MT 0.7(+custom avisynth)a filter to run filters multithreaded. Usefull for SMP or HT

[vanessam]

Originally Posted by Manao

OK, thanks a lot Boulder for the debug log.

Here is another dll : http://manao4.free.fr/mt_masktools2.dll

There no guarantee at all that it shouldn't crash, but at least the debug log he sent me pinpoint without a doubt where the crash was occuring. So if that build doesn't work, I still might be able to find the bug.

Yeah !!
No crash !
Thank's
I tested it tonight

[Boulder]

Playing the avs script doesn't crash but when I close the file in VDubMod, the program vanishes after a while. I don't know if it's actually MT that causes it though. Maybe tsp could also provide a debug build so that could be double-checked.

[Manao]

Ok. Well, after all, it was - as always - a silly mistake. I was compiling the masktools with the single thread windows library. What amazes me is that pre a18 were also compiled in that fashion, but didn't crash.

Anyway, I'll clean the mess I've done with the debug log, fix the YUY2 crash, add another of tsp's optimization for the median computation, and release a new, proper version.

Thanks a lot Boulder and vanessam for the help.

[Boulder]

A new item that doesn't like SetMTMode(2) is Cedocida, the open source DV codec. See the sticky thread in the DV forum for its own thread if you're interested. Using SetMTMode(2) produces weird artifacts which disappear when you use for example ffdshow for decoding DV content.

[tsp]

Boulder:
Does it help to use something like this:

SetMTMode(5)
Avisource("c:\foo.avi")
SetMTMode(2)
filter()
filterMore()

[Boulder]

I couldn't reproduce the problem with SetMTMode(2) again but I now started encoding a long DV clip with SetMTMode(5).AVISource().SetMTMode(2) and see if there are any artifacts. When they appeared, they looked like the ones you get with a broken stream although they were not green as usually.

[Boulder]

OK, I didn't see any artifacts in the one encoded with SetMTMode(5) in the beginning so that's what I'm going to use from now on - in all of my scripts just to be safe

[J-Wo]

Just wondering but has there been any further development on this wonderful filter just to iron out some of the bugs? Thanks!

[tsp]

Due to lack of spare time I did not worked very much on MT but in the next couple of days I will get a dualcore opteron 165 so that should speed up development. Most of the "bugs" is incompatiple filters though and it is difficult for me to fix that.

[aberforthsgoat]

I'm almost positive that I've already seen some posts about this, but I can't find them:

When I run this filter I get a thin hoizontal line across the middle of my screen:

SetMTMode(2)
MT("LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=3,strength=60,overshoot=7)")

When I run it without MT, as follows, I don't get a line:

LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=3,strength=60,overshoot=7)

Can someone help me figure out where it's coming from? It's pretty subtle and isn't a big bother - but I'd still like to get rid of it.

Thanks!

Mike

P.S. Did I mention how awsome this thing is? I have a 2.8 Northwood, and I can't run LMSF + a denoiser in realtime without major league stuttering - until I put MT in the mix. Wow.

[foxyshadis]

You shouldn't mix MT() with mt-enabled avisynth. Instead, change SetMTMode until you get to a mode that works for the filter; each step higher is more compatible.

MT is more of a workaround if you can't install mt-avisynth.

(Since I run 2.5.7-pre, I should try MT, that might've sped up my last 18-hour filtering...)

Edit: TSP, have you considered using Tritical's avisynth releases to base your filter on? They include some nice patches that aren't in cvs (such as lifting autoload restrictions, \n in subtitle, and a bunch of others I forgot about). That would be most excellent. =D

[shpitz]

Originally Posted by foxyshadis

You shouldn't mix MT() with mt-enabled avisynth. Instead, change SetMTMode until you get to a mode that works for the filter; each step higher is more compatible.

MT is more of a workaround if you can't install mt-avisynth.

(Since I run 2.5.7-pre, I should try MT, that might've sped up my last 18-hour filtering...)

so 2.5.7 has no MT capability?

[aberforthsgoat]

Originally Posted by foxyshadis

You shouldn't mix MT() with mt-enabled avisynth. Instead, change SetMTMode until you get to a mode that works for the filter; each step higher is more compatible.

Hmm. No. I must be doing something wrong here.

When I try to drop the MT() part and just use the SetMTMode command, I loose my multithreading. CPU two maxes out while CPU one stays stuck very low.

However, I have realized that I can run MT() without the SetMTMode line. Multithreading comes back to normal. However, the horizontal line also remains in the screen.

Could it be an installaton error? I reinstalled avisynth 2.5.6, copied the modified file into my system32 directory and copied in the MT plugin - all to no avail. Same as before.

Bother.

Mike

P.S. Could it be that I installed the wrong version of avisynth 2.5.6? There seems to be a 2.5.6a and a 2.5.6 RC2 on the avisynth site.

[foxyshadis]

so 2.5.7 has no MT capability?

I doubt it will unless it works for all filters. 2.5.7 is supposed to mostly be a bugfix and cleanup update, 2.6 is where the main features go. But if it's comptible enough, who knows.

When I try to drop the MT() part and just use the SetMTMode command, I loose my multithreading.

Odd. Hope tsp can help, but LSF is a pretty complex function. Not surprising that MT fails on it, since it doesn't have good border handling, so try overlap=4 or so. That should prevent the artifacts.

[aberforthsgoat]

Originally Posted by foxyshadis

Odd. Hope tsp can help, but LSF is a pretty complex function. Not surprising that MT fails on it, since it doesn't have good border handling, so try overlap=4 or so. That should prevent the artifacts.

Hmm. Well, I removed every lst plugin from my plugin folder and downloaded all new stuff, and the line is gone. So I must have a had a funky version of mt_tools or removegrain in there.

However, the SetMTMode command still won't do it for me. MT() is working nicely though.

Thanks for the tips!

Mike

[tsp]

Originally Posted by foxyshadis

You shouldn't mix MT() with mt-enabled avisynth. Instead, change SetMTMode until you get to a mode that works for the filter; each step higher is more compatible.

the latest version of mt 0.5 requieres my custom avisynth.dll that also includes setmtmode.

MT is more of a workaround if you can't install mt-avisynth.

I would more say that mt works better with some filters like fft3dfilter while setmtmode is better suiter for others like deinterlacers. Also using mt without my custom avisynth.dll would slow down the filter so it is nearly the same speed as running without mt (that was the reason I requires the custum avisynth.dll)

(Since I run 2.5.7-pre, I should try MT, that might've sped up my last 18-hour filtering...)

Edit: TSP, have you considered using Tritical's avisynth releases to base your filter on? They include some nice patches that aren't in cvs (such as lifting autoload restrictions, \n in subtitle, and a bunch of others I forgot about). That would be most excellent. =D

maybe

[foxyshadis]

Oh! That makes sense, sorry about that.

aberforthsgoat,

Ive found that line to be a problem between versions of MT and masktools - I cant tell you which combination causes it - only that ive fixed it previously on my system by the same method (re-installing versions of each)

Cheers,
Bitey

[aberforthsgoat]

Originally Posted by Mr.Bitey

Ive found that line to be a problem between versions of MT and masktools - I cant tell you which combination causes it - only that ive fixed it previously on my system by the same method (re-installing versions of each)

Odd. Very odd. BTW, are you able to run LimitedSharpenFaster with the SetMTMode thing or do you MT() it?

Mike

I do both...

SetMTMode at the beginning of the avisynth script, then MT with hqdn3d and limitedsharpen faster inside the call to MT..

Cheers,
Bitey