Avisynth 2.6 MT

[Stealth5325]

wOxxOm, thank you for that idea.
I tried to have a look with process monitor, but the folder's not even being accessed.
Even deleting the folder and creating a dummy plugins folder didn't yield results. Any other ideas?

[Stealth5325]

More information:
I've also tried using the SVP project at www.svp-team.org. It works fine with MPC-HC, because it's using the 32 bit FFDShow and 32 bit filters. The minute I install FFDShow x64 and run the SVP manager, it crashes. I'm assuming that manager is running something through AVISynth.
I'm basically feeling like I can't use this 2.6 AVISynth on 64 bit FFDShow. Can anyone help?

[Keiyakusha]

you right you can't have 64bit ffdshow. make sure all your playback chain is 32 bit, you lose nothing anyway.

[Stealth5325]

I've found basically the same thing. 64 bit AVISynth isn't working for me.
I am, however, wanting to use Windows Media Center to run everything, and of course, I use 64 bit Windows, because 4GB RAM doesn't cut it nowadays.
Any way to make 32 bit FFDShow work with 64 bit Windows Media Center?

[Keiyakusha]

Oh I didn't though of WMC. No, you can't make x32 software work with x64...

[forclip]

SEt, can you confirm that your recent build includes this (http://avisynth2.cvs.sourceforge.net/viewvc/avisynth2/avisynth/src/core/cache.cpp?r1=1.38&r2=1.39&pathrev=MAIN) commit? Because I get this problem (http://forum.doom9.org/showthread.php?p=1581796#post1581796) again.

[SubJunk]

Wish they would start using SVN or Git...

[SEt]

Stealth5325
This build is 32 bit only, so you need 32 bit player. 64 bit builds of Avisynth 2.6 are unlikely.

forclip
Yes it does. Btw, in MT mode it worked this way all the time. If you want to investigate further be sure what mode (MT or not) is activated as they use different caches.

[forclip]

If you want to investigate further be sure what mode (MT or not) is activated as they use different caches.
Thanks, very interesting. When I start my script with SetMTMode >=1 this problem seems to go away. Non-MT cache is broken?!

[lansing]

i have a rare problem here, I can't get the mt running on my machine.

I'm using Windows 7 64bit, and installed everything correctly. A fresh official avisynth with no additional plugins, and then replace the avisynth.dll in the Syswow64 folder with the mt one. And when I called SetMTMode() whatever in the script, it works with no error. However when I tried on different filters, the speed boost I was supposed to get was not there, it just seems like the mt part was not even in the picture. I also tested with filters that were reported to work like trim(), limitedsharpenfaster().

When I ran a script like this on AVSmeter, the line that reports "active MT mode" was 0.
SetMtMode(2)
LimitedSharpenFaster()

[Groucho2004]

When I ran a script like this on AVSmeter, the line that reports "active MT mode" was 0.
SetMtMode(2)
LimitedSharpenFaster()

Post the Avisynth version AVSMeter reports and your complete script.

[lansing]

avsmeter reports:
Avisynth 2.60 build:August 28 2012


MPEG2Source("mt_test.d2v")

SetMtMode(2)
LimitedSharpenFaster()

[IanB]

A SetMtMode statement must be the first line of the script.

[Groucho2004]

I just tried several MT modes with the latest MT version and AVSMeter reports them correctly. No idea why it does not work for you.

Edit: Ouch, it seems that Ian was the first to see the forest for the trees. :o

[lansing]

tried this, but still nothing's change and avsmeter still reports 0. Must be something wrong with my computer but I can't find the problem.

SetMTMode(1,0)

MPEG2Source("mt_test.d2v")

SetMtMode(2)
LimitedSharpenFaster()

[Groucho2004]

Just put "SetMTMode(5)" in the first line.

[lansing]

Just put "SetMTMode(5)" in the first line.

still the same

[Groucho2004]

Sorry, I'm pretty much out of ideas. Probably a long shot - Maybe something in your "autoload" directory is interfering?

[lansing]

Sorry, I'm pretty much out of ideas. Probably a long shot - Maybe something in your "autoload" directory is interfering?

ok i finally got the mt working. After uninstalling Avisynth, I need to delete the entire folder that was left in the program files directory, and that was what got it back working. Previously I tried only empty the plugins folder and that didn't work.


UPDATE:

After further investigation, I was able to find the problematic avs script in the plugins folder, named tbcv01.avs, one i got back in 2008. Removing it will solve the problem. I'll leave a note in the avisynth plugin wiki about this issue.

[fbs]

Which mvtools2.dll should I use with this? I've getting lots of crashes after some hours of processing..

[martin53]

Please allow me to reveal myself as novice. I am using 2.60MTalpha_SEt 8/28/12 and obviously, DeleteFrame(0,0) does not delete two frames, but just one.

Question: is there, and if so where, a bugtracking list, and is this issue maybe already known?

Sorry for causing effort with this probably misplaced post.

[IanB]

DeleteFrame(0,0) is only supposed to delete 1 frame.

The input list is internally sorted, deleting duplicates. The coding choices were to ignore duplicates or throw an error, it was decided by a very small margin that ignoring duplicates would be a more useful implementation.

To delete the first 2 frames either do :-
DeleteFrame(0, 1)
or
DeleteFrame(0).DeleteFrame(0)

[martin53]

Thanks a lot, reasonable.

[mark0077]

Hi, does anyone know how to get around the problem of mpc-hc not closing down properly when using avisynth. I have read on some threads that this is due to a problem somewhere in avisynth / mt? I am using SVP Interframe scripts with the latest avisynth from this thread, and whenever avisynth is enabled in ffdshow, mpc-hc refuses to gracefully shut down about 30% of the time. Sometimes it silently stays there in task manager, other times it will give a program not responding windows dialog.

Any hacked workarounds that has proven to work for anyone or does anyone know the reason for this?

[ney2x]

Hi, does anyone know how to get around the problem of mpc-hc not closing down properly when using avisynth. I have read on some threads that this is due to a problem somewhere in avisynth / mt? I am using SVP Interframe scripts with the latest avisynth from this thread, and whenever avisynth is enabled in ffdshow, mpc-hc refuses to gracefully shut down about 30% of the time. Sometimes it silently stays there in task manager, other times it will give a program not responding windows dialog.

Any hacked workarounds that has proven to work for anyone or does anyone know the reason for this?

What build and revision of ffdshow-avisynth are you using? Have you tried ffdshow builds from xvidvideo.ru or clsid rev4486? I am also experiencing freeze or memory leak of MPC-HC / BE whenever I used xhmikosr ffdshow-avisynth latest builds.

[mark0077]

Hi ney2x, I try to keep upto date with ffdshow builds from xhmikosr's page here http://xhmikosr.1f0.de/ (currently using build 4487)and avisynth 2.6 MT builds from this thread (currently using 2012.08.28: http://www.mediafire.com/file/4dm34k...rk/avisynth.7z). I'll try a different build of ffdshow now from xvidvideo.ru and report back.

EDIT: Wow you're right, I havn't been able to reproduce with the xvidvideo one at least. Opened a video 20 times and got 0 crashes of mpc-hc. Usually I would get at least 6 or 7. I'll ping xhmikosr

[ryrynz]

Yeah I already mentioned this in the ffdshow thread, his ICL12 builds are broken.

[nibus]

Which mvtools2.dll should I use with this? I've getting lots of crashes after some hours of processing..

Me too.. tried a bunch of mvtools2.dll's with no luck, all with different crashes. Reverting back to the last version of MT from this thread to see if that helps.

[bcn_246]

Is there any way to get the MT.dll to work with AviSynth v2.6... or plans to build one... I have tried v0.7 and it (understandably) throws up an error regarding the wrong avisynth.dll version. SetMTMode() gives a speed improvement, but MT("filter") is vastly faster. This is the only reason I still use v2.58.

[SEt]

If you recompile MT.dll it'll probably work the same.

[IJM]

@SEt. Can I offer a suggestion, please? If read ahead buffering is required to make many of these filters (especially temporal ones) work properly in an MT environment, would it not be a good idea to add such a feature into Avisynth 2.6 MT? That way it would be native to the environment, instead of relying on a third party's software being at a reliable patch level? (In fact I'm not keen on being forced to use ffdshow, so I'm running single threaded instead.) Besides, wouldn't it be better to have all the required functionality self contained as part of your Avisynth MT "package"?

For instance, have a SetMTBuffer(framesback int, framesahead int) function. Or build the buffering functionality into existing Avisynth source filters such as avisource() and directshowsource().

I imagine this might not be a trivial piece of work for you, but I thought I'd put it out there as a suggestion to see what you and others think.

[SEt]

I have several interesting ideas but they require quite some time and effort, so no promises when.

[jb_alvarado]

Hello SEt,
I will use your Avisynth.dll in a Batchfile. This Batch check if is avisynth install and when not it will download it via WGet.exe. The Problem is, that WGet can not download from your link. Is it ok for you when I upload your dll to my webspace and share the link?

[SEt]

Avisynth is free software under GPL license, so of course it's allowed.

[Asmodian]

But it is still polite to ask when using someone's build. :)

[bcn_246]

Not really much of a coder or I would re-compile the MT.dll (v0.7) myself. Could anybody modify it to work with AviSynth v2.6 (from what I can see it is just a version check that stops it working - but there may be more).

Regards,

Ben

[Vesdaris]

I suppose there is not much sense in switching to Avisynth 2.6 MT from standard AVS if i don't use on most of my blurays encodes any specific filters, just standard crop\resize?
Would it increase perfomance on interlaced sources?

[bcn_246]

No not really, and quite the opposite if you use any filter other than the ones you mentioned. v2.5.8 with the ability to use MT("*) will be much faster than v2.6 (can only do SetMTMode(*) at the moment, due to the MT.dll not being compatible). The main advantage (IMO) of v2.6 is support for high-depth colour spaces (YV24 etc...) which doesn't really affect Blu-Ray (the spec won't allow higher than 4:2:0, YV12).

[Keiyakusha]

you never thought that mt.dll still not compatible because setmtmode may be better and noone bothered to mess with suboptomal stuff

The main advantage (IMO) of v2.6 is support for high-depth colour spaces (YV24 etc...)
you wanted to say "support for other colorspaces". high bitdepth is not supported.

[bcn_246]

you never thought that mt.dll still not compatible because setmtmode may be better and noone bothered to mess with suboptomal stuff

I wouldn't say MT(* was 'suboptimal' for a lot of filters. Yes, there are filters that processing parts of the image separately is a bad idea for (debanding over large areas, filters that change the number of frames...).

However, provided the user chooses an appropriate orientation and a reasonable amount of overlap, then checks for any visible artefacts were the split parts are joined and adjusts the parameters accordingly) MT(* still provides huge speed benefits with little/no quality loss for a huge number of filters (deblocking, sharpening, de-noising, anti-aliasing...).

Comparing (preview w/VDub, source 1280x720@24 uncompressed YV12 AVI stored on 2xSSDs in RAID-0)
MCTemporalDenoise()
with
SetMTMode(2,8)
MCTemporalDenoise()
with
MT("MCTemporalDenoise(),threads=8,overlap=16,splitvertical=true)
On my i7@4.2GHz the single-threaded version rendered at ~4fps, SetMTMode rendered at ~6fps while MT(... rendered at ~13fps. I couldn't see any visible artefacts (and an overlap of 16 is overkill, for most video an overlap of 4 would have been fine).

Its one small test, but I've used both enough times now to know that splitting the image is considerably faster in every circumstance I have seen and causes issues (when used properly) in very few circumstances.

I haven't had any issues mixing the two with v2.5.8, even when switching multiple times. For example...

SetMTMode(5)
AVISource("blah.avi")
MT("ConvertToYV12",threads=4,overlap=0)
SetMTMode(2,16)
QTGMC(FPSDivisor=2)
SetMTMode(5)
MT("DeBlock_QED(quant1=16,quant2=17)",threads=4,overlap=8)
MT("LSFMod(60)",threads=4,overlap=16)
SetMTMode(2)
GradFun2DBmod(radius=3,str=0)
SetMTMode(5)
Crop(4,6,-2,-8)
Spline36Resize(960,544)
MT("AddGrain(1.2)",threads=4,overlap=0)
MT(* also does work with some filters change the amount of frames. I have used MT("QTGMC",threads=4,overlap=16,splitvertical=true) and have yet to see any 'joins' (or artefacts around the joined area) on the output. It's not like the speed difference between the two is small, in many cases MT(* can be several times as fast as SetMTMode(2).

you wanted to say "support for other colorspaces". high bitdepth is not supported.

Your right, my bad... I was referring to 4:4:4 support...

[SEt]

You are obviously doing something wrong. With 1280x720@24 source running on i7-930@4.1, VDub preview:
DGSource("test.dgi")
MCTemporalDenoise()
4.5 fps

SetMemoryMax(1024)
SetMTMode(3)
DGSource("test.dgi")
SetMTMode(2)
MCTemporalDenoise()

16 fps

[bcn_246]

Just tried again with AviSynth 2.6.0 a3 (110525)...

Got about ~11fps with the same file, which is a big improvement over v2.5.8 RC5 (081230) which gave ~6fps.

Still, not quite as fast as splitting the picture (got ~13fps with MT("MCTemporal... but I can understand were you guys are coming from.

I'll keep playing around, possibly one of the filters it uses is sub-optimal (I have been using Vit-Mod filters were available, and do check for updates pretty often).

Regards,

Ben

P.S.
Set, do you have MultiThreading enabled (in your BIOS) for your over-clocked 930 (so it detects 8 cores)?. I have found it crashes (KERNALBASE error in VDub) with it on...

[SEt]

You need to tweak SetMemoryMax to your case to get the optimal speed and memory consumption. Larger values are not always better though.

Yes, it's fully enabled, so 8 threads. You get crashes in VD likely because its exe isn't LargeAddressAware (google it, it can be easily corrected).

[bcn_246]

...You get crashes in VD likely because its exe isn't LargeAddressAware...

You hit the nail on the head!

I rebuilt the VirtualDub.exe (v1.10.2) PE header to make it LargeAddressAware. Now when running ('preview input'):
SetMemoryMax(1024)
SetMTMode(3,8)
DGSource("720p24test.dgi")
SetMTMode(2)
MCTemporalDenoise()
I am getting 18fps+ on my 1280x720p24 sample.

Thanks for taking the time to help a noob like me (can't even begin to say thanks for everything else you have done here). :D

For others experiencing similar issues here is a ZIP containing LargeAddressAware patched versions of:
VirtualDub v1.9.11 build 32842 (x86)
VirtualDub v1.10.2 build 34807 (x86)
VirtualDubMod v1.5.10.2 build 2542 (x86)

Download Link 1: http://www.mediafire.com/?afapcj020qcra4n
Download Link 2: http://www.sendspace.com/file/borm0d
Virus Scan*: http://virusscan.jotti.org/en-gb/scanresult/8f0031402d6e4603f7b6a5514a1a1b30b8656758

*The files are safe, ClamAV is just complaining about VirtualDubMod being compressed with UPX

[GMJCZP]

Thanks for the input bcn_246.
I have a question: could not do the same with the version of VirtualDubMod 1.5.10.3 build 2550? :thanks:

[bcn_246]

Here ya go:
http://www.mediafire.com/?99z6k97picyo1s8

[djonline]

What about official x64 support ?

[Carpo]

Would be nice if there was official x64 support, the dev's of MeGUI are planning on stopping development of the x64 version as there is no x64 Avisynth support

[SEt]

x64 support for 2.6 is unlikely.

[Poutnik]

Just tried again with AviSynth 2.6.0 a3 (110525)...

Got about ~11fps with the same file, which is a big improvement over v2.5.8 RC5 (081230) which gave ~6fps.

Still, not quite as fast as splitting the picture (got ~13fps with MT("MCTemporal... but I can understand were you guys are coming from.

Is not there a general advice that spatial MT() slicing is a bad idea for MC ?
MT(MCwhatever) speeding up wrt SetMTMode() can be caused by sideeffect of spatial limiting of MC scope, decreasing MC quality.