Yuku free message boards

>        subq.w #2,%d0 | subtract 4 from phase counter

It's a good thing you comment your code so well; otherwise, I might think that subtracting 2 only subtracted 2.

Sorry, I'm in a bad mood this morning. I lost way too much sleep last night fighting with Flash Studio to work around one of those damn compiler bugs. Then I tried to speed up some floating point trig by looking for a "better" algorithm in Numerical Recipies, only to find that when I implemented it, it took the _exact_ same amount of time. I wouldn't have minded much even if it was slower, as long as there was a clear winner. But somehow the two radically different algorithms took the exact same amount of time, and that's really frustrating me.

        -Scott

Actually, I think I prefer my original reply, which I scrapped in favor of the one you see above. It went something like this:

> subq.w #2,%d0 | subtract 4 from phase counter

Wow, my understanding of 68k must be even more mistaken then Kevin said!

        -Scott

Back when I first got HW2 grayscale working, someone made an interesting modification to my code - rather than copying the plane to the LCD in chunks, he did an interlaced copy. Instead of copying the first half, then then second half, it copied the even lines, then the odd lines.

This is a bit difficult to do line-by-line -- 30 isn't divisible by 4, and you can't do 2 lines at a time because the movem would require 15 registers, but we need to use 2 of the 16 for source/destination, leaving only 14. If you cut the movem from 13 registers down to 12, you can interlace in chunks of 4 lines at a time (that's how we originally did it), although those are fairly large chunks.

Right now, it looks like the HW2 routine is copying the data to LCD_MEM in what - three chunks of 1280 bytes? You might want to consider an interlaced approach instead. It might not make a huge difference, but it wouldn't hurt to try.

Also, are you using identical copy routines for the 89 and 92+? A 89-specific version that only copies the 89 portion of LCD_MEM might be able to work faster and save a lot of flicker on the 89.

        -Scott