I was wandering the 'net and somehow kept noticing that more than a few 65xx programmers were taking more effort than necessary to sign-extend integers. I've been playing with 65xx assembly for at least 39 years, and I really can't remember if I got this little ditty from somewhere else long ago and simply forgot from where, or if it's an original idea of mine. Either way, here it is, all ten bytes and 13 cycles worth:



[If you're feeling a bit kinky, you can replace the SBC #$7F with ADC #$80 ...]

Comments, critiques, suggestions, corrections, insults?

_________________
Got a kilobyte lying fallow in your 65xx's memory map? Sprinkle some VTL02C on it and see how it grows on you!

Mike B. (about me) (learning how to github)

With CMOS instructions, I can cut the 8->16 case down to 9 bytes with the same average cycle count (either 11 or 15 cycles, depending on sign): You could avoid the ASL A if the flags already reflect the contents of A, and branch on BPL instead of BCC. This saves 1 more byte and 2 cycles. This even has the virtue of preserving the value in A.

For the 16->32 case: This is 11 bytes, best-case 14 cycles, worst-case 15 cycles. Not quite so good - but if we delete the final STA (thus extending the sign by only one byte), we are back to 9 bytes and 11-12 cycles.

The Sign-extend section the Nesdev Wiki's synthetic instructions page offers a similar technique for a constant-time replacement of A with its sign extension, but this one uses no branches. If one bookends it with a STA zp and a STA zp+1, it's almost (but not quite, by 1 byte and 1 cycle) as good as yours. It might be useful if you don't want to store A in the zero page, though, since it doesn't rely on re-using the original value of A as you need to do with your CMP.

_________________
Curt J. Sampson - github.com/0cjs

As always, there's more than one way…

Cute tricks, I like them. Will come in handy for me fairly soon.