MVTools-pfmod

[feisty2]

Originally Posted by GMJCZP

Hi pinterf, and Windows XP problem?
Anyway thank you for your effort.

just drop xp already, turn to Linux if you dislike windows10 or windows7

[tormento]

Originally Posted by Reel.Deel

Very nice. Thank you pinterf!



Yes, both QTGMC and SMDegrain are quite bloated due to the planar YUY2 hack. A while back I started modding QTGMC to remove the hack, I never finished but MVTools supporting YV16 makes it a lot easier.

Please make the SMDegrain mod available when finished

[Motenai Yoda]

It's an hard work to add y8 and 16bit support too?

[feisty2]

Kind of, some components like SATD were coded in asm and incompatible with high bitdepth, gotta rewrite them completely
And 16bits support is sort of practical still
Floating point support is much harder, not sure if it's possible to implement floating point support and integer support both within a single plugin
vectors are like, int vx, int vy, int sad for both uint8_t and uint16_t
And SAD(sum of absolute differences) for floating point inputs are also floating points obviously, gotta do some reinterpret_cast tricks here

[pinterf]

Originally Posted by Motenai Yoda

It's an hard work to add y8 and 16bit support too?

y8 is possibly easy, possibly working with three planes is hardcoded now.

16 bit/float support is hard (=takes time), but even this modding to YV24 took days and was a nightmare.

I have already put high bit depth supporting code here and there and my last internal version has a working MSuper in 16 bits as a first step.

And I think I will not cut the possibility of the float version, just for my curiosity (inspired by feisty2)

[pinterf]

Since 2.7.0.22c I have found 3-4 other use of harcoded div 2 for horizontal subsampling (affects only YV24), but will make a fix only if I have time.
Till then you are still encouraged to test it, of course.

[feisty2]

you're gonna get into the floating point SAD trouble if you do plan to mess with fp support,
I did a nasty reinterpret_cast trick to stuff fp SAD in vectors

static inline auto _fakeint(float a) {
	return reinterpret_cast<int32_t &>(a);
}

static inline auto _back2flt(int32_t a) {
	return reinterpret_cast<float &>(a);
}

maybe there's a better way to do this, but...

[Motenai Yoda]

Originally Posted by feisty2

Kind of, some components like SATD were coded in asm and incompatible with high bitdepth, gotta rewrite them completely

IIRC both x264 and x265 should yet sports high bitdepth SATD in asm.

[amayra]

can you do something about SVPflow i try to updated but i failed

[Reel.Deel]

Originally Posted by tormento

Please make the SMDegrain mod available when finished

I never said that I was modding SMDegrain. Ask Dogway if he's interested, I'm sure he can do more efficiently since it's his script. A while back I gave him a few suggestions (post 87-89) but nothing came of it.

Originally Posted by amayra

can you do something about SVPflow i try to updated but i failed

Please do not cross-post. SVPflow has nothing to do with this thread.

[tormento]

Originally Posted by Reel.Deel

I never said that I was modding SMDegrain

Oh... I am so sorry.

I confused you with real.finder.

[GMJCZP]

The version 2.7.0.22c not working in XP 32 bits. Access violation.

[pinterf]

Sad. I was compiling them with xp in mind.
One week holday begins, I'll disappear.

[real.finder]

Originally Posted by pinterf

P.S.
Could not try it with SMDegrain.avsi, because the script internally assumes YV12 for planar YUV format.

it didn't

planar  = IsYUY2(input)       ? true : false

planar in SMDegrain mean YUY2, I was surprised too

so it should work like qtgmc in yv16 and yv24, especially with http://pastebin.com/JtWFN124

[pinterf]

Originally Posted by real.finder

it didn't

planar  = IsYUY2(input)       ? true : false

planar in SMDegrain mean YUY2, I was surprised too

so it should work like qtgmc in yv16 and yv24, especially with http://pastebin.com/JtWFN124

This modified smdegrain.avsi works!
The old version gave "image heights don't match" when using lsb=true because it assumed the dimensions of YV12 subsampling.

[pinterf]

After-before holiday release for early adopters

• general: supporting Y8, YV16, YV24 format besides YV12 and YUY2
• Some additional fixes for YV24
• Allow greyscale input (Y8) - I hope
• dct modes >= 5 now use SATD again (so far it was in dead code, contrary to 2.5.13.1 remarks)
• fftw 3.3.5 support (changed function names, 3.3.4 still OK), see http://www.fftw.org/download.html
• XP support again
• compiled to SSE2 with optional SSE4.2 paths

Download: 2.7.0.22d for 32 and 64 bit (20160816)

[Sparktank]

Originally Posted by pinterf

[*]fftw 3.3.5 support

Oh cool, didn't even notice an update for FFTW.
Looking at the Release Notes, it's not critical to little old me.

[pinterf]

Originally Posted by Sparktank

Oh cool, didn't even notice an update for FFTW.

There are other nice topics on doom9

[Sparktank]

Originally Posted by pinterf

There are other nice topics on doom9

I'm too far behind to pick up on VS scripting to make regular use of VS.

But, it's good to know that the update is making rounds.
For future reference, I'll make note of it.

[yup]

pinterf!

I see some slowdown for dct=5, approx. 10%.
FFTW updated.
yup.