Smudge optimizations

[LukasT.dev@gmail.com]

Hello,

I'm working on optimizing the smudge. 

In smudge we have these buffers:
dab (the brush)
device (we copy pixels from the actual layer)
m_srcdev (we decrease opacity here but we first copy pixels from layer there)
dabAsSelection - copy of dab so that bitBlt works with selection saved in fixed 
m_target - KisPaintDevice which will be bitBlt on layer

Cyrille proposed to remove the dabAsSelection, because in valgrind it showed 
some performance problems.

So I created KisFixedPainter so that I can composite KisPaintDevice on 
KisFixedDevice. This is needed when coping pixels from device to m_srcdev.
Because I wanted to have m_srcdev as fixed device. 

I also made this method
void KisPainter::bltFixed(qint32 dx, qint32 dy,
                          const KisFixedPaintDeviceSP srcDev,
                          const KisFixedPaintDeviceSP selection,
                          qint32 sx, qint32 sy,
                          qint32 sw, qint32 sh)

so that we can bitBlt with fixed device as selection. This way we don't 
have to copy fixed device into KisSelection so that we can bitBlt with 
selection saved in fixed device on KisPaintDevice.

Problem is that m_srcdev has to be KisPaintDevice in the end. It has to grow 
according pressure.

You can resize fixed device but you invalidate the data and you actually need 
them for next iteration. I also tried to allocate double size of the buffer 
because I supposed that the device would grow maximally twice. But that is not 
good assumption because with e.g. dyna tool the paint device starts with size 
1 and grows to 10, so 10 times in that case. You can't guess maximum scale 
size, it is dynamic so we need to use KisPaintDevice anyway for m_srcdev.

Now I can't use KisPainter::bltFixed as the m_srcdev is not fixed anymore.
The solution would be to code bitBlt with KisFixedDevice as selection.

BitBlt on KisPaintDevice with source KisPaintDevice and selection 
KisFixedDevice is quite crazy to code. Let's take a look at bitBlt function 
and you will get the idea. 

So what do you suggest to do ?
Should I start to code bitBlt with fixed selection or have you got better idea?







Here is the patch
http://lukast.mediablog.sk/patches/smudgeop.patch