Hi there,

I am trying to implement 32 bit addition and subtraction routines but I
have some problems with the carry management.

Before reinventing the wheel I have done a search (on the piclist
archive, on the Mchip AppNotes and on some Web sites) just to see if
there is something ready made. I have found nothing appropriate: just
some 16 bit routines (I know is out there but I can't find it. The Web
is really frustrating!).
Ok. I will take a 16 bit add routine and I will extend it to 32 bit. I
found a Microchip AppNote (the AN617: Fixed Point Routines) that has,
among others, an "expandable" 16 bit add. Here it is:

ADD     MOVF    AARGB1,W
        ADDWF   BARGB1

        MOVF    AARGB0,W
        BTFSC   _C
        INCFSZ  AARGB0,W
        ADDWF   BARGB0

And this is the text following the routine:

    The four instructions after the initial add/subtract, can
    be easily concatenated for operations involving more
    than two bytes. Because addition and subtraction are
    required in standard algorithms for multiplication and
    division, these issues permeate the implementation of
    both fixed and floating point algorithms for the
    PIC16C5X/PIC16CXXX families.

So essentially I sum the LSB, if there is a carry (as in
0x90+0x80=0x10+carry) I have to increment the LSB+1 of AARG before
summing it BARG. The writer of the cited note says me I can repeat the
process on next bytes because at every sum I have the carry set if the
operation result overflows.
Right? I don't think so!.
Try this sum: 0x0001ff90+0x00013080=0x00033010
We have (from LSB to MSB):
      take 0x90, sum 0x80, store 0x10 + set the carry flag
      take 0xff, increment to 0x00, sum 0x30, store 0x30 + RESET THE
CARRY FLAG because 0x00+0x30 doesn't overflow
      take 0x01, sum 0x01, store 0x02 + reset carry
      take 0x00, sum 0x00, store 0x00 + reset carry

The final result is: 0x00023010!! Quite different from the correct
answer. The problem arises for every AARG with a 0xFF byte (except for
LSB).

Am I completely wrong or the Microchip AppNotes are buggy? (I heard some
rumors about that :)

I have tried to solve the puzzle but I can't work out an elegant
solution. I am sure someone had solved this problem already.

PLEASE, do you have some good advice to spare?

Thanks
   Marco

BTW: The subtraction routine had the same structure and comment. I think
I actually have two problems.

----
Marco DI LEO                  email: m.dileo@sistinf.it
Sistemi Informativi S.p.A.    tel:   +39 6 50292 300
V. Elio Vittorini, 129        fax:   +39 6 5015991
I-00144 Roma
Italy