Saturday, February 20, 2016

Squeezing and compacting

I've been plowing away at the game, adding some nicer graphics on the subscreen, fixing a bunch of minor issues, getting some bugs worked out with items, and starting on the overworld.

When I got to the overworld, I realized that my first couple screens were really boring.  Here is the hero standing outside of the cave that leads back to the first dungeon. I want to add some trees or something to the bottom half, but the way my engine works, each room only has 1 wall color.  I stared at this for awhile, but just wasn't happy with it.


Thus I started a new quest to squeeze yet one more thing into my display kernel. This seemed like crazy talk to me, as I already was having trouble getting all my timings right during the kernel. But Atari programming is all about crazy, so I stayed up WAY too late last night trying to find places to slightly tighten up the speed of my kernel code, so that I could squeeze in just a couple more instructions to update the color every few lines.

Turns out, by squeezing part of the calculation on one scanline, and the other half on the next, and by getting my timings just right so that I could eliminate a WSYNC, I got it working. (WSYNC is the instruction where you tell the Atari/TV that you are done with the current scanline, and it should start on the next. It takes 3 clock cycles -- but you can omit it, and the next scanline will just start when it's time. You often use it to force the timing to reset on each line, but if you have everything timed right, you can omit it and just start on the next line).

So now I have the ability to change the playfield color every 4 scanlines. The problem now is that this means that I have to devote 23 bytes of ROM to each different color layout that I want. And my 4k bank where the kernel and graphics live is already getting dangerously full.

Which meant that I started another crazy quest to tighten up the ROM space of the kernel code. Originally I had a lot of loops that were unrolled for the sake of speed, and no subroutines (which waste 12 clock cycles during the jump and return, which I generally can't afford in the kernel)  So now I'm going back and selectively re-rolling loops and moving things around, trying to free up a few bytes here and there. I've managed to free up about 200 so far (which is about 5% of the space of this ROM bank, so that's something), and have some ideas about places I can free up more.

So....progress is slowly happening. Although if I keep thinking of new features, I might no longer be able to squeeze things in anymore! (I'm already down to only 160 bytes of ROM left in my "main" bank, although that's less timing-intensive, so I can jump to other banks for subroutines if necessary)

No comments: