The similarity of bit mixers is an example of convergent evolution in code. Many people have done brute-force searches for bit mixers with a good balance of se

The similarity of bit mixers is an example of convergent evolution in code. Many people have done brute-force searches for bit mixers with a good balance of several quality criteria and performance, the latter being somewhat hardware dependent. If you restrict yourself to basic ALU operations, you pretty much always end up in the same part of the bit mixer phase space even allowing for moderate differences in selection function. There are two design elements of interest that are more difficult to search for using brute force, though possible in principle. The first is taking advantage of ALU parallelism, which can have a large performance impact but is more challenging to model. The second, particularly in the context of hashing, is taking into account that the input may be partially pre-mixed. A generally optimal bit mixer may be slower than one that is specifically optimized from the assumption that the inputs have been partially pre-mixed by a specific algorithm.