New ffdshow build (?)

[bob0r]

http://download.microsoft.com/download/9/4/2/9428d199-bf6b-4bf4-8c45-89a439199278/dxsdk_dec2005.exe

Might be useful.

[_xxl]

http://rapidshare.de/files/9627377/ffdshow-20051221.zip.html

[_xxl]

http://rapidshare.de/files/9636739/ffdshow-20051221.zip.html

[DeathTheSheep]

A faster, better mirror for your build, drevil ;)
(drevil_xxl ICL9+GCC4.0.2) ffdshow-20051221.zip (http://www.myfilehut.com/userfiles/5120/CollectedBuilds/ffdshow-20051221.zip)

Your last ICL9 build (12-08) had the noise shaping feature working perfectly, but this one shows the crashing behavior I mentioned before. Perhaps if you compiled the noise shaping code with ICL9... :D

[Kurosu]

I had a build automation script for years (when Athos was still around) and I finally put enough effort into it to have it running more or less clean. What it does:
- get ChangeLog
- do cvs update
- compile the targets set in the configuration files
- generate the HTML page linked below
- upload files

You can check the result over at http://kurosu.free.fr/ffdshow.htm

Prerequisits for it to work are probably:
- gcc 4.0.2 from sherpya or your own compile
- mingw (my version is a patchwork, so I can't be more precise)
- msys and its DTK (mostly for perl)
- a large upload space available through FTP
- NSIS. 2.02 probably
- upx in the path
- apply the patch listed here (http://sourceforge.net/tracker/index.php?func=detail&aid=1379941&group_id=53761&atid=471491) as some libraries wouldn't be remade without it
- a FFDShow CVS tree; its folder name must be ffdshow, and this script should be put in its parent folder

I tested it both through the "task scheduler" in Windows, and a cron job + cross-compilation under linux, and it worked at some point. I don't know if it does now.

There would be many additional things to say about how to use and set it up, but I'm just going to say, like many funny people: the documentation is the code, so enjoy. As far as I'm concerned with this script, I'm done.

[thuan]

Kurosu I can't find your Athlon 1225 build?

[canuckerfan]

Kurosu, I noticed that your builds also have SSE3. However, your builds don't work on my system. The option dialogues for ffdshow don't open and whenever I play something in MPC, it says something along the line of cannot render and try downloading a codec.

[MacAddict]

I've got the same problem as canuckerfan above. No properties dialog and MPC gives the rendering error, then XviD Decoder kicks in and works. That problem only existed with the K8 A64 build though. I tried the 12-21 P4 build and it seems to work great.

Many thanks

[IvS]

Same here with a Venice core Athlon 64 3200+.

[DeathTheSheep]

Same problem as those above, but I don't have SSE3 (...probably. Celeron D 2.93ghz with 533FSB). But still, it's the exact same problem as the folks above have. :)

[issa]

Latest CVS build,

Here is Intel C++ Compiler build with '-arch:SSE2' (some dll compiled with gcc 4.0.2).

http://rapidshare.de/files/10074282/ffdshow-20051230-icc-sse2.exe.html

Could someone to test it working or not on non-Intel CPU?

[IvS]

issa: It works.

[cyberyeye]

Hi :)

Sorry to ask this if this already answer but what build is fine or optimized for A64 (3700+, san diego, 90nm) ???

Gcc build post on http://x264.nl/ seems broken to me because seeking has some "freeze" during 10 sec. this not systematic but really boring.

Don't have this problem with xvid decoder from 1.1.x only with ffdshow.

so wich build is better for A64 (on win32) ?

ps: happy new year to everyone ^^

39 pages to read, maybe it's time to put this info "wich build for..." on sticky or at least on the 1st page no ?

[ExtraEye]

cyberyeye,
im using that build with no problems.
also my processor is athlon64 3000+.

from what i recall the best build for athlon64 is gcc with sse2 (and also from what i understood sse3 woun't improve the performence).
so what i do is stick to bob0r builds(unless someone else made a more updated gcc with sse2 ability) that use sse2.

[cyberyeye]

Thanks for your reply :)

Bob0r build = http://x264.nl ???
mean for example this build: http://mirror01.x264.nl/x264/ffdshow-20051221-gcc4.0.2-sse-x264.nl.exe
build linked on x264.nl

but this build is "marked" sse, and not sse2...

If i'm wrong where's sse2 bob0r build on 264.nl I see only sse build ?

[ExtraEye]

sometimes he doesn't make an sse2 build.

btw
nice nick ;)

[DeathTheSheep]

What, are we mustn't be sheepy about our eyes--they can lead to our untimely death ;)

[cyberyeye]

Thanks for the info ExtraEye ^^
Now we should just wait for SSE2 build from bob0r !

:goodpost:

[videomixer9]

Libraries only, replace with existing ones in ffdshow directory, no ffdshow.ax

GCC 4.0.3 for Athlon 64 SSE3 non-x64 (may also work on SSE2 only)
ffdshow-20060104-gcc4.0.3-athlon64-sse3-x86.7z (http://rapidshare.de/files/10396650/ffdshow-20060104-gcc4.0.3-athlon64-sse3-x86.7z.html)
GCC 4.0.3 for Athlon XP SSE
ffdshow-20060104-gcc4.0.3-athlonxp-sse-x86.7z (http://rapidshare.de/files/10399396/ffdshow-20060104-gcc4.0.3-athlonxp-sse-x86.7z.html)
both contain x264, libmad, libmplayer, libavcodec, libdts, liba52, theora, tremor, realaac, faad

[cyberyeye]

GCC 4.0.3 for Athlon 64 SSE3 non-x64 (may also work on SSE2 only)
ffdshow-20060104-gcc4.0.3-athlon64-sse3-x86.7z
Thanks I tried and it works fine ; replaced over 21122005 gcc build from x264.nl, now sse3 in ffdshow should be activate and now no "freeze" when seeking an xvid or divx files :-)

[ExtraEye]

thx videomixer but how do i check if it works now? i mean the libraries and the SSE3?
(it doesn't have a box for sse3)

[Rash]

Celtic Druid is not releasing builds anymore?

Oh, and just a question. I notice here SSE, SSE3, Athlon64... releases. So that information available on "Info & Debug" while a video is playing, that shows which optimizations are available, are worthless?

[Romario]

Celtic Druid is not releasing builds anymore?

Sad to hear that:(

[signatory]

Can someone be so kind to share their ffdshow settings for good xvid playback ? I used the default but get a bit jumpy playback. (frame skippin' ) hit the bug switches, a little bit better. But still not perfect. libavcodec is codec I don't want to install xvid stand alone codecs cos they suck for me.

What IDCT should I use ? Any other ideas? much appreciate help

EDIT: I figured it out. A bug in ffdshow vfw.

[issa]

Libraries only, replace with existing ones in ffdshow directory, no ffdshow.ax

GCC 4.0.3 for Athlon 64 SSE3 non-x64 (may also work on SSE2 only)
ffdshow-20060104-gcc4.0.3-athlon64-sse3-x86.7z (http://rapidshare.de/files/10396650/ffdshow-20060104-gcc4.0.3-athlon64-sse3-x86.7z.html)
GCC 4.0.3 for Athlon XP SSE
ffdshow-20060104-gcc4.0.3-athlonxp-sse-x86.7z (http://rapidshare.de/files/10399396/ffdshow-20060104-gcc4.0.3-athlonxp-sse-x86.7z.html)
both contain x264, libmad, libmplayer, libavcodec, libdts, liba52, theora, tremor, realaac, faad

Where did you find the GCC 4.0.3?

[Inventive Software]

http://forum.doom9.org/showthread.php?p=764952#post764952. Hopefully this will be a more up to date catalogue of ffdshow builds.

[ExtraEye]

good idea
now what will become of this thread...

[videomixer9]

Where did you find the GCC 4.0.3?

www.reactsoft.com

had this idling on my disk for month so ... doesn't seem to be official but the compiled code said 4.0.3 too so I just used that ... guess they made it 4.1.0 later or maybe it's only late bugfixes ... whatever it worked fine so I kept using it.

[Romario]

Is SSE3 FFDSHOW really work? I doubt it.

[videomixer9]

just used those compiler options, I doubt it uses SSE3 really though, however there weren't much (if any at all) changes anyways but it seemed to do the job in not crashing in certain situations, only reason I did it and tagged it that way to make only ppl with those athlon cores get it :) not that i'll do any fruther compiles anymore as CoreAVC totally killed my reason for using ffdshow ...

[cc979]

has anybody tried compiling gcc-4.1 yet ?
and compiling ffdshow with it ?

[bob0r]

Can you compile gcc 4.1 on mingw/msys then?

[cc979]

Can you compile gcc 4.1 on mingw/msys then?

not had time try, well do tonight

[cc979]

Can you compile gcc 4.1 on mingw/msys then?

rumours that is possible, with some earlier builds but not with svn

to many errors for me

just did a test rebuild of ffdshow, nsis2 give me a strange error and theres no dts - help

[DigitalDivide]

I downloaded the latest Milan build and seem to be having a problem with the Preset Automatic Loading.

I have two presets. One for my DVD's and one for avi's. One loads on Filename Match with wildcards using *.avi and the other video_ts*. For some reason which ever preset is highlighted is the one that gets loaded regardless of the filename. I've tried uninstalling and reinstalling and no matter what the same preset always loads regardless. I know I used to have this working with an earlier one but I don't remember what version I had installed previously. Anyone have this problem? Any idea how to fix it.

[lithoc]

I'm using my friend PC the other day.

It's a Celeron 500mhz + Intel i810g chipset(s370, integrated graphics), WinXP SP2, 128MB SDRAM.


Having tried,

1.) ffdshow-20051221-gcc4.0.2-sse-x264.nl.exe from x264.nl.
2.) ffdshow-20051109.exe from x264.nl.
3.) (drevil_xxl ICL9+GCC4.0.2) ffdshow-20051221.zip

Crashed when:
1.) Playing audio dshow is use. Happen to any audio codec.
2.) Opening up the video dshow configuration screen. It play the video tho.
3.) Any filter in video dshow will crash.

Need expert oppinion on this.

Do anybody build ffdshow for old machines (eg Pentium-MMX, Pentium 2, older Celeron)?

I still have Pentium-MMX(overclocked) to watch movies sometimes(for sake of torturing it :devil: )
It plays perfectly on Linux + mplayer via VESA 2.0 output.

[issa]

I'm using my friend PC the other day.

It's a Celeron 500mhz + Intel i810g chipset(s370, integrated graphics), WinXP SP2, 128MB SDRAM.


Having tried,

1.) ffdshow-20051221-gcc4.0.2-sse-x264.nl.exe from x264.nl.
2.) ffdshow-20051109.exe from x264.nl.
3.) (drevil_xxl ICL9+GCC4.0.2) ffdshow-20051221.zip

Crashed when:
1.) Playing audio dshow is use. Happen to any audio codec.
2.) Opening up the video dshow configuration screen. It play the video tho.
3.) Any filter in video dshow will crash.

Need expert oppinion on this.

Do anybody build ffdshow for old machines (eg Pentium-MMX, Pentium 2, older Celeron)?

I still have Pentium-MMX(overclocked) to watch movies sometimes(for sake of torturing it :devil: )
It plays perfectly on Linux + mplayer via VESA 2.0 output.

Try mplayer on windows with directx output, it use less resources then DirectShow filter.

[issa]

Latest CVS build,

SSE2 build (icl 9 + gcc): http://rapidshare.de/files/11624452/ffdshow-20060123-icc-see2.exe.html

[Egh]

Latest CVS build,

SSE2 build (icl 9 + gcc):

Were there any serious changes in ffdshow cvs last 3-4 weeks or so? ^^

[clsid]

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

No updates since Dec 21, 2005 ... until today. I guess Milan is back from vacation :)

[videomixer9]

anyone willing to do a more general build?

[clsid]

Look in this topic for ffdshow builds:
http://forum.doom9.org/showthread.php?t=105386

[Lemonzest]

with the icl9 builds are the non-pentium checks removed?

[clsid]

Yes. GenuineIntel check has been patched.

[lithoc]

Yes. GenuineIntel check has been patched.

Audio filters is not working. Output statics when enabling it.

[cc979]

anybody get this error building todays cvs,

TglobalSettings.cpp: In member function 'virtual CodecID TglobalSettingsDecAudio::getCodecId(DWORD, FOURCC*) const':
TglobalSettings.cpp:994: error: 'WAVE_FORMAT_DSPGROUP_TRUESPEECH' was not declared in this scope
make[1]: *** [ffdshow_settings_all.o] Error 1
make: *** [lib] Error 2

[OutsiderX]

issa,:thanks: for your effort to supply us with the lastest build of ffdshow...

but I found that ffavisynth.dll is not workin' at all....maybe b/c your compiled it with gcc
for anyone who has the same problem like me
you can use
this
http://files.x264.nl/force.php?file=./ffdshow/ffdshow-20051021/icl9.0/ffavisynth.dll
or this one
http://files.x264.nl/force.php?file=./ffdshow/ffdshow-20051021/msvc7.1/ffavisynth.dll

[clsid]

Audio filters is not working. Output statics when enabling it.
I did some quick tests. Only Freeverb doesn't seem to work. However if I enable Dolby Decoder too, then Freeverb works.

[lithoc]

I did some quick tests. Only Freeverb doesn't seem to work. However if I enable Dolby Decoder too, then Freeverb works.

Try Equalizer, anything higher than 0db will output statics.

Volume filter works, but the Normalize percentage is not correct.
Sound weird as well, it's not smooth. The sound level suddenly jump from soft to loud and vice versa.

[issa]

issa,:thanks: for your effort to supply us with the lastest build of ffdshow...

but I found that ffavisynth.dll is not workin' at all....maybe b/c your compiled it with gcc
for anyone who has the same problem like me
you can use
this
http://files.x264.nl/force.php?file=./ffdshow/ffdshow-20051021/icl9.0/ffavisynth.dll
or this one
http://files.x264.nl/force.php?file=./ffdshow/ffdshow-20051021/msvc7.1/ffavisynth.dll

Latest CVS build (icl 9, msvc, gcc [libmplayer, libavcodec]): http://rapidshare.de/files/11775390/ffdshow-20060125-icc-sse2.exe.html

[OutsiderX]

I just tried your new build and I still get the same error...But 10X for trying to fix it though.
I just did some search and found that I'm not the only one who has this problem,Thank God for that :D

http://forum.doom9.org/showthread.php?t=104633

[Esc]

ffdshow-20060125-icc-sse2.exe says "Error while registreing ffdshow.ax" on installation. Where do I address that?

[Razorholt]

If you have any version already installed, uninstall it, reboot and try again.

- Dan

[namchik]

ffdshow-20060125-icc-sse2.exe says "Error while registreing ffdshow.ax" on installation. Where do I address that?
Make sure you have MS runtime libraries in your sys32 or ffdshow folder... To be more precise i guess you need "msvcr71.dll"

[signatory]

ffdshow-20060125-icc-sse2.exe says "Error while registreing ffdshow.ax" on installation. Where do I address that?

I have encountered this error message too, in the past and with this one.
OK it's easy to fix by uninstalling the old ffdshow but it would be pretty terrible if one would forget to save the ffdshow.reg settings first....

[LoRd_MuldeR]

Latest CVS build (icl 9, msvc, gcc [libmplayer, libavcodec]): http://rapidshare.de/files/11775390/ffdshow-20060125-icc-sse2.exe.html

Is there a build for non-SSE2 CPUs aviable yet?

[issa]

ffdshow-20060125-icc-sse2.exe says "Error while registreing ffdshow.ax" on installation. Where do I address that?
Sorry for the problem.... Here is the lastest CVS build and the installation should be fixed.

Updated 02-02-2006, all the DLL outside the ffdshow directory, libavcodec & libmplayer are compiled with gcc.
SSE2 Build (icl 9, msvc 8, gcc): http://rapidshare.de/files/12367610/ffdshow-20060202-icc-sse2.exe.html
SSE Build (icl 9, msvc 8, gcc): http://rapidshare.de/files/12429903/ffdshow-20060203-icc-sse.exe.html

[LoRd_MuldeR]

Sorry for the problem.... Here is the lastest CVS build and the installation should be fixed.

SSE2 Build (icl 9, msvc 8, gcc [libavcodec & libmplayer]): http://rapidshare.de/files/11916301/ffdshow-20060126-icc-sse2.exe.html

and SSE1 ???

[CeeJay.dk]

Sorry for the problem.... Here is the lastest CVS build and the installation should be fixed.

Installation should be fixed.
SSE Build (icl 9, msvc 8, gcc [libavcodec & libmplayer]): http://rapidshare.de/files/12073437/ffdshow-20060129-icc-sse.exe.html

This gives me the error R6064.
"An application have attemted to launch the C Runtime library incorrectly" .. or something like that.
It keeps bringing up that errormessage when ffdshow tried to load and if I click OK it keeps comming back .. forcing me to reset the PC.
So the errormessage might not be exact word for word.. that just what i remember it said.

[issa]

This gives me the error R6064.
"An application have attemted to launch the C Runtime library incorrectly" .. or something like that.
It keeps bringing up that errormessage when ffdshow tried to load and if I click OK it keeps comming back .. forcing me to reset the PC.
So the errormessage might not be exact word for word.. that just what i remember it said.

Sorry for the error of the last build. It may case by installing msvcr8.dll on system32 directory on the last build. The lastest build will not do that since those dll were compiled with gcc and will not depended on msvcr8.dll.
Please try the lastest build upstair.

[CeeJay.dk]

I tried the new SSE build .. it crashes MPC when I try to watch something.

It also crashed on me , when I click the Output or the Grab or the section of the Video decoder configuration of ffdshow

[LoRd_MuldeR]

Updated 02-02-2006, all the DLL outside the ffdshow directory, libavcodec & libmplayer are compiled with gcc.
SSE2 Build (icl 9, msvc 8, gcc): http://rapidshare.de/files/12367610/ffdshow-20060202-icc-sse2.exe.html
SSE Build (icl 9, msvc 8, gcc): http://rapidshare.de/files/12373865/ffdshow-20060202-icc-sse.exe.html

I tried the latest SSE build (ffdshow-20060202-icc-sse.exe) and it seems to work fine so far ;)

Nevertheless it seems to me like the audio quality is not as good as in the older builds. Anyone noticed the same? (I noticed it with MP3 audio in an AVI video)

[issa]

I tried the new SSE build .. it crashes MPC when I try to watch something.

It also crashed on me , when I click the Output or the Grab or the section of the Video decoder configuration of ffdshow
What's kind of cpu you have?

[LoRd_MuldeR]

What's kind of cpu you have?

I was able to reproduce the crash CeeJay.dk mentioned :(
CPU is AMD AthlonXP (Barton)

[issa]

Fixed SSE CVS Build.

SSE Build (icl 9 + msvc 8, gcc): http://rapidshare.de/files/12429903/ffdshow-20060203-icc-sse.exe.html

I did tried the installation. It's seem to work fine, no crash. Sorry for the problem on last build.

[CeeJay.dk]

I was able to reproduce the crash CeeJay.dk mentioned :(
CPU is AMD AthlonXP (Barton)

So is mine.

[CeeJay.dk]

I found out why it crashes (still does) MPC.
I had decoder set to XVID because IMO the quality of the decoder is better (better deblocking mostly)
When I set it to libavcodec it does not crash MPC.

The Xvid decoder relies on the xvid.dll .. might be a problem with your compile of that dll.

[CeeJay.dk]

I've tested all the options briefly and I can only find one other flaw besides Xvid not working.

When you rightclick in the codecs list in the audio configuration you get the option of setting all to either libavcodec or Xvid .. which is probably carried over from the video configuration.
That's hardly a fault with the compile but rather with the sourcecode.

[Matrics]

A suggestion: issa, can you change the version from title bar? I'm still seeing ffdshow-20051216...

[LoRd_MuldeR]

Fixed SSE CVS Build.

SSE Build (icl 9 + msvc 8, gcc): http://rapidshare.de/files/12429903/ffdshow-20060203-icc-sse.exe.html

I did tried the installation. It's seem to work fine, no crash. Sorry for the problem on last build.

THX :D

[issa]

I found out why it crashes (still does) MPC.
I had decoder set to XVID because IMO the quality of the decoder is better (better deblocking mostly)
When I set it to libavcodec it does not crash MPC.

The Xvid decoder relies on the xvid.dll .. might be a problem with your compile of that dll.

Did you try my latest build? The xvidcore.dll were compiled by msvc 8 with SSE support turn on.
I did try it on my computer; however, I didn't get crash. I am using intel cpu.

[clsid]

Why include xvidcore.dll with ffdshow? If someone wants to encode with XviD, they should install XviD. And for playing XviD with ffdshow that dll isn't needed.

[issa]

Why include xvidcore.dll with ffdshow? If someone wants to encode with XviD, they should install XviD. And for playing XviD with ffdshow that dll isn't needed.
I would like to know how many people want my build to include xvidcore.dll with ffdshow installation. If there is more people perfer not to include/should not include it, I will remove it on next build.

[hellfred]

I would like to know how many people want my build to include xvidcore.dll with ffdshow installation. If there is more people perfer not to include/should not include it, I will remove it on next build.
I do not need xvidcore.dll. I usually dl latest xvid binaries from Koepi and ffdshow usually finds those. Therefor I prefere to just have one dll and smaller download size for ffdshow.

Hellfred

[CeeJay.dk]

Did you try my latest build? The xvidcore.dll were compiled by msvc 8 with SSE support turn on.
I did try it on my computer; however, I didn't get crash. I am using intel cpu.

Yes , I used ffdshow-20060203-icc-sse.exe.

clsid : AFAIK It is needed if you want to decode with Xvid's decoder instead of libavcodecs .. and Xvid's decoder provides better quality.

But as hellfred just let us know if the xvidcore.dll is not included ffdshow will find xvidcore elsewhere on the computer (like in my Xvid directory).
I just tried this and it works .. the koepi xvid i'm using causes no problems.

Issa , I think you should leave xvidcore.dll out .. ffdshow will find and use xvid if its installed anyways.
If you want to compile xvid then just make a xvid install just like koepi's.

[Lemonzest]

what's the latest build? i cant dl from any of the rapidshare mirrors, and would like a updated gcc sse2 build. thanks.

Lemonzest

[LoRd_MuldeR]

what's the latest build? i cant dl from any of the rapidshare mirrors, and would like a updated gcc sse2 build. thanks.

Lemonzest

What's the problem with downloading from Rapidshare ???

[Lemonzest]

well my isp uses transparent proxies and it says I'm already downloading a file, or i have reached my limit.

[japouleripeur]

well my isp uses transparent proxies and it says I'm already downloading a file, or i have reached my limit.Yes, same problem for me too:confused:

[LoRd_MuldeR]

To the guys that can't download from Rapidshare:

Try Free-Codecs mirror. Up-to-date builds can be found there:
http://www.free-codecs.com/FFDShow_download.htm

[Lemonzest]

thanks, any word on the gcc sse2 builds?

[canuckerfan]

^my comp likes gcc better than icc:rolleyes:

[Lemonzest]

for gcc builds i sometimes get them off here

http://kurosu.free.fr/ffdshow.htm

last few builds used to crash when using the OSD but todays seams fine (using the P4 sse/sse2 Build)

[B.F.]

I would like to know how many people want my build to include xvidcore.dll with ffdshow installation. If there is more people perfer not to include/should not include it, I will remove it on next build.
I don't need a xvidcore.dll in ffdshow.
Actually I also don't need any encoding dll's in ffdshow.

[cc979]

@bob0r

i just successfully built gcc-4.1 if ya interested, i can you the binaries - if it fails for you

[LoRd_MuldeR]

But as hellfred just let us know if the xvidcore.dll is not included ffdshow will find xvidcore elsewhere on the computer (like in my Xvid directory).
Issa , I think you should leave xvidcore.dll out .. ffdshow will find and use xvid if its installed anyways.
If you want to compile xvid then just make a xvid install just like koepi's.

I don't have XviD installed, since I now use x264 for encoding. Therefore I think it's fine to have the xvidcore.dll included with ffdshow :D
If there is any problem on compiling that DLL, why not just include the one form koepi's build?

[cc979]

@bob0r

been doing some compiling, `WAVE_FORMAT_DSPGROUP_TRUESPEECH' still broke for me

baseclasses not build using gcc-4+, but builds fine with gcc-3.4.5

any ideas ?

[issa]

@bob0r

i just successfully built gcc-4.1 if ya interested, i can you the binaries - if it fails for you

Did you get all the DLLs compiled with gcc-4.1?

Also, I will remove xvidcore.dll from my build. Sorry for any trouble causing from my build.

[Eretria-chan]

I noticed in ffdshow-20060203-icc-sse.exe, that SPP Deblocking is still extremely slow and my computer is barely able to handle fast spp deblocking. In windowed, it's fine - but in fullscreen is another matter. It starts to lag sometimes.
But I suppose this might be due to my weak Athlon XP 2800+ processor and my high resolution of 1920 x 1200.

[CeeJay.dk]

SPP deblocking have always been slow, the new fast SPP is a welcome addition but as you point out, even that can be too much for some computers.

My 2500+ barton is clocked at 2170 mhz and while it can just handle fast SPP i don't use it, because I prefer having noise- and sharpening- filters on instead.

This is hardly a problem with the compile - To make real improvement in this area, someone will have to code a faster SPP deblocker.

[Eretria-chan]

I know it probably isn't a problem with the compile, but worth mentioning anyway...
There are always faster methods to do something, although it doesn't mean it has to be implemted...

[cc979]

Did you get all the DLLs compiled with gcc-4.1?

Also, I will remove xvidcore.dll from my build. Sorry for any trouble causing from my build.

not had much look, does not even compile the baseclasses

[bob0r]

I have not tried gcc 4.1 pre builds yet.
Simple compile guide for mingw should be enough.

I myself are clueless about fixes.
Just report the bugs like you did on sourceforge and hope Milan awakens soon :)

Untill Milan is active, i will just lay low and use CoreAVC for h264 decoding :p

[cc979]

I have not tried gcc 4.1 pre builds yet.
Simple compile guide for mingw should be enough.

I myself are clueless about fixes.
Just report the bugs like you did on sourceforge and hope Milan awakens soon :)

Untill Milan is active, i will just lay low and use CoreAVC for h264 decoding :p

i've just tried cygwin - gcc 4.1 compiles no problems at all
seems fairly fast compiling to, been trying to do a cross-platform compile using cygwin to make gcc for mingw - no luck yet

also been trying to compile ffdshow with cygwin gcc 4.1 - cant compile the baseclasses not sure how to get the dxsdk into cygwin any hints ?

[cc979]

@bob0r could you upload the last compile-able version of ffdshow i've lost it, thanks

[cc979]

@bob0r could you upload the last compile-able version of ffdshow i've lost it, thanks

think i managed to get it of cvs

[max-holz]

I got this error during compilation of gcc-4.1

gcc.exe: gengtype-lex.c: No such file or directory
gcc.exe: no input files
make[2]: *** [build/gengtype-lex.o] Error 1
make[2]: Leaving directory `/home/Proprietario/gcc-4.1/obj/gcc'
make[1]: *** [stage1_build] Error 2
make[1]: Leaving directory `/home/Proprietario/gcc-4.1/obj/gcc'
make: *** [bootstrap] Error 2

Anyone could help me? I downloaded the source from an official mirror, so I don't understand

[cc979]

I got this error during compilation of gcc-4.1

gcc.exe: gengtype-lex.c: No such file or directory
gcc.exe: no input files
make[2]: *** [build/gengtype-lex.o] Error 1
make[2]: Leaving directory `/home/Proprietario/gcc-4.1/obj/gcc'
make[1]: *** [stage1_build] Error 2
make[1]: Leaving directory `/home/Proprietario/gcc-4.1/obj/gcc'
make: *** [bootstrap] Error 2

Anyone could help me? I downloaded the source from an official mirror, so I don't understand

you will probably need bison and flex

[Egh]

I used Kuroshi last build (for Athlon, SSE1 only) but recently discovered problems with it. It's either build problem or new updated libavcodec in ffdshow. Essentially the problem is some kind of instability in ffdshow audio playing vorbis streams. E.g. using new (test) haali splitter causes crashes in ffdshow audio (more accurately libavcodec.dll) on playback. Haali checked it out and seems it's not his problem :) Works with 20060130 splitter but fails with new version, though he didn't change anything which could cause crashes since last official build.

So has anybody did any other successful builds of ffdshow in january? Would like to try them to see if it's really libavcodec probem (and need to be reported) or just particual build.

[B.F.]

I used Kuroshi last build (for Athlon, SSE1 only) but recently discovered problems with it. It's either build problem or new updated libavcodec in ffdshow. Essentially the problem is some kind of instability in ffdshow audio playing vorbis streams. E.g. using new (test) haali splitter causes crashes in ffdshow audio (more accurately libavcodec.dll) on playback. Haali checked it out and seems it's not his problem :) Works with 20060130 splitter but fails with new version, though he didn't change anything which could cause crashes since last official build.

So has anybody did any other successful builds of ffdshow in january? Would like to try them to see if it's really libavcodec probem (and need to be reported) or just particual build.

Same here.
Some Vorbus steam in video cause player to crash.
Also instead first keyframe in XviD video I see the second one two times.

[bob0r]

I got this error during compilation of gcc-4.1

gcc.exe: gengtype-lex.c: No such file or directory
gcc.exe: no input files
make[2]: *** [build/gengtype-lex.o] Error 1
make[2]: Leaving directory `/home/Proprietario/gcc-4.1/obj/gcc'
make[1]: *** [stage1_build] Error 2
make[1]: Leaving directory `/home/Proprietario/gcc-4.1/obj/gcc'
make: *** [bootstrap] Error 2

Anyone could help me? I downloaded the source from an official mirror, so I don't understand

GCC snapshots are not made mingw ready.

However
ftp://ftp.nluug.nl/mirror/languages/gcc/prerelease-4.1.0-20060219/gcc-core-4.1.0-20060219.tar.gz
ftp://ftp.nluug.nl/mirror/languages/gcc/prerelease-4.1.0-20060219/gcc-g++-4.1.0-20060219.tar.gz
Are made mingw ready, i was able to compile this gcc 4.1.0-20060219 with gcc 4.0.2 (which already compiled on mingw)

So i guess soon we can expect gcc 4.1.0, hopefully not before i have submitted a x264 make fprofiled bug :P

For cygwin i don't know

[Egh]

Same here.
Some Vorbus steam in video cause player to crash.
Also instead first keyframe in XviD video I see the second one two times.

Using kuroshi builds? Btw i didn't found difference in that aspect between athlon and p3 (both sse1 only) builds. I tried both 0123 and 0127 (which are the last builds after Milan dissappeared :)

[zambelli]

Here's an interesting bug I found in ffdshow:

If you use the Avisynth module in ffdshow to resize the video (i.e. Lanczos(width*2, height*2) ), ffdshow never updates the video image size when passing it on to the renderer. If using a player such as WMP or MCE, this will result in the video being zoomed in and cropped off because the renderer will perform scaling based on the old video resolution.

Can someone else try to repro this? Just enable the Avisynth module, enter "Lanczos(width*2, height*2)" as the script, and then play some video handled by ffdshow in WMP10.

[Egh]

Here's an interesting bug I found in ffdshow:

If you use the Avisynth module in ffdshow to resize the video (i.e. Lanczos(width*2, height*2) ), ffdshow never updates the video image size when passing it on to the renderer. If using a player such as WMP or MCE, this will result in the video being zoomed in and cropped off because the renderer will perform scaling based on the old video resolution.

Can someone else try to repro this? Just enable the Avisynth module, enter "Lanczos(width*2, height*2)" as the script, and then play some video handled by ffdshow in WMP10.

Hmm, first, not Lanczos but LanczosResize :) Second, yeah i've seen some difficulties on resizing. But which build are you using? In my case the video stops updating after resizing turned on. (But size displayed is correct one). Though it's usually enough to seek video to bring back video updating. Btw problems with updating occur here with MPC (don't use other players :P).

[zambelli]

Hmm, first, not Lanczos but LanczosResize :) Second, yeah i've seen some difficulties on resizing. But which build are you using? In my case the video stops updating after resizing turned on. (But size displayed is correct one). Though it's usually enough to seek video to bring back video updating. Btw problems with updating occur here with MPC (don't use other players :P).
Sorry, I was writing shorthand... :)
MPC is one of the few players that handles this well, actually, although I can see it update its window size once the first video frames are decoded. It starts with a video window size based on the input file information, but then it adjusts its window size to whatever the correct processed video size is.

The reason why I'm pretty sure this is a bug in ffdshow and not in the players is because ffdshow's own internal resizing module doesn't exhibit these problems. I haven't checked ffdshow's code, but I'm guessing the image properties are updated throughout the pipeline. Apparently, however, this doesn't happen after the video goes through Avisynth.

[vlada]

There is also a bug in FFDShow when switching on and off 4:3 letterboxing in subtitle filter. It is especially annoying, if FFDShow doesen't load subtitles automatically. If I load them manually then, the video gets vertically squeezed, because neither MPC nor ViPlay will resize their window to 4:3 AR.

Also how far is implementing of DVD support? Is Milan working on it? Does anybody know?

[LoRd_MuldeR]

Also how far is implementing of DVD support? Is Milan working on it? Does anybody know?

ffdshow already decodes MPEG-2 as well as AC3/DTS
What more is needed ???

[Zarxrax]

There is a bug in ffdshow, such that when it is installed, it causes Adobe After Effects to crash when starting! If you select to install the Video For Windows part as well, then it fixes the problem.

I'm not sure why installing the VFW part would fix it, but regardless of that, ffdshow shouldn't be causing applications to crash!

Could anyone check into this?

[vlada]

ffdshow already decodes MPEG-2 as well as AC3/DTS
What more is needed ???

It is impossible to use FFDShow as a DVD decoder. Try it yourself. The DVD graph is not a common DirectShow graph.

Choose Video decoder -> Codecs -> MPEG-2. In the right bottom corner you'll see a checkbox saying DVD decoding (not working yet). I talked about supporting DVD decoding with Milan before. He said he'll be working on it, but unfortunately it is still not finished yet.

[LoRd_MuldeR]

It is impossible to use FFDShow as a DVD decoder. Try it yourself. The DVD graph is not a common DirectShow graph.

Choose Video decoder -> Codecs -> MPEG-2. In the right bottom corner you'll see a checkbox saying DVD decoding (not working yet). I talked about supporting DVD decoding with Milan before. He said he'll be working on it, but unfortunately it is still not finished yet.

I can playback Video-DVDs in MPC using ffdshow for audio- and video decoding...

[zambelli]

I can playback Video-DVDs in MPC using ffdshow for audio- and video decoding...
That's probably because MPC is using internal DVD source filters.
I'm pretty sure ffdshow can't handle CSS either.

[vlada]

Yes, it works in MPC, but it crashes ViPlay. But now I tried it in WMP10 and it works. But with libav as MPEG-2 decoder, the video is very choppy (uses 50% CPU). Mith libmpeg2 it plays smoothly, but still uses 50% CPU on my AthlonXP 2000+. But VLC is using 50% CPU with this DVD too, don't know why.
So it seems the DVD support is already implemented, but not very stable. ViPlay works flawlessly with other DVD (MPEG-2) decoders, my default is DScaler 5.

[bob0r]

February 28, 2006
GCC 4.1.0 has been released.

ftp://ftp.nluug.nl/mirror/languages/gcc/releases/gcc-4.1.0/

Let the testing/support begin! :D

[Poutnik]

What FDDShow release/binary would you recommend me for W2000 SP4 + SSE Celeron 600 ?

I have recently installed on my PC W2k SP4. I tried to see a movie - at that time it was still standard mplayer2 v 6.4 (Now using my favorites MV2playerRC2 or MPClassic ).
Without FFDShow it played fine.
I had fresh install of one of latest FFDShow ( not remember time of release yet, downloaded more ones ) with default settings. I thought W2K is said to be the most stable MS OS, but it caused complete OS freezing.
I know FFDShow is known to cause trouble at some circumstances so I would like you to advice me a good one by your experience.

[Egh]

What FDDShow release/binary would you recommend me for W2000 SP4 + SSE Celeron 600 ?

I had fresh install of one of latest FFDShow ( not remember time of release yet, downloaded more ones ) with default settings. I thought W2K is said to be the most stable MS OS, but it caused complete OS freezing.


You could easily have some odd build. Try using last "official" one i.e. Milan's build from sf.net. and try kuroshi P3 build as well.

[Lemonzest]

how about a cvs all gcc 4.1.0 sse/sse2 build?

[Egh]

how about a cvs all gcc 4.1.0 sse/sse2 build?

I second :goodpost:

But anyhow cvs is last updated on 27th of jan, so no newer features or bugfixes since then.

[cc979]

I second :goodpost:

But anyhow cvs is last updated on 27th of jan, so no newer features or bugfixes since then.

just tried with gcc 4.1, no luck with the latest ffdshow cvs

[LoRd_MuldeR]

ffdshow 2006-02-26 (http://www.free-codecs.com/FFDShow_download.htm) ;)

[clsid]

ffdshow 2006-02-26 (http://www.free-codecs.com/FFDShow_download.htm) ;)
The free-codecs .com site is bullshit. Latest CVS is still January 27th.

[Egh]

The free-codecs .com site is bullshit. Latest CVS is still January 27th.


they apparently put date of build, not actual CVS snapshot used, which is quite sensible.

[ExtraEye]

ah this build isn't very good for me. makes video and audio desync when i use the settings i had on my former ffdshow build.

[Matrics]

What you say about this page (http://www.paehl.com/open_source/?FFDSHOW)?

[LoRd_MuldeR]

ah this build isn't very good for me. makes video and audio desync when i use the settings i had on my former ffdshow build.

Same problem here!

Must be the audio-part of ffdshow that is broken in 2006-02-26 build, since I used CoreAVC for the video decoding...

[cc979]

looks like there's an offical pre-release of gcc 4.0.3 going to try it

http://www.mirrorservice.org/sites/sources.redhat.com/pub/gcc/prerelease-4.0.3-20060303/

[haruhiko_yamagata]

Hello. This is my first post.

For multi-thread resize support, I have modified swscaler(libmplayer.dll). Thank you for your thread I have succesfully build libmplayer.dll. It is fast on dual-core and working as far. I sent E-mail to milan cutka a week ago, waiting for the reply.

I tryed to build ffdshow.ax with MSVC 2003.net standard edition, which does not support optimize, I got very slow one. It looks like "simd.h" is not compiled to MMX.

Next, I tryed with GCC-MingW 4.0.3. ,which was very complicated thing to me. I struggled for 3 days and finaly got ffdshow.ax. It was faster than the former, but slower than build20051221 or 20060202.

Would you kindly guide me how to compile better?

I installed Direct X SDK February 2006. I am a beginner to MSYS+MingW.

My environmental variable
C_INCLUDE_PATH = CPLUS_INCLUDE_PATH = C:\mingw\include;C:\Program Files\Microsoft DirectX SDK (February 2006)\Include;C:\Program Files\Microsoft Visual Studio .NET 2003\Vc7\PlatformSDK\Include
LIBRARY_PATH = C:\MingW\lib;C:\Program Files\Microsoft DirectX SDK (February 2006)\Lib\x86

Please help me.

[cc979]

check this guide: compiling gcc for mingw

http://forum.doom9.org/showthread.php?t=108215

[haruhiko_yamagata]

Thank you, cc979.
I updated include files and library.

Oh shit, wrong was my code. I modified the code meant to be faster, but it was much slower. I could compile Original code poroperly, I think.

Thank you.

[zilexa]

I recently discovered FFDShow makes my CPU go crazy when playing 720p XViD, while I can play it
normally with lower CPU load using XViD codec instead of FFDShow.

Howcome?
Could this be because I use the version from the official ffdshow site (dates nov 2005)?
http://ffdshow.sourceforge.net/tikiwiki/tiki-index.php?page=Getting+ffdshow
So perhaps using a cpu specific build would solve it?

[Egh]

I recently discovered FFDShow makes my CPU go crazy when playing 720p XViD, while I can play it
normally with lower CPU load using XViD codec instead of FFDShow.

Howcome?
Could this be because I use the version from the official ffdshow site (dates nov 2005)?
http://ffdshow.sourceforge.net/tikiwiki/tiki-index.php?page=Getting+ffdshow
So perhaps using a cpu specific build would solve it?

I play 720p XVid very nicely indeed, and in VMR9 and RGB32.
CPU is only 1800+ XP Athlon (sse1 only build of ffdshow, of course).

[max-holz]

I have compiled succesfully but I have problem during the creation of the installer, I got error from line 981 of the ffdshow.nsis2:

!insertmacro testcodec ffdshow XVID ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow DIV3 ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow DIVX ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow DX50 ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow MP43 ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow MP42 ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow MP41 ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow WMV1 ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow wmv2 ${IDFF_MOVIE_LAVC} 0
!insertmacro testcodec ffdshow _3iv ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow mpg1 ${IDFF_MOVIE_LAVC} 0
!insertmacro testcodec ffdshow mpg2 ${IDFF_MOVIE_LAVC} 0
!insertmacro testcodec ffdshow h263 ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow h264 ${IDFF_MOVIE_LAVC} 1
!insertmacro testcodec ffdshow mjpg ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow dv ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow hfyu ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow png1 ${IDFF_MOVIE_LAVC} $R0
!insertmacro testcodec ffdshow rawv 1 0

[Romario]

What's going on, no FFDSHOW new build for a long time.

[sillKotscha]

What's going on, no FFDSHOW new build for a long time.

hey Dude,

I haven't read all of your posts but the last 2 I have seen (including this one) really sound a little bit too demanding respectively too disrespectful (http://forum.doom9.org/showpost.php?p=802855&postcount=4).

just my 0,02 €

[celtic_druid]

Must be related to Kostarum Rex Persia or something.

[Egh]

still wondering where Milan dissappeared ^^

[videomixer9]

prolly just not anymore in the mood for ffdshow development with all the stupid stuff about SMP etc. or busy with sth. else. Say bye bye to ffdshow, hehe :O

[Egh]

prolly just not anymore in the mood for ffdshow development with all the stupid stuff about SMP etc. or busy with sth. else. Say bye bye to ffdshow, hehe :O

bye to development, but not to ffdshow itself. I don't think in the nearest future anything superior to it appears. Yeah, CoreAVC might be better than ffdshow already, but that's commercial anyway. In terms of free software ffdshow might be on top for months to come.

[Liisachan]

Exactly, it was weird to see ffdshow was not in the list in the voting for 2006 Community Choice Awards.
maybe I will make some donation for ffdshow too anyway.
(I might have paid some money for CoreAVC too if it were free. It might be really nice free software...)

[ExtraEye]

does CoreAvc do everything ffdshow does? i mean post-processing, all the filters, etc?

[dk75]

No.
CoreAVC is H264 decoder only. Nothing less, nothing more.

[videomixer9]

i'd wish there was a clean directshow implementation for libavcodec, ffdshow has all the stupid junkfilters from mplayer and more implemented that are totally useless to me or maybe proper SSA/ASS support for libav players and them not just stopping playback when moving the window, having major crappy GUIs or other annoyances?

[Romario]

i'd wish there was a clean directshow implementation for libavcodec, ffdshow has all the stupid junkfilters from mplayer and more implemented that are totally useless to me or maybe proper SSA/ASS support for libav players and them not just stopping playback when moving the window, having major crappy GUIs or other annoyances?

Well, ffdshow certainly have few bugs, but I like ffdshow very much.:) Full directshow implementation, however, sounds very good. But, we will see what Milan thinks about that.:(

[Egh]

i'd wish there was a clean directshow implementation for libavcodec, ffdshow has all the stupid junkfilters from mplayer and more implemented that are totally useless to me or maybe proper SSA/ASS support for libav players and them not just stopping playback when moving the window, having major crappy GUIs or other annoyances?

most filters in ffdshow are useless for most indeed, but you can just disable all of them. Lack of proper styled subs support is more concerning, I agree.

[bob0r]

19:34 yesterday ffdshow Commit by milan_cutka :: ffdshow/src/codecs/wmv9/ff_wmv9.vcproj:
don't require strmiids.lib #
18:32 yesterday ffdshow Commit by milan_cutka :: ffdshow/ (34 files in 14 dirs):
few 64-bit fixes #
18:25 yesterday ffdshow Commit by milan_cutka :: ffdshow/src/simd.h:
updated libmpeg2 #
18:22 yesterday ffdshow Commit by milan_cutka :: ffdshow/src/codecs/libmpeg2/libmpeg2/ (4 files):
updated libmpeg2

I wonder if he will look at the old bug reports...

[breez]

ff_wmv9.vcproj

There is a WMV9 decoder in works? Or what is it?

[clsid]

There is a WMV9 decoder in works? Or what is it?I think it is used for encoding WMV9.

[videomixer9]

there is a wmv9 decoder in ffdshow even for longer now, in the registry add under HKEY_CURRENT_USER\Software\GNU\ffdshow a value type REG_DWORD named "wmv3" and set it 1 to enable ffdshow wmv3 decoding, this worked for me with older versions. ffmpeg's wmv9 decoder isn't called wmv3 or wmv9 in their sourcetree, it was either vc1 or vc9 ...

[Shapierian]

I noticed that ffdshow still seems to build if you replace h264.c, h264idct.c and h264dsp_mmx.c with files from the latest ffmpeg. This may be useful for people looking for a quick update to ffdshow.

[dk75]

Or they could look at http://www.paehl.com/open_source/?FFDSHOW

[videomixer9]

prolly has only the updates from last cvs changes though, and that was no update to the libavcodec code. But beware, there is also h264 assembler code, last time I tried replacing the stuff it didn't work smoothly, but well maybe i should try again

I still wonder about the text describing ffdshow though nowadays ... DivX 6.1 decoder without postprocessing doesn't really use more cpu nowadays for me at least than ffdshow and also supports smp ...

[max-holz]

Where is possible to find a good tutorial how to build ffdshow and create the installer? Like the one that is present in the x264 compilation thread for eample.

[videomixer9]

I tried updating the libavcodec code from ffdshow with new one from ffmpeg project and there were still a lot of problems, I then replaced quite some part in the ffmpeg code and used the ffdshow makefile for it with the original ffmpeg but there were always many linker errors in the end with missing links. Even just replacing the few parts about h264 failed.

Whatever I found out that GCC 4.1.1 produced a libavcodec that had a bit better performance than my old one ... lousy result of trying for hours how to fix up libavcodec for ffdshow ... but you also have to add skals code etc.

[Negi]

Is milan the only coder on ffdshow?

[videomixer9]

yes ... and he even though it is open source there seem to be some mysterious about some things, like there could be a patch or batches or notes about what to do to e.g. modify libavcodec ... but it's not like open source would mean that you get insides on anything without trying to figure it out all by yourself really. After all open source is usually well known for bad documentation upto none at all.

[BoNz1]

CVS looks abandoned to me. However, there is some recent activity in the SVN (http://svn.sourceforge.net/viewcvs.cgi/ffdshow/)

[Liisachan]

@BoNz1
Right... and what was stopped at Revision 2476 (http://svn.sourceforge.net/viewcvs.cgi/ffdshow?rev=2476&sortby=date&view=rev) @ Jan 27 2006 restarted on Mar 29 2006 as Revision 2477 (http://svn.sourceforge.net/viewcvs.cgi/ffdshow?rev=2477&sortby=date&view=rev), and since then updates have been quite frequent, several times a day; as of writing this, the newest one is Revision 2496 (http://svn.sourceforge.net/viewcvs.cgi/ffdshow?rev=2496&sortby=date&view=rev)

[Egh]

@BoNz1
and since then updates have been quite frequent, several times a day; as of writing this, the newest one is Revision 2496 (http://svn.sourceforge.net/viewcvs.cgi/ffdshow?rev=2496&sortby=date&view=rev)

yet that last one happened about 16h ago already.

Plus most those revisions are quite marginal, mostly dealing with x64bit tweaks (changing data format definitions and such). Seems Milan was able to make ffvdub and ffvwf 64bit as well :) And, of course, updating libavcodec and libmpeg.

So can anyone try to build the project now? :P

[Liisachan]

yet that last one happened about 16h ago already. Come on, milan has to sleep and eat and do real life too... When I said "updated several times a day" I didn't mean "evey 3 horus"...

[Negi]

So milan took a 2-month leave? Good thing he's back on the job. I was getting worried that he'd abandon FFDSHOW. Thanks milan!

[PatchWorKs]

So milan took a 2-month leave? Good thing he's back on the job. I was getting worried that he'd abandon FFDSHOW. Thanks milan!

I can't see the problem: isn't ffdshow open source ?
So, everyone can update it... :goodpost:

[Negi]

Anyone can contribute to the CVS but an admin (milan) has to commit the patches. If someone wanted to continue it without milan, they'd have to fork it. I think. I could be wrong.

[Egh]

Come on, milan has to sleep and eat and do real life too... When I said "updated several times a day" I didn't mean "evey 3 horus"...

Yeah, I understand that. But, last change atm is 37 hours ago :) I was merely wondering if milan came back for a long time or just to do some quick fixes :)

[Liisachan]

i bet 'for a long time'
svn was updated again anyway (twice as of now)

[videomixer9]

updated libmpeg2 to 0.5.0, the rest is vs2005 project changes.

is there a page on CIA for SVN projects?

[Liisachan]

http://ffdshow.faireal.net/bakaupdates.php
^^;

[LoRd_MuldeR]

updated libmpeg2 to 0.5.0, the rest is vs2005 project changes.

is there a page on CIA for SVN projects?

This (http://cia.navi.cx/stats/project/ffdshow) one ???

[videomixer9]

amazing, suddenly also svn changes appear ... ? ... checked a while ago and it did show only cvs

[Liisachan]

true, that changelog had been stopped & I un-bookmarked it a few days ago. then it is alive again...

[Egh]

Good build :approved:

http://kurosu.free.fr/ffdshow-20060403-23H17-athlon.exe

Kuroshu is superior again:0 So far I don't have any problems with it (and yes, it uses latest SVN)

(well latest for 3rd april of course :P I think upto 2496 revision is included)

[Shapierian]

I noticed that for the Makefile based build systems -msse2 seems to be on by default. This seems kind of dangerous to me considering Athlon-xp doesn't support sse2. Am i missing anything?
Also why is -msse on but it is set to -march=i586? I don't thunk there are any 586 class cpus that support sse?

[videomixer9]

I usually replace this and used this here:

-D__SSE__ -D__MMX__ -O3 -ftree-vectorize -ffast-math -funsafe-math-optimizations -msse -m3dnow -mmmx -mfpmath=sse -march=athlon-xp -mtune=athlon-xp -fomit-frame-pointer -finline-functions -finline

and for others i get rid of the sse2 part as it otherwise would crash. Dunno about other ppls libavcodec etc. compiles. Dunno what milan thinks with those settings, but I think that it's made for general public, and the optimizations are runtime detected mostly anyways.

[Egh]

Updated build :approved:

Kuroshu r00x0r :P

http://kurosu.free.fr/ffdshow-20060405-21H28H42-athlon.exe

[Kostarum Rex Persia]

Updated build :approved:

Kuroshu r00x0r :P

http://kurosu.free.fr/ffdshow-20060405-21H28H42-athlon.exe

Please, post change log for that build.

[Egh]

Please, post change log for that build.

upto ffdshow SVN rev 2501, which is just one before the current one (2502, which apparently added "20 bit lpcm support" :confused:).

So, if you need, check http://cia.navi.cx/stats/project/ffdshow and see the entry named "new vs2005 projects configurations", that's the revision which was used for that last atm Kuroshu build.

P3 build (no SSE2 (?)) http://kurosu.free.fr/ffdshow-20060405-21H28H42-P3.exe

Athlon64 build (gcc+sse1+sse2)
http://kurosu.free.fr/ffdshow-20060405-21H28H42-k8.exe

[max-holz]

Is there any updated build for P4?

[videomixer9]

is this a game like "whoever has the most stupid question wins?" or are you ppl just way too lazy to just visit http://kurosu.free.fr/ffdshow.htm and check for yourself. Some people are just too much for me ...

[magnusr]

My athlon 64 x2 4200+ supports sse3 also.

But im sticking with the k8 version. Waiting for a athlon64 version with sse3 also then ^^.

[videomixer9]

libavcodec nor any of the other libs have sse3 optimizations, autovectorization in gcc is shit and most libs are written for it, so using intel compiler for the libs will give you autovectorization but make you lose most other optimizations. I guess you will see SSE3 once ffmpeg implements it fully, not much of a speedup though as it seems and mostly wasted time.

[madman1980]

Yeah, from what I've read there seems to be little if nothing to gain from SSE3 in this department..

[Egh]

atm iirc there's no point in SSE3 support in ffdshow.

I know very few filters in AVS which actually use SSE3. The difference between SSE1 and SSE2 is quite huge in terms of video processing, but SSE3 can hardly add much gain to that, imo.

[Kostarum Rex Persia]

Commit by milan_cutka :: r2504 /trunk/ (120 files in 22 dirs):

64-bit strides in x64 build #

Commit by milan_cutka :: r2503 /trunk/ (7 files in 4 dirs):

add "release unicode" configuration to vs 2003 solution,
compute ac3 and dts average bps only from valid frames


Very good, finally he add necessary 64-bit strides. So, someone can compile new build today.

[foxyshadis]

magnusr, even if ffdshow was optimized for sse3, a64 still executes sse2/3 as (roughly) multiple sse instructions anyway. :p

[videomixer9]

SSE, SSE2 and SSE3 are lousy instruction sets to begin with, the vector units in most x86 cpus are quite much of a joke. powerpc or ultrasparc have far better vector units ...

[Rash]

Humm, I think they've changed that. Apple showed some benchmarks and it seems P4 is really faster than PowerPC nowadays.

[foxyshadis]

Anyone can make up benchmarks that show that one processor is better than another, or vice versa, depending on what tests you choose or how optimized the software (and underlying libraries) is for the instruction sets. That has nothing to do with the sanity and ease of programming for them, and x86 is a boar. Even x64 is only marginally better (much less named register pressure).

[ckjnigel]

I just watched a high quality XviD encode with
ffdshow-20060407-19H04-k8.exe
and that's the first time ever that ffdshow rendering seemed a tad better than what I get using whatever is Koepi's latest XviD.
I'm on Win x64 -- a 3.3+ GHz 64-bit-enabled Sempron with an nVidia 6600GT.
Thanks, kurosu!

[videomixer9]

first ... 3300+ doesn't mean the CPU has 3.3 GHZ, not even near. Second, that build is for Athlon 64, but it's no x64 compile.

[Shirokuu]

Vm9, he just means he has a 3.3 Ghz EQUIVALENT Sempron. No need to moan about that.

[videomixer9]

not sure about that, plenty of people think the ratings are equal to mhz. Also the rating was measured in comparision to other CPUs performances, for a while it was the thunderbird, I think AMD changed that though. So saying it's a 3.3 ghz equal, and especially as it is a Sempron I wouldn't count on that comparing to Intel CPUs e.g.

[LigH]

:(

Most happy people out there get their well optimized build for their current CPUs.

Only very few have to stick with older, but not too optimized (or "generic") builds (having a CPU that supports MMX and ISSE only) until milan may publish the next one...
__

But in general, good to see that some people are interested in keeping this project up-to-date. Well done, go on!

:thanks:

[videomixer9]

there are plenty of builds optimized for MMX and SSE only, many of them though also for 3dNow! as Athlon-XP e.g. has MMX/SSE and 3dnow. And if you got an Intel CPU you're alone out there as most it seems that most computer nerds and encoding folks have AMD. At least in my little world of people not buying crappy complete PCs from Intel-blackmailed companies like Dell. Other than that P3 optimized builds should also have SSE ...

Besides clsid's build of rev 2405 should be MMX only. Nothing changed since then. We're still at rv. 2405, and if you don't even have MMX I don't think you have a PC that can really decode modern videos ...

Optimization is kinda useless anyways as GCC doesn't do much of a vectorization, it's just hand optimized code in libavcodec that can adjust to anything really. So optimized builds for any CPU are kinda overkill anyways. Most speed is gained simply by using compiler that produce better code. Thus newer GCC usually mean a tad more speed, only Intel Compiler can do proper and good vectorization, GCC verbose mode on this still finds way less parts that can be optimized. ICL also has the bonus that it can produce code optimized for any CPU into a single binary. You just have to enable them all with /Qax, I'm still waiting for GCC to do this, along with automatic vectorization you would have larger ffdshow binaries, but universal ones. However, libavcodec can make use of all units like SSE/SSE2/MMX etc. without optimization switches while compiling, just like mplayer can with automatic detection (the reason to optimize is that there's other things than just libavcodec involved).

[celtic_druid]

Actually on my poll currently it is
Intel Pentium 4: 655
AMD Athlon-XP/MP: 326
AMD K8 (Opteron, Athlon64, Athlon-fx): 301
Even if you combine the XP's and 64's the P4 is still infront.

[LoRd_MuldeR]

:(

Most happy people out there get their well optimized build for their current CPUs.

Only very few have to stick with older, but not too optimized (or "generic") builds (having a CPU that supports MMX and ISSE only) until milan may publish the next one...

There are checkboxes in ffdshow to disable CPU features like SSE2 or 3DNow!. Don't know if they have any effect in optimized builds, but have you tried to simply disable the features not supported by your CPU ???

[videomixer9]

Question is, were those poll participants just viewers or just the random turds happily buying intel as they love shitty core design. Or maybe it's just that my surrounding on irc etc. has too many AMD users ... only Intel users are usually people that buy complete boxes.

[bob0r]

filename: ffdshow-20060413-gcc4.0.3-sse-x264.nl.exe
url: http://x264.nl

sse2 version: http://files.x264.nl/ffdshow/ffdshow-20060413-gcc4.0.3-sse2-x264.nl.exe

[LoRd_MuldeR]

filename: ffdshow-20060413-gcc4.0.3-sse-x264.nl.exe
url: http://x264.nl

Seems to work fine here. Thx.

[videomixer9]

What is strange with all these compiles is the sound libs ... you can hear a difference in kurosu's athlon compile and this one with all gcc. It's seems that compiling can somehow affect decoding precision and quality, I managed to compile libmad e.g. in a way that channels got swapped O_O Or maybe i'm imagining things ...

[Negi]

filename: ffdshow-20060413-gcc4.0.3-sse-x264.nl.exe
url: http://x264.nl

sse2 version: http://files.x264.nl/ffdshow/ffdshow-20060413-gcc4.0.3-sse2-x264.nl.exe

The SSE2 version breaks on WMV3 (WM9) (decoding it or just running it through uncompressed) with vmr9 while the plain SSE build doesn't.

Athlon 64 3200+ socket 754
Windows XP SP2
DirectX9
Media Player classic with all internal filters off.

[Egh]

What is strange with all these compiles is the sound libs ... you can hear a difference in kurosu's athlon compile and this one with all gcc. It's seems that compiling can somehow affect decoding precision and quality, I managed to compile libmad e.g. in a way that channels got swapped O_O Or maybe i'm imagining things ...

first kuroshu is all gcc as well iirc, just 3 different versions used to build a whole package :P

As for sound, what sounds better and with what codecs? vorbis i suppose? tremor or libavc?

[Shapierian]

filename: ffdshow-20060413-gcc4.0.3-sse-x264.nl.exe
url: http://x264.nl

sse2 version: http://files.x264.nl/ffdshow/ffdshow-20060413-gcc4.0.3-sse2-x264.nl.exe


Are these with low accuracy tremor?

[bob0r]

@Negi
Submit bugs @ ffdshow project:
http://sourceforge.net/tracker/?group_id=53761&atid=471489

@Shapierian
Test and find out.... i compile ffdshow direct via ffdshow/src > make.
I add only directx stuff, i do not edit anything.

[celtic_druid]

Unless things have changed, you have got low accuracy. You need to edit ogg_os.h or whatever it is.

[breez]

Is there anything wrong with using libavcodec for Vorbis? Any practical difference to tremor (when not in low accuracy mode)?

[videomixer9]

iirc I wanted to fix this in ffdshow when compiling but it wasn't defined. Not sure.

[Liisachan]

Is there anything wrong with using libavcodec for Vorbis? Any practical difference to tremor (when not in low accuracy mode)?
Using livavcodec for decoding vorbis is ok.

The problem, if anything, is:
(0) Many of those evil codec packs use ffdshow to decode vorbis without thinking carefully.
(1) If vorbis decoding is enabled while installing ffdshow, the default decoder which will be used will be tremor.
(2) Playing vorbis, he or she will then feel "Hmm, Ogg Vorbis is not as high-quality as they say... it's even noisy... vorbis sucks..."

That's not "my" problem, but I don't want to see that happening.

[celtic_druid]

So perhaps changing:

!insertmacro testcodec ffdshow_audio VORBIS ${IDFF_MOVIE_TREMOR} 0

to:

!insertmacro testcodec ffdshow_audio VORBIS ${IDFF_MOVIE_LAVC} 0

would work better?

[Liisachan]

Would that change the default vorbis decoder? If so, Yes, I think so. I assume almost everyone here agrees that lavc is safer than tremor at least for now.

[dimzon]

Would that change the default vorbis decoder? If so, Yes, I think so. I assume almost everyone here agrees that lavc is safer than tremor at least for now.
Agreed 100%

[Egh]

g00d :) Milan actually now even corrects bugs from time to time:P

But i don't recall anyone recently reporting this one:

Revision 2506 - Directory Listing
Modified Tue Apr 18 19:20:55 2006 UTC (3 hours, 51 minutes ago) by milan_cutka
fixed mplayerc crash on exit when audio OSD enabled

[haruhiko_yamagata]

I uploaded a patch to support multithreading of video playback. Please test.

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

[Egh]

Lots of updates recently to ffdshow SVN, including updated x264 and libavcodec. So maybe time for a new build? :P

[bob0r]

SSE: http://x264.nl
SSE2: http://files.x264.nl/ffdshow

[Slave01]

Many people use ffdshow lanczos resize to obtain better quality expecially on great displays. However unfortunately this ffdshow resize eats great cpu so why don't put a shader window like media player classic in which insert small shaders to be executed by graphics card? Is this possible?

Greetings
John Slave

[CEC]

Many people use ffdshow lanczos resize to obtain better quality expecially on great displays. However unfortunately this ffdshow resize eats great cpu so why don't put a shader window like media player classic in which insert small shaders to be executed by graphics card? Is this possible?

Greetings
John Slave

Great idea!! I have a 21 inch monitor! Can this be done???

[haruhiko_yamagata]

Many people use ffdshow lanczos resize to obtain better quality expecially on great displays. However unfortunately this ffdshow resize eats great cpu so why don't put a shader window like media player classic in which insert small shaders to be executed by graphics card? Is this possible?

Greetings
John Slave

I don't konw if it is possible. If you have dual core CPU my patch may accelarate resize.

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

[Egh]

Many people use ffdshow lanczos resize to obtain better quality expecially on great displays. However unfortunately this ffdshow resize eats great cpu so why don't put a shader window like media player classic in which insert small shaders to be executed by graphics card? Is this possible?

a) This question is old and at least on bugtracker Milan already responded to that. No, ffdshow is NOT a video renderer per se, thus currently it doesn't have access to shaders.

b) Can't you use Haali renderer with proper bicubic resizer? The difference between lancz and bicubic (with proper adjustment) is quite marginal in many cases. Haali renderer does entirely shaders-based resizing.

c) On my FX5500 and Athlon XP (with just SSE1 !) lanc resizing in software is considerably faster than bicubic in hardware. I dont' haver trouble playing 720p videos and applying lanczos to them (just xvid though, not h264 :P). Thus on anything less than some good 6xxxx or better ge-force the speed in hardware will still be worse compared to CPU-based one.

[foxyshadis]

bob0r, would it be possible to make a build with that patch included? I'd like to give it a workover but I can never get ffdshow to compile properly.

[Rash]

Thanks b0bor, your builds are great. ;)

[Egh]

Thanks b0bor, your builds are great. ;)

Kurosu counter-released though :P


April 16
* Using accuracy over speed for tremor and libmad (not tested whether that matters, but ffdshow always report low accuracy mode - code shows it's hard coded)
* Patching source whenever possible; modify FPATCHES in update.cfg


So is it higher accuracy now for vorbis decoding by tremor in this kurosu build or not? :confused:

And what kind of patches were applied?

of course his builds are available at http://kurosu.free.fr/ffdshow.htm for those who still don't know :P I use sse1 only http://kurosu.free.fr/ffdshow-20060420-14H04-athlon.exe

[bob0r]

bob0r, would it be possible to make a build with that patch included? I'd like to give it a workover but I can never get ffdshow to compile properly.

gcc -c -I. -I.. -I../imgFilters -Ilibvo -w -ffast-math -mno-cygwin -mdll -mthreads -pipe -DNDEBUG -UDEBUG -DWIN32 -D_WIN32 -O3 -march=i586 -mtune=i686 -fomit-frame-pointer -finline-functions -finline -o postproc/swscale.o postproc/swscale.c
postproc/swscale.c:80:30: error: swscale_multicpu.h: No such file or directory
make[1]: *** [postproc/swscale.o] Error 1
make[1]: Leaving directory `/home/user/ffdshow_gcc_mt/src/mplayer'
make: *** [../bin/libmplayer.dll] Error 2

http://google.com/search?q=swscale_multicpu.h
Your search - swscale_multicpu.h - did not match any documents.

Doh!

[issa]

SVN REV.2523 Build with gcc 4.1.1 (20060421), please test it to see if the gcc 4.1 produce faster code.

SSE: ffdshow-20060421-gcc-4.1.1-sse.exe (http://rapidshare.de/files/18549439/ffdshow-20060421-gcc-4.1.1-sse.exe.html)
SSE2: ffdshow-20060421-gcc-4.1.1-see2.exe (http://rapidshare.de/files/18549758/ffdshow-20060421-gcc-4.1.1-see2.exe.html)

The following build compiled with auto vectorize turn on,

SSE: ffdshow-20060421-gcc-4.1.1-sse-vectorize.exe (http://rapidshare.de/files/18554087/ffdshow-20060421-gcc-4.1.1-sse-vectorize.exe.html) [vectorize information log (http://rapidshare.de/files/18554177/vect.sse.7z.html)]
SSE2: ffdshow-20060421-gcc-4.1.1-sse2-vectorize.exe (http://rapidshare.de/files/18550046/ffdshow-20060421-gcc-4.1.1-sse2-vectorize.exe.html) [vectorize information log (http://rapidshare.de/files/18575684/vect.sse2.7z.html)]

Please someone give it a try, and test to see if the auto vectorize break any part of the ffdshow.

[videomixer9]

As ICL GCC can tell you when it does vecotrization, would've been nice to have the compile logs for this build to see if it actually vectorized something useful, last GCC 4.1.1 build from march had horrible auto vectorization. Cannot test yours, don't have SSE2. I only noticed that TomsMoComp and KernelDeint dramatically increased compile time with auto vectorization on.

[ExtraEye]

SSE: http://x264.nl
SSE2: http://files.x264.nl/ffdshow

bob0r for me your build doesn't run mpg, only avi.
changing to libavcodec makes it run again but libmpeg doesn't work as far as can tell.

im using the sse2 version if it matters.

EDIT: same happens with issa's vectorize build. im trying the vectorize off version to see if it has anything to do with it.

[videomixer9]

thanks issa, as i can see from the log there wasn't much vectorization done. I hoped for some advance in the vectorization stuff for this new version of gcc 4.1.1. Intel Compiler still vectorizes much more successful.

Other than that it is like the time I tried the -ftree-vectorize switch, the decoder doesn't really crash, but it just doesn't work. It builds the graph several times and fails and then falls back to another decoder or none.

So the vectorized build is unusable.

[foxyshadis]

Drat about the SMP. It isn't in my MSVC either, so maybe it's ICL.

Vectorized SSE2 runs fine on mine. Core Duo, XP SP2, latest MPC. (And I did ensure that it's actually ffdshow doing the decoding.) I have no idea how useful it actually is, though, haven't benchmarked anything. Except wma audio, which I causes a crash every time. (Not sure if it's 7, 8, or 9.)

libmpeg is totally broken, but is there a compelling quality or speed reason to use it? Since it seems to break every other compile I've always left it on lavc.

[haruhiko_yamagata]

Hello foxyshadis and bob0r.
Thank you for testing and report.
I'm very sorry. It was a terrible mistake.
I uploaded the fixed version.

[ExtraEye]

libmpeg is totally broken, but is there a compelling quality or speed reason to use it? Since it seems to break every other compile I've always left it on lavc.
from my perspective there is. and a pretty big one.
i use zoom player and i use it to remember the last place i got to on the file im watching (like - closing the movie on 31:20, and when loading it again it starts from 31:20).
when using that option with libavcodec, the player crashes. with libmpeg it doesn't.
it's true that it's not rare for builds to have broken versions of libmpeg but that's just a reason not to move to them.
right now i changed back to "ffdshow-svn2503-20060407-unicode" build and it works so if you want look for it somewhere on this thread.

[bob0r]

used patch: http://sourceforge.net/tracker/index.php?func=detail&aid=1472926&group_id=53761&atid=471491

SSE and SSE2: http://files.x264.nl/ffdshow/ look for -mt- in the filenames.

[ExtraEye]

works for me except for the problem with libmpeg2 already mentioned before

[foxyshadis]

Hmm, I've been pushing it as hard as I could (surprisingly difficult!), resizing, enabling many different filters & avisynth plugins, and it just refuses to use more than 50% cpu, even when it's stuttering. I used "ffdshow-20060421-gcc4.0.3-sse2-mt-x264.nl.exe".

[igor1st]

works for me except for the problem with libmpeg2 already mentioned before
I made quick test today for using ffdshow with dvd-video (it crashs on navigating menu year ago).

MPC rev592, ffdshow rev2523 (bob0r sse build).
libmpeg2 with selected "DVD decoding" works almost perfectly.
libavcodec as a decoder didn't work properly at all in this case.

[haruhiko_yamagata]

Thank you, bob0r.
As foxyshadirs said, it seems multithreading is not used in the build.
I checked out svn newly, applyed the patch, and just typed make for gcc 4.0.3 on MSYS and created an instraller. No problem found.
I use directX9 header and library.

[TheShadowRunner]

Hey all, I have a bug to report for :
http://kurosu.free.fr/ffdshow-20060420-14H04-P3.exe
Each time i open a video with "Subtitles" > "Vobsub" enabled it crashes the player (upon exit for WMP or as soon as the video starts with Zoom Player).
Previous build didn't have this bug.
Later,

TSR

[haruhiko_yamagata]

Hello, bob0r.

I found ffdshow-20060421-gcc4[1].0.3-sse-mt-x264.nl.exe
works on multi-thread.

[foxyshadis]

Yep, I can confirm SSE/MT works (should have tried that one too). It maxed out around 75%, could be because of using limitedsharpenfaster or just some combination of filters that aren't multithreaded yet. But at least it works! Horray!

[haruhiko_yamagata]

Originally Posted by foxyshadis
Yep, I can confirm SSE/MT works (should have tried that one too). It maxed out around 75%, could be because of using limitedsharpenfaster or just some combination of filters that aren't multithreaded yet. But at least it works! Horray!

Thank you for testing.
Ok it works. I feel relieved.
Go beyond 75% is difficult because threads depend each other. Optimization is required for scenarios used by many people.

[bob0r]

LOL, you both missed the "look for -mt- in the filenames." part :)

Very good, because i was sure i did everything right.

Now lets see what Milan thinks of it eh? :cool:

[Liisachan]

LOL, you both missed the "look for -mt- in the filenames." part :) What do you mean? You made -sse2-mt- and -sse-mt- and apparantly the only -sse-mt- works.

Now lets see what Milan thinks of it eh? :cool: I see no reason not to merge... If the patch makes it faster even by .5% it's really precious :D

[foxyshadis]

LOL, you both missed the "look for -mt- in the filenames." part :)
I used "ffdshow-20060421-gcc4.0.3-sse2-mt-x264.nl.exe".
???

I'm sure that's the one I used, since I tried to redownload it and reinstall it when it didn't spill over 1 cpu, just in case I had screwed up and used the wrong one. Blargh.

[JarrettH]

what is auto vectorize? :devil:

[JarrettH]

is there something i have to do to enable multithreading in the ffdshow options? mine never goes above 50%

also...

these latest SSE2 builds (13th, 20th) crash Media Player Classic and Got All Media when I try to use my image settings. the SSE versions work fine though

[Liisachan]

is there something i have to do to enable multithreading in the ffdshow options? mine never goes above 50% It is supposed to work automatically.

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

2. Configuration

There is no configuration to stop multithreading
currently. If your CPU is single core and non hyper-
threading, multithreading is avoided.

[zilexa]

Just wanted to say, THANK YOU KUROSU for creating (and maintaining)
THE ffdshow downloadpage... (http://kurosu.free.fr/ffdshow)
It is the only website I found listing the CPU-specific Ffdshow builds so clearly!
_______________________________________________



Do have one q: Can't there be just 4 builds?
instead of:# Build Athlon XP (has 3DNow, SSE1 but no SSE2), 20060420-14H04 (3826 KB)
# Build Pentium4 (has SSE1 and SSE2), 20060420-14H04 (3644 KB)
# Build Pentium 3 (only has SSE1), 20060420-14H04 (3994 KB)
# Build Prescott (has from SSE1 up to SSE3), 20060420-14H04 (3680 KB)
# Build Athlon 64 (has 3DNow, SSE1 and SSE2), 20060420-14H04 (3843 KB)for AMD:

3dNow! & SSE1
3dNow! & SSE & SSE2

for INTEL:

SSE1
SSE1 & SSE2

--> Why is there a build for Prescott? ffdshow doesn't use sse3.
And couldn't there be just 1 build? One build with 3dNow! & SSE1 & SSE2 if it would be used on AthlonXP then the SSE2 optimizations would not be used. If it would be used on a Pentium, 3dNow! wouldn't be used, if it would be used on a P3 only SSE1 would be used.
Seems more logical to me :)

[LotharZ]

"ffdshow-20060420-14H04-k8" continue causing problems with MPEG1/2 files if SSE2 option is enabled.
I dont have this problem with other compiles.

[bob0r]

used patch: http://sourceforge.net/tracker/index.php?func=detail&aid=1472926&group_id=53761&atid=471491

SSE and SSE2: http://files.x264.nl/ffdshow/ look for -mt- in the filenames.

[foxyshadis]

zilexa, it doesn't define what hand-optimizations are used; those still go through a cpu-checking process and use whatever you have no matter what build you run. (It would be an interesting size optimization to modify the makefiles to chop out asm code above or below the "defined" level.)

It's the masses of straight C code we're attempting to autovectorize into iSSE/2/3. Of course, as seen here, gcc's autovectorize (as little as it does compared to ICC's powerful one) is still somewhat buggy. For the most part it makes only a minor difference, if you're suffering performance problems and need every possible boost.

Thanks bob0r, I'll give it another shot.

[multiblitz]

I think it is simply wonderful that you guys strted to work on the MT-capability of ffdshow. Thanks a lot.

Here is what I see currently using the latest MT-Version from x264 with my x2 3800:

With Milan's version from 20051129 one CPU maxes out at 100%, the other at 10-15%, total around 55-60%.

My settings are De-noise 0.5,0.5.0.5,5, HQ on; Resizing at 1920*1080 Lanszos 4 with Luma sharpening at 2; Sharpening with un-sharp-mask at 20

Picture-wise, with ZP and Dscaler in YV12, every 15 sec the picture stutters.

Same setting with the ffdshow-20060423-gcc4[1].0.3-sse2-mt-x264.nl.exe version:

One CPU at 90-95%, the other at 35-40%; total at 65-75%, so some MT haopens; Stuttering is much worse that in Milan's version of 20051129, every 5 secs.

[LoRd_MuldeR]

Question:

Does the MT patch effect multi CPU systems only?
Can I expect any improvements for my single CPU too?
Or could the MT patch even lower performance for a single CPU?


//EDIT

Okay, did some testing with that "super high resolution" Superman trailer (x264). With the previous ffdshow build the video was stutterting extremely. Sometimes no new frame for several seconds. And audio/video was completely out of sync. With the new one there's a difference: Video is running more fluid, but still not very well. Audio/Video is now in sync, but I get heavy "clicking" sounds all the time. So I'm not sure what is better...

CPU usage with *new* build (single CPU):
http://img522.imageshack.us/img522/7056/cpu0cp.gif

[JarrettH]

SSE2 version still doesn't work when I try to use image settings:scared:

[JarrettH]

Hm, I'm just not happy with these SSE and SSE2 builds

I kind of regret updating now. I remember having a build from Feb 6, 2006 ICL9 unicode. Is there somewhere I can get the latest ICL9 build?