Whilst having lunch. (the 3 bite decrementing sandwich), and
pondering what in the world was I thinking when I reversed
the byte (bite) order for the 24 bit decrement code I posted
yesterday... here is the third bite..

;
; The starting point. 16 bit decrement
; 4 cycles doesn't use W and is isochronous
; thanks to Chip Weller for pointing out the TSTF macro
; which just does MOVF XXX,F, thus preserving W
;
        TSTF    LB16
        SKPNZ
        DECF    HB16,F
        DECF    LB16,F

Now how can this be extended to 24 bits cleanly?
;
; The straight forward extension to 24 bits would be
; something like this.. 5 cycles when LOW is non zero
; and 8 cycles when LOW is zero, and doesn't use W
;
        TSTF    LOW
        BNZ     lbnz
        TSTF  MID
        SKPNZ
        DECF    HIGH,F
        DECF    MID,F
lbnz    DECF    LOW,F

If you were just counting stuff, then 255 times out of 256 it
will take just 5 cycles. So I claim an average execution time of
    (5*255+8)/256 = 5.0117 cycles.... (just kidding)

Testing, the above code decrements monotonically for all the
case I have checked, however comments and bug reports are
always welcome.

And yes, I too think Dmitry's solution is the best!

Ray Gardiner ray@hdc.com.a