Instruction-level Parallelism in AES Candidates
Craig S. K. Clapp · 1999
We explore the instruction-level parallelism present in a number of candidates for the Advanced Encryption Standard (AES) and demonstrate how their speed in software varies as a function of the execution resources available in the target CPU. An analysis of the critical paths through the algorithms is used to establish theoretical upper limits on their performance, while performance on finite machines is characterized on a family of hypothetical RISC/VLIW CPUs having from one through eight concurrent instruction-issue slots. The algorithms studied are Crypton, E2, Mars, RC6, Rijndael, Serpent, and Twofish.