> This will not be slower. It compiles down to a single multiplication. Thanks Mike, in the meantime I've done some quick tests and the results are surprising: 1) Lua implementation[1] -- 230 MB/s 2) C implementation[2a] and test[2b] -- 140 MB/s 3) C implementation called via ffi -- 140 MB/s However, I must be doing something wrong since they quote 2G/s on a core2 and I have an E5200 which is certainly not 10x slower. Tests continue... [1] http://code.google.com/p/lua-files/source/browse/murmurhash3.lua [2a] http://code.google.com/p/smhasher/source/browse/trunk/PMurHash.c [2b] http://code.google.com/p/lua-files/source/browse/csrc/pmurhash/pmurhash_test.c