Dr. Douglas Jones wrote a pretty great article on the subject, https://homepage.cs.uiowa.edu/~jones/bcd/divide.html. Depending on the divisor it can be a bit messy, since reciprocal multiplication sometimes needs extra precision. For example, on an 8-bit machine, x/6 == (x*42>>8), which is nice because 42 fits in 8 bits. But x/105 == (x*313>>15) needs 9 bits to hold 313. On some systems (thinking 8-bit AVR), it'd probably be faster to branch on the 3 cases rather than perform a full 16-bit multiplication.