GUI lag

[markzyu]

Using trees in GUI code seems too much.

[markzyu]

And we could also consider hardware acceleration: http://wiki.osdev.org/Accelerated_Graphic_Cards

But QEMU defaults to emulate "cirrus" graphics card, without much support for acceleration... It must be our drawing algorithm that could be enhanced.

After some experiments, I found that making better use of cache might help. But that requires nested loops.

I am working on loop exchange techniques in Compositor::drawSingle

[markzyu]

Faster in fded70c

Now the lag seems to be due to long critical sections.

Also after exchanging the loops, some new bugs appear in compositor (redrawing mouse will affect the windows beneath it).

Maybe we should not exchange the loops but check the conditions in advance, and use a temporary list of elements to draw.