I can now demonstrate that the hash implementation is around as fast as guile internal implementation. But for some workloads like if we have a hash with millions of elements. And lookups a working set of about 10000 key hash pairs the speed is doubled compared to guiles internal. Now this is maybe not as interesting. But the implementation is decoupled and allows one to construct custom hashing algorithms that are fast compared to guile's current speed when it comes to the hash tables (hash,eq,assoc) is never called from the inside of the hash function which means that any custom implementation lives in scheme totally. The fast execution was enabled by implementing the hash function as a vm operation and compiling it with a size 1<<60 (% is efficiently optimized as a logand) and hence we got essentially the same speed using a vm-op for hashq. Next I will see how fast we can make hashtables when pushing the hash-ref equivalent to a vm-operation. With this we can now get a hash algo that mimics the python algo.
