07:34
I need to clip against the surface bounds rather than somewhere off to the right. This way the drawing operations for the most part can be very direct which would be much faster.
08:41
If I go for a completely top/down view, I can have it where I only need to draw a single sprite and use rotations of that sprite for angles and such. Otherwise I will need to draw rotations of an oblique view. I believe for simplicity, I will draw just a single sprite and rotate that. That will reduce the work that is required to be done. In the long run, with the number of units and buildings available it will save me much time. Also all of the images will cost size where none may be available. I believe for sounds I will for 8-bit sounds at a low rate such as 11kHz. For music if I do add it will just be MIDI because that uses the least amount of space and should be simple to implement, no need to worry about decompression algorithms and such.
09:55
Images drawn off the left side remain their size, but become skewed and smear to their image sizes. Not sure why this happens.
10:02
It is the AWT code which smears, because it draws the whole slice each time. The smudgy parts are just pixels that are left over. So I just need to draw with the width of the limit.
10:08
Actually the AWT code is scaled.
10:13
At least on my current system, drawing all those tiles in the background is quite CPU intensive. It will need to be optimized so that the background tiles are drawn as fast as possible. Since the viewport might not actually change, except when the view is changed, I can draw to a secondary image. Then that image can be blitted to the display. Right now each tile is 32 scans, so a view width a width of 640 has 20 scans per row, with a height of 480 that means 9600 scans are drawn. The terrain is not going to change at all unless it needs to be redrawn due to position changes. It just uses lots of CPU time in general. Another optimization that can be performed would be tile groups. Essentially instead of drawing individual tiles of 32x32, I would cache images for a 64x64 pixel group of tiles, or 2x2 tiles.
10:24
This would mean less images to draw at the cost of extra memory. However, one
other thing I can do is split the view window into 4 quadrants. Essentially
what will happen is that when the view is shifted, two, three, or all four
quadrants will be shifted off accordingly. However, something that would be a
bit more optimized is if the screen were split into 9 rectangles. But, these
might be overly complicated. What I really just need to do is cache the
MegaTile
tiles into a single image (where the caches can be cleared by the
GC too). Then I can just blit the whole megatile at once. Since megatiles are
8x8 groups this makes them 256x256 pixels. So in general, these blits would be
more efficient in the long run. The only thing after this point would be fog.
I would need some way of efficiently drawing fog without requiring that
megatiles be redrawn whenever fog changes (which will be very often).
14:31
The XPM images have transparent areas in them, which means blending is possible.
14:39
I wonder if I can lookup table alpha blending.
16:38
AWT is rather slow when it comes to rendering it seems.
18:35
So the next thing will be keybinds. But for the JIT, base work on setting up the register allocator.