some years ago, I had a similar problem. At this time, I wrote some 
subroutines for multiply and divide, by only in 16bit.

But I think, it will help you increase it to 32bit.

The underlaying idea is the same, like you calculate this on papier, 
like in the first years of school:
example:
                  nbr#1        nbr#2
                  0100'1010 : 0001'0010
shift nbr#2 one cycle LESS than  nbr#1<nbr2:
                  0100'1010 (nbr#1)
                  0100#1000 (nbr#2*)
subract:        ----------------------
                 .....
shift right...

Of course it is, Integer-Divison  is (almost) always rounded downwards - 
and the next whole number smaller than 1 is 0 -- there are no ints 
between them ;)
You could use some kind of fix-point arithmetics, where the LSB 
represents i.e. 1/256 (or 0.001, or ...) rather than 1. But in that case 
you have to take care that your program can cope with that (put the "." 
in outputs and things alike) or you google for a Floating-point Library 
( but that sounds like an assembler-nightmare to me).

sry, my (written) english is a little rusty ;)

hth. Jörg