real.finder's Avisynth Stuff

[gugglu]

Originally Posted by real.finder

if anyone interested, I made an update to mpp (MP_Pipeline) using last code from official github https://github.com/SAPikachu/MP_Pipeline

there are two changes aside from support old avs mt, one is "Add support for specifying cores", and the 2nd one is "Fix crash on systems with many CPUs"

you can get it from here https://github.com/realfinder/MP_Pipeline/releases

Can you please write one example how to use core future?

this is my script i've tried with ### cores:4 at the end of script, can you clear this to me am i ordering it correct or not , thanks.

o = last

mb1 = o.minblur(1)

mb2 = o.mt_adddiff(mt_makediff(mb1,mb1.removegrain(4)))

ss = seesaw(o, mb2, ssx=1.30, ssy=1.30)

rep = ss.repair(o,1,1,1,false)

mt_lutxy (o, rep, expr="x y < y 1 - x y > y 1 + y ? ?", y=3, u=3, v=3)

Can you please check my 8-bit process and i think it needs some correction. if possible pls look below mt_adddiff(mt_makediff(mb1,mb1.removegrain(4))) i tried seesaw / repair / mt_lutxy i think they need swapping up or down with each other i tried replacing them what not sure where thats how they should be placed or need correction. i would be greatly thankful to you if you clear this illusion in script.i have 14Core/28thread cpu i need an script example for

thanks a ton ,your efforts are very much appreciated.

[real.finder]

it's just the old

### lock threads to cores

but now you can set the thread/core like this

mp_pipeline("""
### platform
1st block here
### lock threads to cores 1
### ###
### platform
2nd block here
### lock threads to cores 2
### ###
""")

and so, and ofc you can use another threads/cores as you like

Regarding 8-bit process Question, it's off topic but anyway seesaw need denoised clip in 2nd clip input, o.mt_adddiff(mt_makediff(mb1,mb1.removegrain(4))) give sharp noisily output

aside from that why you still use lsb not the native clean and fast HBD in avs+?

[gugglu]

Originally Posted by real.finder

it's just the old

### lock threads to cores

but now you can set the thread/core like this

mp_pipeline("""
### platform
1st block here
### lock threads to cores 1
### ###
### platform
2nd block here
### lock threads to cores 2
### ###
""")

and so, and ofc you can use another threads/cores as you like

Regarding 8-bit process Question, it's off topic but anyway seesaw need denoised clip in 2nd clip input, o.mt_adddiff(mt_makediff(mb1,mb1.removegrain(4))) give sharp noisily output

aside from that why you still use lsb not the native clean and fast HBD in avs+?

Hello, and sorry for late reply (it is my working hour very long).thanks for the example.i have done two test on my laptop both works.

mp_pipeline("""

### platform:win32
### dll:avisynth.dll

Setmemorymax(1536)

DGDecode_mpeg2source("C:\Users\Kalik\Videos\VTS.d2v", cpu=4, info=3)

ColorMatrix(hints=true, threads=0)

crop(2, 66, 0, -66)
### prefetch:128,32
### lock threads to cores: 2
### ###

### platform:win64
### dll:avisynth.dll

setmemorymax(2048)

SMDegrain(Tr=3)
### prefetch:64,16
### lock threads to cores: 2
### ###
""")

or

mp_pipeline("""

### platform:win32
### dll:avisynth.dll

Setmemorymax(1536)

DGDecode_mpeg2source("C:\Users\Kalik\Videos\VTS.d2v", cpu=4, info=3)

ColorMatrix(hints=true, threads=0)

crop(2, 66, 0, -66)
### prefetch:128,32
### lock threads to cores 2
### ###

### platform:win64
### dll:avisynth.dll

setmemorymax(2048)

SMDegrain(Tr=3)
### prefetch:64,16
### lock threads to cores 2
### ###
""")

AVS+ Native HBD perform better gives nicer output, but i don't know any Debanding filter which works with Avs+ Native HBD so i can stick that just after smdegrain this is why i use lsb thing some time(but i tried Avs+ Native HBD and to be honest i like the result more than the lsb ting).

Please can you correct this if u have some free time , i've core-i9 7940x 14cores/28threads only thing where i get confused ### lock threads to cores , do i have to type cores number or threads number? thanks for your precious work i really like mp_pipeline and i use it every time i encode any clip.

mp_pipeline("""

### platform:win32
### dll:avisynth.dll

Setmemorymax(1536)

DGDecode_mpeg2source("C:\Users\Kalik\Videos\VTS.d2v", cpu=4, info=3)

ColorMatrix(hints=true, threads=0)

crop(2, 66, 0, -66)
### prefetch:128,32
### lock threads to cores 14
### ###

### platform:win64
### dll:avisynth.dll

setmemorymax(2048)

SMDegrain(Tr=3)
### prefetch:64,16
### lock threads to cores 14
### ###
""")

[real.finder]

from the old read me of mpp

### lock threads to cores

Slave processes with this statement will lock their worker threads into a core. This may improve performance for some scripts.

Core selection is on a round-robin manner, with awareness of hyper-threading.

the new change is

Affinity: Add support for specifying cores

so every block should has different number (or using some threads/cores that has no heavy load, mean you can balance that in which way make the script run faster), and those numbers should be up to cores counts (or maybe threads, you should ask a Developer ) in the cpu

the old mpp as the read me said, does it automatically I think, maybe even in the new one still does that if not specified by user


and Regarding

### lock threads to cores 2

vs

### lock threads to cores: 2

I think the 1st one is the correct as I understand from the Source code, the 2nd one may act like old method (ignore the number) or maybe work as it should, see what you got in the task manager in windows

[real.finder]

Originally Posted by gugglu

AVS+ Native HBD perform better gives nicer output, but i don't know any Debanding filter which works with Avs+ Native HBD so i can stick that just after smdegrain this is why i use lsb thing some time(but i tried Avs+ Native HBD and to be honest i like the result more than the lsb ting).

you can use flash3kyuu_deband, it's can use interleaved format which is like Native in speed, https://f3kdb.readthedocs.io/en/late...arg-input-mode with using ConvertToDoubleWidth() in avs+

[real.finder]

SMDegrain update
new divide parameter (Dogway) and some edits for avs+ and avs neo cases and some others changes

---

there are many of my scripts already use some method for avs+ detection since years, but unfortunately it not work with avs neo so updates will come soon for those, and from now on you will need SMDegrain.avsi more than ever since it have the IsAvsNeo(), IsAvsPlus() and AvsPlusVersionNumber() functions. why?

[real.finder]

almost all filters updated, and SMDegrain.avsi is needed now for many scripts since it have the IsAvsNeo(), IsAvsPlus() and AvsPlusVersionNumber() functions

[StainlessS]

Thankyou RF for you continued work, I'm gonna have to steal some of your code, so thanx in advance.

[ChaosKing]

Thx for the update(s).

[Vitality]

Originally Posted by real.finder

also AKA A.SONY's Avisynth Stuff

note:- SMDegrain.avsi (v3.1.2.101s or up) is needed now for many scripts since it have the IsAvsNeo(), IsAvsPlus() and AvsPlusVersionNumber() functions. why?

DeHalo_alpha_MT2

I have SMDegrain v3.1.2.101s but I still get the AvsPlusVersionNumber() missing error with DeHalo_alpha_MT2.

[real.finder]

I did update dfttestMC (now "dfttestMC and MC things")

there are new mcKNLMeansCL(), you can also try with sMCwarper(prefilter=1,"SMDegrain_KNLMeansCL(d=2)")

edit: there was some bug, re-download it

[tormento]

Originally Posted by real.finder

I did update dfttestMC (now "dfttestMC and MC things")

Well, we need a OpenCL DFTtest

[Natty]

i recently upgraded to ryzen + radeon. i am getting weird frames while using knlmeans prefilter in smdegrain like this

https://i.imgur.com/YL2rHpH.jpg

it didnt happen in intel + nvidia... is there any solution ?

[real.finder]

Originally Posted by Natty

i recently upgraded to ryzen + radeon. i am getting weird frames while using knlmeans prefilter in smdegrain like this

https://i.imgur.com/YL2rHpH.jpg

it didnt happen in intel + nvidia... is there any solution ?

other prefilter's work fine? if not then it's mvtools problem, if they work fine then it's knlmeans

[Natty]

Originally Posted by real.finder

other prefilter's work fine? if not then it's mvtools problem, if they work fine then it's knlmeans

yes prefilter 2 and 3 work fine.. even prefilter 4 works good in normal noised sources.. craps out on too noisy sources.