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

[J-Wo]

okay, after some further testing it seems setmtmode(3) was the fastest for me, other modes either locked up cce or were much slower. Before adding any of these new filters, encoding speed was at 97fps. Placing setmtmode(3) at the top brought that up to 108fps using the following script:

LoadPlugin("D:\Program Files\DVD-RB PRO\DGDecode.dll")
SetMTMode(3)
mpeg2source("F:\BABYLON5_SEASON2_DISC1\D2VAVS\V01.D2V",idct=6)
trim(5,7384)
Crop(16,16,-16,-16)
DeGrainMedian(limitY=5,limitUV=5,mode=3)
AddBorders(16,16,16,16)
ConvertToYUY2(interlaced=true)

But as a test I tried going back to just mt.dll and found it was even faster, at 112fps, using this script:

LoadPlugin("D:\Program Files\DVD-RB PRO\DGDecode.dll")
mpeg2source("F:\BABYLON5_SEASON2_DISC1\D2VAVS\V01.D2V",idct=6)
trim(5,7384)
Crop(16,16,-16,-16)
MT("DeGrainMedian(limitY=5,limitUV=5,mode=3)",2,2,false,true)
AddBorders(16,16,16,16)
ConvertToYUY2(interlaced=true)

But if I try to add a combination of setmtmode and mt as tsp suggested the encoding speed becomes worse than without these filters. Interesting...

[mrcleeo]

will this work with just a straight mpeg2 in dvd-rb encode with no filters?

LoadPlugin("C:\Program Files\DVD-RB PRO\DGDecode.dll")
mpeg2source("F:\HP WORKPATH\D2VAVS\V01.D2V")
trim(0,465)
ConvertToYUY2(interlaced=true)
AudioDub(BlankClip())


LoadPlugin("C:\Program Files\DVD-RB PRO\DGDecode.dll")
SetMTMode(2,2) <----?
mpeg2source("F:\HP WORKPATH\D2VAVS\V01.D2V")
trim(0,465)
ConvertToYUY2(interlaced=true)
AudioDub(BlankClip())


i have an amd x2 4400 and i was trying to get it to utilize the whole cpu.

ive tried to get it to work a few times and when it encodes it makes the video with red lettering of errors

[tsp]

It should work if you use my custom avisynth.dll from the first post. What error do you get?

[mrcleeo]

when i go to look at the video it encoded, the video says

"Script Error: There is no function named SetMTMode"

("F:\HP WORKPATH\D2VAVS\v01000000001001.avs line 6")

[tsp]

ok download this file:
http://www.avisynth.org/tsp/avisynth256MT3.zip
It contains a file called avisynth.dll. Make a backup of your existing C:\windows\system32\avisynth.dll and exctract this file to c:\windows\system32

[mrcleeo]

ok that worked

but im not seing a speed increase :/

thanks for your help

----

actually its a bit slower

[tsp]

well it is a rather simple script you are using so that might be why.

[mrcleeo]

well i dont use filters on retail dvds

what else could i add to try to make it speed up?

i get a speed of 4 in cce without MT and with MT i get 3.2


--------

this is the full avs from dvd rebuilder

#------------------
# AVS File Created by DVD Rebuilder
# VOBID:01, CELLID:01
#------------------
LoadPlugin("C:\Program Files\DVD-RB PRO\DGDecode.dll")
SetMTMode(2,2)
mpeg2source("F:\HP WORKPATH\D2VAVS\V01.D2V")
trim(0,465)
ConvertToYUY2(interlaced=true)
AudioDub(BlankClip())

[Revgen]

I believe that CCE takes advantage of multi-threading, right?

If so, then SetMT may be taking processing power away from CCE, while being unable to speed up what is a relatively simple script.

Both Mt.dll and SetMTmode are better used to speed up resource hungry avisynth plugins. If you want more speed you should contact the CCE devs and ask what they are doing to speed up multi-threading routines.

I typically use SetMTmode or MT.dll to speed up slow filters like FFT3DFilter or resource hungry deinterlacers like the EEDI2/TDeint combo. This can turn an encode that may take 24 Hrs. and cut it down to 12 hrs without sacrificing quality.

[Boulder]

CCE doesn't utilize multithreading like for example VDub does. I've mostly seen CPU usage range between 50 and 65 percent. On all my tests, using MT has provided the biggest performance gain.

[tsp]

new release of MT and a new avisynth.dll that are based on the final 2.5.6 version. Also includes some fixes to sharpen and blur so that they work correct with mt and SetMTMode. So please update both. Get them from the first post

[Boulder]

The new version doesn't seem to work correctly with Didée's MCNR_simple2.




The top one is with the new package, the bottom one with v0.41 and a vanilla AVS v2.5.6.

The function is here for example : http://forum.doom9.org/showthread.ph...898#post766898

And the script:

DirectShowSource("i:\men.avi",pixel_type="YUY2",audio=false)
Trim(23798,42353).FadeIn(5) ++ Trim(46787,63550).FadeOut(250)
AssumeTFF()
TMCBob()
ConverttoYV12()
Crop(12,4,-4,-4,true)
BilinearResize(656,544)
MT("MCNR_simple2(frames=2,removdirt=true,lprad=2.0)",2)
AssumeTFF()
SeparateFields()
SelectEvery(4,0,3)
Weave()

[tsp]

try increasing the overlap like this:

DirectShowSource("i:\men.avi",pixel_type="YUY2",audio=false)
Trim(23798,42353).FadeIn(5) ++ Trim(46787,63550).FadeOut(250)
AssumeTFF()
TMCBob()
ConverttoYV12()
Crop(12,4,-4,-4,true)
BilinearResize(656,544)
MT("MCNR_simple2(frames=2,removdirt=true,lprad=2.0)",2,overlap=4)
AssumeTFF()
SeparateFields()
SelectEvery(4,0,3)
Weave()

[Boulder]

I had to use overlap=16 due to MVTools, but it doesn't help. The similar problem exists although not at the same frame.

[tsp]

what removedirt script do you use and is it only some of the frames where this occurs(and is it random)?

[Boulder]

Originally Posted by tsp

what removedirt script do you use and is it only some of the frames where this occurs(and is it random)?

The function apparently uses RemoveDirt at its default settings. I'll try disabling it when my current encode ends and see if it helps.

(EDIT: It's the old removedirt.dll that's used, i.e. the one which has the RemoveDirt() function only)

The problem cannot be seen at every frame but it is the same frames for the same clip every time.

[Boulder]

Nope, the problem still remains even without RemoveDirt.

[tsp]

could you upload a 10 frame clip with the problem frame in the middle to ftp://tempclips%40avisynth%2Eorg:QfJY(86m@avisynth.org/ so that I can try to reproduce the error.

[Boulder]

OK, the sample's there. I had to upload a larger clip (~13MB) as I've got another encode going on which will last several hours. Let me know if you absolutely need a small clip. It's encoded as ffdshow mjpeg.

[tsp]

Boulder: I couldn't reproduce the error but that might be because I had to disable half the filters to get it to work on my non-SSE celeron. Could you try to find out part of the MCNR_simple2 script that causes the error so I can take a closer look at the sourcecode for that filter.