> From:    Scott Dattalo <sdattalo@UNIX.SRI.COM>
>
> A week or so ago Andy W. made the claim:
>
> > I have in front of me a 16-bit square-root routine that takes only
> > 126 cycles to execute on a PIC16C5x.

[snip]

> Well, I've been prodded. However, I've come up a little short. My solution
> requires:
>
>    RAM:  4 registers, including the two that hold the 16-bit input. (same)
>    ROM:  28 words.  (4 words too big)

Nice work, Scott! How many cycles is your routine, worst case?

> Perhaps someone may see where 4 words can be squeezed out. But, realize that
> the slightest change has many ramifications. There are many inconspicuous
> dependencies (i.e. tricks).

I believe you could save one instruction as follows (std peephole):

>         BTFSC   s1,6            ;If this is the first time through the loop
>           goto  L1              ;then N is < s1+s2
>
>     ;N may still be greater than s1+s2, so we need to check MS bit of N
>         BTFSS   N_lo,0          ;If MS bit is zero, then N < s1+s2
>           goto  L1

   =>

         BTFSS   s1,6            ;If this is the first time through the loop
           BTFSS   N_lo,0        ;If MS bit is zero, then N < s1+s2
             goto  L1

 -- Martin

Martin Nilsson                           http://www.sics.se/~mn/
Swedish Institute of Computer Science    E-mail: mn@sics.se
Box 1263, S-164 28 Kista                 Fax: +46-8-751-7230
Sweden                                   Tel: +46-8-752-1574