A sufficient condition for a code to be optimum on discrete channels with finite input and output alphabets is given, where being optimum means achieving the minimum decoding error probability. This condition is derived by generalizing the ideas of binary perfect and quasi-perfect codes, which are known to be optimum on the binary symmetric channel. An application of the sufficient condition shows that the code presented by Hamada and Fujiwara (1997) is optimum on the q-ary channel model proposed by Fuja and Heegard (1990), where q is a prime power with some restriction. The channel model is subject to two types of additive errors of (in general) different probabilities.