_druid]

New build: http://mirror05.x264.nl/celtic_druid/force.php?file=./ffdshow-20050630.exe
ICL9

[Musafir_86]

@Celtic_Druid,

-Can you always post the changelog for ffdshow? Or at least bundle together with the binary package; this will help to know whether someone specific problem was solved or not.

Thanks.

-Musafir_86.

[celtic_druid]

Changelog?
http://cvs.sourceforge.net/viewcvs.py/ffdshow/ffdshow/?sortby=date#dirlist

[Musafir_86]

-Thanks, but wouldn't it be nicer if they come together, so one don't have to browse the CVS each time a build is up. If you don't have time, then it's okay...

Musafir_86.

[celtic_druid]

Well someone has to browse the cvs, because as far as I know there is no simple changelog, otherwise I would just copy and paste it.

[Liisachan]

@celtic_druid
I downloaded the file 3 times from the .nl server, 2 times from your server.
The file from the .nl server is always 1 byte larger (3192268 Bytes), it has extra 0x20 at the end of the file. The same happened to me for the pervious build. Maybe firefox is not good at downloading files??????

I mirrored the 'smaller' one (3192267 Bytes)

[celtic_druid]

Don't know. Maybe the server? ftp shows that the file on their is 267bytes, so it isn't an upload issue.

[celtic_druid]

Ok changelog since last release:

fixed ICL 9 compilation (ffdshow.ax compiled with ICL9 to)
updated libavcodec
faster h264 deblocking,
updated x264 - custom quantization matrices (not supported by libavcodec yet)
saving of JM quant tables
removed min_order and max_order,
GCC hash_map implementation, the old one was buggy
adaptive FFT length in convolver depending on impulse file number of samples
SSE2 avisynth yuy2 <-> yv12 conversion
SSE2 bitmap blending and soften
SSE2 colourize
SSE2 gradual denoise
SSE2 hue and saturation
SSE2 luminance offset and gain
SSE2 mplayer noise
SSE2 unsharp
SSE2 warpsharp bump
recreate configuration pages when changing presets
working on direct communication between filters and configuration pages
trying to fix HW overlay filter

taken from cvs log -rHEAD -d"lastbuild<thisbuild". Not quite as easy as a cut and paste.

[JoeBG]

Ok changelog since last release:

fixed ICL 9 compilation (ffdshow.ax compiled with ICL9 to)
updated libavcodec
faster h264 deblocking,
updated x264 - custom quantization matrices (not supported by libavcodec yet)
saving of JM quant tables
removed min_order and max_order,
GCC hash_map implementation, the old one was buggy
adaptive FFT length in convolver depending on impulse file number of samples
SSE2 avisynth yuy2 <-> yv12 conversion
SSE2 bitmap blending and soften
SSE2 colourize
SSE2 gradual denoise
SSE2 hue and saturation
SSE2 luminance offset and gain
SSE2 mplayer noise
SSE2 unsharp
SSE2 warpsharp bump
recreate configuration pages when changing presets
working on direct communication between filters and configuration pages
trying to fix HW overlay filter

taken from cvs log -rHEAD -d"lastbuild<thisbuild". Not quite as easy as a cut and paste.


Thanks a lot :)

When are you visiting europe? Didn´t you plan holidays? :)

[celtic_druid]

No, didn't plan anything, but yeah I leave Monday. So ~4 days 2 hours.


2005-06-19

another try on fixing b frame delay
common function to search for pin interface
fixed DwString::sEmptyBuffer initialization,
less explicit casting in binds and options lists
md5sum OSD item,
more comptrQ,
show coded frame size in OSD,
simple right-click menu in codecs list
updated x264 and libavcodec - lossless h.264
using function pointer in TomsMoComp
writing OSD info to file,
moving binds to Twindow (can't be compiled now)
working on direct communication between configuration pages and filters

2005-06-11

attempt to use libavcodec computed PTS when available
better unicode subtitles support
cabac fixes
don't create keyboard, remote and tray icon threads when in explorer.exe,
merged OSD and Info,
removed IDFF_max constant,
remove ff_realaac.dll and start menu shortcuts on unistall,
updated minilzo
updated vc6 project

2005-06-09

always display levels curve,
don't remove all characters surrounded by { },
fixes in separate luma and chroma resizer
new info panel,
makefiles update and italian setup translation by Gianluigi Tiesi
offer all font charsets (detected are displayed bold in combobox)
updated libavcodec,
updated x264,
zones support for x264,
working on full unicode subtitles support

2005-06-07

autoincrement build version
autoload preset on decoder match
included baseclasses can be compiled by GCC without modifications
include stripped baseclasses library
fixed GCC ffvdub compilation
ffdshow can be compiled by GCC: image filters are unoptimized, crashes immediatelly at start
QueryInterface -> comptrQ
support for reading theora from RadLight ogg splitter
support for cross-compilation by Gianluigi Tiesi
updated libavcodec
cosmetic,
trying to compute correct font width for italic subtitles,

[LigH]

I have some small issues as well: b0bors server tells the MIME type "application/force-download". That's some nonsense, in my opinion: *.exe files usually shall be sent using "application/octet-stream".

Most probably due to this strange MIME type, Opera 7.54 suggested to save this file with the extension "*.ace" (WinAce archive)?! Don't know why Opera once registered this extension to this MIME type - I had to edit this list entry...

[yaz]

Sure, VLC is good. But i prefer MPC by far. FFdshow has made everything work except this, so I hope the developers can learn something from the WinDVD MPEG2 decoder with regards to defaulting to the correct aspec ratio.
may i suppose u use mpc w/ffdshow for svcd/dvd playback ? if so, how do u manage to force ffdshow to be the mpeg2 decoder. i've tried tricking in mpc a/o in ffdshow but the only way is killing all the possible candidates (nero, windvd, pdvd, aso, aso)

as regards AR.
- u set 'use overlay mixer' in 'output' and 'overlay a.r.' to default in 'resize...', don't u ?
- does it work w/other stream types ? u may find some anamorphic example files here in the forums w/which u can test it.
just ask, because that part worx correctly for me.

thx
y

[Leak]

Well someone has to browse the cvs, because as far as I know there is no simple changelog, otherwise I would just copy and paste it.

Well, this could help with the recent changes:

http://cia.navi.cx/stats/project/ffdshow

np: Markus Guentner - Sleep Well (Audio Island)

[celtic_druid]

Speaking of leaks, I noticed earlier that there is a leak in subtitles reader. I'll see about putting up a new build.

[Sharktooth]

mirrored.

[iron2000]

Why does the last version install at 6.89mb and the new version install at 11.77mb?
Not that I'm pressed for space, just curious.
(A bit off 10mb is ok but don't go to 20+mb)

[Alvy]

Latest Ffdshow audio decoder doesn't work with reclock together. After downgrading to the last version it works again.

By the way thanks for this great program !!!!

[ViCroié]

Is this me, or ffdshow?

I'm using ffdshow-20050630 (celtic) and i wanted to use ffdshow to be able to deinterlace video from my TVcard (YUV2 Interlaced) in Media Player Classic.

I used the AVIsynth feature from ffdshow to do this. and deinterlacers like: TomsMoComp and FieldDeinterlce work. The problem comes when i use a BOB deinterlacer (wich is the best way of watching this footage imho). When i enable it it will only show 25 of the 50 FPS. When i do thesame thing in avisynth itself it works great.

[celtic_druid]

2005-07-01

fixed few possible crashes when processing small images
fixed memory leak in subtitles reader,
positional controls apply to embedded bitmap subtitles
Italian translation by Dario Brama
smoothing method and scale apply to embedded bitmap subtitles,
SSE2 32-bit to 16-bit integer samples conversion
SSE2 original and avih noise
SSE2 SPP store_slice

Probably the last build from me for a month.

[Sharktooth]

CD, enjoy your holydays and thanx for your builds :)

[yaz]

@celtic_druid
waving good-by is always sorrow, but u made it a bit easier ;) have a nice holiday !
thx a lot
y
ps. the climate is a bit 'armageddonian' here in europe (it's pouring like hell, actually), so don't forget your life-belts :D

[Yama4050242]

maybe a bug,
both ffdshow-20050630 ffdshow-20050701 crash on playback clip encoded with megui(x264 cli build 270) setting subpixelrefinement RDO and ME Algo Exhausitive

ffdshow-20050619 is fine for playback

[Leak]

I used the AVIsynth feature from ffdshow to do this. and deinterlacers like: TomsMoComp and FieldDeinterlce work. The problem comes when i use a BOB deinterlacer (wich is the best way of watching this footage imho). When i enable it it will only show 25 of the 50 FPS. When i do thesame thing in avisynth itself it works great.

I fear this is a limitation of ffdshow - it doesn't notice if you change the framerate, and thus doesn't propagate it down the DirectShow filter graph...

I took the liberty to file a bug on it... (http://sourceforge.net/tracker/index.php?func=detail&aid=1231101&group_id=53761&atid=471489) ;)

np: Luke Vibert - Gwithian (Lover's Acid)

[madman1980]

maybe a bug,
both ffdshow-20050630 ffdshow-20050701 crash on playback clip encoded with megui(x264 cli build 270) setting subpixelrefinement RDO and ME Algo Exhausitive

ffdshow-20050619 is fine for playback

Thanks for letting us know. I'll stay with the .19 version until there is one out that doesnt seem to introduce new problems.

[celtic_druid]

Just tested a clip with RDO and Exhausitive ME and it played back fine here.

[Yama4050242]

@celtic_druid
http://mail.e8007.com/viewsharenetstore.php?sessid=2765442c51873b4700&userid=snow4050242&filename=L3NhbXBsZS9zYW1wbGUucmFy&isdir=0

the above link is small parts from my crashed clip.the file sever is not stable, hope you can get my file

[celtic_druid]

Looks like it is probably an ICL9/VC6 issue. The build from the 19th was compiled with MVC7.1.

[bob0r]

@celtic_druid

I assume you got the test file?
Can you upload the video file to mirror05 (maybe latest ffdshow too)?

And before you go siesta, compile ffdshow with all possible tools and we will figure out which is best :sly:

[Palikrovol]

OT: CD if you come to Madrid, think here we are at 36 degrees at 15:42 and sometimes at 28-30 at 23:00-24:00 (Hotter if you go to the south, Sevilla for instance 38 degrees at 16:00) so do no bring too much clothes with you :).

Happy hollydays here in HOT Spain

E=MC^2
España = Mucho Calor^2 :)

[celtic_druid]

Yeah, I left the test file downloading over night and somehow managed to get it.
MSVC7.1 and ICL9 builds are up. ICL9 crashes on the sample, but maybe is faster?

Packed a couple of pairs of shorts and tshirts/singlets. So should be ok.

[wata]

the OSD remaining time doesn't work for all builds after ffdshow-20050527.exe
it turn to 00:00:00 before movie end.

[Dark_Angel_PT]

I have submitted this request at Sourceforge. What do you think of this?

http://sourceforge.net/tracker/index.php?func=detail&aid=1231905&group_id=53761&atid=471492

[Defiler]

Anyone feel like making a non-SSE/SSE2 build?
I'd do one myself, but I haven't had a lot of luck getting the ASM parts working in Visual Studio.

[LigH]

ffdshow by celtic-druid works well on my AMD Duron (which is not SSE capable).

[Defiler]

ffdshow by celtic-druid works well on my AMD Duron (which is not SSE capable).
Really? That's good news. Does it melt when you enable things like denoise3d, that have been SSE2-optimized?

[vlada]

I fear this is a limitation of ffdshow - it doesn't notice if you change the framerate, and thus doesn't propagate it down the DirectShow filter graph...

I took the liberty to file a bug on it... (http://sourceforge.net/tracker/index.php?func=detail&aid=1231101&group_id=53761&atid=471489) ;)

np: Luke Vibert - Gwithian (Lover's Acid)

I did some tests with FFDShow and it's deinterlace filters some time before. No one of them is able to produce double framerate (in player or in VirtualDub). Also the quality is rather poor and they aren't very fast as well. I tried some deinterlacers from DSclaer and they didn't work as expected neither. I would like to see at least one smart bob filter, which would be able to deinterlace to 50 fps realtime on 2,0 GHz CPU. But I don't know if I'm not asking too much.

If anyone's interested, I did a comparison of different deinterlace filters. It is located at http://jech2.webz.cz/gallery/index.php. I think I already posted the link here some months ago. My commentaries are written in czech, but the pictures say usually enough.

[LigH]

Does it melt when you enable things like denoise3d, that have been SSE2-optimized?
:D Melt - no. But it "crumbles", e.g. SPP deblocking is impossible. And -- Denoise3D? Where is that one hidden?!
__

Ah - found it... No, doesn't crash. I'd guess that SSE optimizations are used when possible, but probably either skipped or applied with the unpotimized version if not.

[Px]

And what about of ffdshow build for Windows x64?

[multiblitz]

One question: Why needs the latest celtruid version 10% more cpu load (Athlon 64 3500@2400mhz) that the normal sse2-build from Oktober 12th ?

[Isochroma]

Using MPC 6.4.8.4, all builds of ffdshow with matroska container. Aspect ratio embedded in matroska container for the video stream is totally ignored if the video stream is decoded by ffdshow (any supported codec). Using the real codec (ie. xvid, etc.) aspect is shown fine. Why?

[LigH]

Do you have "use overlay mixer" enabled in the "output" register?