In IRC, Sorunome mentioned how he wanted to maybe do an 8-level grayscale routine for Reuben. This somehow got me thinking about one of my older projects, a 9-level grayscale routine, except this time, I wanted to optimize it far too much. So I did. Here are the results, with a demo pattern. It would look better if I put it into an interrupt routine, but I actually think it looks okay at least without.  Bonus points to anyone who can figure out how it works without looking at the spoiler. Clear exits the demo program.

A screenshot would be nice. In all cases I've seen, the higher number of shades you go the less difference there are between them. Ie, the darkes shades tend to look too similar to one another as do the light ones. I think about 6 shades seem to be the most discernible.

Yes, that is true. Unfortunately, I'm without a link cable at the moment.

Uuuuh, here it is that sometimes when i run it the gray is only on the left 64x64pxl box and the rightmost 16pxl column is just static o.O

I'll have to test this out when I get home, looks promising .

This definitly sound interesting. I wonder however how the r register is used. It look like you manually set it up with your value (46h). However, won't that cause some part of RAM to stop refreshing and goes random ?

That only applies to DRAM. The calculator uses SRAM which maintains state without the need to refresh.

----
I just d/l'd the file and sent it to the emulator. Looks pretty nice with steady freq settings, though you can see the patterns. I know this isn't representative and I should endeavor to get it onto actual hardware ... as soon as I can pry the cable from my brother, who is using it on his PS3 controller to keep it charged.

Yep, the R register is free, and you can use it for any loop which always has the same number of instructions (branching is only allowed if you're very very careful), of up to 128 loops. if you're using it for >64 loops, you need to have an odd number of instructions in your loop because of divisibility.

The I register, designed for interrupt routines is also free if IM 2 is not used, or it is saved, but is not as useful because arithmetic is not really possible, and it doesn't change by itself like R.