please test master ( -> performance topic)

Fri Aug 31 07:40:30 UTC 2012

On Thursday 30 August 2012 Aug, David Revoy wrote:
> Hi,
> 
> Thanks for your work Boud about performances. I *really* appreciate all 
> the work that could be done in this direction. This is the most 
> important topic in my opinion for Krita ( this and of course polishing 
> existing features , as text / navigation / textures / etc... ) I tested 
> a before / after the change ; using a set of rules* I'm used to do for 
> comparing digital painting apps. I saw a very subtle advantage in openGL 
> ( but maybe it also comes from the Nvidia updates ) like 1/10 better. 
> This mainly affect brush with smudge/dulling. For the no-openGL canvas ; 
> I experienced no changes.

At the moment, performance work is going on for mask calculation (which is the slowest part of painting by a long chalk) and for the canvas. Sven is working on masks, Arjen is working on a new opengl canvas that should be much faster and much more flexible than the old canvas implementations.

> Here I also installed a dual boot with a Win7 to test performances 
> differences between four digital painting softwares** and Krita. The 
> difference is like going from 1 to 5. Watching this statement was a bit 
> hard, and make me sad for the slow FLOSS tools.

Yeah, on Windows we have some unique challenges. The good thing is that we'll have to meet those head-on and that we've got the tools to ferret out the bottlenecks. The bad news is that Windows is still a strange and terrible country...

The target hardware there is 32 bit win8 tablets and 64 bit win8 ultrabooks, mostly dual-core with 2-4 gb of memory.

> Where all the computing effort of my hardware goes ? :-)
> I can't barely demo Krita on my laptop , a dual core 2Ghz , 3GB of ram 
> for painting smoothly a 2484x1200pixel concept art previz at 50% of 
> viewport. For performing this, I must use my workstation , a icore7 
> 2,3Ghz , with a Nvidia GT220 and 8Gb ram ; and I'm happy I can, but it's 
> exclude a lot of user also.
> 
> Of course, I'm deeply aware than back in 2009 while doing concept for 
> Sintel, Krita couldn't paint smoothly on a 1000x1000px canvas with the 
> simpliest rounded brush***. A long and wondefull road was made, there is 
> not a single doubt about it, and thanks to all developers involved into 
> this field. But , I think you'll understand it is reasonable to ask 
> Krita to be five time more fast.

Yes... I'm sort of astonished that I can paint a 7000x5000 image image with a 1000px brush in realtime on my workstation! Not bad for an application that was started back in the day when a 1024x768 image was considered big! But there's still more space to improve, though how soon we'll arrive at the point we'll have to reconsider our basic architecture again, I cannot say. I hope not, for some time.

> I'm also aware than tweaking the deep engine of Krita is not a little 
> work , and rare people can do it. I assume it's also a no-fun job, 

Well, that's not the case :-). Many people find hacking on the core and finding speed improvements way more fun than fiddling with the gui or tools like the transform tool :-).

> complex. Developer(s) on this task should be paid for to can invest full 
> time on it.
> Community could contribute around with a fund rising about that , as in 
> 2009 again ***.

For that, we first need a plan and a candidate. We still have money in the tea kitty to allow Dmitry to work on Krita for some time instead of taking a student job, but it would be good to start thinking of another fund raiser.

I'm also working (really, I've been asking notaries about prices) on the Krita Foundation idea.

Finally, I still hope that I can make Krita a commercial proposition on Windows and for high-end users, while still, of course, keeping Krita free software under the GPL. That should enable more funding for development in the coming years.

> -David
> 
> 
> ( * personnal set of rules : my test is only about how it feels to paint 
> a sketch with various brush on a A4-300dpi ; at 25% view port zoom , 
> with large brush ; I repeat the test on a multiply layer ; under the 
> multiply layer ; merging layers speed ; and color action as 'level' / 
> 'hue/sat" change ... for the brush I use a simple hard rounded with no 
> visible 'spacing' and pressure on opacity ; and also brush mask ; and 
> also brushmask + dulling or smudging ; I'm used to judge software like 
> that ... it's a sort of my personnal calibration to how get how a 
> software feels. )

It's a good test. Maybe we should have a video or something comparing various applications using that test :-)

> 
> (** digital painting softwares tested on Win7 : last OpenCanvas , Corel 
> Painter 12 , SAI Tools , Photoshop CS2 )
> 
> (*** 2009 story : 
> http://dot.kde.org/2009/12/02/krita-team-seeking-sponsorship-take-krita-next-level 
> , can't a similar fund rising be done again ? )

-- 
Boudewijn Rempt
http://www.valdyas.org, http://www.krita.org, http://www.boudewijnrempt.nl