Dividing both dividend and divisor by 2 (right shift) until both are reduced to managebale 8-bit numbers, and then usng the 8-bit divide gives a fairly accurate result (in my experience, if both values are comparable in bit length, the quotient/integer part of the result is on the dot)
Done properly the error can be reduced to +/- 1 bit.
If the error is too much,you can rectify using MUL and ADD operations

e.g. Try this:
if A and B = two 32-bit values
Do
{a = A/2
b = B/2}
until (a and b = 8-bit)
q1 + r1/b = a/b ;tentative result

e = A - (q1*b + r1)
;e is the error in the first division
Or u can try :
f = A -(q1*B + r1*(B/b))

you can manipulate these values to give amazingly accurate results, and highly optimize the algorithm for your operating range.

Kundi