Improved fast software implementation of block ciphers

This paper improves the fast DES implementation in software proposed by Biham at the 4-th Fast Software Encryption Workshop. That is, we propose a new algorithm which reduces the number of instructions for computation of S-boxes, which is the factor dominating the performance of Biham's implementation. When we apply our algorithm to DES S-boxes, we need only 87.75 instructions in average. We can reduce the number of instructions for 1 round of DES to 942, and in total, in terms of the number of instructions, our implementation is expected to be about 8% faster than Biham's implementation.