Is AviSynth 2.5 faster?
 

With AviSynth 2.5 filters, has anyone noticed an increase in speed?

Re: Is AviSynth 2.5 faster?
 

It's supposed to be up to 35% faster, but I assume this is without any filtering. When we stack up our filter chain, I'm sure the overall encoding speed will sink just like if you use 2.07 :D

-kwag

You'll get the big speed increase only when you convert the DVD to DivX or XviD with VirtualDubMod and fast recompress enabled. This way you'll do no colorspace conversions, it goes straight from YV12 to YV12. When doing KVCDs, there's a different path, I think it goes YV12->RGB24->YV12.

If TMPGEnc would encode in YV12 colorspace, the speed increase would be huge :idea:

I downloaded and did a test with the trial version of CCE using Avisynth 2.5. I only used LegalClip and FaeryDust as the filters. The speed using 2.07 was .15, using 2.5 it averaged at about .22 and got as high as .26. So a definite speed increase. I wish TMPGEnc would add in a display of speed like CCE.

Not all filters are optimized for YV12 yet. You should at least be able to reach the same speeds as 2.0, by converting your material to YUY2 colorspace, if that's your destination format anyway.

I'm currently working with Donald Graft on optimizing the decomb package for YV12. Currently all code is unoptimized, and this is probably one of the most used filters.


(wow - 1st post!)

Welcome sh0dan :wink:

-kwag

oh yes. Welcome sh0dan. What a pleasure :D

hi sh0dan..

8O suprise to see you here. I'm glad you found a place here too he he..

Anyways, back to post..

I have question for ya w/ respect to Donald Graft's great decomb.dll
filter.. first, I'm glad you're re-working it. Thanks' (and Donald too.

Are there any plans to optimize it's Interlace detection ?? ?? ??
I've ben processing LOTS of IVTC type projects in my encodes and I still
see those blastid lines - rats !!
I say, ..I notice them when a scene changes. The pic:
* gets distored PRIOR to scene change (you notice blochy'ness) and,
* the next frame after scene change, you see the left-over Interlace - rats !!

If I try and tweak the settings, I loose some detail (too much for my taste)
and/or, the video becomes too software or blurry - rats again !!
I had some thoughts to throw around, but sometimes asking such questions over at
d9 is like asking for a sledge hammer. It's nice and convenient to ask you such
questions here :) Anyways.. I will continue of course, to use the decomb.dll filter
in my AVS projects that require IVTC.

Ok, one more.. I would love to see some work, for broken IVTC processes
handled better ie, ST-NG is shot on Film 24, then Teleined to 29.970, however,
the pattern is not your usual 3,2 3,2 3,2..

It's more like 3,2 3,2 3,2 3,2 3,2 3,2 2,2 3,2.. and somewhat repeats.
Would be nice for a feature like decomb(pattern=3,2,7,2) where the 7 is the
repeat position where the 3,2 breaks down to a 2,2
and the x,x,x,2 would be the amount of Interlace. Since we already know
our source's pattern, it would probably be more accurate to intervent manually ??

The above filter syntax would be just for those broken IVTC processes.
I have other issues, but this is probably not the place for such questions
at this time..

Anyways..
In adition to kvcd templates and things, I also dabble in the divX field
of encodes, cause its just plain fun and it's a part of my hobby to try new ideas.
That's why I'm here at KVCD and loving every minute of it !! ..sorry
Kwag.. didn't mean to kiss your ZZ. Anyways.. i'm glad you're here
too, and hope to beat on you some he he.. I mean, hope to see you more often
here he he..

Have a good evening all.
-vhelp
ps: i've noticed that the AVIsynth Area is mising a MOD. :wink:

I wasn't aware that there was such a large AviSynth user base here, so I haven't dropped by before now. I see a lot of people are still using the old 2.0 line - let's get something done about that. :)

I let Donald do the basic algorithm, and I just optimize it. I have some proposals for how chroma is handled, but the basic algorithm remains in control of Donald - decomb is quite complex, and I would probably break more things by trying to re-code stuff.

On the other hand I'm sure Donald can help you detecting these cases you describe. Drop by the doom9 forums, or direct him here.

Hey, nice to see Avisynth Guru in this forum often...! :)

I'll use 2.5 when my favourite filters are ported (dust, peach etc). Does anyone know in Convolution3d is stable in 2.5 and suitable for mpeg encoding?

Jim

It IS possible (by using a plugin) to use 2.0 plugins in 2.5!

This thread describes how. Of course they can only be used in YUY2 colorspace.

Anyway - plugin authors should easily be able to do a version for 2.5 - for most filters it is "only" a matter of 15-30 minutes of work, to do a native plugin for 2.5.

Convolution3D is perfectly stable for 2.5 - I have seen no bug reports.

Well i'm converted! I'll get 2.5 installed over the weekend!

Jim

Thanks sh0dan!!! Come around here more often!!!

I already consider myself a regular user :)

Correct me if I'm wrong, but I think none of Sansgrip's filters have been written for 2.5 yet, right? And since he is the filter "god" for KVCDs are we ready for the switch?

don't know if all the Sansgrip's filters,
but we use other filters and some of them don't be writen for 2.5!

:wink:

Some of SansGrip's filters are 2.5x ready. Look at the "readme" files here: http://www.jungleweb.net/~sansgrip/avisynth/

-kwag

Thanks Kwag. Taking a quick look at the filters it looks like NoMoSmooth and LegalClip are not 2.5 ready, and these two filters are used in your latest scripts (in fact LegalClip is like in EVERYONE'S scripts). So you're still using 2.08, right?

2.0 LegaClip = 2.5 Limiter, which is now builtin.

The filter was renamed to avoid potetial conflicts.

How about NoMoSmooth? I noticed that Kwag is using it now for his latest scripts incorporating mergechroma and mergeluma.

Update:
Okay guys, so I tried to make the switch to 2.5 but man I am so confused! It's just a real headache trying to figure out which filters are legacy and which are not. And I really don't know what's wrong but when I load my AVS file in TMPGenc it says the resolution is 864x54 even though I have a GripCrop(480x480) statement. Here's the script I'm using right now, can anyone help me with it?

Hi J-Wo,

It gives you this resolution because there is an error in your script: watch it with the tmpgenc's preview pane, and you'll have the error!

Salu2.

Ya I know there is an error in my script, that was pretty obvious. So do you know what's wrong? What should I fix?

Hi,

I think you have to add a
LoadPlugin("D:\Program Files\AviSynth 2.5\plugins\blockbuster.dll
line... But I insist: avisynth displays the error as output video stream!, so you can see yourself what is the error (or it changes since the 2.07?!)

Salu2.

Open the script in VirtualDub and it will tell you what the problem is.

You should add at the end of your script
ConvertToRGB24()

Oops you are right, I did forget that

I know what you're talking about but no I don't see that! Because the resolution is only like 54pixels high, I just see a very thin black rectangle with no error message.

Whenever I do that VirtualDub just crashes and exits! I tried both VirtualDub and VirtualDubMod (I tried both open video file and open via avisynth). I also tried uninstalling 2.08 and reinstalling 2.5, no dice.

I wasn't exactly sure where "at the end" to put it, I tried both before Gripborders() and after the last Limiter() but I saw no change to my problems.

Any other ideas on what the hell I'm doing wrong??? :? 8O :?: :?:

Hopefully sh0dan will come up with something. We've just about gone through every possible cause here :evil:

Edit: Have you tried the script without dubbing the audio in it? You could also try commenting out the filters and then adding them one by one and see which one blows the script up in VDub.

DustV5 isn't a 2.5 plugin IIRC. Try commenting out the filter, and the loadplugin. Is "nomosmooth" a 2.5 plugin?

Oh...my...GOD. I've never done so much debugging in my life, I feel almost like a computer programmer :D . Okay so I commented out every single line and then uncommented them one..by..one. May I point out that 2.5 does NOT show you error messages in the preview screen??? How annoying!

So firstly, I figured out I had mispelled the fluxsmooth plugin in the LoadPlugin line. Then I figured out the only filters giving me problems in my script were Convolution3D and Blockbuster. Both of these plugins were 2.5 versions, so I figured out I needed a ConvertToYV12() line before them. Well it turns out that worked for Convolution3D but not for Blockbuster! I have the latest v0.7, but I tried downgrading to 0.6 and 0.5 (all 2.5 versions) with no luck. I then decided to load the non-2.5 Blockbuster, but that meant I needed a ConvertToYUY2() line before Blockbuster, but it worked!!! By that I mean I can preview the movie through Tmpgenc without errors.

Anyway here's my script, could you guys take a look at it and see if it's all good? I feel like I have to add a LOT of ConvertTo lines and I wonder if that would negate any speed advantages of using 2.5. I would rearrange the order of my filter use but I'm pretty sure they're being used in this order for a reason.

You should not need the Blockbuster filter if you are using the DCTFilter, they basically do the same thing, get rid of DCT blocks.

No, they are not the same thing :idea:

If you want to get rid of DCT blocks, use Blockbuster but I'm not that sure if it does any good at that resolution.

With the current KVCD "Notch" matrix, there's hardly any need to use Blockbuster(noise) anymore :wink:

-kwag