2x 128 bit hash instead of 2x 64 bit hash. Why?

[aratz-lasa]

I noticed that for computing multiple hashes, you make use of the work of Less Hashing, Same Performance, which is calculated by: gi(x) = h1(x)+ih2(x) . For that you generate a 256 bits long hash partitioned into 4 uint64s. So I wonder why you decided to generate a 256 long hash partitioned into 4 uint64s. instead of 128 long hash partitioned into 2 uint64s 🤔 Wouldn't it be the same regarding hashing, but with a performance improvement?

[lemire]

Would you be willing to produce a pull request ? Note that we want to preserve backward compatibility so that the hash function is not allowed to change. However, if you have a more efficient implementation of the same hash function, we would love to have it.