lemire

6 ms·

1. 1.▲
   by lemire 17y ago

   Implementing vectorized hash tables is no joke, so I don't think it is silly to have merely evoked the possibility rather than implementing it. It would have taken me days of work, possibly, and even then, people could have questioned my co
2. 2.▲
   by lemire 17y ago

   Please consider vectorization. Natively, modern processors can multiply several 16-bit integers in the time it takes to multiply one pair of 64-bit integers.
3. 3.▲
   by lemire 17y ago

   I am not sure optimization, cache misses, and interpreter overhead would obviously impact the computational complexity. "You're assuming that people are not aware that multiplication is not always a constant-time operation." I made no such
4. 4.▲
   by lemire 17y ago

   I'm giving an example (vectorization) where experiments would be needed: "Am I being pedantic? Does the time required to multiply integers on modern machine depend on the size of the integers? It certainly does if you are using vectorizatio
5. 5.▲
   by lemire 17y ago

   My post was specifically about vectorization. In which case, I do claim that it matters.
6. 6.▲
   by lemire 17y ago

   My post specifically refers to vectorization where you may use the fact that you can multiply 4 pairs of 16-bit integers in the time it takes to multiply a pair of 64-bit integers. So, you could operate four 16-bit hash tables in the time i