DCTFilter

[Chikuzen]

I got some requests to write 64bit version DctFilter from VFR-maniac (he wants to run DeblockQED on 64bit) and a person who does not know well.

I read the original trbarry's code and decided to rewrite from scratch.
Because that code is a heap of asm which doesn't know the meaning brought from xvid project.

DCTFilter (https://github.com/chikuzen/DCTFilter)

binary (https://github.com/chikuzen/DCTFilter/releases)

Latest version: 0.5.0

Requirements
- Windows Vista sp2 or later.
- Avisynth2.6.0 / Avisynth+MT r2172 or later.
- Microsoft VisualC++ Resdistributable package for VisualStudio 2015.
- SSE2 capable CPU.

Usage
see README

Currently, DCT/iDCT are processed by single precision floating point LLM algorithm.
Therefore, this filter will output more accurate results than original version.

[Reel.Deel]

I got some requests to write 64bit version DctFilter from VFR-maniac (he wants to run DeblockQED on 64bit)

Thanks Chikuzen, I've been waiting for a 64-bit DCTFilter as well. :) One step closer to a complete 64-bit workflow.

Edit: I do not see any binaries in the releases section.

[Chikuzen]

I do not see any binaries in the releases section.
oops, it seems that uploading was failed.
fixed.

[Wilbert]

Great! Does that mean that the stripe bug is not present? see http://avisynth.nl/index.php/DctFilter#Known_Issues
Is it possible to support avs 2.6 too?

[Chikuzen]

Great! Does that mean that the stripe bug is not present?
DctFilterD is not implemented yet.
so, I don't know.
Is it possible to support avs 2.6 too?
Possible.
If IanB fixes that alignment bug, I'll support avs2.61.

[Chikuzen]

ok, I confirmed that stripe bug.

https://dl.dropboxusercontent.com/u/19797864/forum/dctfilter_buf.jpg

seems the bug was gone.;)

updated to 0.1.0.

[Chikuzen]

updated to 0.2.0

- add SSE4.1 code path and more optimizations.

It seems that 0.2.0 with SSE4.1 and DCTFilter_015 is same speed mostly on my box(i7-4790).

[Wilbert]

DctFilterD is not implemented yet.
so, I don't know.

Possible.
If IanB fixes that alignment bug, I'll support avs2.61.
He promised to fix that for 2.62.

[Chikuzen]

He promised to fix that for 2.62.

ok, updated.

When is the next official avs release?

[Motenai Yoda]

Is this v0.0.1.5's bug fixed too?
http://forum.doom9.org/showpost.php?p=1770001&postcount=16

[Chikuzen]

Is this v0.0.1.5's bug fixed too?
http://forum.doom9.org/showpost.php?p=1770001&postcount=16

I've not read anything but interface of 0.1.5 and readme.txt.
Main processing of this filter is unrelated to trbarry and xvid.
If a bug is found, that's the new bug perfectly.

[Chikuzen]

updated to 0.4.0.

- add 4x4 DCT mode.

Now, DCTFilter allows mod 4 width/height planes without crop or padding.

[GMJCZP]

Thank you Chikuzen.
Possibilities of a version for Windows XP?

[Chikuzen]

Thank you Chikuzen.
Possibilities of a version for Windows XP?

I don't support OS which has a possibility that InternetExplorer8 is installed.

Please ask someone who installed VS2015 with XP components.

[Chikuzen]

updated to 0.5.0

- fix a bug when using with avs2.6
- add avs+'s new colorspace support (all planar YUV(A) and RGB(A)P)

btw, I received a bug report that this filter can't work as MT_NICE_FILTER, but I couldn't reproduce that.
Does anyone have more infomation?

and, don't send me a bug report in PM.
Use this thread or github issue tracker, please.

[Motenai Yoda]

K! so with deblock_qed() avs+ x86 it will give me this (last version 0.4.0)
but today I updated some stuff and avs+ too and I can't reproduce it anymore.

http://6.t.imgbox.com/3E1gn2hU.jpg (http://imgbox.com/3E1gn2hU)