A selection of Hash functions coded in C
Specifically intended for 8-bit microcontrollers, hence optimised primarily for speed (assuming 8 bit architecture) and very low RAM impact with code size a further factor. Will likely be inefficient on e.g. 32 bit machines.
For instance, when shift is of fixed width known in advance, uses this knowledge to only act on certain bytes rather than whole 32 bit word.
Overall
Of this set and specific implementation, MD5 has lowest code size and is quickest. However MD5's security is relatively weak.
SHA-1 and SHA-256 have almost identical speed, with SHA-1 having larger code size and lower security. Hence SHA-256 preferred. SHA-1 does however use less RAM.
RIPEMD-160 is slower than the SHA implementations and larger than SHA-256 (code and RAM)
The reason SHA-256 has less code is because its algorithm is more consistent across different rounds.
Testing
Extensively tested natively on Atmel microcontroller (100,000+ random hashes each) compared with same hash calculated in python using hashlib. Used LFSR to generate hashes from 0 to 1499 pseudo random characters and transmitted the LFSR start point, length, and result over the network.
For subset of above, calculate hash of message repeated 500 times. i.e. largest is c. 1,500x500 characters, i.e. 750,000 characters and test versus Python result.
Testing checked for false positives by e.g. setting one byte of hash to arbitrary value and observing that only c.1/256 now match.
However tested using Little endian ATMega328P. Big endian version not tested.