Speeding up Serpent

We present a method for finding efficient instruction sequences for the Serpent S-boxes. Current implementations need many registers to store temporary variables, yet the common x86 processors only have 8 registers, of which even fewer are available for computations. The instructions are also destructive, replacing one input with the output. Alternative versions of the S-box instructions are presented, requiring only 5 registers and also utilizing parallelism. Speedup of C language implementations of 24% is shown on the Pentium Pro Processor, and 42% on the Pentium, both compared to the previously fastest known implementation of Serpent.

• Type: Conference
• Year: 2000
• Where: AES
• Authors: Dag Arne Osvik
• Keywords: AES 2000, Common X86 Processors, Cryptology, Efficient Instruction Sequences, Instructions