Neil Conway <nconway@klamath.dyndns.org> writes:
> I haven't looked at the implementation of hash joins; if they happen to
> use this hash function as well, that would be another informative
> situation to benchmark.
Hash joins use some chosen-at-random hashing code of their own; see
hashFunc() in src/backend/executor/nodeHash.c. One of the things on my
to-do list has been to replace that with the datatype-specific hash
functions used for indexing/caching, since the latter seem better
engineered (even before your improvements).
BTW, I don't particularly approve of the parts of this patch that
simply remove unused arguments from various routines. You aren't
going to save very many cycles that way, and you are reducing
flexibility (eg, the changes to remove nkeys would interfere with
trying to make hash indexes support multiple columns).
regards, tom lane