I have added comments to each line...


> From: Andrew Warren <fastfwd@IX.NETCOM.COM>
>
> John Payson <PICLIST@MITVMA.MIT.EDU> wrote:
>
> > How about this one [figure out what it does yourself :-)]
> >
> >         movf    Num1,w      w = n1
> >         movwf   Result      r = n1
> >
> >         movf    Num2,w      w = n2
> >         subwf   Result,w    w = n1 - n2
> >         btfss   C
> >          subwf  Result,f    if (n2 > n1) r = n1 - (n1 - n2) = n2
> >
> >         movf    Num3,w      w = n3
> >         subwf   Result,w    r = r - n3 = max(n1,n2) - n3
> >         btfss   C
> >          subwf  Result,f    if (n3 > max(n1,n2)) r = n3

thus result = max((unsigned chars) n1, n2, n3)

> >
> > Anyone know of any smaller way?
>
> John:
>
> Does this work?
>
>         MOVF    NUM1,W        w = n1
>         SUBWF   NUM2,W        w = n2 - n1
>
>         MOVF    NUM1,W        w = n1
>         SKPC
>         MOVF    NUM2,W        if (n1 > n2) w = n2
>
>         MOVWF   RESULT        r = min(n2, n1)
>
>         SUBWF   NUM3,W        w = n3 - min(n2, n1)
>         SKPC
>         ADDWF   RESULT        if (min(n2,n1) > n3)
                                  r = min(n2,n1) + n3 - min(n2,n1) = n3

thus result = min(n1, n2, n3).  Not exactly the same, Andy, but we
assume you meant SKPNC instead of SKPC.  I know I forget this
wierd Microchip convention all the time -- I'm so used to carry
meaning the answer was negative after a subtraction.

This is reminiscent of the XOR technique of swapping 2 or more bytes.
It basically works because there are operations (op) such that given
Z = X op Y, then if you know any two of X, Y or Z, then the third
value can be reconstructed.  ADD, SUB and XOR are such op's.  AND
and OR, on the other hand, fundamentally destroy information.

Regards,
SJH
Canberra, Australia

PS: I missed your subsequent post with 'the answer' so forgive me
if I'm correcting an already corrected oversight.