New ffdshow build (?)

[Liisachan]

Yet another flavor :D

ffdshow-20060423_gcc4.1_speed.exe (http://www.paehl.com/open_source/?FFDSHOW)





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?
That -mt- patch is not for stability. There are many builds recently, I know about 15 builds for rev.2523. What are you looking for. According to celtic_druid, "Infact MSVC8 seems to be the most stable (talking about ffdshow.ax). ICL I guess would be ok for most of the plugins. Then you have an MSVC8/ICL/gcc build though." If what you are looking for is ffdshow-20060203-icc-sse or ffdshow-20060202-icc-sse2 i put them here (http://ffdshow.faireal.net/mirror/Misc%20(not%20by%20celtic_druid)/).

[JarrettH]

The one CLSID posted works pretty good actually.

SVN: 2006-04-20 (rev. 2523)
Compilers used:
- MSVC71 (ffdshow.ax, etc)
- GCC 3.4.5 (libavcodec.dll, mplayer.dll, kerneldeint.dll)
Minimum CPU requirement: MMX
Download: click here (http://rapidshare.de/files/18585145/ffdshow-svn2523-20060420.exe.html)

I noticed using Live TV the SSE and SSE2 builds were definitely not as fast as what I had been using and looked a bit more aliased in places. Does ICL9 mean it could have been SSE/SSE2? It didn't specifically say in About ffdshow.

Ah it must have been Feb 3rd then ;)
You are so kind for uploading that for me!!:D

[haruhiko_yamagata]

Thank you very much, bob0r. They works nicely on my computer.
Originally Posted by multiblitz
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.
Hello, multiblitz. I will investigate. This may take several days. Thank you.

[haruhiko_yamagata]

Hello, LoRd_MuldeR.
Originally Posted by 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?
MT patch is not supposed to affect performance on single CPU. You can not expect any improvements for single CPU. MT patch check CPU count (just read from memory) and this is some overload. But this can be ignored.

In regard to bugs, there is room for them. The critical session, m_csReceive, is locked at different timing.

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):

Thank you for detailed report.
I have no idea why.
Strange CPU usage curve with up and down. Does that occur on other builds without MT patch?

To investigate this, I would like to know

the application and its version ( and How to get)
how to get Superman trailer
the setting of ffdshow and application.

[haruhiko_yamagata]

I found a multithreading related bug.
It crashes Zoom Player!!

Confession : I have not tested with Zoom Player.

[foxyshadis]

I'm pretty sure the cpu curve is caused by avisynth. When I enable limitedsharpen on mine, cpu usage is ~70%, but occasionally dips to zero for a second or two and freezes, before resuming. Other ffdshow-specific maniplations (such as resize, subtitles, etc) don't show that behavior whether they use 20% or 80% total cpu.

Rawr, vsfilter crashes on here but ffdshow refuses to show the first line of softsubs, so I'm all out of luck for subs. ;_;

[LoRd_MuldeR]

Hello, LoRd_MuldeR.

To investigate this, I would like to know

the application and its version ( and How to get)
how to get Superman trailer
the setting of ffdshow and application.

Thanks for info. But does MT patch do any harm on single CPU or is it just without any effect and I can use the new builds with no risk?

The application is "Media Player Classic" with output set to "VMR7 (Windowed)". This setting seems to give best results on my machine. I use latest build from Celtic Druid. And I have *no* filters enabled in ffdshow at all. Excpet I did some testing with resize-filter: Downsizing to 640x384 (Fast Bi-Linear filter) seems to improve performance a little compared to full resolution of 1920x1088, but of course image quality suffers and it still does not play good...

You can download that trailer right here:
http://jfl1974.free.fr/upload/superman.mp4

[haruhiko_yamagata]

Originally Posted by LoRd_MuldeR
Thanks for info. But does MT patch do any harm on single CPU or is it just without any effect and I can use the new builds with no risk?

There is some risk even if you use single CPU. Just relatively safe.

Thanks for the detail. I'm downloading superman.

[LoRd_MuldeR]

There is some risk even if you use single CPU. Just relatively safe.

Thanks for the detail. I'm downloading superman.

So I hope there will be some new(!) builds *without* MT patch available...

[multiblitz]

It would be great if the most used filters in ffdshow would have separate threads, i my case this would be:

- (maybe de-interlace)
- Denoise 3d in HQ-Mode
- Lanszos Resizing at 4 tap+ and Lanszos Sharpening
- Sharpening with Unsharp-Mask

and this in combination with VMR9+zoomplayer. It would be simply wonderful if MT allows us to have this quality (ideally with limitesharpening / avisynth) in RGB32 at 1080p, which is today totally unfeasible (BTW, my x2 3800 runs at 2450 mhz)

[Egh]

So I hope there will be some new(!) builds *without* MT patch available...

Exactly. Especially taking into account that two new revisions were made to SVN already since 2524.

And kuroshu builds with patches are rather buggy, even casual sse1 build. In my case, for instance, b0b0rs build plays mpeg files ok, but kuroshu build with same settings doesn't do that properly. Both are same revision and both are sse1 only.

So for single CPU systems I would still like to get not patched builds :P

[clsid]

SVN: 2006-04-24 (rev. 2526)
Compilers used:
- MSVC71 (ffdshow.ax, etc)
- GCC 3.4.5 (libavcodec.dll, mplayer.dll, kerneldeint.dll)
Minimum CPU requirement: MMX
Download: click here (http://rapidshare.de/files/18838575/ffdshow-svn2526-20060424.exe.html)

[Liisachan]

Thanks clsid!!!

@Egh
following the svn version is one thing, experimenting with mt is another. Each has its own value.

[Liisachan]

celtic_druid returns!

ffdshow-rev2526-SSE2.exe (http://ffdshow.faireal.net/mirror/ffdshow/ffdshow-rev2526-SSE2.exe)

[asasadad_1]

clsid's ffdshow build always works well on my machine(especially in libmpeg2).:thanks:

i cann't install celtic_druid's new ffdshow build (when register ffdshow.ax, there is error).

[videomixer9]

the main trick is not using gcc 4.x :O

[celtic_druid]

I used ICL9 for libmpeg2.

Build is unicode and requires SSE2. Other than that it should work.

[LoRd_MuldeR]

I used ICL9 for libmpeg2.

Build is unicode and requires SSE2. Other than that it should work.

Too sad I don't have no SEE2 support :mad:

[celtic_druid]

For about €100 you should be able to get a Sempron 3000+ and a cheap MB. So basically just sacrifice a night of drinking and you can have SSE2. No idea what Sempron's are like at video though esp. socket 754 ones. Probably not worth it.

[asasadad_1]

http://img246.imageshack.us/img246/9563/snap17ca.jpg
yes,my cpu support sse2,xp sp2

[Liisachan]

celtic's one installed on Win2k but doesn't install for me on winxp, with Runtime Error R6034 (http://ffdshow.faireal.net/tmp/r6034.png).

Normal (svn2526 clsid) vs. mt (20060423-sse-mt-x264.nl)

YES! mt does work, more or less. For a testing purpose, I used something I've never used:
- Postprocessing (Presets mplayer)
- Resize to 640x480 to 800x600 Lanczos
And this is what I almost always use:
- output colorspace RGB32, hq conv

Player is MPC celtic_druid Rev 604, VMR-9 Renderless, subpic buffer disabled. Tested on Prescott 3.4 GHz HT enabled, Windows XP SP2

http://ffdshow.faireal.net/tmp/ffdshowmt.png

EDIT: I captured Task Manager at a random time while playing the same file (I should have done that at the exactly same timing but I didn't...) So "Normal" and "mt" can't be compared side by side, but we can roughly conclude that mt is working, i.e. one of the two (shown as the left pane) is nearly equal to 0% with the normal version, but significantly non-zero with the mt version.

[celtic_druid]

Problem with manifest files again. Win2k doesn't care about them so it installs. XP I don't know. I tried with name='Microsoft.VC80.CRT' and it still won't load the runtime library.

[videomixer9]

I WANT TOO BUILD! all the newest freaky stuff, still generic build though... o_O

ffdshow-rv2526.exe: click here (http://rapidshare.de/files/18859754/ffdshow-rv2526.exe.html) or here (http://www.megaupload.com/de/?d=HZDO6OF7)

[LotharZ]

@videomixer9

libmpeg2 no works with your new build, even disabling SSE2.
Running now with your "ffdshow-20060408_msvc8_gcc41_sse" smoothly.

@celtic_druid

And yours cant be installed under Win2003, error with Visual C++ libraries.

[videomixer9]

bummer ... updated links to hopefully fixed version ...

http://www.megaupload.com/de/?d=HZDO6OF7
http://rapidshare.de/files/18859754/ffdshow-rv2526.exe.html

[LotharZ]

Problem solved, everything works perfectly.

thx

[Rash]

Dammit! The superman trailer pwned my computer! It is an Athlon 64 3200+ with a GeForce 6800GT. Where are you guys running this video?

[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.
This is going to be reality (http://sourceforge.net/tracker/index.php?func=detail&aid=1468239&group_id=53761&atid=471492). When asked to use the high-quality mode of Tremor, milan replied:


I'd like to leave tremor as it is as fast but not so high-quality
option for those with slower computers and make libavcodec
the default decoder for vorbis.
Do you or others have vorbis samples which aren't decoded
correctly with libavcodec vorbis decoder? I'd like to fix
them if there are any before making libavcodec vorbis
decoder the default.


Would lavc for vorbis decoding be 100% ok already? Afaik, yes, but does anyone know anything?

[celtic_druid]

I made lavc default for my installer. Just a pitty that it only works under win2k due to the manifest/dependancy problem. Now that I think about it I think I had the same problem with 64bit builds under X64.

[clsid]

Decoding audio doesn't use much cpu power in general. So I think high-quality is the best option, even for slow computers.

Btw, is high accuracy equivalent to the 32-bit mode of CoreVorbis?

[Liisachan]

That's what I suggested, but what milan says makes sense too (for really slow CPU). Tremor vs. libavcodec vs. CoreVorbis etc etc...that's a really difficult question, like "Which MP3 decoder is the best?" I have a vague impression that CoreVorbis is the best, but without proof. As for Tremor low vs high, I have a solid ABX proof.

I asked the same question in HA.org and it looks like no one got any problem with lavc.
http://www.hydrogenaudio.org/forums/index.php?showtopic=43983
Only, I was asked "why writing another independent Vorbis decoder in libavcodec, possibly buggy, when there is already proper one delivered by Xiph...?" which I cannot answer for sure. I thought maybe ffmpeg devs needed to make it GPL, not using a BSD-style license of libvorbis...but that's just a random guess. I could say "maybe because they hate xiph" but saying that in HA.org would be a bad idea, i guess.

[Yong]

ffdshow libavcodec-vorbis decoder have problem decoding 6-channels Ogg vorbis file.

[videomixer9]

ffdshow rev. 2527

requires: a CPU, with SSE preferably
compilers: GCC 4.1.1 + MSVC 8
specials: hopefully high accuracy tremor, maybe libavcodec default vorbis decoder
links: here (http://rapidshare.de/files/18909341/ffdshow-rv2527.exe.html) and here (http://www.megaupload.com/de/?d=B69NO8JR)

[Egh]

ffdshow rev. 2527


I downloaded the one from megaupload.


Error while registering ffdshow
Is shown during the installation.

I got Win XP SP2. b0b0r version (svn2523 iirc) is still nice.

I think is it actually installers code problem? :O

[foxyshadis]

Oh right, I need to report a bug in the mt builds. Seeking is screwy - it sometimes works, and sometimes just quits decoding video after the seeked-to frame. Maybe it has to do with landing on the wrong type of frame. It won't even restart at the next I-frame, I waited to make sure. Normal builds always seek just fine.

[Liisachan]

ffdshow libavcodec-vorbis decoder have problem decoding 6-channels Ogg vorbis file. I can't test it since my system is just stereo. Are you sure? and is Tremor OK with that? do you happen to know would libvorbis ok with 6-chan vorbis?

[Liisachan]

I think libavcodec does have problems with 6ch vorbis, and Tremor works better in this case.

Can anyone confirm?
Sample: xvid+6ch_vorbis.ogm (http://ffdshow.faireal.net/tmp/xvid+6ch_vorbis.ogm) 2.5MB

-Edit-
This is what I'm hearing. (channel 1 Front Left)
tremor_FL.wav (http://ffdshow.faireal.net/tmp/tremor_FL.wav) at least decent
lavc_FL.wav (http://ffdshow.faireal.net/tmp/lavc_FL.wav) "bubbly" artifacts

[celtic_druid]

Looks like a problem with the manifest again. Had the same problem with MPC. XP didn't like MSVC8's autogenerated manifest file. Remove it or replace it and it would run. MPC doesn't require Microsoft.VC80.CRT though. Haven't figured that one out yet.

Anyone got a MSVC8 built ffdshow.ax to register?

edit:


<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
<assembly xmlns="urn:schemas-microsoft-com:asm.v1" manifestVersion="1.0">
<assemblyIdentity version="1.0.0.0" processorArchitecture="X86" name="ffdshow.ax" type="win32" />
<description>WindowsExecutable</description>
<dependency>
<dependentAssembly>
<assemblyIdentity type='win32' name='Microsoft.VC80.CRT' version='8.0.50608.0' processorArchitecture='x86' publicKeyToken='1fc8b3b9a1e18e3b' />
</dependentAssembly>
</dependency>
<dependency>
<dependentAssembly>
<assemblyIdentity type="win32" name="Microsoft.Windows.Common-Controls" version="6.0.0.0" processorArchitecture="X86" publicKeyToken="6595b64144ccf1df" language="*" />
</dependentAssembly>
</dependency>
</assembly>


That works. If you have MSVC2005 you can just open the ax and replace the manifest. External manifests are ignored when registering, so they won't help.

I'll put a new build up later.

[videomixer9]

I downloaded the one from megaupload.

Is shown during the installation.

I got Win XP SP2. b0b0r version (svn2523 iirc) is still nice.

I think is it actually installers code problem? :O

Install VC8 runtime libs! get vcredist_x86.exe from MS site. Last time had the same with people that were all missing the runtime libs.
http://faux.warwickcompsoc.co.uk/vcredist/vcredist_x86.exe

[videomixer9]

ffdshow rev. 2529

requires: a CPU, with SSE preferably
compilers: GCC 4.1.1 + MSVC 8 (ffdshow.ax)
specials: high accuracy tremor
tested: yes, working
links: here (http://rapidshare.de/files/18954480/ffdshow-rv2529.exe.html) or here (http://www.megaupload.com/de/?d=ZVPYYON9) (identical files just other hoster)

needs MSVC8 runtime libs, otherwise fails registering!!!

[LoRd_MuldeR]

ffdshow rev. 2529

requires: a CPU, with SSE preferably
compilers: GCC 4.1.1 + MSVC 8 (ffdshow.ax)
specials: high accuracy tremor
tested: yes, working
links: here (http://rapidshare.de/files/18954480/ffdshow-rv2529.exe.html) or here (http://www.megaupload.com/de/?d=ZVPYYON9) (identical files just other hoster)

needs MSVC8 runtime libs, otherwise fails registering!!!

Here is one problem I found:
If I play WMV content, go to the ffdshow controls and switch to the "Info" page, then MPC just crahes and disappears without error message...

[celtic_druid]

WMV1 is working fine here, including the info page.
I did a SSE build this time to. Although I haven't tested it. Should install fine now though.

[videomixer9]

I don't use ffdshow for WMV usually, especially not WMV3. I have no real clue why it crashes except gcc messing it up. I wouldn't recommend ffdshow for WMV3 in any case though. I'll look into that for the next build, though I need to acquire an WMV video for that first as I usually avoid those. Guess I have to do my own somehow as WMV1/2 is not really used anymore since quite some time, if the crash is produced by WMV3 then I skip on this as WMV3 and ffdshow is stupid game, it works then it maybe doesn't and so on, even depends on the encoding settings and more :O

Another story about these runtime libs for MSVC 2005 (v8), the runtime libs should have external manifests so just getting dlls isn't enough usually.

[Liisachan]

lavc+vorbis6ch=buggy is confirmed by the 3rd party too.
vorbis 6ch = problematic on ffmepg too so probably it's not just ffdshow's pb but the pb is in lavc vorbis.c itself.
Thank you very much Yong, for this critical report :) Without you things might be foobared without knowing that problem.

Anyway celtic_druid's new builds:
- ffdshow-rev2529-SSE.exe (http://ffdshow.faireal.net/mirror/ffdshow/ffdshow-rev2529-SSE.exe)
- ffdshow-rev2529-SSE2.exe (http://ffdshow.faireal.net/mirror/ffdshow/ffdshow-rev2529-SSE2.exe)

Let's hope they will install on XP :)

changelog (http://ffdshow.faireal.net/bakaupdates.php)

[LoRd_MuldeR]

WMV1 is working fine here, including the info page.
I did a SSE build this time to. Although I haven't tested it. Should install fine now though.

Approved. Thanks for the SSE build :D

However it still crahes on the "Info" page during WMV playback.
That's not a big problem, because I avoid WMV too.
But sometimes I get WMV files and since ffdshow supports it, why not use it then?
The fewer M$ filters we need, the better :cool:

[Liisachan]

celtic's SSE & SSE2 now both install on XP for me.
that mt patch is not included, or if it is, it's not working for me this time.
no other obvious things so far.

[celtic_druid]

No mt patch in either build. WMV containing what? I only tested WMV1.

Who uses 5.1 vorbis anyway?

[LoRd_MuldeR]

Dammit! The superman trailer pwned my computer! It is an Athlon 64 3200+ with a GeForce 6800GT. Where are you guys running this video?

I just noticed that it helps a lot to change AAC decoder from "libfaad2" to "realaac". At least the heavy "clicking" sounds are gone and audio is okay now. A/V sync is also pretty much okay. Only video is still stuttering...

[LoRd_MuldeR]

No mt patch in either build. WMV containing what? I only tested WMV1.

WMV3 only I think. Problem is there with *all* WMV3 files I tested. With some WMV1 file I tested there's no problem.

But: The only problem is the "Info" page in ffdshow controls. Playback itself works fine for me...

[thuan]

LoRd_MuldeR you're probably playing a WMV3 file with ffdshow WMV3 codec enable by hand in the registry. I have the same problem when I do it, but then again WMV3 is not offically supported, and if I'm right ffdshow just load Microsoft decoder to do it anyway, so there's no point in enabling it.
BTW it doesn't crash with ffshow WMV3 decode with MPC here just Zoom crashes.