BLAKE and 256-bit advanced vector extensions

Intel recently documented its AVX2 instruction set extension that introduces support for 256-bit wide single-instruction multiple-data (SIMD) integer arithmetic over double (32-bit) and quad (64-bit) words. This will enable Intel’s future processors—starting with the Haswell architecture, to be released in 2013—to fully support 4-way SIMD com­ putation of 64-bit ARX algorithms (32-bit is already supported since SSE2). AVX2 also introduces instructions with potential to speed-up cryptographic functions, like any-to-any permute and vectorized table lookup. In this paper we show how the AVX2 instructions will benefit the SHA-3 finalist hash function BLAKE, an algorithm that naturally lends itself to 4-way 32or 64-bit SIMD implementations thanks to its inherent parallelism. We also wrote BLAKE-256 assembly code for AVX and AVX2, and measured for the former a speed of 7.62 cycles per byte, setting a new speed record.