Perfomance Comparison Kokke
At kokke's behest and just for fun...(TLDR ifknot faster than kokke) This is a direct comparision noting that there are a number of AES speed optimizations but kokke's objective was always to favour small memory footprint over performance.
(Not least of which is to use AES-NI extensions to the x86 instruction set architecture for microprocessors from Intel and AMD proposed by Intel in March 2008.)
Graph 1. - HP Z600, Linux Clang (Release) execution time for 1 million encrypts sampled 50 times. Despite a small group of slow outliers for both kokke and ifknot AES implementations ifknot AES-CPP17 is faster than kokke tiny-AES-c.
Graph 2. - HP Z600, Linux Clang (Release) execution time for 1 million decrypts sampled 50 times. Decryption is twice as slow for both kokke and ifknot AES implementations - this is likely due to the Galois Field inverse multiplication in a finite field for the inverse mix column directions. This could be optimized with a lookup table or simply favouring Counter Cipher Block Mode, regardless ifknot AES-CPP17 is faster than kokke tiny-AES-c.