ffdshow tryouts project: Discussion & Development

[clsid]

^ committed

[foxyshadis]

Originally Posted by Leak

If you're using TortoiseSVN, you can just select two revisions in the "Show log..." dialog, then right click on one and select "Show differences as unified diff" - that yields all changes between to revisions in one diff file.

Or right click and select "Compare revisions" to get a list of all changed files, then double click on one to get a diff of just those two files in TortoiseSVN's diff viewer, which displays both files side by side.

Quite useful, IMHO.

That's exactly what I do. I'll admit it - Trac is just prettier. >.>

Originally Posted by TheShadowRunner

Hey All,
Just a simple request pertaining to .ASS type subs.
Those don't seem to be affected by the font size selection under the "Subtitles" > "Font" section.
Could this be fixed/implemented ?

No. SSA/ASS is based on fixed font sizes. But since I think this was an extremely dumb decision on the designers' parts - typical publisher mentality, whether it's brochures or fansubs - I'd be quite willing to propose that the size doubles as a scale.

But how to scale the values inside without forcing them all to the same size? Just multiplying them won't really work. I say pick the most common style (adding up the lines for each) and fix it at the selected font size, scaling everything else according to the ratio. Annoying, but that's what you get when protocols are designed badly.

Thoughts?

Later,

TSR
Edit: Also, what exactly is this "autosize" checkbox supposed to do? (still on the "Subtitles" > "Font" section)

Sizes based on the width of the text, iirc. Again, only currently works for srt.

[Kado]

Originally Posted by foxyshadis

Wrong order of operations, imho. Keep the mixer before the Winamp plugin filter always, then you can downmix if necessary without affecting the plugins. The mixer is your mono/dual/multichannel switch.

If I do it that way I'll end up with stereo output because Enhancer only supports 16bit stereo, I use volume to normalize audio then enhancer or dfx to enhance the audio then the mixer to up/downmix to quad but if source audio is more than 2 channel winamp dsp will be bypassed on purpose. What I was asking is that an option would be added so that audio with more than 2 channels be dumped to winamp dsp.

Originally Posted by haruhiko_yamagata

Multi-channel audio is excluded from ffdshow-winamp plugin on purpose.

bool TaudioFilterWinamp2::is(const TsampleFormat &fmt,const TfilterSettingsAudio *cfg)
{
 return super::is(fmt,cfg) && fmt.nchannels<=2;
}

If I remove <=2, it works, but some other plugins does not work, gets very noisy.
I tested winamp2, it seems it's the same.

Milan probably did that because there were no multichannel winamp dsp plugins at the time, but now it's time to embrace the new technologies!

[squid_80]

Hmm I haven't been keeping up with this thread and all the 64-bit developments. A few things spring to mind:

• xp64 does come with a 64-bit media player, but it sucks badly: c:\windows\system32\mplay32.exe
• directshowsource plugin is (and has been for quite some time) available for avisynth64, but I am yet to hear of anyone using it.
• 64-bit splitters for mkv and mp4 are available from celtic_druid ( http://ffdshow.faireal.net/mirror/Me...20filters/x64/ ). Just download and regsvr32 them.

[jidelite]

Originally Posted by Leak

Here's another update of my patch:

* Skip ahead in AviSynth instead of reinitializing everything when the cache needs to be disregarded - that should prevent longer pauses when seeking caused by AviSynth filters getting initialized again. Note that the filter of course still gets reinitialized when the output format changes or playback is stopped.

* Made sure that timestamps produced are strictly increasing (Overlay Mixer is not too fond of frames that end before they start)

* Removed the "Current" buffer setting as it's been folded into the "ahead" setting, especially when the correct setting for that can be gotten from the OSD (make sure "AviSynth info" is checked on the OSD page) - be warned that the detection can't be perfect if AviSynth filters malfunction when not enough buffers are set, so perhaps start out with a setting of 10/10...

* Added a "Use Current" button to fill in the current max back/ahead buffers shown in the OSD in the config dialog for convenience

Download (MSVC build of patched rev. 1093)
Patch against rev. 1093

P.S.: I'm pretty happy with the patch as it is now - any chance of it getting rolled into SVN?

np: cLOUDDEAD - Physics Of A Unicycle (Ten)

Leak,
I like your patched version as it allows me to have alittle more headroom for my P4 3.0 with regular DVD viewing. However, I still come across a stutter or 2 (that I don't see with an un-patched version) and was wondering if you could provide a way to 'bypass' your patch in case I want to do regular AviSynth filter processing to test with?

jidelite

[Eno / Omni]

What's the main difference between your and xxl release ? =)

[Leak]

Originally Posted by jidelite

Leak,
I like your patched version as it allows me to have alittle more headroom for my P4 3.0 with regular DVD viewing. However, I still come across a stutter or 2 (that I don't see with an un-patched version) and was wondering if you could provide a way to 'bypass' your patch in case I want to do regular AviSynth filter processing to test with?

Could you perhaps tell me what script you were using that produced those stutters, and when? Were those one-time stutters or did they start and then last until you seeked or paused or something like that?

(Do note that the extra buffering isn't meant to smoothen playback or "give headroom", like you wrote - for that, activate ffdshow's queue on the "Queue & misc" page... it's meant to allow AviSynth filters to work correctly that need more than just the current frame or that add/remove frames. Or, if I were completely selfish I would say it's meant to make TIVTC work on the fly... )

Anyway, to bypass almost anything just uncheck "Apply pulldown" and set both buffer settings to zero - that should do it, except that it probably still does an extra blit when compared to the old implementation - that's next on my to-do list to fix.

[Kado]

Originally Posted by Eno / Omni

What's the main difference between your and xxl release ? =)

In terms of usage and performance there are none. There are more details in another page of this tread.

Anyway, I suppose you are talking about clsid and xxl's builds, Leak's build has additional AviSynth tweaks and optimizations.

[jidelite]

Originally Posted by Leak

Could you perhaps tell me what script you were using that produced those stutters, and when? Were those one-time stutters or did they start and then last until you seeked or paused or something like that?

(Do note that the extra buffering isn't meant to smoothen playback or "give headroom", like you wrote - for that, activate ffdshow's queue on the "Queue & misc" page... it's meant to allow AviSynth filters to work correctly that need more than just the current frame or that add/remove frames. Or, if I were completely selfish I would say it's meant to make TIVTC work on the fly... )

Anyway, to bypass almost anything just uncheck "Apply pulldown" and set both buffer settings to zero - that should do it, except that it probably still does an extra blit when compared to the old implementation - that's next on my to-do list to fix.

Leak,
Yes, you are right, perhaps 'headroom' was a bad choice of words. Your patch allows me to run filters that would
otherwise stutter on a non-patched ffdshow builds.
I posted this in another thread but here goes - I'm using:
Abit AG8 Motherboard
P4 530J 3.0 GHz using hyperthreading
GeForce 6600
1 Gb ((2) 512 ddr-sdram)
TheaterTek software player

At the moment I've only been using the 1 filter/Avisynth script (nothing else is check in ffdshow):
Buffers BACK=1 AHEAD=3 (The OSD indicates this filter only needs 1-back & 1-ahead.)
SetMemoryMax(224)
SetMTMode(2)
SeeSaw(NRlimit=1, NRlimit2=3,Sstr=1.4,Slimit=5,Spower=2,Sdamplo=2,SdampHi=20,Szp=10,Bias=65,SootheT=10)

I sat down finally to watch the R1 version DVD of 'Blood Diamond' and 1:17 into the first chapter the video freezes and
stutters for about 2-3 seconds and then it is fine - something that doesn't happen with an unpatched version of ffdshow. I didn't watch it all the way thru with your patched version but the first couple of chapters seem to work ok after that so it
appears to be a one time thing. I tried unchecking "Apply pulldown" and set both buffer settings to zero but I get the
same results. I also reported (in another thread) the same thing happening with the montage on my Video Essentials DVD. I'll have to watch more DVD's with your patched version to see if I can get more examples.

Thanks,
jidelite

[Avish]

Hey Leak!!

Can you please tell me how to get your build working? I mean "TIVTC on the fly" thingy you keep talking about!!!

I've MPC, your latest ffdshow build, avisynth 2.5.7 & latest TIVTC dlls. Now I wanna know how to get it work?? What script should I use? what options should be ticked or unticked?? etc etc...

Thanks in advance!!

[Leak]

Originally Posted by Avish

Hey Leak!!

Can you please tell me how to get your build working? I mean "TIVTC on the fly" thingy you keep talking about!!!

I've MPC, your latest ffdshow build, avisynth 2.5.7 & latest TIVTC dlls. Now I wanna know how to get it work?? What script should I use? what options should be ticked or unticked?? etc etc...

Yeah, I know, it's still a bit lacking in the "documentation" department. But I'm currently a bit swamped with other work, so it'll take a bit to write.

Anyhow, in your case a script like

TFM(order=1) # add chroma=false for a bit of speedup
TDecimate(mode=1,hybrid=1,conCycle=1) # and maybe chroma=false again

should work, although you might want to tweak that - I'd suggest consulting TIVTC's documentation anyhow...

- Check "Add ffdshow video source", which adds a "ffdshow_source" command to the beginning of the script, which is an AviSynth filter provided by ffdshow that acts like AVISource and feeds AviSynth the frames ffdshow gets

- Check "Apply pulldown", which should always be checked for IVTC as it makes sure MPEG2 and other video that has interlace flags gets turned into proper interlaced 30FPS video like you'd get when watching a DVD on a TV, instead of the 24/30 FPS mix it might be on a PC

- Check the four colorspace checkboxes; you could uncheck the two RGB settings since TIVTC only supports YUY2 and YV12, but you can always just do the conversion yourself in the AviSynth script

- Use values for back and ahead that the "AviSynth info" in the ffdshow OSD spits out while playing video using the above script; for me that's 0 back and 12 ahead - you might have to do that again if it reports a higher value after seeking around a bit.

np: Sage Francis - Smoke And Mirrors (Personal Journals)

[Leak]

Originally Posted by jidelite

Leak,
At the moment I've only been using the 1 filter/Avisynth script (nothing else is check in ffdshow):
Buffers BACK=1 AHEAD=3 (The OSD indicates this filter only needs 1-back & 1-ahead.)

Hmmm... are you really using my latest build? There was a bug in the previous version that would often report more needed back buffers than there really were - now it's quite rare, but not impossible to get a back value other than 0. (I guess I'll have to try SeeSaw myself sometimes... )

Still, if the OSD says "1 ahead", I'd really recommend using just 1 for the buffers ahead setting, since the greater that value the greater the pause (and CPU load) between starting/seeking and the beginning of playback.

Originally Posted by jidelite

I sat down finally to watch the R1 version DVD of 'Blood Diamond' and 1:17 into the first chapter the video freezes and stutters for about 2-3 seconds and then it is fine - something that doesn't happen with an unpatched version of ffdshow. I didn't watch it all the way thru with your patched version but the first couple of chapters seem to work ok after that so it appears to be a one time thing. I tried unchecking "Apply pulldown" and set both buffer settings to zero but I get the same results. I also reported (in another thread) the same thing happening with the montage on my Video Essentials DVD. I'll have to watch more DVD's with your patched version to see if I can get more examples.

Hmmm... could you perhaps extract the first two minutes or so of your Blood Diamond DVD (using DVD Decrypter and Chopper XP or something like that) and upload it somewhere so I can try this out myself?

Also, how much CPU load do you have while playing the DVD? And does this also happen if you don't use SetMTMode? (It's not going to help much with HyperThreading anyway...)

np: Sage Francis - Pitchers Of Silence (Personal Journals)

[Avish]

Originally Posted by Leak

Yeah, I know, it's still a bit lacking in the "documentation" department. But I'm currently a bit swamped with other work, so it'll take a bit to write.

Anyhow, in your case a script like

TFM(order=1) # add chroma=false for a bit of speedup
TDecimate(mode=1,hybrid=1,conCycle=1) # and maybe chroma=false again

should work, although you might want to tweak that - I'd suggest consulting TIVTC's documentation anyhow...

- Check "Add ffdshow video source", which adds a "ffdshow_source" command to the beginning of the script, which is an AviSynth filter provided by ffdshow that acts like AVISource and feeds AviSynth the frames ffdshow gets

- Check "Apply pulldown", which should always be checked for IVTC as it makes sure MPEG2 and other video that has interlace flags gets turned into proper interlaced 30FPS video like you'd get when watching a DVD on a TV, instead of the 24/30 FPS mix it might be on a PC

- Check the four colorspace checkboxes; you could uncheck the two RGB settings since TIVTC only supports YUY2 and YV12, but you can always just do the conversion yourself in the AviSynth script

- Use values for back and ahead that the "AviSynth info" in the ffdshow OSD spits out while playing video using the above script; for me that's 0 back and 12 ahead - you might have to do that again if it reports a higher value after seeking around a bit.

Thanks man for the quick n' detailed reply!!!
I'll try it out immidiately.

[jidelite]

Leak,

Hmmm... are you really using my latest build? There was a bug in the previous version that would often report more needed back buffers than there really were - now it's quite rare, but not impossible to get a back value other than 0. (I guess I'll have to try SeeSaw myself sometimes... )

Still, if the OSD says "1 ahead", I'd really recommend using just 1 for the buffers ahead setting, since the greater that value the greater the pause (and CPU load) between starting/seeking and the beginning of playback.

My mistake , I am using your latest build and the back value IS 0

Also, how much CPU load do you have while playing the DVD? And does this also happen if you don't use SetMTMode? (It's not going to help much with HyperThreading anyway...)

My CPU load is about 30%. I removed the SetMTMode and I get the same results.

Hmmm... could you perhaps extract the first two minutes or so of your Blood Diamond DVD (using DVD Decrypter and Chopper XP or something like that) and upload it somewhere so I can try this out myself?

I have a 30 sec (29Mb) that you can get here: http://files.filefront.com/Blood_Dia.../fileinfo.html

Thanks,
jidelite

[Leak]

Originally Posted by jidelite

I have a 30 sec (29Mb) that you can get here: http://files.filefront.com/Blood_Dia.../fileinfo.html

Ummm... is the scene where it stuttered in there, and where?

EDIT: Scratch that, it studders if I poke it enough - somehow I managed to get the timestamps messed up; turn on "Frame duration" in ffdshow's OSD and you'll see they oscillate between 41 and 75 ms, instead of a steady 41 ms...

I'll look into it this evening. As a workaround, you can seek back a bit once or twice, it should fix itself then...

np: Contriva - Unhelpful (Separate Chambers)

[jidelite]

Originally Posted by Leak

Ummm... is the scene where it stuttered in there, and where?

EDIT: Scratch that, it studders if I poke it enough - somehow I managed to get the timestamps messed up; turn on "Frame duration" in ffdshow's OSD and you'll see they oscillate between 41 and 75 ms, instead of a steady 41 ms...

I'll look into it this evening. As a workaround, you can seek back a bit once or twice, it should fix itself then...

np: Contriva - Unhelpful (Separate Chambers)

Glad you could spot it . I turned on the Frame Duration in the OSD and I did see it oscillate. While I'm loading clips, here is a 1 min (70Mb) clip from my Video Essentials disc that also gives me problems - http://files.filefront.com/Video_Ess.../fileinfo.html...

Thanks,
jidelite

[FredThompson]

Would somone please add a selective setting to the deinterlacing such that it is only enabled if the source is above a use specified set of X and Y dimensions? I have kernel deinterlacing on by default (I work with a lot of different camcorder source) and have noticed podcast-sized video looks horrible. Maybe this is part of how kernel deinterlacer works. There doesn't seem to be much negative effect while viewing large video but 320x240 progressive looks horrible.

[FredThompson]

wrt my comment about kernal deinterlacer, it seems to only be a problem with Apple-encoded mp4v. It's really odd.

[Leak]

Another update:

* Added some comments to the code
* Don't always reallocate the buffers
* Fix b0rked code to get the decoder's filters in BeginFlush, since that's not gonna work in the encoder, which will then likely crash or act strangely
* Really make timestamps strictly increasing; frame durations in the last version could start to oscillate, turning playback into a stutter-o-rama

Download
Patch (against rev. 1103)

One thing I still need to do is make the filter not buffer anything at all if buffering is disabled, but I didn't have the time to squeeze this in today.

@jidelite: The last change listed above should hopefully fix your problem, could you please try it? At least I wasn't able to make playback of the sample you posted stutter again, no matter how hard I tried seeking around in it, which always triggered it after a few tries with the last build...

@haruhiko_yamagata: Are you still getting green frames with this version? I think I made sure that no frames that AviSynth may have cached (and discarded, or zeroed, or whatever) are used after a seek anymore.

np: Richie Hawtin - TZ Entry Point (DE9 - Transitions)

[jidelite]

Originally Posted by Leak

Another update:

* Added some comments to the code
* Don't always reallocate the buffers
* Fix b0rked code to get the decoder's filters in BeginFlush, since that's not gonna work in the encoder, which will then likely crash or act strangely
* Really make timestamps strictly increasing; frame durations in the last version could start to oscillate, turning playback into a stutter-o-rama

Download
Patch (against rev. 1103)

@jidelite: The last change listed above should hopefully fix your problem, could you please try it? At least I wasn't able to make playback of the sample you posted stutter again, no matter how hard I tried seeking around in it, which always triggered it after a few tries with the last build...

np: Richie Hawtin - TZ Entry Point (DE9 - Transitions)

Leak, tried rev. 1103 and I still get same results. By the way, my videos oscillates between 41ms & 33ms when I have a problem (if that makes a difference) - and always in the same location. Thanks for looking into this - if you cannot reproduce it and no one else is reporting a stuttering problem (this doesn't happen in every movie) then this could very well be isolated to my PC.

jidelite