Re: Measurement data
Posted: Sat May 06, 2017 10:22 am
For the Shellsort version, the moves are sorted right after move generation, so that's before doing anything with the hash tables.
Selection sort is performed right before iterating over the node moves, so each move has a chance to get dealt with by the hash table lookup in the next depth level as that is the first thing that is tried.
The hybrid is a mixture - the first MVA/LVV move can be a hash hit, the other moves are sorted before entering the next depth level.
However, I don't think this makes a big difference because the main hash tables have only 2*4098 entries. They are more useful towards the endgame.
Selection sort is performed right before iterating over the node moves, so each move has a chance to get dealt with by the hash table lookup in the next depth level as that is the first thing that is tried.
The hybrid is a mixture - the first MVA/LVV move can be a hash hit, the other moves are sorted before entering the next depth level.
However, I don't think this makes a big difference because the main hash tables have only 2*4098 entries. They are more useful towards the endgame.